Building Knowledge across Language Systems: The Role of Audio and Visual Supports in Bilingual Learning through Self-Derivation

Abstract

Children who enter the US education system as emerging bilinguals (or English Learners) show significant gaps in test scores and graduation rates compared to their English monolingual peers. Dual-language education programs may provide an instructional context that capitalizes on emerging bilingual children’s strengths and supports their academic performance while they acquire English. However, prior research has shown that integrating semantic knowledge across language systems poses a challenge. Supports may mitigate these challenges. Thus, across two studies, we examined instructional practices that may facilitate the integration of academic content across English and Spanish in elementary-age children enrolled in dual-language education programs. In Study 1, we examined whether facts presented through reading-while-listening or children’s silent self-paced reading more effectively supported learning the facts and subsequent integration for dual-language students in grades 3 and 4 (n = 56; M_age = 9.54 years). In Study 2, we examined whether combining graphics with text was more effective in supporting fact learning and cross-language integration compared to facts presented as text alone in dual-language students in grades 4 and 5 (n = 67; M_age = 10.48 years). Overall, the studies replicate the benefits of reading-while-listening and graphics for learning directly taught facts, but underscore the difficulty in integrating semantic knowledge across lessons and languages.

1. Introduction

The US has a history of educational inequity for children who enter school speaking non-English languages. Children who enter the system as emerging bilinguals (referred to as English Learners by the US government) show significant gaps in test scores and graduation rates compared to their English monolingual peers [1,2]. Bilingual education programs may provide an instructional context that capitalizes on emerging bilingual children’s strengths and support their academic performance while they acquire English. However, there is a lack of research investigating the educational practices that best support integrating academic content and building semantic knowledge across lessons and language systems in dual-language education. For example, if a child learns in one lesson that mammals are the only animals with hair, then learns in another lesson that dolphins have hair, they have the opportunity to integrate those facts in memory to self-derive the knowledge that was not directly taught that dolphin are mammals. Without memory integration across academic lessons, learning is limited to that which is directly provided, without the connections that build conceptual knowledge. Prior research has shown that integrating knowledge across not just lessons, but also languages poses a particular challenge. Thankfully, there is some indication that the challenges can be mitigated by appropriate supports at the time the facts are presented [3]. Two recommended teaching practices to support students’ learning of semantic knowledge are reading-while-listening (providing audio support as students read expository information; [4]) and conveying expository information with graphics [5]. These strategies are thought to be particularly helpful to emerging bilingual students, regardless of their home language and the second language [6,7]. Just as these strategies are thought to help with learning semantic knowledge in the classroom, they also may help mitigate the challenges of integrating semantic knowledge to generate new knowledge across lessons and languages. However, this has yet to be examined. Thus, across two studies, we investigated whether the instructional practices of reading while listening and presenting facts through graphics facilitate the learning and integration of academic content across lessons and languages in elementary dual-language classrooms.

2. Memory Integration across Lessons and Languages

Memory integration and other memory processes support knowledge acquisition and play a crucial role in cognitive development [8,9,10]. When information is not integrated into the knowledge base, conceptual understanding cannot develop because lessons remain separate. Integrating memories bridges the gaps between lessons that have been distributed over time. Several measurement paradigms for capturing memory integration exist, such as transitive [11,12] and associative [13,14] inference. These paradigms gauge whether successful memory integration has occurred based on correct inferences. For example, if A is larger than B and B is larger than C, memory integration would be evidenced by the production of the inference that A is larger than C [15]. A characteristic feature of these paradigms is that the to-be-integrated information is arbitrary and intended to be ephemeral.

In contrast to other memory integration paradigms that use transient associations, self-derivation through memory integration teaches novel facts that are meant to be retained over time and contribute to the knowledge base. Self-derivation through memory integration is the cognitive process of deriving new knowledge by integrating separate facts and experiences [16]. When integrated, the related facts can be used to produce new knowledge that was not directly taught. For example, a student may learn that “liquid expands when heated” in a classroom lesson and then that “thermometers are full of liquid” from a book. If they integrate these two facts in memory, they could then derive the correct response to the question “How does a thermometer work?” (the liquid inside expands when heated). Thus, self-derivation through memory integration is measured with a paradigm that uses factual information intended to be retained over time, making it well suited to the study of learning.

In addition to the factual stimuli, the paradigm of self-derivation through integration is particularly well suited to examine classroom learning for at least four reasons. First, facts derived through this process are rapidly integrated into the knowledge base [17]. Second, the knowledge derived through self-derivation is retained over time [18,19]. Third, the process has been observed in both laboratories and classrooms over a range of content, including stimuli aligned to the elementary curriculum, and from pre-school children through adulthood [16,19,20,21]. Fourth, and possibly most importantly, performance on self-derivation through memory integration tasks predicts academic performance in elementary school and college students [19,22]. Thus, self-derivation through integration is an excellent model of classroom learning and relevant to academic performance.

In order to integrate separate but related facts or lessons, children must first learn the facts (referred to as encoding) so they can be retrieved in memory and then recognize that the facts are related [23]. Recognizing the relation between facts is challenged by situations with low surface similarity. Surface similarity refers to the details of the presentation that are not specific or necessary to the content meaning. For example, if the goal is to teach children about polygons, surface similarity would refer to non-pertinent information such as color (instead of the number of sides or angles). Low surface similarity between related facts challenges self-derivation through integration [8] and other productive processes [24,25]. Language of instruction can be considered a surface-level feature because it is the mechanism of delivery rather than the focus of instruction (except in language classes). In a previous study of cross-language memory integration in dual-language classrooms, third- and fourth-graders were tested on same-language (high surface similarity) and cross-language (low surface similarity) self-derivation through integration [3]. Third-graders showed significantly lower performance in the cross-language condition, and across all conditions and grades, students showed poor performance compared to laboratory studies. This indicates that memory integration in classrooms is difficult, and especially so across languages. However, dual-language instruction requires exactly this; the language of instruction often alternates, requiring students to integrate content across lessons presented in different languages to build conceptual knowledge.

In summary, memory integration is critical to building knowledge and concepts. Self-derivation through integration is particularly relevant to academic performance, making it well-suited as a paradigm to investigate memory integration in educationally relevant work. However, classroom conditions also challenge memory integration, including learning models that teach through different languages. While these bilingual education models have many benefits, it is crucial to examine the pedagogical strategies that best support memory integration across lessons and languages for students in bilingual education.

3. Supporting Encoding Supports Integration Processes

It is generally agreed that memory functions with at least three processes: encoding (bringing information into the memory system), storage (retention of the information for later use), and retrieval (bringing stored information back into consciousness). Encoding the facts is the first step of the memory process and the first step of the knowledge integration process [23]. Previous research investigating encoding found that if children were given more time to learn individual facts, their self-derivation through integration performance improved [16]. Thus, targeting interventions at the encoding phase has previously been successful in facilitating the integration of separate but related facts across lessons. However, the challenges of integrating across lessons and languages in classrooms may need pedagogical support beyond additional exposure.

