Neural Nets for NLP 2018 - Neural Semantic Parsing

Explore neural semantic parsing in this comprehensive lecture from Carnegie Mellon University's Neural Networks for Natural Language Processing course. Delve into tree structures of syntax, semantic representations, and various meaning representations. Examine special-purpose representations, query tasks, command and control tasks, and code generation tasks. Learn about tree-based parsing models, code generation with abstract syntax trees, and the challenges of weakly supervised learning. Investigate first-order logic, Abstract Meaning Representation, and other formalisms. Discover techniques for parsing graph structures, CCG parsing, and neural module networks. Gain insights into neural models for semantic role labeling and the use of deep highway LSTM taggers.

Tree Structures of Syntax
Representations of Semantics
Meaning Representations
Example Special-purpose Representations
Example Query Tasks
Example Command and Control Tasks
Example Code Generation Tasks
A Better Attempt: Tree-based Parsing Models • Generate from top-down using hierarchical sequence- to-sequence model (Dong and Lapata 2016)
Code Generation: Handling Syntax • Code also has syntax, e.g. in form of Abstract Syntax Trees
Problem w/ Weakly Supervised Learning: Spurious Logical Forms . Sometimes you can get the right answer without actually doing the generalizable thing (Guu et al. 2017)
Meaning Representation Desiderata (Jurafsky and Martin 17.1)
First-order Logic
Abstract Meaning Representation (Banarescu et al. 2013)
Other Formalisms
Parsing to Graph Structures
Linearization for Graph Structures (Konstas et al. 2017)
CCG and CCG Parsing
Neural Module Networks: Soft Syntax-driven Semantics (Andreas et al. 2016) . Standard syntax semantic interfaces use symbolic representations . It is also possible to use syntax to guide structure of neural networks to learn semantics
Neural Models for Semantic Role Labeling . Simple model w/ deep highway LSTM tagger works well (Le et al. 2017)