Research on the literacy and language development of Spanish-speaking English language learners
VIAS Subproject 3
Developing Vocabulary in Spanish-Speaking Children: The Acquisition of Vocabulary in English Project (AVE)
Summary
This subproject consisted of a series of research studies that built on each other to address the research goals. Study 1 investigated third grade ELL and English proficient students’ knowledge of high frequency academic vocabulary. Findings indicated large significant differences favoring English proficient students compared with ELL students in decoding and vocabulary. For both groups, Living Word Vocabulary (LWV) grade level and conceptual complexity were significantly related to difficulty in acquiring vocabulary.
Study 2a investigated the relative effectiveness of two methods of vocabulary instruction (direct instruction and paraphrasing) on second-grade Spanish-speaking ELL students’ acquisition of high frequency content vocabulary. Study 2b tested the effectiveness of an engaging multi-modal, text-based approach to teaching connectives to the same sample of second-grade students.
Findings indicated significant pre- to posttest gains on both subtests of the Gates-MacGinitie and on a researcher-designed assessment that tested all intervention content words and connectives.
Study 3 investigated the effectiveness of an intervention that combined the best elements of Study 2 and added reinforcement and cognate conditions. Findings indicated the intervention was effective in developing students’ knowledge of targeted vocabulary as measured by standardized and researcher-developed measures of vocabulary. Higher levels of reinforcement helped ELLs acquire abstract and concrete non-cognates but this was not the case for abstract and concrete cognates. Young ELLs acquired these words at similar rates without additional reinforcement, indicating that these young Spanish-speaking ELLs were drawing on first language resources to acquire English vocabulary, when the target English vocabulary shared cognate status with Spanish.
Our goals in this subproject were to describe the vocabulary knowledge of English language learner (ELL) students and their English proficient peers (Study 1), and to use that information to design and evaluate two vocabulary interventions for Spanish-speaking ELL students (Studies 2 and 3).
Description of Project Work and Findings
Study 1: Investigating Word Characteristics that Impact Vocabulary Acquisition
The goal of Study 1 (implemented in the 2008–2009 school year) was to explore third-grade students’ knowledge of high frequency English academic vocabulary and to determine whether word difficulty varied as a function of word attributes. To answer these questions, we developed an assessment called the Word Inventory. Part 1 of that assessment consisted of eight 13-item mini-tests that measured high frequency content word knowledge for students in Grades 1-6 drawn from the Educators’ Word Frequency Guide (Zeno, Ivens, Millard, & Duvvuri, 1995). Word meanings were assigned to word forms based on grade-appropriate definitions in the Living Word Vocabulary database and then were coded for attributes that might relate to word difficulty, including part of speech, cognate status (English/Spanish), degree of polysemy, and conceptual complexity. Part 2 of the Word Inventory consisted of 22 items that measured students’ knowledge of connectives (e.g., therefore, however); connectives were coded for type of relationship (additive, temporal, adversative, or causal).
A total of 335 third grade students of whom 153 were ELL students participated in Study 1. Findings indicated that English-only students outperformed ELL students on the Word Decoding subtest of the Gates-MacGinitie as well as on the Word Knowledge subtest. There were large significant differences between English proficient students and ELL students, favoring English proficient. While English proficient students did better than ELL students on cognates, within group analyses suggested that ELL students performed better on the cognates relative to non-cognates, whereas this was not the case for the English proficient students. Again, there were significant group mean differences in favor of the English proficient students.
To determine what makes high frequency content vocabulary difficult to acquire, we examined our data using a general linear model. For the full sample, item percent correct was uniquely predicted by Living Word Vocabulary level, conceptual complexity, and cognate status. Additionally, there were significant form and order effects, suggesting that fatigue/lack of time could be interfering. We also conducted an item analysis using IRT to obtain a latent indicator of difficulty for each item. This IRT difficulty parameter is a different way of indexing difficulty, as compared to item percent correct, and suffers from less measurement error. Results were similar for the IRT difficulty parameter. However, conceptual complexity and cognate status just missed being significant, whereas the number of letters was a significant unique predictor. Again, results were somewhat similar for the two subgroups. In total, the data suggest that Living Word Vocabulary level and conceptual complexity are related to difficulty in acquiring the content words. Further, the number of letters and cognate status may also be important factors.
We also examined what makes connective words hard to acquire. For both ELL and English proficient students, the Living Word Vocabulary level and the frequency level were significantly related to knowledge of the connectives. Further, findings suggest that ELL students perform better on additive and temporal connectives when compared with adversative connectives, whereas English proficient students perform better on temporal connectives than on adversative connectives and show no difference on additive and adversative connectives.