In a learning environment with only one language, students can use the cue of language to help access previously learned material. Indeed, in a study with adults, the language of the cue word significantly predicted memories initially encoded in the same language [26]. Having knowledge stored does not mean we will access it (retrieval) when needed and this can be challenged by a dual-language learning environment. There is also evidence that skills learned in one language will transfer to the other language. Cross-linguistic transfer has been extensively studied in the area of literacy, particularly in regard to whether reading comprehension and decoding skills learned in one language will transfer to reading in another language [27,28,29,30,31]. Generally, evidence supports transfer, although this depends on the languages, individual differences, and pedagogical strategies. Pedagogically, there is mounting evidence that strategies, such as bridging, that directly point out differences between language systems are helpful and possibly even necessary for the successful transfer of literacy skills (see [27], for overview). Overall, the literature investigating cross-linguistic transfer in literacy has underscored that while transfer can and does happen, it also has challenges and likely requires pedagogical support. In the current investigation, we are not examining reading or literacy skills. However, we are interested in how learners are making connections across languages. Specifically, we are investigating how semantic content can be encoded, stored, and then retrieved to integrate with related semantic content to generate new knowledge that was not directly taught. Similar to the supports needed to develop literacy skills across languages, we are investigating whether supports can facilitate integration of semantic information across languages.

One previous study indicates that supports can facilitate learning and subsequent self-derivation through integration of information conveyed through different languages. In Study 1, children in second grade were provided with facts portrayed through illustrated stories [3]. Stories were read in the same language (English) or through two different languages (English and Spanish) in separate lessons. Children showed evidence of self-derivation through integration in both conditions and there were no differences between conditions—meaning children did not face additional challenges in the cross-language condition. This was interpreted to mean that the familiar story structure and illustrations may have supported the encoding of the information and subsequent self-derivation through integration. Greater exposure to the facts [23] in a monolingual study and greater support through the presentation mode [3] are both thought to support encoding. Thus, we are focusing on support at encoding given prior indications that support at the time of encoding facilitates fact learning and subsequent integration performance [3,23].

The current study examines whether self-derivation through integration across lessons and languages benefits from pedagogical support at encoding. Previous research indicates that support at encoding can facilitate subsequent self-derivation of new knowledge through memory integration. We investigate two potential encoding supports: reading-while-listening and the use of graphics.

4. Reading-While-Listening

Reading-while-listening refers to the simultaneous reading of text while listening to corresponding audio. Webb and Chang [32] noted that reading while listening supports superior comprehension compared to silent reading, freeing attentional resources for unknown words. It also promotes chunking text segments to comprehend literary meaning, make connections, and infer the meaning of unknown words. Thus, reading while listening is recommended for language comprehension and fluency, particularly for struggling readers and Emerging bilinguals [4].

There are at least three theoretical frameworks to consider regarding whether reading while listening will benefit learning. First, Dual Coding Theory posits that information is processed differently, yet simultaneously, with visual and audio representations [33]. Thus, presenting the information in visual and audio formats allows more information to be processed because both “channels” are used [34]. Second, the Simple View of Reading states that reading is the product of decoding and comprehension [35]. In this framework, students who are struggling with decoding are essentially blocked from comprehending. Listening to the text could support the decoding process for better comprehension [36]. Third, Cognitive Load Theory is based on evidence that attention and working memory are limited resources [37]. Reading-while-listening from a Cognitive Load Theory perspective may overwhelm limited resources with redundant information [38], making reading-while-listening more difficult than silent reading. However, it could be beneficial if reading-while-listening supports comprehension and reduces the cognitive load associated with decoding unfamiliar text. Although we are not testing reading or reading comprehension, we posit that greater comprehension will support learning the facts and that this successful encoding will facilitate integration. Thus, we are looking at reading-while-listening as a support for encoding when children are asked to read to learn.

Reading-while-listening has mixed support in the literature. A meta-analysis of reading-while-listening studies that included children through adults found that the effect of reading-while-listening compared to silent reading on comprehension was trivial [39]. This is certainly not the only study to investigate reading-while-listening and find no beneficial effect. For example, Rogowsky and colleagues [40] investigated reading-while-listening compared to silent reading with adults in an immediate and delayed (retention) test and found no benefit. However, a meta-analysis examining whether reading-while-listening benefitted comprehension compared to silent reading with struggling readers did find a benefit [41]. Indeed, several studies with struggling readers have found benefits [42,43,44]. Additionally, another meta-analysis found that younger participants, struggling readers, and emerging bilinguals benefited the most from reading-while-listening [4].

One previous study has compared memory integration performance of facts presented through reading-while-listening or silent reading [45]. This study was conducted in English in elementary classrooms with third- through fifth-graders. The results showed an advantage of reading-while-listening for third-grade students but a benefit of silent reading for fifth-grade students. The results were interpreted as a developmental shift from requiring decoding support to being hindered by redundant information when that support was no longer necessary. These results highlight that reading-while-listening is not going to benefit all learners and emphasize the need to examine whether it will be a helpful or harmful strategy for dual-language learners. It may be that the benefits of reading-while-listening are restricted to specific groups.

Dual-language students and emerging bilinguals are attempting to learn content presented through texts written in a language they are still learning, adding substantially to the challenge of academic success. Reading-while-listening may improve reading comprehension and vocabulary learning, improving learning outcomes from expository texts. Indeed, several studies investigating second-language learning have shown advantages in reading-while-listening over silent reading in comprehension [46], fluency [36], and vocabulary learning [47]. The benefit can be explained by how adding audio changes reading behavior. In an eye-tracking study with adult second-language readers, Conklin and colleagues [6] found that silent reading in the second language resulted in longer fixations than first-language reading. These differences, however, were eliminated with the addition of audio in the reading-while-listening condition. Reading while listening supported fluent reading and comprehension of second language materials. Thus, evidence supports reading-while-listening for reading comprehension and vocabulary acquisition.

In summary, higher comprehension of text and better vocabulary learning from texts are evidence that reading-while-listening supports encoding of new factual information. Encoding is a necessary first step to memory integration (see [23] for discussion). Additionally, the support in fluency and decoding, as indicated by eye-tracking [6], may lower the cognitive load associated with reading in the second language to free up more attentional resources for learning the target material and subsequently integrate new information with prior knowledge. Thus, we predict that reading-while-listening will support self-derivation through integration in dual-language students across lessons and languages through better encoding of the new information.

5. Graphics

The use of graphics in the classroom is widely recommended and used in science materials [48,49]. In fact, IES What Works Clearinghouse recommends the use of graphics/visuals (with conditions) to promote learning ([50]; for guidelines, [51]). Additionally, integrating across text and graphics is one of the education standards in the United States as seen in the Common Core State Standard, https://www.thecorestandards.org (accessed on 15 August 2024). Graphics in the classroom, especially when accompanied by text, are thought to improve comprehension by drawing attention to important information, depicting abstract concepts in a concrete way, and facilitating the encoding of the information [48,49,52,53].

