NLP: Hybrid-Architecture Symbolic Parser and Neural Lexicon (HASPNeL)

Research Question

While the syntactic structures of internal language are hierarchical, their externalizations are linear due to sensorimotor constraints. This loss of information can lead to ambiguity, which is inefficient for communication. How can an AI system model human grammatical competence to identify all lexical and structural ambiguities within a given sentence?

Abstract

HASPNeL is a biologically-feasible computational model of human language cognition that comprises a Minimalist parser-lexicon system combining NN learning capabilities and symbolic AI procedures and representations that evaluates grammaticality and recognizes structural ambiguity in any natural language. This parser will be useful in industry, education, and as a basis for further development of technologies in human language modeling, inference and understanding

Objectives

1. Execute categorical grammaticality judgements and detect lexical and structural ambiguity by producing all possible syntactic representations for a given utterance.

2. Learn any natural language lexical and externalization parameter settings from small training corpora.

3. Model human language cognition and test current syntactic theory by implementing a grammar formalism satisfying conditions of learnability, evolvability, and universality.

Presentation Video

Research Poster