Cognitive Factors

From LearnLab
Revision as of 11:32, 24 October 2008 by Koedinger (Talk | contribs)

Jump to: navigation, search

The research in this thrust is aimed at understanding cognitive learning—changes in knowledge—that result from instructional events. It builds on work in the learning sciences field at large and on research carried out in the PSLC over its first four years within the Refinement and Fluency cluster and part of the Coordinative Learning cluster, thereby merging two themes that organized the first phase of the PSLC. Each of these clusters was concerned with identifying instructional events that produce robust learning. They differed mainly in that the relevant theme within the Coordinative Learning cluster had a specific focus on instructional events that included more than one input. (A second theme within the Coordinative Learning cluster was on instructional events that provoke learning events involving more than one reasoning method and this theme will be continued in the Metacognition and Motivation thrust). In the fifth year of the PSLC, we carry forward research from each of these clusters, while making a transition to an additional set of research questions. Although we frame this section in terms of the new Cognitive Factors thrust, the research carried out during the 5th year has been initiated in the current Refinement and Fluency and in part of the Coordinative Learning clusters.

Our work on cognitive factors encompasses a triangulated set of events around learning: learning events, instructional events, and assessment events. Anything from a lesson to an entire curriculum can be considered a sequence of events whose durations vary from seconds to semesters. The hypotheses of the Cognitive Factors Thrust concern how instructional procedures (e.g., decisions about the learner’s task, materials, practice, feedback) affect learning events and thus the outcomes of learning. Learning involves the acquisition of knowledge components, an increase in the feature validity and the strength of these components, and the integration of these components through practice. Our basic hypotheses include the following:

  • Explicitness: Instruction that draws the learner’s attention to valid features that support the relevant knowledge components leads to more robust learning than instruction that does not.
  • Assistance: The degree of assistance in the instruction affects learning in relation to student knowledge on specific knowledge components.
  • Practice: Practice schedules can be optimized using models of learning based on memory activation assumptions.
  • Integration: Knowledge components that are integrated during learning and practice lead to more robust learning and fluent performance across different tasks.

The research plan tests these hypotheses across knowledge domains, as exemplified by the following projects:

Language background factors in L2 learning. This work illustrates the synergies that develop in the PSLC’s LearnLab context, in this case between English as a second language (ESL) director Alan Juffs and other PSLC language researchers. In a prior cluster meeting, Juffs presented ESL classroom data that compared various L1 background students in their performance on transcribing their own speech, a standard piece of instruction in the ESL curriculum. The result that caught the interest of PSLC researchers (Dunlap, Guan, Perfetti) was the very poor spelling performance of Arabic-background students, relative to Spanish, Korean, and Chinese ESL students, despite comparable levels of spoken language performance. Furthermore, Juffs identified this discrepancy as a long-standing one in ESL instruction. Although one might hypothesize that a key factor is orthographic differences between L1 and L2, this seems unlikely here. Spanish to English is closer, but Chinese to English is farther in L1-L2 orthographic similarity. The first steps toward a new study have been taken with the help of a PSLC summer intern, who coded the errors made in spelling by all L1 background learners. The pattern of errors can be characterized as qualitatively similar, differing across languages quantitatively, suggesting a generalized English spelling problem. This analysis has led to the hypothesis that feature focusing—attention to full spelling patterns—is different across the L1 backgrounds, which we will test in a training experiment that focuses attention on spelling patterns.

Second language vocabulary learning. Another new project originating within the Refinement and Fluency cluster will study English vocabulary learning using REAP. Based on recent research by Balass on the trade-offs between explicit (dictionary-based) and implicit (inferences from text) instruction in learning new words by monolingual subjects (Bolger et al, 2008), the new work will apply this tradeoff idea to second language learners. The hypothesis is that allowing learners to view definitions is more effective after they have read a sentence containing the word to be learned. This hypothesis reflects ideas about assistance (giving a definition versus inferring it) and the assumption that learning word meanings from context depends on the overlapping memory traces established by specific encounters with the word (Bolger et al, 2008). REAP allows us to use authentic texts for studies with students of various L1 backgrounds learning English through reading texts in their areas of interest. In our experiments, we will vary the availability of definitions provided on-line as part of the text reading.