Similar to reading-while-listening, the use of graphics can be understood through Dual Coding Theory and Cognitive Load Theory. Just as seeing the text while hearing audio uses both “channels” to allow for the processing of additional information, graphics and other forms of multimedia learning capitalize on separate streams [34]. However, just as Cognitive Load Theory might predict that text and audio overload limited working memory resources, graphics could also tax limited attention resources [38]. Alternatively, if graphics support comprehension, facts presented through graphics could be less taxing on limited resources. The research appears to support the latter, but with conditions. These theories also informed the development of the Cognitive Theory of Multimedia Learning [54], which guides the development and use of effective multimedia materials.

A particularly attractive feature of graphics is that they are thought to transcend language, meaning that some or all the important information can be gleaned without requiring the learner to know a given language. This means that graphics might be a particularly well-suited tool for conveying new factual information in classrooms of bilingual learners. Graphics are already being used in classrooms to aid bilingual education. Schall-Leckrone [55] conducted qualitative research and found that teachers regularly use four different types of scaffolds to support bilingual learners’ acquisition of academic content: visuals (graphics), vocabulary instruction, graphic organizers, and annotated text. Wong and Samudra [7] found that 4- and 5-year-old children learned more second-language vocabulary words when they were presented with graphics in addition to audio than when they were presented with audio alone. Indeed, including visuals in early childhood education has a robust impact on making content more comprehensible [56,57], thus supporting learning of new semantic knowledge.

Graphics are considered to be beneficial to learning. However, graphics can also be difficult to understand. Indeed, graphics often require support for students to extract the intended information accurately (for discussion, see [54]). Thus, graphics can aid learning, but careful design and support might be needed for students to get the most information from them. The use of graphics could facilitate comprehension and encoding of target facts, therefore supporting subsequent cross-language memory integration. However, the graphics could also tax limited attention resources, be misunderstood or interpreted, or distract with seductive details, making integration across languages even more challenging. A previous study ([58], Study 3) found that children in classroom settings performed better on an integration task when half of the information was presented in a graphic format compared to all the information being presented in text-only format, but only when there was support provided at encoding. The children in this study were native English speakers being instructed in English. Thus, graphics with appropriate supports can facilitate learning and subsequent integration across lessons, but we do not yet know if graphics support cross-language self-derivation through integration.

In summary, the use of graphics is recommended in teaching academic content, such as science, because it is believed that graphics can transcend a language barrier to facilitate encoding, making it particularly interesting as a strategy in dual-language classrooms. However, prior research has indicated that graphics are often misinterpreted or misunderstood. They do not always support encoding factual information [59,60], let alone the self-derivation of new knowledge through memory integration. Indeed, integrating in a same-language format with graphics can pose a challenge ([58], Study 2). It is unknown if adding graphics would support encoding and subsequent cross-language integration or be an additional burden to limited attention resources.

6. The Current Research

In the current research, we conducted two studies to examine whether effective pedagogical practices for encoding directly provided facts would support self-derivation through memory integration in emerging bilingual elementary students. In Study 1, we examined whether self-paced reading or reading-while-listening better supports integrating facts presented in separate lessons. Facts were presented in English, Spanish, or one in each language (cross-language). Dual-language education students (M = 9.54 years) in grades 3 and 4 participated in their classrooms in this within-person experiment. In Study 2, we examined whether using graphics to support text would mitigate the challenges of cross-language integration. We used best practices for presenting content through graphics and text to examine whether graphic presentations would facilitate cross-language integration. Dual-language education students (M = 10.63 years) in grades 4 and 5 participated in their classrooms in this within-person experiment testing English-only and cross-language integration performance in text–text and text–graphic integration trials.

The research was made possible with the collaboration of a school system offering dual-language education. The school system is in an area of rural poverty with racial, ethnic, and linguistic diversity. The work was all completed within dual-language education (Spanish/English) classrooms. Thus, the research was conducted with a population and in an area that is underrepresented in research.

In both studies, we experimented with support at encoding based on previous research that shows encoding is a necessary first step to self-derivation through memory integration (for discussion, see [3,23]). Additionally, difficulty in encoding could use attentional resources needed to integrate the new information with prior knowledge. Based on prior research on memory integration, reading-while-listening, and the use of graphics, we predicted that both reading-while-listening and graphics at encoding would facilitate target fact encoding and subsequent memory integration performance as indexed by self-derivation through integration.

7. Study 1

The primary aims of Study 1 were to examine whether independent silent reading or reading-while-listening more effectively supported the integration of semantic content presented in English, Spanish, or a cross-language condition. We examined these questions with dual-language students in classrooms in a within-person experimental design.

7.1. Methods

7.1.1. Participants

The participants were 56 children (29 females, M = 114.46 months, SD = 7.48 months) across grade 3 (n = 25; 12 girls; M = 107.32 months, SD = 3.64 months) and grade 4 (n = 31; 17 girls; M = 120.23 months, SD = 3.91 months). The participants were drawn from a larger study sample of 476 children. Children were included if they were in the target grades within dual-language education classrooms and had the critical measure of self-derivation through integration. All children attended the same school in a rural southern state and were enrolled in the dual-language program, which is a strand within the school. Each grade level has two dual-language classrooms. The dual-language program followed a 50/50 model that switched classrooms for English and Spanish instruction half-way through the day and alternated language of content instruction by unit. The school system holds two lotteries for acceptance into the dual-language program. One lottery is for native Spanish-speaking children who speak primarily Spanish in the home. The other is for native English-speaking children who do not speak Spanish as the primary language in the home. Half of the available spots are given to each group such that the classrooms are composed of 50% Spanish-dominant and 50% English-dominant speakers. The majority of Spanish dominant participants were recent immigrants from Mexico although there is a subpopulation from Columbia. Consent forms were sent out via parent communication folders and only data from children whose parents/guardians returned signed consent forms were included in the analysis (approximately 54% of the population).

The sample was racially, ethnically, linguistically, and socioeconomically diverse, as recorded in the family demographic survey. Families identified as 9% Black, 70% Hispanic/Latinx, 5% multiracial, and 16% White. Primary caregiver education was reported as 63% with a high school education or less, 9% with some training beyond high school, 11% with a technical or associates degree, and 11% with a bachelor’s degree or additional education beyond a college degree. Approximately 6% did not report education. According to the school system, approximately 86% of children in the community qualified for federally funded school lunch assistance during the year of data collection.

Participating teachers were thanked with a USD 20 gift card, parents with a USD 10 gift card, and participating children with a small school supply item (e.g., an eraser). The Institutional Review Board and participating school system School Board reviewed and approved all study protocols and procedures.

