Parsing Model Evolution
Transitioning to sequence-to-sequence models simplifies the development process by eliminating the need for explicit alignment, relying instead on an attention mechanism. However, this shift significantly expands the search space, making it more complex to generate programs. While the assumption is that powerful models like LSTMs can handle this complexity with basic search techniques, there are concerns about the limitations of this approach as datasets evolve.In this clip
From this podcast
NLP Highlights
91 - (Executable) Semantic Parsing, with Jonathan Berant
Related Questions