This dissertation describes a process called best-fit constructional analysis and the associated implementation called the Constructional Analyzer. The constructional analyzer takes an utterance as input and performs deep semantic analysis, mapping the utterance onto its most likely interpretation. The best-fit constructional analyzer is unique because it combines Embodied Construction Grammar with the power of best-fit processing. This combination enables the constructional analyzer to be both a cognitive model of interpretation and a practical semantic analysis system.

The best-fit constructional analyzer performs incremental unification grammar parsing, using a factored probabilistic model over syntax and semantics to guide interpretation. The constructional analyzer has been applied to a range of applications:
a) It is a tool for building and testing construction grammars.
b) It is a psychologically plausible model of human interpretation that makes predictions about reading time that match experimental evidence.
c) It is a practical system for semantic analysis that has been tested on a corpus of Mandarin child-parent dialogues.




Download Full History