7.1.2. Stimuli

The stimuli were 16 novel “target” facts (eight fact pairs) that could be integrated to create eight novel “integration” facts aligned to the curriculum of each participating grade level (eight additional fact pairs were included for the larger study, but are not included in this analysis). Within both presentation formats (independent silent reading and reading-while-listening), two of the four were presented in the same language condition (both target facts in English or both target facts in Spanish), and the other two were presented in a cross-language condition (one target fact in English and the other in Spanish). Facts were presented as written black text on a white background projected through PowerPoint. The reading-while-listening condition also included an audio recording of the facts read aloud. A native Spanish speaker from Columbia made the Spanish audio recording, the same country of origin as the majority of the Spanish teachers in the dual-language program. Teachers in the program reviewed audio to ensure the accent was consistent with classroom exposure. The English audio recording was made by a native English speaker using a General American accent [61] and was also reviewed by classroom teachers for comprehensibility. Although English accents varied across classrooms, the General American accent was often heard given its prevalence in media, including educational materials used in the classrooms such as BrainPOP. Pilot testing was conducted to establish that the target facts and integration facts were novel to children in the participating grades and that both target facts were necessary to generate the integration fact. Pilot testing was conducted in classrooms from schools not participating in the research as part of a research outreach program. In addition, the testing paradigm without the presentation format manipulations was validated in previous laboratory and classroom studies with the same age range [22,62,63]. Stimuli are available by request, https://scholarblogs.emory.edu/bauerlab/ (accessed on 15 August 2024).

7.1.3. Measures

We assessed these cognitive domains: verbal comprehension (English and Spanish), silent reading, working memory, and nonverbal intelligence.

Verbal Comprehension. The Woodcock-Muñoz Language Survey-Revised Normative Update (WMLS^®-R NU; [64]) is a measure of verbal comprehension normed in both English and Spanish. We administered two subtests: Vocabulary (Test 1) and Analogies (Test 2) in both languages. Participants received one point for each correctly answered item and the test was discontinued when six consecutive items were answered incorrectly. Scores were summed for both Test 1 and Test 2 in each language, resulting in one verbal comprehension score for English and one for Spanish. Tests were administered individually.

Silent Reading. The Test of Silent Reading Efficiency and Comprehension (TOSREC) is a brief, group- or individually administered timed reading test that assesses silent reading of connected text for comprehension [65]. Examinees read sentences from a grade-level test booklet and decide whether sentences are true or false (e.g., “A cow is an animal.”). The testing time is approximately three minutes, and there are four equivalent forms. One form was translated into Spanish to assess Spanish silent reading. We used the TOSREC scores as a measure of reading comprehension with English and Spanish outcome scores based on number correct. The test was administered to the class as a group.

Working Memory. We measured working memory with a visuo-spatial working memory task that was not reliant on English or Spanish vocabulary knowledge: the backward Corsi block task [66]. Participants view a screen of squares that light up in a sequence. Participants then touch the squares in reverse order. The sequence begins with two blocks in each trial and increases by one block after two correctly completed trials. The task terminates if neither trial at a given level is completed correctly. We recorded the total score of the correct number of touches made during the task. The test was administered individually.

Nonverbal Intelligence. The TONI₄ [67] is a test of non-verbal intelligence specifically designed to be language-free. Individuals are shown a series of patterns and asked to choose an image that completes each puzzle. We administered the task in group format by showing the full class each image and then providing four to five options (labeled with letters that corresponded to clicker buttons) to complete the puzzle (the original task has four to six options). Children responded individually with their “clickers” within the time limit. The score was the sum total of correctly completed puzzles. The test was administered to the class as a whole.

Positionality

The PI is a cognitive developmental psychologist and former elementary dual-language teacher. The motivation for this work came from the PI’s own experience in the classroom and a desire to know more about how children build semantic knowledge across languages and if classroom strategies can facilitate cross-linguistic semantic knowledge integration. Research assistants were all bilingual students of cognitive psychology and/or education studies learning about educational research through their participation.

7.1.4. Procedure

Children were tested in two sessions in March of their academic year. The first session was administered in groups within their classrooms, while the second session took place individually one week later. Participation in Session 1 included all children in the classrooms, but data analysis was limited to those for whom parents provided written consent. The classroom sessions, each lasting 45 min, were structured in three phases. Phases 1 and 2 presented the target facts. Phase 3 tested children’s self-derivation through integration performance and target fact recall.

During Phase 1, students heard and saw the text for the first target fact in each of the eight integration pairs. The text was projected on the classroom screen (approximately 4′ by 6′). For those presented as reading-while-listening, prerecorded audio files also read the text aloud. Children then completed the TOSREC in Spanish and English. This also served as a buffer activity for approximately 10 min. Phase 2 started immediately after the buffer activity. In Phase 2, children saw (and heard for reading-while-listening) the second of each of the 16 target facts from each of the eight integration pairs. They next completed the non-verbal intelligence measure, which served as the 10-min buffer activity. The displayed text and associated audio advanced automatically for both phases to ensure consistent timing across classrooms. The target facts were counterbalanced such that integration fact pairs were represented equally across classrooms in the silent reading and reading-while-listening. In addition, English and Spanish fact pairs were counter-balanced across classrooms.

Phase 3 started immediately after the non-verbal intelligence test. Children were presented with eight open-ended integration questions in written format. While all children received the same questions, these were presented in four different predetermined random sequences. This variation ensured that adjacent children did not have identical question versions, thus minimizing the potential for sharing answers. Each child silently read the questions and wrote down their responses. After completing the open-ended testing phase, the integration questions were presented in a forced-choice format using PowerPoint. An experimenter read each question aloud, and children were instructed to select the “best answer” from three provided options, where only one option was correct (resulting in a 33% chance of guessing correctly by chance). All answer choices were drawn from presented facts so children could not correctly answer based on familiarity. Each child used individual response devices, and their answers were recorded using TurningPoint Cloud software. After the integration questions were asked, children were asked a subset of target fact recall questions in forced-choice format, answering using their response clickers just as in the integration test. All children were thanked for their participation with a mechanical pencil.

In Session 2, children with parental consent completed Spanish and English verbal comprehension tests and working memory one-on-one with a research assistant in a separate quiet area. This was part of a test battery for the larger study. A team of 10 bilingual research assistants completed Session 2. All research assistants were extensively trained in working with children and helping them to feel comfortable.

7.1.5. Scoring

Children received one point for each correct answer to the integration questions in each language and presentation format for a total of six scores (silently read English, silently read Spanish, silently read cross-language, reading-while-listening English, reading-while-listening Spanish, reading-while-listening cross-language) in open-ended and forced-choice format. Children answer target fact forced-choice questions from both languages and presentation formats, resulting in four scores (silently read English, silently read Spanish, reading-while-listening English, reading-while-listening Spanish). Scores are reported as percent accuracy.

7.2. Results

The results are reported in three sections. First, we examined silently read compared to reading-while-listening self-derivation through integration performance between English facts, Spanish facts, and cross-language provided facts (performance reported in

Table 1. Study 1 integration performance accuracy.

