Search

Code tracing involves simulating at a high level the steps a computer takes when it executes a computer program. This is a fundamental skill needed for programming activities, but one that novices find challenging. Thus, work is needed on how to support novice programmers in this activity. We conducted an experimental study with university students (N = 56) learning to code trace in two conditions, namely peer tutoring and solo code tracing. Our primary outcome variable was learning. However, since how students feel is also an important factor in educational settings, we also measured student emotion. Contrary to prior work in other domains, there was no significant benefit of peer tutoring and self-reported levels of emotion were similar in the two conditions; Bayesian statistics provided evidence for the null model in the majority of cases.

Computer programs correspond to instructions that tell a computer how to perform a task. Because computers do not understand human languages, programs have to be written in a specialized programming language. Such languages have unique syntax and semantics that on the surface appear quite different from human languages. To foster understanding what a program means, novices need to translate the programming language instructions into a familiar human language. We developed a tutoring system that provides translation activities, from the language Python into plain English, and feedback on student answers. One version of the tutoring system provides open-ended input boxes that students type translations into. A second version instead asks students to generate answers to the translation using drop-down menus. The latter provides more scaffolding but also may encourage shallow behaviors such as guessing and checking. To shed light on how these designs influence learning, we evaluate them with novice programmers. Here, we report on the results of the evaluation and next steps.

This article reviews literature on worked examples in the context of programming activities. We focus on two types of examples, namely, code-tracing and code-generation, because there is sufficient research on these to warrant a review. We synthesize key results according to themes that emerged from the review. This synthesis aims to provide practical guidance for educators and shed light on future research opportunities. While there is established work in some areas (e.g., dynamic code-tracing examples in the form of program visualization tools, utility of subgoals in code-generation examples, and incomplete examples in the form of Parsons puzzles), there are also gaps. Thus, the article concludes with directions for future work on examples in computer science education.

A notional machine is an abstract representation of how a computer works. Prior research shows that students often form incomplete or inaccurate mental models of the notional machine, reflecting misunderstandings and causing programming errors. To date, however, there is very little work on how to help students acquire accurate models of the notional machine. To fill this gap, this research focuses on novice programmers and experimentally investigates the impact of direct instruction of the notional machine compared to standard instruction that does not include the notional machine. The instruction is embedded in an online computer tutor that we implemented and compared to a version of the tutor that does not refer to the notional machine but otherwise includes the same materials. In particular, both tutor versions include instructional videos followed by self-explanation prompts. We will present data from an evaluation of both tutor versions.

A fundamental skill needed to learn to program is code tracing, which involves simulating the steps a computer takes when executing a program to predict how it will behave and what output it. While code tracing helps students understand and write programs, students find this activity challenging. However, to date, little work exists on how to best support novice programmers in this activity. To address this gap, the high-level goal of our research is to investigates the utility of peer-tutoring for learning to trace programs. In our prior work, we conducted a preliminary case study exploring qualitative features of peer tutoring interactions. As the next step, we conducted an experimental study with two conditions: code tracing in a peer tutoring context vs. code-tracing alone. We present results from this study, focusing on quantitative results related to learning outcomes in each condition.

Programming is a complex problem-solving domain, requiring the coordination of different types of knowledge and skills. The present study investigates expert and novice programming problem solving by analyzing talk-aloud transcripts and the code generated. Based on this analysis a set of basic goal and step components used by novice and expert programmers are identified, which will inform on the generation of cognitive models in the next phase of this research.

Productive failure (PF) is a learning paradigm that reverses the standard order of instruction by asking students to solve problems prior to instruction. This paradigm has been shown to be effective for fostering student learning. To date, however, the role of student emotion in productive failure has not been investigated. In other paradigms, there is some evidence that failure elicits negative emotions and that these emotions can interfere with learning. This leads to a conundrum given productive failure’s positive effect on learning. To shed light on this, we report on results from a study (N = 48) in the productive failure paradigm. For the analysis, we used a mixed-methods approach to investigate the distribution of emotions in productive failure, how these changed across different instructional activities, and the relation between emotions and posttest performance.

