Few-shot Adaptation For Parsing Contextual Utterances With Llms

Lin Kevin, Xia Patrick, Fang Hao. Arxiv 2023

[Paper]    
Few Shot Fine Tuning In Context Learning Pretraining Methods Prompting Reinforcement Learning Training Techniques

We evaluate the ability of semantic parsers based on large language models (LLMs) to handle contextual utterances. In real-world settings, there typically exists only a limited number of annotated contextual utterances due to annotation cost, resulting in an imbalance compared to non-contextual utterances. Therefore, parsers must adapt to contextual utterances with a few training examples. We examine four major paradigms for doing so in conversational semantic parsing i.e., Parse-with-Utterance-History, Parse-with-Reference-Program, Parse-then-Resolve, and Rewrite-then-Parse. To facilitate such cross-paradigm comparisons, we construct SMCalFlow-EventQueries, a subset of contextual examples from SMCalFlow with additional annotations. Experiments with in-context learning and fine-tuning suggest that Rewrite-then-Parse is the most promising paradigm when holistically considering parsing accuracy, annotation cost, and error types.

Similar Work