There's no computable way to detect 'similar' descriptions in any fashion that would be even vaguely reliable.  For HL7-defined extensions, once the FMM rules come into play, there's at least an expectation that the work group(s) that manage the extended resource(s)/data type(s) must review and approve the extension.  That should capture many (though not all) duplicate/overlapping extensions in the HL7 space.  Doing so outside of the HL7 space is nigh impossible. Also, we wouldn't want to only check extensions with the same context, as ideally we'd like to detect extensions with the same purpose but different contexts.  All that said, if you have a specific technical solution that you think is workable, feel free to resubmit with specific details of how the tooling would reliably and usefully detect 'similarity' in a way that would have a decent signal-to-noise ratio.