	Presentation Format	Language	N	Min	Max	Mean (SD)	p-Value	Skewness		Kurtosis
								Statistic	Std. Error	Statistic	Std. Error
Open-ended testing	Silent reading	English-English	51	0	1.00	0.11 (0.25)		2.35	0.33	4.95	0.66
		Spanish-Spanish	51	0	0.50	0.02 (0.10)		4.89	0.33	22.83	0.66
		Cross-language	51	0	1.50	0.22 (0.38)		1.69	0.33	2.08	0.66
	Reading-while-listening	English-English	51	0	1.00	0.09 (0.26)		2.94	0.33	7.65	0.66
		Spanish-Spanish	51	0	0.50	0.09 (0.19)		1.75	0.33	1.10	0.66
		Cross-language	51	0	1.00	0.10 (0.25)		2.56	0.33	6.05	0.66
Forced-Choice testing	Silent reading	English-English	52	0	1.00	0.51 (0.3)	<.001	−0.03	0.33	−1.06	0.65
		Spanish-Spanish	52	0	1.00	0.39 (0.35)	.19	0.31	0.33	−0.86	0.65
		Cross-language	52	0	1.00	0.57 (0.25)	<.001	−0.002	0.33	−0.11	0.65
	Reading-while-listening	English-English	52	0	1.00	0.54 (0.39)	<.001	−0.14	0.33	−1.36	0.65
		Spanish-Spanish	52	0	1.00	0.61 (0.35)	<.001	−0.31	0.33	−0.86	0.65
		Cross-language	52	0	1.00	0.50 (0.26)	<.001	<0.001	0.33	−0.23	0.65

). Second, we describe the target fact forced-choice performance of silently read compared to reading-while-listening for target facts presented in English and Spanish (performance reported in

Table 2. Study 1 target fact performance accuracy.

	Presentation Format	Language	N	Min	Max	Mean (SD)	p-Value	Skewness		Kurtosis
								Statistic	Std. Error	Statistic	Std. Error
Forced-choice testing	Silent reading	English	42	0	1	0.35 (0.26)	.58	0.47	0.37	0.15	0.72
		Spanish	49	0	1	0.35 (0.35)	.73	0.74	0.34	−0.58	0.67
	Reading-while-listening	English	52	0	1	0.52 (0.34)	<.001	0.05	0.33	−1.14	0.66
		Spanish	51	0	1	0.43 (0.34)	.03	0.25	0.33	−1.05	0.66

). Lastly, we examine the cognitive correlates of self-derivation performance across presentation formats and languages. All analyses were conducted using SPSS Statistics Package (Version 24). All statistical tests were two-tailed.

7.2.1. Self-Derivation through Integration Performance

The forced-choice integration performance data demonstrated normality, whereas the open-ended testing showed minimal response variability, often nearing floor levels. This low performance in open-ended memory integration tests within classroom settings is not uncommon [3,20], but it warrants caution in interpreting results. Consequently, only the data from forced-choice selections are analyzed further.

We first examined whether English-only, Spanish-only, and cross-language (Spanish/English) self-derivation through integration performance was significantly above chance in the silently read and reading-while-listening presentation formats. The one-sample t-tests against chance (0.33) showed that all were significantly above-chance performance with p values less than 0.001, with the exception of the Spanish silent-reading condition which did not differ from chance performance (t(51) = 1.33, p = 0.19).

We next examined Presentation format and Language in a repeated measures 2 × 3 ANOVA with Presentation format (silently read vs. reading-while-listening) and Language condition (English vs. Spanish vs. cross-language) as within-subjects repeated measures. The results revealed no main effects of presentation version (F(1, 51) = 2.91, p = 0.10, η_p² = 0.05) or language condition (F(2, 50) = 0.66, p = 0.52, η_p² = 0.03). There was a significant interaction between presentation format and language (F(2, 50) = 5.50, p = 0.007, η_p² = 0.18). Follow-up analyses revealed that Spanish fact integration performance was higher in the reading-while-listening condition compared to the silent reading presentation format. English integration performance and cross-language integration performance did not differ between presentation formats.

7.2.2. Target Fact Performance

We analyzed target fact performance by first doing a one-sample t-test against chance (0.33). We found that target fact performance for the English silent reading and Spanish silent reading formats did not significantly differ from chance (t(41) = 0.55, p = 0.58 and t(48) = 0.34, p = 0.73, respectively). The target fact performance in the English reading-while-listening and Spanish reading-while-listening conditions were significantly above chance (t(50) = 4.06, p < 0.001 and t(50) = 2.22, p = 0.03, respectively).

We next examined whether target fact performance in a repeated measures 2 × 2 ANOVA with Presentation format (silently read vs. reading-while-listening) and Language condition (English vs. Spanish) as within-subjects repeated measures. The results revealed a main effect of Presentation format (F(1, 39) = 13.77, p < 0.001, η_p² = 0.26), such that reading-while-listening target fact performance was significantly better than silent reading. There was no main effect of Language (F(1, 39) = 0.84, p = 0.37, η_p² = 0.02), nor was there a significant Presentation format by Language interaction (F(1, 39) = 1.63, p = 0.21, η_p² = 0.04).

7.2.3. Cognitive Correlates Related to Self-Derivation

We next examined which cognitive correlates were related to self-derivation performance in each presentation format and language condition. We assessed the relation with English and Spanish verbal comprehension, English and Spanish silent reading, working memory, and nonverbal intelligence. English silent reading integration performance was not significantly correlated to any of the cognitive measures. Spanish silent reading integration performance was significantly correlated with Spanish verbal comprehension (n = 47, r = 0.31, p = 0.03), the test of Spanish silent reading (n = 51, r = 0.42, p = 0.002), Spanish silent reading target fact performance (n = 48, r = 0.34, p = 0.02), and Spanish reading-while-listening target fact performance (n = 50, r = 0.37, p = 0.008). Cross-language silent reading integration performance was also significantly correlated with the test of Spanish silent reading (n = 51, r = 0.34, p = 0.01). English reading-while-listening integration performance was significantly correlated to English verbal comprehension (n = 44, r = 0.44, p = 0.003). Spanish reading-while-listening integration performance was significantly correlated with the test of Spanish silent reading (n = 51, r = 0.32, p = 0.02). Cross-language reading-while-listening integration performance was significantly correlated with Spanish verbal comprehension (n = 47, r = 0.29, p = 0.046). None of the integration measures were significantly correlated with non-verbal intelligence or working memory.

7.3. Discussion

The primary aim of Study 1 was to examine whether reading-while-listening better-supported memory integration of target facts presented with temporal delays and through different languages. Overall, the study did not replicate previously found deficits in cross-language integration compared to single-language integration ([3], Study 2). However, there was a deficit in integrating same-language target facts presented through Spanish. The reading-while-listening presentation format had significantly higher performance for integration of Spanish target facts compared to silent reading. Thus, our hypothesis is partially supported. There does not appear to be an advantage of reading-while-listening for cross-linguistic integration, but reading-while-listening did appear to benefit integration across temporal delays between Spanish facts.