Explicit instruction and practice schedules in algebra and second language learning. Foreign language learning in classrooms has stimulated research on explicit vs implicit instruction, with conclusions favoring the value of explicit instruction (Norris and Ortega, 2000). A major conclusion from PSLC work is that instruction that draws attention to critical valid features—“feature focusing”—is important in acquiring knowledge components for complex tasks. This conclusion has evidence from studies of L2 learning of the English grammar by Levin, Friskoff, Pavlik, studies of radical learning by Dunlap et al and by Pavlik, and by studies by Zhang and MacWhinney and by Liu et al on learning spoken syllables through pin-yin (alphabetic spellings). Projects in French dictation (MacWhinney) and French grammar (Presson & MacWhinney), Chinese dictation (Zhang & MacWhinney), algebra (Pavlik) and arithmetical computation (Fiez) also reflect this theme. Much of this work has been combined with completely general hypotheses about practice, based on Pavlik and Anderson (2005)’s model that describes the trade-off between the benefit of spaced practice and the cost of longer retention intervals brought by spacing. The resulting optimized practice schedule has been tested in several PSLC studies of vocabulary learning in Chinese (Pavlik, MacWhinney, Koedinger; reported in Pavlik, 2006), cues to French gender (Presson, MacWhinney, & Pavlik). Important is the generality of the optimization model. It applies to all domain content and studies in both algebra and second language learning have nee carried out. The new work in second language and in algebra builds on the synergies that have emerged from collaborations between domain researchers (e.g. MacWhinney) and Pavlik around experiments and models for optimizing practice. For Chinese, MacWhinney, Zhang, and Pavlik have developed a tutor for Chinese dictation and vocabulary learning that is being used in 18 sites. Data from these sites will be used to test the results of practice schedules and the form of instructional events (e.g. cues to gender in French) with longer term measures of robust learning. Because each of the tutors logs results to DataShop, the student records are a rich source of data for further study, including researchers beyond the PSLC.

Learning the logic of unconfounded experiments. We will extend our research on college level science topics (chemistry and physics) to middle school science, with a focus on the cross-domain topic of experimental design. The ability to design unconfounded experiments and make valid inferences from their outcomes is an essential skill in scientific reasoning. The key idea here is CVS: the Control of Variables Strategy. CVS is the fundamental idea underlying the design of unconfounded experiments from which valid, causal, inferences can be made. Its acquisition is an important step in the development of scientific reasoning skills , because it provides a strong constraint on search in the space of experiments (Klahr, 2000). The Tutor for Experimental Design (TED), developed by Klahr’s research team, builds on previous work studying the different paths of learning and transfer that result from teaching CVS using different instructional methods that span from direct instruction to discovery (Chen & Klahr, 1999) and show differences along the “physical-virtual” dimension (Triona & Klahr, 2007). We build on this by constructing of a semi-autonomous tutor, then developing a full computer based tutor in Pittsburgh middle school LearnLabs and carrying out in vivo experiments with TED.

Integration of knowledge components. Isolated knowledge components are not sufficient to produce fluent use of knowledge. Integrating knowledge components is important both in authentic practice that follows acquisition of knowledge components but, we hypothesize, also in the initial acquisition of components. Some of our prior work in coordinative learning establishes some of the conditions that favor multiple inputs during learning (e.g., Davenport et al in stochiometry). And experiments on fluency support the value of repeated practice in single-topic speaking as way to support fluency (de Jong, Halderman and Perfetti). In new work we propose to build on progress we have made in the study of fluency in language (de Jong et al) and arithmetic (Fiez). For example, we will follow the discovery by de Jong and colleagues that when L2 speakers repeat a speech on a single topic, their fluency scores increase on a number of measures. We will test the hypothesis that this results from the advantage of retrieving the same conceptual and lexical knowledge and overall speech plan on successive attempts, allowing fluency to increase on procedural components supported by chunking of words to phrases. We are accumulating a large database in the English LearnLab that will support the testing of additional hypotheses. The idea that some relatively simple learning (e.g. 3-5 knowledge components) is supported by integration from the beginning is being tested by Liu, Guan & Perfetti in a study of learning to read Chinese characters. The hypothesis is that when students write unfamiliar characters within the same 60-second time period that they read the character and try to learn its meaning and pronunciation, they will show more robust learning measured by reading tasks. Underlying this hypothesis is the idea that the representation of a character (or other objects that follow structural principles) can be perceptual-motor as well as visual.