Tree to tree transformations for mapping syntax to semantics

This is in the context of applying Cognitive AI to natural language processing.

I’ve found an interesting AAAI paper from 1991 by Alexander Hauptmann (CMU) “From syntax to meaning in natural language processing”, see [1].  He notes the challenges for mapping syntax to semantics, and talks about a constrained tree to tree transformation language that starts with the syntactic parse tree. He is using LISP and Frames, but this isn’t a million miles from the use of chunks.

I already have a set of parse trees for the target dialogue, see the parsing demo in [2], along with the chunks used in the state transition rules in the current demo, see [3][4][5]. The next step will be to explore how to model transformation rules that operate on a tree of chunks. I hope to integrate this with an extended version of the parsing demo.  Open questions include the cognitive plausibility of transformation rules, and how they could be learned through experience.

Note that state transitions are just a starting point for describing behaviour (in this case ordering dinner at a restaurant), and further work is envisaged that uses causal models for greater flexibility. Other work will extend the parser to classify verb phrases in terms of tense, mood, perfect and continuous flags, etc.  Essentially, you first need to learn to crawl, and then to walk before you learn to run and to ride a bicycle!

Your suggestions/comments are welcomed!

[1] https://www.aaai.org/Papers/AAAI/1991/AAAI91-020.pdf <https://www.aaai.org/Papers/AAAI/1991/AAAI91-020.pdf>
[2] https://www.w3.org/Data/demos/chunks/nlp/parsing/ <https://www.w3.org/Data/demos/chunks/nlp/parsing/>
[3] https://www.w3.org/Data/demos/chunks/nlp/dinner/ <https://www.w3.org/Data/demos/chunks/nlp/dinner/>
[4] https://www.w3.org/Data/demos/chunks/nlp/dinner/crules.chk <https://www.w3.org/Data/demos/chunks/nlp/dinner/crules.chk>
[5] https://www.w3.org/Data/demos/chunks/nlp/dinner/wrules.chk <https://www.w3.org/Data/demos/chunks/nlp/dinner/wrules.chk>
Best regards,

Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
W3C Data Activity Lead & W3C champion for the Web of things