It is interesting, though perhaps not surprising, that integrating two Spanish target facts presented in a silent reading condition was more challenging than integrating across languages. The research was conducted in the US in a predominantly English-speaking area. Environmental text (signage, menus, etc.) is in English as are most resources such as library books. In this English print-rich environment, children have limited exposure to Spanish text and may not be as fluent readers regardless of their verbal ability. The higher performance in the reading-while-listening presentation format supports the previous literature indicating that struggling readers or second language learners benefit most from this support [7] but extends this finding from vocabulary and reading comprehension to learning of directly provided facts and subsequent integration and self-derivation of new information.

The target fact performance replicates the previous literature findings that reading-while-listening supports comprehension and vocabulary learning. Although we did not test reading comprehension or vocabulary, vocabulary is a type of semantic knowledge similar to the facts we were presenting. Performance for target facts presented in the reading-while-listening presentation format was significantly higher than in the silent reading presentation format. However, this support at encoding was insufficient to facilitate integration and subsequent self-derivation of new knowledge. There was no main effect of the presentation format, and the benefit of reading-while-listening was limited to the Spanish-only integration pairs. Previous memory integration research has found encoding to be a necessary but not sufficient step in the integration process [8]. If all that was needed to support memory integration in the classroom was stronger encoding, we would see a presentation format benefit for all reading-while-listening fact pairs. Two previous studies have indicated that stronger encoding does benefit integration [3,23]. However, the results of this study indicate that the encoding support offered by reading-while-listening is insufficient.

The correlations replicate previous work examining cognitive correlates of self-derivation through integration. Previous work has found verbal comprehension to be the strongest predictor of successful self-derivation through integration [19,22]. The addition of the silent reading test as a significant predictor is not surprising given the heavy reliance on silent reading in this test paradigm. Overall, the correlations are consistent with prior research.

8. Study 2

The primary aim of Study 2 was to examine whether conveying information through a graphic would support encoding of new factual information and subsequent cross-language memory integration for dual-language students. We examined this question using the same self-derivation through integration paradigm as in Study 1, with modifications to stimuli to present a target fact within each pair through a graphic. Children were fourth- and fifth-graders enrolled in dual-language education programs within the same diverse rural school as Study 1.

8.1. Method

8.1.1. Participants

The participants were 67 children (34 girls, M = 125.8 months, SD = 7.604 months) across grade 4 (n = 38; 18 girls; M = 120.32 months, SD = 4.06 months) and grade 5 (n = 29; 16 girls; M = 132.79 months, SD = 4.77 months). The participants were drawn from a larger study sample of 476 children. Children were included if they were in the target grades within dual-language education classrooms and had been present for the critical self-derivation through integration measure. All children attended a school in a rural southern state and were enrolled in the same dual-language program as Study 1. Consent forms were sent out via parent communication folders. Only the data from children whose parents/guardians returned signed consent forms were included in the analysis (approximately 56% of the population).

The sample was racially, ethnically, linguistically, and socioeconomically diverse, as recorded in the family demographic survey. Based on family reports, the sample was 16% Black, 61% Hispanic/Latinx, 6% multiracial, 13% White, and 4% other or unreported. Primary caregiver education was reported as 55% with a high school education or less, 15% with some training beyond high school, 9% with a technical or associates degree, and 9% with a bachelor’s degree or additional education beyond a college degree, and 12% did not report caregiver education. According to the school system, approximately 84% of children in the community qualified for federally funded school lunch assistance during the year of data collection. Participating teachers, parents, and children were thanked as in Study 1.

8.1.2. Stimuli

The stimuli were 16 novel “target” facts (eight fact pairs) that could be integrated to create eight novel “integration” facts aligned to the curriculum of each participating grade level (eight additional fact pairs were included for a larger study, but are not in this analysis). Text target facts were presented as written black text on a white background projected through PowerPoint, with an audio recording of the facts read aloud (thus, reading while listening). Graphic presentations included a color image with a title and 3–4 labels. The title and the labels appeared as text and read aloud. Within both presentation formats (text–text and text–graphic), two of the four were presented in the same language condition (both target facts in English), and the other two were presented in a cross-language condition (one target fact in English and the other in Spanish). Note that the graphic stimuli were always presented in English. Thus, the two pairs of facts presented in the text–graphic cross-language condition had the text fact in Spanish. The order in which the facts were presented, the presentation format (text–text or text–graphic) in which facts appeared, and the language condition (same-language or cross-language) in which facts appeared were all counterbalanced and appeared equally often across classrooms.

Stimuli were pilot-tested to ensure that facts were novel yet developmentally appropriate for children in the target age range. Pilot testing also assessed that both target facts in each set were necessary to produce or self-derive the integration fact reliably. Pilot testing was conducted in classrooms from schools not participating in the research as part of a research outreach program. In addition, the testing paradigm and stimuli were validated in previous laboratory and classroom studies with the same age range, all in English [59]. Stimuli are available by request https://scholarblogs.emory.edu/bauerlab/ (accessed on 15 August 2024).

8.1.3. Measures

We assessed cognitive domains with the following tasks: Verbal Comprehension (English and Spanish), Visio-spatial Working Memory, and Nonverbal Intelligence. All tasks were the same as Study 1, with the exception that we did not administer a silent reading test.

8.1.4. Procedure

The testing procedures were very similar to those in Study 1, with differences based on the different experimental conditions. As in Study 1, testing occurred across two sessions about a week apart in March. Session 1 was group administered in the classroom and Session 2 was individually administered. Each of the classroom sessions were conducted by four researchers in two-person teams and had three phases: (1) exposure to the first half of the novel target facts, (2) exposure to the other half of the novel target facts, and (3) test for self-derivation through integration of new factual knowledge and target fact recall.

Session 1 occurred in children’s classrooms during a single session with the entire class at once. In Phase 1, the target facts were presented and projected on a screen in the front of the classroom. For facts presented as text, the text appeared on the screen accompanied by audio reading the target fact aloud. For target facts presented as graphics, the graphic appeared on the screen with the title. The title was read aloud and each label appeared one at a time accompanied by audio reading the label. Once all labels had been presented, the researcher asked a question about the target information depicted in the graphic. This question provided support for extracting and encoding the target fact. If a child answered a question incorrectly, another child was given an opportunity to answer. The researcher provided the answer in rare cases where the class did not arrive at the correct answer. A 10-min buffer activity of children listening to several paragraph vignettes followed the target facts.

In Phase 2, children were presented with the paired target facts (text and graphic) projected on the screen accompanied by pre-recorded audio, just as in Phase 1. Across these phases, children were presented with both members of the fact pairs. The stimulus pairs’ presentation order was counterbalanced such that each pair was presented in each condition (text–text, text–graphic; English, cross-language) approximately equally often across participants. The target fact presentation was followed by the non-verbal intelligence test, which served as the 10-min buffer activity.

