Parsing Model
- My doctoral thesis was based on human-like parser.
- It was symbolic and did one pass parsing using preferences
to resolve ambiguity.
- The CABot1 parser used a stack. Neural stack management
is a hassle.
- Instead we followed an activation level approach which kind
of simulates a stack.
- This enabled us to parse in psychological time.
- We used semantics to resolve PP attachment ambiguities.
- We used a frame to generate semantic results.
- For the frame we used binding by STP which
is foundational work.
- Words were activated symbolically.
- The currently active words ignited grammar rules which bound
things to frames.
- There was a two-tier grammar corresponding roughly to x-bar theory.
- The semantics of words was represented by overlapping CAs derived
from WordNet.