Learning to program is challenging, because it involves novel skills. In contrast to the majority of work focusing on code generation, we target the skill of code tracing. Code tracing involves simulating the high-level actions a computer takes when it executes the program, including the flow of execution through it and how variable values change as a result. Code tracing supports program comprehension and generation. However, many students do not code trace effectively. Our project investigates the utility of peer tutoring for the learning of code tracing. While this approach remains untested in this context, it has successfully been used in other domains to improve student learning. We will present qualitative data from a case study involving students tutoring each other to code trace. This work is the first step in the context of a broader project focused on identifying ways to help students learn to code.

The goal of the present study was to use content analysis to gain insight into the process of problem-solving of novice and expert programmers. While classic work on programmers identifies goals / plans as key constructs needed to code, there is relatively little work using protocol analysis. We recruited 7 expert and 12 novice rogrammers who completed up to 3 brief programming problems while providing a talk-aloud of their inner problem solving process. Based on analysis of the transcriptions of this talk aloud data, we identified the goals and steps used, as well as the broad differences between experts and novices in their problem solving process. These differences were formalized into python ACT-R models, and model output was compared to programs written by human participants.

A mental model is an internal representation that explains how something works. Mental model construction is facilitated by self-explanation, the active generation of explanations for oneself. The overarching goal of this research is to empirically investigate the utility of self-explanation for developing mental models when learning to program. Programming is notoriously challenging and, despite evidence of the importance of mental models for learning, little work has focused on mental models of students learning how to program. They need correct mental models of the notional machine, an abstraction of the steps taken by a computer as it processes a program. Because students do not spontaneously self-explain, we are using a user-centered approach to design a computer tutor to prompt for self-explanation about the notional machine. Here, we present qualitative results on students’ interactions with an initial version of the tutor, including the form of their self-explanations and corresponding mental models.

When students are first learning to program, they not only have to learn to write programs, but also how to trace them. Code tracing involves stepping through a program step-by-step, which helps to predict the output of the program and identify bugs. Students routinely struggle with this activity, as evidenced by prior work and our own experiences in the classroom. To address this, we designed a "Code Tracing" (CT)-Tutor. We varied the level of assistance provided in the tutor, based on (1) the interface scaffolding available during code tracing, and (2) instructional order, operationalized by when examples were provided, either before or after the corresponding problem was solved. We collected data by having participants use the tutor to solve code tracing problems (N = 97) and analyzed both learning outcomes and process data obtained by extracting features of interest from the log files. We used a multi-layered approach for the analysis, including standard inferential statistics and unsupervised learning to cluster students by their behaviors in the tutor. The results show that the optimal level of assistance for code tracing falls in the middle of the assistance spectrum included in the tutor, but also that there are individual differences in terms of optimal assistance for subgroups of individuals. Based on these results, we outline opportunities for future work around personalizing instruction for code tracing.

It is established that examples are beneficial for learning, but are certain types of examples more helpful than others?Erroneous examples include errors that students are asked to correct, something that can be helpful in addressing mis-conceptions. One domain that is vulnerable to misconceptions is fraction arithmetic. In the present study, undergraduatestudents solved fraction problems using a tutoring system we designed. Some participants worked with the Erroneous-Example tutor, which supplemented problems with erroneous examples, while other participants worked with a traditionalProblem-Solving tutor that did not include erroneous examples. To evaluate the impact of tutor type on learning andself-efficacy, we analyzed difference scores from pre-test to post-test. While overall participants significantly improvedtheir fraction knowledge and self-efficacy, there was no significant difference between the two groups. Bayesian analysesprovided evidence for the null model, i.e., that erroneous examples were not more beneficial than traditional problemsolving.

