The system targets a skill that decades of second-language acquisition research have identified as foundational: morphological awareness, the ability to slice unfamiliar words into their meaningful building blocks. A learner who recognizes that “international” decomposes into the prefix inter-, the root nat, and the suffixes -ion and -al can decode whole families of words instead of memorizing them one by one. Large language models, the authors note, are remarkably good at performing this decomposition in real time, yet the conversational interfaces through which we access them leave the resulting knowledge trapped in unstructured chat logs, fragmented and impossible to consolidate systematically.

Morphowiki’s architecture is deliberately simple and rests on three layers. A generation layer orchestrates seven Agent Skills, small callable command modules that direct the language model through specific tasks. A data layer stores everything as plain-text Markdown files organized into directories for words, prefixes, roots, and suffixes. A review layer runs a lightweight local Node.js server with a single-page web application and a JSON datastore that tracks spaced-repetition scheduling. Because the entire knowledge base lives in Markdown and JSON rather than a proprietary database, it can be read by humans, parsed by programs, diffed by version-control systems, and shared freely, in sharp contrast to the closed data silos of commercial study platforms.

The workflow follows an “initialize, ingest, expand, review” cycle. After a one-command setup that builds the directory structure and installs dependencies, the learner feeds in an unfamiliar word or an entire sentence. The system tokenizes the input, filters out stopwords, both a default list of roughly 150 English function words and a personalized exclusion list the learner can extend, and then applies what the authors call an architecturally bounded recursive decomposition. Given “international,” the pipeline isolates the root nation (itself nat plus -ion), expands to national, and resolves the target word with the prefix inter-, creating a separate Markdown page for each layer and wiring every page bidirectionally to its constituent morphemes. A visited-set mechanism prevents the recursion from looping endlessly.

Once a word is archived, the /flashcard command enrolls it into the review queue after a strict three-part check: the word’s Markdown page must exist, the word must not be on the stopword list, and no duplicate entry may exist in the JSON datastore. Scheduling follows the classic SuperMemo-2 algorithm, initialized with an easiness factor of 2.5 and a zero interval. When the learner runs the review command, a local server identifies due cards, parses the relevant Markdown on the fly, and presents a four-button evaluation, Again, Hard, Good, or Easy. The authors condensed the traditional six-grade SM-2 scale with a mathematical justification: all failed retrievals reset the interval and repetition counters without touching the easiness factor, so they can safely be merged into a single “Again” button.

Two further commands push the system beyond passive flashcards. The /new-word command scans the learner’s accumulated morpheme pool and synthesizes derivatives that are not yet recorded, for example proposing “nationalize” once the root nat has been mastered, then automatically running the ingestion and linking pipeline. Because language models can hallucinate plausible-sounding pseudowords, the skill specification embeds a validation step requiring the model to confirm that each candidate is an attested modern English word. A /lint command then audits the knowledge base in two modes: a deterministic, token-free structural check of link symmetry and schema completeness, and a model-assisted semantic inspection of individual entries. Crucially, lint never edits anything on its own; it only raises flags, and every state change requires explicit human authorization.

What elevates the paper beyond a software description is its quantitative evaluation of the decomposition step. The authors assembled 950 test items, including a main set of 600 words stratified by vocabulary level and morpheme count, drawn from the intersection of a Taiwanese senior-high-school word list and the SIGMORPHON 2022 morpheme segmentation shared task. Against the unsupervised baseline BPE, which scored an F1 of 29.13, and Morfessor 2.0, which reached 47.26, Morphowiki paired with Claude Opus 5 achieved an F1 of 83.06 and an exact-match rate of 74.83 percent. On a subset of 200 words containing four or more morphemes, the baselines managed essentially nothing, 0.00 and 3.00 percent exact match, while Morphowiki still resolved 46.50 percent correctly.

The most intellectually interesting result comes from a two-step ablation that separates the contribution of the underlying model from that of the Agent Skills scaffolding. With the model held fixed, removing the skill file and then the repository’s project instructions revealed that the two components do very different work. The project instructions, which describe the project in etymological terms but prescribe no segmentation rules, actually degraded every measure on the main set and nearly tripled over-segmentation of simple words, from 6.7 to 23.3 percent. The skill file, by contrast, added more than 52 points of exact-match accuracy on words recorded in canonical form and corrected 17 of the over-segmentation errors the instructions induced. The authors conclude that the skill does not teach the model how to segment strings, a capability it already possesses, but rather imposes canonical discipline: knowing when to restore a base to its dictionary form, when to record an assimilated prefix canonically, and where recursion should stop.

