Skip to main content

LexGraph — RQ1 replication

Type: Independent academic research / reproducibility study.

Research question​

Can vocabulary be assigned a reliable CEFR level while taking the contextual sense of each occurrence into account?

This research starts with a replication of the sense-aware lexical sophistication method reported by Hu, Lu and Hu (2025), rather than claiming an already deployed adaptive language-learning product.

Method and scope​

The private research repository organizes the reproduction into three tracks: verification of released upstream code, a modern Python reproduction, and the paper's dataset/analysis protocol. The comparison is implemented with explicit decision boundaries and repeatable checks.

Current evidence​

The work reached strong core-method replication, but coefficient-level differences, under-specified settings and some stepwise statistical details remain unresolved. It is not claimed as exact paper-level statistical replication.

Further application of the method to other media or a personalized tutor is a separate research phase and should not be reported as already completed.

Availability and research integrity​

The implementation repository is private. Restricted datasets, pretrained resources and other authors' code are not redistributed here. Public technical details should be limited to methods and results for which sharing is permitted.