PIA for LISPers

The PIA is very LISP-like internally:

The main differences are:

The default bindings simply copy elements and text from the input document to the output. The effect is very much like the backquote operator in LISP.
All function bindings are explicit. Unlike LISP, in which the standard functions are precompiled and always present, in the PIA the bindings (tagset) for processing a document is always specified, and more than one set of tag and entity bindings can be in effect. This makes it easier to specify completely different sets of bindings for different purposes (for example, pretty-printing).
In LISP, processing takes place at two times: in READ (when data structures are built and read macros interpreted) and in EVAL. The PIA has an Input (corresponding roughly to READ), a Processor (corresponding to EVAL), and an Output (similar to PRINT) -- processing is more like structure copying than evaluation, and substantial amounts of processing can be done in the Output.
Rather than simply expanding a tree, the Processor copies from a tree traverser (the Input) to a tree constructor (the Output). This lets it operate on documents that are too large to fit in memory.
The Processor can also be made to look like an Output, so pipelines can be constructed. (This is mainly used in tool chains like SAX in which the parser expects to drive an application by calling a function on every node.)