In Phase 3, children were tested for memory integration through open-ended and forced-choice responses to integration questions. This was conducted with the same procedure as in Study 1.

Session 2 was completed by a team of 10 bilingual research assistants with the same protocol as in Study 1. Children again scored an average of 5/5 for engagement and comfort during the individual session.

8.1.5. Scoring

Children received self-derivation through integration scores for both open-ended and forced-choice for tests for all conditions: text–text presentation in the same language condition, text–graphic presentation in the same language condition, text–text presentation in the cross-language condition, text–graphic presentation in the cross-language condition. The maximum number of questions in each condition is two. Children also received scores for target fact recall performance in each of the following conditions: English text, Spanish text, and graphic (always presented in English). For each question, participants received a score of 1 if the answer they provided was correct, and a score of 0 if the answer was incorrect (questions left blank were counted as incorrect). An average accuracy was taken across all the questions in each condition for each participant.

8.2. Results

The results are reported in three sections. First, we examine self-derivation through integration performance across Presentation formats and Languages. Second, we examine target fact recall performance by Presentation format and Language. Lastly, we examine the cognitive correlates of self-derivation performance. All analyses were conducted using SPSS Statistics Package (Version 24). All statistical tests were two-tailed.

8.2.1. Self-Derivation through Integration

As with Study 1, the open-ended data had limited variability and so we proceeded with analyses of the forced-choice data (see

Table 3. Study 2 integration performance accuracy.

	Presentation Format	Language	N	Min	Max	Mean (SD)	p-Value	Skewness		Kurtosis
								Statistic	Std. Error	Statistic	Std. Error
Open-ended testing	Text–text	English-English	49	0	0.5	0.04 (0.14)		3.15	0.34	8.28	0.67
		Cross-language	49	0	0.5	0.02 (0.10)		4.79	0.34	21.83	0.67
	Text–Graphic	English-English	49	0	1	0.11 (0.23)		1.98	0.34	3.32	0.67
		Cross-language	49	0	0.5	0.04 (0.14)		3.15	0.34	8.28	0.67
Forced-Choice testing	Text–text	English-English	60	0	1	0.35 (0.32)	.63	0.38	0.31	−0.66	0.61
		Cross-language	60	0	1	0.31 (0.31)	.59	0.45	0.31	−0.61	0.61
	Text–Graphic	English-English	60	0	1	0.44 (0.37)	.02	0.19	0.31	−1.12	0.61
		Cross-language	60	0	1	0.28 (0.28)	.21	0.32	0.31	−0.85	0.61

for performance data). Forced-choice data showed more variability than open-ended. We began by examining performance against chance (0.33). A one-sample t-test showed that the text–graphic same-language condition was the only condition with above chance performance (t(59) = 2.34, p = 0.023). The rest did not significantly differ from chance: text–text same-language condition performance (t(59) = 0.48, p = 0.633), text–text cross-language performance (t(59) = −0.55, p = 0.586), and text–graphic cross-language performance (t(59) = −1.28, p = 0.205).

A repeated measures 2 × 2 ANOVA with Presentation format (text–text vs. text–graphic) and Language condition (English vs. cross-language) as within-subjects measures revealed no main effects of Presentation format (F(1, 58) = 0.578, p = 0.450, η_p² = 0.010). There was a significant main effect of Language (F(1, 58) = 6.407, p = 0.014, η_p² = 0.098) such that the English-only integration facts had higher performance than the cross-language integration facts. There was not a significant interaction between Presentation format and Language (F(1, 58) = 2.330, p = 0.132, η_p² = 0.038), indicating that the main effect of Language did not differ by Presentation format.

8.2.2. Target Fact Performance

We began our analyses of target fact performance by doing one-sample t-tests against chance (0.33; see

Table 4. Study 2 target fact performance accuracy.

	Presentation Format	Language	N	Min	Max	Mean (SD)	p-Value	Skewness		Kurtosis
								Statistic	Std. Error	Statistic	Std. Error
Forced-Choice testing	Text–text	English	62	0	1	0.48 (0.34)	.002	0.03	0.3	−1.15	0.60
		Spanish	62	0	1	0.35 (0.46)	.77	0.66	0.30	−1.50	0.6
	Text–Graphic	English	62	0	1	0.50 (0.40)	.001	−0.02	0.30	−1.48	0.6

for performance data). We found that target fact performance for the text-only and the text–graphic presentations in English were both significantly above chance (t(61) = 3.321, p = 0.002 and t(61) = 3.392, p = 0.001, respectively). However, the one-sample t-test showed that target fact performance for Spanish text was not significantly different from chance (t(61) = 0.288, p = 0.774). No facts were presented as Spanish graphics.

A paired-sample t-test determined no difference in target fact performance between facts presented through English as text compared to facts presented through English as graphics (t(61) = −0.45, p = 0.652). Similarly, a paired-sample t-test revealed no differences in performance between facts presented as English text and those presented as Spanish text (t(61) = 1.85, p = 0.070).

8.2.3. Cognitive Correlates Related to Self-Derivation

We next examined whether cognitive abilities were correlated to self-derivation performance. We included English and Spanish verbal comprehension, nonverbal intelligence, working memory, and target fact performance. English same-language text–text was correlated with English verbal comprehension (n = 47, r = 0.29, p = 0.048), Spanish verbal comprehension (n = 48, r = 0.286, p = 0.048), and English text target fact performance (n = 60, r = 0.264, p = 0.042). English same-language text–graphic was also correlated with English verbal comprehension (n = 47, r = 0.309, p = 0.035) performance and with working memory (n = 46, r = 0.341, p = 0.020). Cross-language text–text integration performance was related to Spanish proficiency (n = 48, r = 0.305, p = 0.035). There were no correlations with cross-language text–graphic performance or non-verbal intelligence.

8.3. Discussion

The primary aim of Study 2 was to examine whether presenting facts as graphics would facilitate encoding of new factual information and subsequent cross-language integration. The results did not support our hypothesis that graphics would facilitate cross-language memory integration. There was a main effect of Language, such that cross-language integration was significantly lower than English-only integration, but this was not qualified by the Presentation format as predicted.

The results indicate that adding graphics and different languages was incredibly challenging for students. The results do replicate the difficulty of cross-language integration found in previous research [3], but graphics did not offer a mitigating solution. Although this could be due to difficulty in extracting the target information from the graphics, there is no evidence to support this conclusion because performance on the target facts did not differ between text and graphic presentation. They appeared to perform just as well on the graphic-presented facts as text facts. In addition, the only performance condition that was above chance was the single-language text–graphic condition. A second possibility that we also think is less likely is the challenge of reading Spanish facts. It may be that the Spanish facts presented are part of the challenge. We saw in Study 1 that Spanish-presented facts were a challenge. However, there was not a significant difference in performance between Spanish and English presented facts. Thus, it seems less likely that this was the barrier to cross-language integration with graphics. We also did not see an encoding benefit to the graphic facts in the form of significantly higher performance than text-presented facts. Thus, it appears the graphics were not detrimental to encoding as might be the case if they were poorly understood, but they also did not offer the encoding boost that was predicted.

