Currently, feature unification is handled by the PET application or the LKB. Running these requires that the user pull in the entire HPSG framework. It would be nice to instead have a feature unification API. This would enable easier experimentation by providing a fast and reliable unification framework that works independently of HPSG.

This has not been implemented, but there are parts of the cheap parser in PET that likely get us part of the way there.

The cheap source tree can be used to build the following libraries:

• libcheap.a: core feature unification, type hierarchy
• libmrs.a: MRS handling

Inside libcheap.a the unification algorithm is in dag-tomabechi.h: see in particular the "dag_node" and "dag_arc" structures.

"dag_node" includes an index into a type hierarchy structure. This structure is built by flop and loaded at grammar-load time.

YiZhang provided the information about the Cheap libraries.