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

Abstract

Large language models (LLMs) offer transformative capabilities for automated morphological analysis in language learning, yet their generated outputs typically remain fragmented within ephemeral chat interfaces, hindering systematic memory consolidation. To address this, we introduce Morphowiki, an open-source framework built upon a dedicated Agent Skills architecture to construct a persistent, morpheme-centric English vocabulary knowledge base. Driven by a streamlined suite of seven commands, the framework synergistically integrates LLM-powered recursive morphological decomposition, a bidirectionally linked Markdown relational graph, and a JSON-based SuperMemo-2 (SM-2) spaced-repetition algorithm. By instantiating all lexical assets and review states within a localized plain-text data layer, Morphowiki ensures explicitly traceable, version-controllable data orchestration. Furthermore, it embeds non-destructive structural and semantic verification mechanisms to actively mitigate LLM hallucinations. Ultimately, Morphowiki bridges dynamic morphological knowledge compounding with structured memory retrieval, providing researchers in computer-assisted language learning (CALL) with a highly portable, reproducible testbed to empirically evaluate diverse generative models and pedagogical strategies.