The intrinsic integration hypothesis proposes that using coregame mechanisms to teach learning material makeseducational games more fun to play and better for learning.Our study tests the intrinsic integration hypothesis with twoeducational versions of Battleship that were designed for thisexperiment, in the domain of complex numbers. We examinethe learning gains and motivation of 58 participants whointeracted with either the intrinsically-integrated orextrinsically-integrated version of the game. Our resultscontradict previous findings supporting the intrinsicintegration hypothesis: participants reported similar levels ofmotivation from both versions of the game and participantswho interacted with the extrinsically-integrated versionlearned significantly more as measured by pretest to posttestgains. This work contributes empirical data to the debateconcerning intrinsic integration, and it highlights the need foradditional studies exploring the integration of learningmaterial into educational games.

While there is substantial evidence showing that assistanceprovided to students during problem-solving activitiesinfluences learning outcomes, it is not yet clear how to bestdesign educational technologies to maximize learning throughvarious types of assistance. One common type of assistancecorresponds to hints delivered by an educational technology.To date, however, there is little research on the impact ofdifferent types of hints, including high-level hints vs. specificbottom-out hints. Our research takes a step in filling this gap,through an experimental study with an intelligent tutoringsystem we implemented in the domain of algebra (N = 50).We did not find evidence that the type of hint, high level vs.bottom out, influenced learning, with both types of hintsproducing similar outcomes. We did, however, find supportfor the conclusion that the number of hints accessed interactedwith the type of hint to influence learning, and specifically,that accessing more hints was correlated with learning butonly in the high-level hint condition.

Whereas first language (L1) research has demonstrated thatperceptions of creative ability are influenced by the complexityand diversity of language used to answer verbal tests ofcreativity, relatively little is known about the linguisticcomponents of bilingual creative task performance. In thisstudy, we analyze written transcripts of speech produced by466 Japanese learners of English produced during a creativenarrative task for features related to linguistic and cognitivedimensions of creativity. Then, we extract various linguisticfeatures and test whether these features can predict humanperceptions of creativity for the transcripts. Unlike L1 data,results suggest text length and L2 proficiency comprise themost parsimonious explanation of creativity scores in this L2data. At the same time, linguistic features related to positivesentiment explained a significant yet small amount ofadditional variance in perceptions of creativity, suggestingtexts with more positive language were perceived to be morecreative.

When students are solving problems they often turn to examples when they need assistance. Examples are helpful because they illustrate how a problem can be solved. However, when examples are very similar to the problems, students default to copying the example solutions, which hinders learning. To address this, prior work has investigated the effect of manipulating problem--example similarity, showing that learning can be increased by reducing the assistance provided by examples. We contribute to this literature by comparing two types of assistance mechanisms in the context of problem-solving activities: (1) "fade-out assistance," where initially the examples are similar to the problems but over time the problem--example similarity is reduced, and (2) "fade-in assistance" where the opposite is the case (initially the problem--example pairs have reduced similarity but the similarity is increased as more problems are solved). The fade-in assistance condition produced significantly higher learning gains than the fade-out condition and based on eye-tracking data, the fade-in group spent longer attending to the problem, particularly early on in the problem-solving session. Our conjecture that the fade-in group was engaged in more autonomous problem solving instead of copying was confirmed by exploratory analysis on a subset of the data showing that copying was initially reduced in the fade-in condition, as compared to high in the fade-out condition. Overall, our results highlight that initially struggling in a problem-solving activity results in more learning.

Learning to program requires perseverance, practice, and the mindset that programming skills are improved through these activities (i.e., that everyone has the potential to become good at programming). In contrast to an entity mindset, individuals with an incremental mindset believe that ability is malleable and can be improved with effort. Prior research shows that an incremental mindset can be promoted through interventions and that, as a result, individuals report increased belief in the value of effort. Although this is encouraging, the majority of research targets a general mindset, and so little work exists exploring the effect of this construct in the programming domain. The present study (N = 47) used a programming activity to test the effect of an incremental mindset intervention on participants' beliefs, effort, programming behaviors, and performance in an experimental study. The intervention was successful. Compared to the control group, the experimental group shifted significantly more toward an incremental mindset, which resulted in beneficial behaviors related to effort, namely higher time on task and more program creation and modification actions. These positive behaviors, however, did not translate to improvements in programming performance. We speculate the reason for this latter finding may be related to the need for additional domain-based support.

