ChatGPT to Replace Crowdsourcing of Paraphrases for Intent Classification: Higher Diversity and Comparable Model Robustness

TL;DR

This paper investigates replacing crowdsourcing with ChatGPT for paraphrase generation in intent classification, finding ChatGPT produces more diverse paraphrases and results in models with comparable robustness.

Contribution

It demonstrates that ChatGPT can effectively substitute crowdsourcing for paraphrase data collection, achieving higher diversity and similar robustness in intent classification models.

Findings

ChatGPT generates more diverse paraphrases than crowdsourcing.

Models trained on ChatGPT paraphrases are as robust as those trained on crowdsourced data.

ChatGPT can potentially replace crowdsourcing in paraphrase generation tasks.

Abstract

The emergence of generative large language models (LLMs) raises the question: what will be its impact on crowdsourcing? Traditionally, crowdsourcing has been used for acquiring solutions to a wide variety of human-intelligence tasks, including ones involving text generation, modification or evaluation. For some of these tasks, models like ChatGPT can potentially substitute human workers. In this study, we investigate whether this is the case for the task of paraphrase generation for intent classification. We apply data collection methodology of an existing crowdsourcing study (similar scale, prompts and seed data) using ChatGPT and Falcon-40B. We show that ChatGPT-created paraphrases are more diverse and lead to at least as robust models.