This seems very interesting in that the diff engine has some linguistic awareness.
Another candidate for the tool tag.
I filed an issue on the repo. To repeat here for others to consider: this approach would be very valuable for diffing (comparing) semantic structures in different formats. Consider an extreme example: an RDF/XML file (file.rdf) and an NT/triples file (file.nt). Do they mean the same thing (two completely different syntaxes)? If they are similar (e.g. share different types) but have different properties, what is the difference between the two property sets?