The L2 Network: A CEFR-Aligned Knowledge Graph for Grammar Domain Modeling
Proceedings of the Workshop on Structured Linguistic Data and Evaluation (SLiDE)
Abstract
Large language models have renewed interest in the role of structured linguistic data for applications that require controllable, interpretable, and pedagogically aligned language generation. This need is especially visible in intelligent language tutoring, where grammar cannot be modeled as a flat inventory of patterns alone, but must also capture their relations and functions they realize. We present the L2 Network, a machine-readable knowledge graph of CEFR A1-A2 English grammar that encodes formal patterns, functions, and typed relations between them. The resource is grounded in established pedagogical reference materials, combining form inventory and progression information from the English Grammar Profile with a functional layer derived from CEFR descriptors. We further report content validation of the form-function mappings through expert annotation, including agreement analysis and a consensus-filtered core release. The resulting graph provides an explicit schema for representing pedagogically relevant grammatical knowledge and supports downstream uses such as learner modeling, adaptive task selection, and controlled generation in dialogue-based ICALL systems.