The team is candid about the system’s limits. Decomposition quality remains tethered to the reliability of the underlying model, and the deployed configuration still over-segments monomorphemic words at 12.0 percent, a trade-off traced to the etymological priming of the project instructions rather than the skill itself. The reference standard used for scoring is itself noisy, with coverage of the most elementary vocabulary at only 43.1 percent, so the reported accuracies likely underestimate true performance. Cross-model comparison covered only two models from a single provider, the pipeline is calibrated exclusively for English derivational morphology, and cost is nontrivial: roughly 9 to 23 US dollars per thousand words processed, which positions the framework for curated vocabulary construction rather than bulk corpus processing. Pedagogical efficacy, finally, has not yet been tested in classrooms.

Even with those caveats, Morphowiki speaks to a tension now rippling through education research. Recent studies on cognitive offloading and “metacognitive laziness” suggest that instantaneous AI answers can undermine the deep semantic engagement that durable memory requires, a concern rooted in the classic levels-of-processing work of Craik and Lockhart. By forcing learners through active retrieval, spaced review, and human-approved curation, Morphowiki is designed to make the AI a builder of infrastructure rather than a dispenser of answers. The authors frame it as a reproducible testbed on which researchers in computer-assisted language learning can systematically compare models, prompting strategies, and scheduling algorithms on a standardized, transparent data format. If the broader lesson holds, the future of AI-assisted learning may depend less on how cleverly a chatbot explains a word, and more on whether the explanation leaves behind something the learner can keep.

Subject of Research: An open-source Agent Skills framework that uses large language models to decompose English vocabulary into morphemes, build bidirectionally linked knowledge graphs, and schedule spaced-repetition review.

Article Title: Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction

Article References: Chen, Y.-T., Yao, K.-C., Chang, S.-H., & Lin, Y.-C. (2026). Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction. SoftwareX, 36, Article 103114. https://doi.org/10.1016/j.softx.2026.103114

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103114

Keywords: Morphowiki, large language models, morphology, vocabulary acquisition, spaced repetition, Agent Skills, computer-assisted language learning, open-source software, second language acquisition, cognitive offloading, Markdown knowledge graph, SM-2 algorithm

The finding challenges a practice so entrenched in mathematics education that most students and teachers never question it. Open virtually any calculus textbook and you will find chapters organized so that every derivative rule, every integration technique, and every limit-evaluation procedure is practiced in a dedicated block of near-identical exercises. This blocked arrangement feels efficient. Students appear to master a technique quickly, teachers can confirm comprehension at a glance, and homework sessions proceed with a satisfying sense of momentum. But decades of cognitive science have argued that this fluency is largely an illusion, a phenomenon researchers call the illusion of competence: because students know in advance which strategy each problem requires, they never practice the most difficult and most important step—deciding which strategy to use.

Interleaved practice removes that crutch. When problems drawn from different topics appear in mixed order, students must first diagnose the problem—recognizing, for example, whether a given integral calls for substitution, integration by parts, or a trigonometric identity—before they can execute the solution. This diagnostic step, sometimes described as discriminative contrast, forces learners to compare and contrast problem categories rather than repeatedly applying a single memorized template. Laboratory studies dating back to the mid-twentieth century, and more recent classroom experiments in algebra and geometry, have consistently shown that this added difficulty during practice produces substantially better retention and transfer, a counterintuitive pattern known as a desirable difficulty.

What makes the new study consequential is its setting. Calculus is not a laboratory task but a high-stakes, credit-bearing college course that serves as a gateway to degrees in engineering, the physical sciences, economics, and medicine. It is also a course with a well-documented attrition problem: students who arrive with weaker preparation are disproportionately likely to fail or withdraw, and those failures ripple outward, discouraging students from pursuing scientific careers altogether. Demonstrating that a low-cost, curriculum-neutral adjustment to homework design can improve outcomes in this environment matters far beyond the psychology of memory. It suggests that part of the achievement gap in STEM may be an artifact of instructional convention rather than an inevitability of prior preparation.