We investigated the impact of assistance on learning and affect during problem-solving activities with a computer tutor we built using the Cognitive Tutor Authoring Tools framework. The tutor delivered its primary form of assistance in the form of worked-out examples. We manipulated the level of assistance the examples in the tutor provided, by having similar problem-example pairs in one version of the tutor (high-assistance condition) and reduced similarity problem-example pairs in the other version (reduced-assistance condition). The reduced-assistance condition resulted in significantly higher learning, without increasing negative affect like frustration.

The question of what type of attention is associated with creative potential has predominantly been examined using divergent thinking tasks and without the benefits of cognitive neuroscience methods. To address this gap, we rely on a neural approach by indexing attention using the P50 ERP component. The P50 measures sensory gating, which is the brain's ability to filter out or inhibit responses to repeated stimuli. We analyze the relationships between neural responses to P50 and performance on two tasks designed to measure creative potential: a convergent task and a divergent task. While we did not find evidence of a relationship between attention and the originality component of divergent thinking, our results show that leaky attention was positively associated with the fluency component of divergent thinking but negatively associated with performance on the convergent thinking task. This result highlights that the relationship between attention and performance on creative tasks may depend on the type of task.

When students solve problems with access to examples show-ing worked out solutions, they often resort to shallow methodslike copying that do not result in learning. An open questionis therefore how to encourage deeper processing in this type ofinstructional context. To address this question, in the presentstudy, we investigate the impact of manipulating problem-example similarity over the course of a problem-solving ses-sion in several ways, including faded assistance (high to lowsimilarity), reverse faded assistance (low to high similarity),and a control group with high, constant assistance. We foundthat the reverse faded assistance condition resulted in the great-est learning gains. We analyzed the gaze behaviours to shedlight on this finding and found that participants in this condi-tion focused significantly more on the problem solution, sug-gesting more cognitive processing during problem solving thanin the other conditions.

The Iowa Gambling Task (IGT) is an established toolused for evaluating the role of emotional learning underconditions of uncertainty. To date, however, themajority of studies have not explicitly manipulated theemotional content within the IGT or examined the effectof doing so on different populations. We address thisgap in the present study, focusing our analysis on twogroups: low vs. high psychopathy individuals insubclinical populations. Our findings demonstrate thatemotional content boosted learning for the high but notthe low psychopathy group.

People report solving division problems by mentally recastingdivision problems as multiplication (e.g., 72 ÷ 8 à 8 × [?] =72). Mediation of division by multiplication occurs mainly onlarger problems. Eye tracking data was used to determinewhether patterns of gaze durations on division problemsprovided support for mediation. Adults solved divisionproblems in two formats: traditional (e.g., 72 ÷ 8 = [ ]) andrecasted (e.g. 72 = 8 × [ ]). Processing of individual problemelements was compared across formats. Results providesupport for mediation. Processing patterns for traditionally-formatted problems were more similar to those for traditionaldivision in earlier work (72 ÷ 8) whereas problems in recastedformat (72 = 8 × [ ]) were more similar to patterns foundwhen participants solved multiplication problems (e.g., 8 × 9).These findings provide a novel source of support fordifferential processing of problems across presentationformats.

As computational thinking becomes increasingly important for children to learn, we must develop interfaces that leverage the ways that young children learn to provide opportunities for them to develop these skills. Active Learning Environments with Robotic Tangibles (ALERT) and Robopad, an analogous on-screen virtual spatial programming environment for educational Human Robot Interaction (HRI), have been developed. Evaluations of these in the context of free play and open-ended learning activities show that both systems afford opportunities for young children to engage in spatial programming, creating improvisational and sequential programs that mediate interactions between the environment, robots, and humans in responsive and creative ways. These systems demonstrate innovative opportunities for advancing mixed reality spatial programming activities as a form of HRI that fosters engaging seamless cyberlearning experiences, across formal and informal environments.

