Researchers Automate Schema Language Conversion Across Data Systems

Modern applications often need the same data model represented across multiple schema languages—JSON Schema for web APIs, XSD for data interchange, SHACL for knowledge graphs. Keeping these versions synchronized as models evolve is a persistent engineering challenge, especially since converters between formats are scattered, inconsistent in quality, and often lose information.

Researchers have tackled this problem by treating schema languages as a network and converters as pathways that can be automatically discovered and combined. Their approach, published as an arXiv paper, models schema conversion as a graph problem: when no direct converter exists, the system finds intermediate conversion paths, attempts them in ranked order, and reports outcomes with full provenance.

The Schema Conversion Orchestrator#

The team built the Schema Conversion Orchestrator as an open-source tool integrated into MetaConfigurator. Rather than relying on a single "perfect" converter, it orchestrates imperfect, heterogeneous black-box converters into reliable automated workflows.

Testing on 60 real-world conversion tasks across five schema languages, the system successfully produced usable results for 43 tasks. The researchers used agent-assisted, human-reviewed quality annotations to measure conversion fidelity. Failures in the remaining 17 cases revealed concrete gaps—specific missing converters or unsupported language features that limit the current ecosystem.

Implications#

The work illustrates both the potential and limits of orchestrating existing tools. By exposing where the converter landscape falls short, it offers tool builders insight into which new converters would have the highest impact. It also frames a practical question: how should conversion quality be measured when multiple imperfect paths exist?