The study’s headline result is that mixing problems raised performance across the entire distribution of student ability. Higher-achieving students, who might have been expected to gain the least from a change in practice structure, still benefited from the interleaved format, consistent with the broad laboratory literature on spaced and varied retrieval. But the effect was not uniform. Students at the lower end of the achievement spectrum showed the largest improvements, a pattern with significant implications for equity in mathematics education. Interventions that lift the whole class while disproportionately lifting struggling students are rare in educational research, where the most common outcome is that advantage compounds: students who start ahead pull further ahead.

Why would low achievers gain the most? The authors’ explanation, grounded in established learning theory, centers on what blocked practice conceals. Under blocked conditions, struggling students can complete an entire assignment by mechanically repeating the worked example from the top of the page, without ever engaging in genuine problem-solving. The feedback signal is delayed until the examination, when the support of topical grouping disappears and the weakness is exposed too late. Interleaving converts that hidden failure into immediate, low-stakes feedback: students discover early which distinctions they cannot yet make, and instructors can see and address misconceptions while there is still time to correct them. In effect, mixed practice functions as a continuous diagnostic instrument woven into ordinary homework.

The mechanics of the improvement are worth spelling out, because they illuminate why the effect appears in calculus specifically. Calculus is a subject of bewildering surface variety concealing a relatively small set of underlying procedures. Two problems that look nothing alike—a related-rates word problem and an implicit differentiation exercise—may rely on the identical chain-rule computation, while two problems that look nearly identical may demand entirely different tools. Blocked practice teaches students to classify by surface features, which fails the moment an exam mixes contexts. Interleaved practice compels classification by mathematical structure, which is precisely the skill that expert mathematicians deploy automatically. The mixed format thus trains the categorization process itself, not merely the execution of procedures within a category.

The practical barriers to adoption are modest, which adds to the study’s policy relevance. Interleaving does not require new technology, smaller classes, additional instructional hours, or retraining in novel pedagogy. It requires reordering existing problem sets so that review of earlier material is distributed throughout the course rather than concentrated in a single pre-exam scramble—a change that also delivers the well-established benefits of spaced repetition as a side effect. Textbook publishers and online homework platforms could implement the restructuring at scale, and instructors can begin immediately by pulling a handful of problems from prior weeks into each week’s assignment. The chief obstacle, the literature suggests, is perceptual: interleaved practice feels harder and slower to students, and performance during practice sessions often looks worse, which can dissuade teachers who rely on short-term performance as evidence of learning.

That perceptual hurdle is also why studies conducted in real courses, with real grades and real students, carry more weight than laboratory demonstrations. Laboratory experiments on interleaving typically use artificial materials and short retention intervals, and skeptics have reasonably asked whether the effects survive contact with the messy realities of motivation, attendance, and competing coursework. By showing the effect in an authentic college calculus setting—and by showing that it operates most powerfully for the students whom standard instruction serves least well—the new research strengthens the case that desirable difficulties are not merely a laboratory curiosity but a practical lever for improving learning in the courses where the stakes are highest.

The broader message for students, teachers, and curriculum designers is a lesson in intellectual humility about what learning feels like. Performance during study is a poor proxy for durable knowledge, and the teaching practices that feel smoothest often produce the shallowest results. Mixing problem types makes practice harder, slower, and less comfortable—and that discomfort is the signature of the brain doing the comparative, structural work that long-term mathematical competence requires. If the findings generalize across institutions and course levels, as the underlying cognitive theory predicts, then one of the cheapest reforms available to mathematics education may also be one of the most equitable: stop telling students which tool to use before asking them to solve the problem, and let the mixed problem set do the teaching.

Subject of Research: The effect of interleaved versus blocked practice problems on student performance in college calculus

Article Title: Mixing problems increases performance of all students but especially of low-achieving ones in college calculus

Article References: Bennoun, S., Yan, V. X., & Xu, A. (2026). Mixing problems increases performance of all students but especially of low-achieving ones in college calculus. npj Science of Learning. https://doi.org/10.1038/s41539-026-00450-6

Image Credits: AI Generated

DOI: 10.1038/s41539-026-00450-6

Keywords: interleaved practice, blocked practice, college calculus, desirable difficulties, mathematics education, STEM equity, low-achieving students, retrieval practice, spaced repetition, cognitive science of learning, problem-solving, achievement gap