Study 2: Investigating Extended and Embedded Vocabulary Instruction for Young English Learners: Moderating Effects of Word Characteristics and Method of Instruction
The second study investigated two interventions (Studies 2a and 2b). The goal of Study 2a (implemented during the 2009–2010 school year) was to examine the relative effectiveness of two methods of vocabulary instruction for improving the content vocabulary knowledge of second-grade Spanish-speaking ELL students. The first method, direct instruction, provided more attention to individual words, while the second, paraphrasing in text, provided only enough instructional time for fast mapping. We also included a condition in which students were merely exposed to words. The nine-week intervention was delivered via high quality children’s literature (in Big Book form) read aloud to children. In each week, 18 words were targeted—six through direct instruction, six through paraphrasing, and six through exposure only.
Because the intervention involved vocabulary instruction that would help children comprehend challenging text, the target words for all three conditions were the kinds of words, based on data from Study 1, that second-grade students were not likely to have in their vocabulary and that appear frequently in school texts. Words were stratified by conceptual complexity and cognate status because Study 1 findings indicated that these factors contribute to ease of acquisition. The research question was the following: Are there differences in the learning probabilities of words as a function of method of instruction and of the word characteristics identified in Study 1?
The second intervention (Study 2b) tested the effectiveness of an engaging multimodal approach to teaching connectives to the same sample of second-grade students. Students were introduced to the connectives through the same shared book reading approach used in Study 2a and encouraged to use and manipulate these words through a variety of oral and written activities. Connectives that appeared in the Educators’ Word Frequency Guide at grades 1-6 and had a Living Word Vocabulary meaning ranked between fourth and sixth grade were included in the instructional condition. Connectives that were too easy according to the Study 1 assessments (i.e., known by more than 80% of the students) were excluded from the intervention. As with Study 2a, connectives were stratified according to Living Word Vocabulary levels and type of connective and assigned to the intervention or exposed-only condition.
Our sample consisted of 187 second-grade ELL students in 10 second grade classrooms in four high poverty elementary schools in a school district in the Southwest United States. We administered two subtests of the Gates-MacGinitie (the Word Decoding subtest and the Word Knowledge subtest) and a researcher-developed assessment—the Acquisition of Vocabulary in English Vocabulary Assessment (AVEVA)—which tested students’ knowledge of content words and connectives that appeared in the interventions. We also used an observation protocol to collect information on teachers’ fidelity of implementation and facility in Spanish and English; and we administered surveys to collect information on teachers’ background and self-reported facility in Spanish and English, teachers’ satisfaction with the intervention curriculum, students’ schooling history, and students’ home language and literacy practices.
For all variables, posttest scores were greater than pretest scores. To assess the impact of the intervention, we conducted within-group t-tests comparing pre- and posttest means for Gates-MacGinitie vocabulary, all blocks of content words, and connectives. Blocks of content words consisted of four types of words (abstract cognates, abstract non-cognates, concrete cognates, concrete non-cognates) instructed under three conditions (directly instructed, paraphrased, and exposed). In all instances, we controlled for family-wise error rate. Using this significance criterion, there were significant pre- to posttest gains on both subtests of the Gates-MacGinitie and on the AVEVA for all types of words under the directly instructed and paraphrased conditions, and for instructed connectives.
In order to compare gains on types of words using the directly instructed and paraphrased instructional conditions, we computed gain scores for the four types of content words in both the directly instructed and paraphrased conditions. Next, we compared these gains using t-tests. Results indicated that the gains were statistically greater for the directly instructed words in all conditions except the concrete cognates. In this instance the mean difference was in favor of the directly instructed condition, but not statistically significant.
Study 3: Investigating Rich Vocabulary Instruction in a Between-Subjects Design and the Role of Reinforcement within the Treatment
In Year 4, we implemented Study 3, the second intervention study. Development of the intervention was informed by what we learned in Study 2. For content words, we selected the same word types and created one instructional method that combined the best elements of the two instructional methods used in Study 2. The direct instruction approach was slightly reduced and the paraphrasing approach was enhanced. Additionally we explored how much reinforcement is necessary for the various types of words (concrete cognates, concrete non-cognates, abstract cognates, abstract non-cognates) by assigning words to two conditions—one with twice as much reinforcement as the other. We also explored whether students who were taught about cognates and practiced working with cognates were able to use this knowledge to figure out the meanings of English cognates they had not been taught. Finally, we continued with the connectives intervention with some slight modifications to strengthen it.
