Traditionally, the early stages of learning to read were considered a process of essentially memorizing the large set of rules for translating the marks on the page into sounds that can be linked to meaningful ideas and concepts. This, after all, is the way we typically teach our children to read. However, while these rules are clearly an important educational tool, many psychologists now view learning to read as a process of encoding the statistical relationships between written letters and sounds rather than explicit rules. The brain is capable of identifying complex correlations between diverse sets of information even when these correlations are not easily describable. Successful reading strategies are likely to rely heavily on this ability. As a result, learning to read is more of a skill than a set of knowledge -- it may be more like learning to shoot a basketball than like memorizing a list of items. A research team led by Robert McMurray & Eliot Hazeltine at the University of Iowa will investigate how factors known to affect skill learning can be harnessed to improve reading ability. They will manipulate the ways in which words are organized and presented to students to find the approach that makes the critical statistics used by the relevant learning mechanisms explicit.
The experiments will identify approaches that can be used to improve how we teach reading to children. Reading is a foundational skill, laying the groundwork for the entire curriculum that follows. Thus, improvements in children's reading ability can have widespread and lasting impact for their performance in school and in society in general. Additionally, because reading is an elaborate and nearly universal behavior, it provides an excellent model task for studying how humans learn complex tasks more generally.
Our proposal examined how first-graders learn reading skills like how to pronounce letters as a function of their surrounding context (e.g., MAT vs. MATE). We sought to determine how principles from learning theory scale to systems like reading, and if they can be applied in the classroom. We had two aims. 1) To develop new training tasks based on words’ meanings (most tasks used to teach these skills focus on spelling/sound) and to determine how these relate to existing spelling/sound (decoding) tasks. 2) To determine whether similarity or variability among words used to train phonics rules promotes better learning and generalization. To accomplish this, we needed a platform in which to run short-term, controlled learning studies on students in their classrooms. Thus, we reprogrammed a commercial phonics intervention, Access Code (Foundations in Learning Inc., 2010), to teach a smaller number of phonics regularities in a shorter time and to manipulate theoretically interesting factors. Access Code uses a collection of tasks to teach children phonics, delivered over the internet to their classroom. In these tasks, students might select a missing letter from an auditorially presented word, or select the word that matches what they hear. Experiment 1. As Access Code focuses on phonics, none of these tasks involve meaning. Our long-term goal is to examine how children learn both to map letters to sound, and letters to meaning, so we developed eight new tasks, that were similar to Access Code's tasks but used pictures of a word's meaning. This experiment asked if first-graders could perform these tasks, and if these tasks would correlate with the sound-based tasks (implying one general skill), or if they were more independent (implying separate skills) . We ran 97 first-grade children at two elementary schools in West Des Moines, IA, delivering our training directly to their computer labs. Students performed well. However, we needed 130-150 participants for our correlation analyses, and were not able to achieve this during the study period. This year, we have obtained additional funds to test an additional cohort and begin data collection in April, 2012. Experiment 2. An important debate in reading is whether spelling/sound regularities are explicit rules (E is prounced as /?/ in contexts like BED ), or probabilistic associations. While intuitively they seem rule-like, rules never fullly apply and there are strong subregularites within exceptions. For example, EA is typically prounced with an /i/ as in BEAD, but there are groups of exceptions like DEAD and THREAT). One prediction from the statistical framing is that principles from laboratory-learning paradigms may improve reading instruction. However, these principles often conflict . Experiment 2 used the remodeled Access Code framework to contrast two principles. We looked at vowels, where surrounding consonants are irrelevant to their prounciation (e.g., BED, WRECK and GEM are all pronounced similarly). Laboratory work conflicts on what should promote learning. Some studies argue that variable consonant frames will highlight more invariant pronunciations (Gómez, 2002; Rost & McMurray, 2010); others, that similar frames force children to deeper levels of analysis (Ahissar & Hochstein, 2004). Students started with a pre-test on six phonics rules. They then completed several days of training, followed by post-test. Pre- and post-tests assessed trained words and tasks, and generalization to new words and tasks. Children were randomly assigned to recieve either variable or similar consonants in the words used in training. The results were startling. Variability significantly improved performance across the board, in both trained and novel words and tasks. In some conditions, no learning at all was seen with similar words, but significant learning was observed with variable words (for example, in the short vowels, or the boys in our sample). Existing phonics curicula emphasize tasks like word-families that promote similarity (McCandliss, Beck, Sandak, & Perfetti, 2003). Our results suggest better results with variability. This could easily be applied beyond our computer-based implementation, for example, guiding word lists used in worksheets, reading and spelling textbooks, and other educational materials. Methodologically, it suggests a powerful way to study learning in the classroom, while retaining the control of the laboratory. Theoretically, it is one of the first studies to clarify the nature of the learning mechanisms that underly early reading. References Ahissar, M., & Hochstein, S. (2004). The reverse hierarchy theory of visual perceptual learning. Trends in Cognitive Sciences, 8, 457-464. Foundations in Learning Inc. (2010). Access Code. Iowa City, IA. Gómez, R. (2002). Variability and detection of invariant structure. Psychological Science, 13, 431-436. McCandliss, B., Beck, I., Sandak, R., & Perfetti, C. (2003). Focusing Attention on Decoding for Children With Poor Reading Skills: Design and Preliminary Tests of the Word Building Intervention. Scientific Studies of Reading, 7, 75-104. Rost, G., & McMurray, B. (2010). Finding the signal by adding noise: The role of non-contrastive phonetic variability in early word learning. Infancy, 15, 608.