There are at least two potential explanations for the poor integration performance and failure of graphics to mitigate the challenge. First, the combination of graphics and different language stimuli was cognitively taxing. The limited cognitive resources were overwhelmed trying to track and attend to target facts and left no resources for integrating facts. This explanation can be understood through the framework of Cognitive Load Theory. Cognitive Load Theory posits three types of load: intrinsic is the processing needed for the characteristics of the task; extraneous cognitive load to the processing needed for the, often poor, design of instructional materials; and germane cognitive load is used to integrate new information into the knowledge base [68,69]. If the extraneous cognitive load is too high, there are not enough cognitive resources left to devote to the germane load. If this is the case, simplifying the presentation could result in higher performance and potentially more benefits of using graphics. A second related possibility is that the combination of differences in languages and presentation format amplified the differences between related facts, making it more difficult for children to recognize that the facts were related to each other and exacerbating the challenge of cross-language integration rather than mitigating it. If this is the case, support at encoding would be insufficient and more support would be needed at retrieval and in recognition that the facts were related to each other. These are not mutually exclusive explanations and both have similar solutions: simplify the presentation and offer additional supports at retrieval to help connect related facts.

Integration performance in Study 2 was overall poor. Indeed, only the text–graphic single language condition was above chance. Given the overall poor performance and that the only successful evidence of integration was in a text–graphic condition, it does open the possibility that graphics are a support. However, the lack of a main effect shows that much more research is needed to understand whether and how best to use graphics without overwhelming cognitive resources and pushing facts cognitively further away from each other in memory.

9. General Discussion

The aims of the current research were to identify whether supports at encoding for learning new facts would facilitate the productive process of self-derivation through memory integration. We examined reading-while-listening and graphics as potential supports for learning new information and instructional strategies in dual-language programs to facilitate memory integration across lessons and languages, mitigating challenges found in previous research. Overall, we found evidence to support reading-while-listening and, to a lesser extent, graphics as instructional strategies for supporting learning of directly provided information. However, there was little evidence in support of these strategies facilitating integration across lessons and languages.

In both Study 1 and Study 2, we did find differences between memory integration performance conditions. In Study 1, this was between presentation format for Spanish-only fact pairs and in Study 2 this was a main effect of Language. In both studies, we concentrated our support on encoding with the hypotheses that supporting encoding of the facts through reading-while-listening and the inclusion of graphics would support subsequent self-derivation through memory integration. Our hypotheses are partially supported in that the target fact performance was better in the reading-while-listening compared to silent reading presentation format conditions (Study 1). The reading-while-listening presentation format does appear to have supported superior encoding compared to silent reading. We interpret this as better comprehension led to better learning of the target facts. In Study 2, we did not find a presentation format advantage, but we also did not find a deficit. Thus, although our support concentrated on encoding, encoding is a necessary but not sufficient step in self-derivation through integration, and the support at this step in the process was not enough to overcome the barriers to integration across lessons and languages. Prior research [8] found that self-derivation through integration performance improved when 6-year-old children were given the hint to “use what they learned before” during the integration test. In this case, no additional help with encoding was provided. Instead, guidance on how to use the information they had encoded was needed. The current research offers further evidence that encoding is not sufficient and encourages research into pedagogical strategies that support recognizing relations and integration.

The interventions worked on directly provided facts (especially in Study 1), but did not provide the needed support at integration. From our theoretical framework, this might indicate that the cognitive load of integrating across lessons and languages was too high. The task did not leave enough cognitive resources for integrating novel information and self-deriving new knowledge. We do not know from this study where the process of self-derivation through integration broke down, but the target fact performance indicates it was not at encoding. Further research can work to identify whether the challenge is in realizing the relevance of related facts to each other, retrieving facts from memory, selecting appropriate facts, or the act of self-deriving itself. Although we do not know where in the process memory integration is failing, we do know that providing additional support at encoding is not sufficient to support memory integration across lessons and languages in classrooms. Pedagogical strategies focused on encoding are likely to not be sufficient to help dual-language students build conceptual knowledge over time and language systems.

An interesting finding is the lack of consistent challenges in cross-language integration. Whereas there was a significant main effect of language in Study 2 indicating an advantage for English-only compared to cross-language integration, Study 1 found no such difference. Indeed, for some comparisons, cross-language integration was nominally higher than English-only integration. The children in Study 1 were younger than in Study 2 and the same age as previous research that also found the effect, making it unlikely that the difference is developmental. The stimuli were based on grade level curriculum in both studies and were consistent (except for the presentation format manipulations) for grade 4 across Study 1 and Study 2. Perhaps the requirement to silently read some facts required greater attention or provided additional motivation that was not present in Study 2 or previous research [3]. It is an interesting area for future research to identify what factors supported cross-language integration in Study 1.

The current study is not without limitations. Both studies were part of a larger investigation and, thus, time in the testing sessions and the number of stimuli that could be included was limited. One result of this limitation is that we did not provide the text–graphic presentation format in a Spanish-only language condition. We also did not have enough trials available to manipulate presentation order by language in Study 2. Previous work in classrooms has provided guidelines for the number of stimuli children can attend to and this limited the stimuli presentations. Expanding on the manipulations described here is an interesting direction for future research. Given the results of Study 1, future research should investigate integration in same-language conditions other than the majority language of the community. In addition, the studies took place in a rural community in the southeastern United States with a large Spanish-speaking population. The results should not be assumed to generalize to other communities with different socio-economic, racial, ethnic, and linguistic demographics. Although we have no specific hypotheses regarding how the demographics of the sample may impact the findings, we expect that children in a resource-rich environment who are already performing at high levels may not benefit as much from interventions to support performance as those in an area of rural poverty that often lacks access to resources. The sample limits the generalizability of the findings; however, it is also a strength of the study as it is more representative of the US population compared to if the study were conducted within an affluent community.

In conclusion, we found little evidence that support at encoding in the form of reading-while-listening or graphic support mitigated the challenges of integrating across lessons and languages. Although our directly taught facts indicated that the presentation format manipulations were successful, the benefits to encoding did not lead to higher self-derivation performance. Indeed, even when encoding was well-supported, integrating across lessons and languages remained a challenge for these dual-language students. Integration puts a higher demand on cognitive resources. It requires that students not only learn a fact, but recognize the relation between that fact and other lessons or prior knowledge to generate inferences. Future research may want to examine if supports at other parts of the memory integration process are a more effective pedagogical strategy for dual-language students and all students who are struggling to integrate across lessons and languages. Teaching facts is not enough. We need to support productive processes that help children build their own knowledge and concepts.