Study 3 was carried out in 22 classrooms in eight high poverty schools in the Southwest United States. All schools and teachers were new to the study and thus were not exposed to the intervention implemented in Study 2. Randomization was conducted at the teacher level within schools. This study had two major components, a between-subjects component and a within-subjects component. The between-subjects component focused on differences between the treatment group and control group on acquisition of the targeted vocabulary and growth in oral language proficiency as indexed by a standardized oral language development measure—the Test of Language Development (TOLD). The within-subjects analysis focused on the treatment group only and examined the differences in outcomes for four types of words (concrete cognates, abstract cognates, concrete non-cognates, and abstract cognates) in three different instructional conditions (exposed only, instructed with some reinforcement, instructed with double reinforcement).
To address the question of differences between treatment and control groups we conducted a repeated measures ANOVA with time (pre/post) and word type (abstract cognates, concrete cognates, abstract non-cognates, concrete non-cognates) repeated within classroom being predicted by group status (treatment/control). On the AVEVA, the researcher-developed measure of content vocabulary, there were significant differences between treatment and control groups with treatment classrooms outperforming control classrooms on all types of vocabulary. A repeated measures ANOVA was also conducted for the connectives with time (pre/post) repeated within students being predicted by group status. Results indicated that treatment classrooms gained more connectives knowledge than control. For the oral vocabulary subtest of the TOLD, the results just missed significance, with the treatment classrooms having a higher oral vocabulary mean. For the other two subtests of the TOLD and a passage comprehension subtest, there were no statistical differences between treatment and control groups.
Within-subjects analyses were conducted for the treatment group to determine if certain types of content words were learned more readily and if twice as much reinforcement of instructed words resulted in greater gains. To investigate this question we conducted a repeated measure ANOVA with time, instructional condition, complexity, and cognate status repeated within treatment students. In examining the results, we found that several factors were significant. Students made gains over time, double reinforcement resulted in greater gains, and gains were greater for conceptually simple words and for non-cognates. To determine if instruction and reinforcement impacts gains on words and types of words we fit a random intercepts model for the treatment group only. In this model, classroom performance was at the between level with performance on word type X instructional type (e.g., exposed concrete cognates, reinforcement plus abstract non-cognates, etc.) being predicted by instructional type, cognate status, complexity, and wave.
There is a significant wave X instruction interaction, which can be interpreted to mean that across time certain instructional methods are better than others. To investigate which specific methods lead to better results, we conducted follow-up comparisons, first for all words, then for each specific type of word. Overall both some reinforcement and double reinforcement are superior to exposure only, and double reinforcement is superior to some reinforcement. However, in examining the specific word types, double reinforcement is only superior to some reinforcement for abstract and concrete non-cognates. For words overall, abstract non-cognates, and concrete non-cognates the slope of words for the exposure condition is the least steep followed by some reinforcement, followed by double reinforcement. However, for concrete cognates and abstract cognates the slopes for some reinforcement and double reinforcement are much closer to being parallel. Thus, it appears that extra reinforcement, relative to regular reinforcement, may only improve performance on non-cognates.
In year 5, we collected delayed posttest data on a sample of students in treatment and control groups who participated in the intervention study in year 4. Seven students per classroom received the TOLD oral vocabulary subtest and a subset of words from the AVEVA. Analyses of these data are ongoing. To investigate if students retained the information they learned in the intervention we fit an explanatory IRT mixed model, with classroom at the between level, and item level performance being predicted by wave and treatment. There was a significant wave by treatment interaction. In examining the least squared means, it appears that the treatment group is not performing as well as they were, one year after the conclusion of the intervention, but they are still performing substantially higher than the control group. Additional analyses are underway to investigate these data in relation to word type and amount of reinforcement.
In the final year, we conducted additional analyses on data collected during the two intervention studies and continued to present findings at national conferences and prepare manuscripts for publication.
Selected Dissemination Activities
The Principal Investigator and AVE team have disseminated subproject 3 findings at national conferences including American Educational Research Association Annual Meetings (2010, 2011, 2012, 2013), SRCD Meeting (2011), Society for the Scientific Studies of Reading (2011).,National Council of Teachers of English (2013), TESOL International (2010, 2012, 2013), Ms. Artzi, Ms. Massoud, Dr. August, and Dr. Barr presented validation evidence regarding the AVEVA at the Language Testing Research Colloquium (2012). Dr. August, Ms. Artzi, and Ms. Massoud have presented workshops based on project findings at TESOL International (2010, 2012, 2013), California Association for Bilingual Education’s Annual Conference (2012), and National Council for Teachers of English Annual Conference (2012). Dr. August also presented findings at a meeting convened by National Research Council in July 2012 focused on bridging the early childhood years to the early grades and in a keynote speech at the Children’s Learning Institute, UT Health Sciences Center in May 2013.
Reference
Zeno, S. M. , Ivens, S. H., Millard, R. T., & Duvvuri, R. (1995). The educator's word frequency guide. New York: Touchstone Applied.
Subproject 3: Table of Instruments Used in the Study Download the PDF.
back to top