Creativity is commonly assessed using divergent thinking tasks, which measure the fluency, flexibility, originality, and elaboration of participant output on a variety of different tasks. This study assesses the degree to which creativity can be identified based on linguistic features of participants' language while completing collaborative divergent thinking tasks. To this end, 78 participants' conversational dialogs (i.e., 39 dyads) within a chat environment were collected while completing three open-ended problem-solving tasks. Expert raters scored the dialogs in terms of fluency, flexibility, elaboration, and originality, as well as three types of creative language (metaphor and simile, humor, and word play). Factor analyses indicated that these scores captured two main constructs (creativity and elaboration). The linguistic features of the participants' language (captured computationally using natural language processing tools) accounted for significant amounts of variation in both the creativity (R[superscript 2] = 0.640) and elaboration (R[superscript 2] = 0.550) scores within linear mixed effect (LME) models. These results highlight specific linguistic features that can be used to explain large amounts of variance in constructs related to creativity.

Idea generation is an important component of most major theories of writing. However, few studies have linked idea generation in writing samples to assessments of writing quality or examined links between linguistic features in a text and idea generation. This study uses human ratings of idea generation, such as "idea fluency, idea flexibility, idea originality", and "idea elaboration", to analyze the extent to which idea generation relates to human judgments of essay quality in a corpus of college student essays. In conjunction with this analysis, linguistic features extracted from the essays are used to develop a predictive model of idea generation to further understand relations between the language features in an essay and the idea generation scores assigned to that essay. The results indicate that essays rated as containing a greater number of ideas that were flexible, original, and elaborated were judged to be of higher quality. Two of these features (elaboration and originality) were significant predictors of essay quality scores in a regression analysis that explained 33% of the variance in human scores. The results also indicate that idea generation is strongly linked to language features in essays. Specifically, the use of unique multiword units, more difficult words, semantic but not lexical similarities between paragraphs, and fewer word repetitions explained 80% of the variance in human scores of idea generation. These results have implications for writing theories and writing practice.

This article describes research results based on multiple years of experimentation and real-world experience with an adaptive tutoring system named Wayang Outpost. The system represents a novel adaptive learning technology that has shown successful outcomes with thousands of students, and provided teachers with valuable information about students' mathematics performance. We define progress in three areas: improved student "cognition," "engagement," and "affect," and we attribute this improvement to specific components and interventions that are inherently "affective," "cognitive," and "metacognitive" in nature. For instance, improved student "cognitive outcomes" have been measured with pre-post tests and state standardized tests, and achieved due to personalization of content and math fluency training. Improved student "engagement" was achieved by supporting students' metacognition and motivation via affective learning companions and progress reports, measured via records of student gaming of the system. Student "affect" within the tutor was measured through sensors and student self-reports, and supported through affective learning companions and progress reports. Collectively, these studies elucidate a suite of effective strategies to support advanced personalized learning via an intelligent adaptive tutor that can be tailored to the individual needs, emotions, cognitive states, and metacognitive skills of learners.

A promising instructional approach corresponds to" learning by observing others learn" (i.e., by watching tutorial dialogue between a tutor and tutee). However, more work is needed to understand this approach's pedagogical utility. Thus, in 2 experiments we compared student learning from collaborative observation of dialogue with 2 other instructional contexts: 1-on-1 human tutoring and collaborative observation of monologue. In Study 1 (N = 50), there was no significant difference in learning outcomes between the dialogue and tutoring conditions, while the dialogue condition was superior to the monologue condition. Study 2 (N = 40), which involved a younger population than in Study 1, did not replicate these results, in that students learned less from observing dialogue than from being tutored, and there was no significant difference between the dialogue and monologue conditions. To shed light on our results, we analyzed the verbal data collected during the 2 experiments. This analysis showed that in Study 1, the dialogue observers generated more substantive contributions than did the monologue observers. In contrast, in Study 2 there was no significant difference between the observers in terms of substantive contributions; moreover, the total number of contributions was modest, which may have hindered observer learning in that study. In general, our findings suggest that collaboratively observing tutorial dialogue is a promising learning paradigm, but more work is needed to understand how to support young students to effectively learn in this paradigm.