Constrained Language Models Yield Few-Shot Semantic Parsers

  • 2021-11-16 23:33:56
  • Richard Shin, Christopher H. Lin, Sam Thomson, Charles Chen, Subhro Roy, Emmanouil Antonios Platanios, Adam Pauls, Dan Klein, Jason Eisner, Benjamin Van Durme
  • 0

Abstract

We explore the use of large pretrained language models as few-shot semanticparsers. The goal in semantic parsing is to generate a structured meaningrepresentation given a natural language input. However, language models aretrained to generate natural language. To bridge the gap, we use language modelsto paraphrase inputs into a controlled sublanguage resembling English that canbe automatically mapped to a target meaning representation. Our resultsdemonstrate that with only a small amount of data and very little code toconvert into English-like representations, our blueprint for rapidlybootstrapping semantic parsers leads to surprisingly effective performance onmultiple community tasks, greatly exceeding baseline methods also trained onthe same limited data.

 

Quick Read (beta)

loading the full paper ...