DC Field | Value | Language |
---|---|---|
dc.contributor.author | Kim, Jeonghwan | ko |
dc.contributor.author | Hong, Giwon | ko |
dc.contributor.author | Myaeng, Sung-Hyon | ko |
dc.contributor.author | Whang, Joyce Jiyoung | ko |
dc.date.accessioned | 2023-12-27T00:00:59Z | - |
dc.date.available | 2023-12-27T00:00:59Z | - |
dc.date.created | 2023-12-24 | - |
dc.date.created | 2023-12-24 | - |
dc.date.issued | 2023-12-08 | - |
dc.identifier.citation | The 2023 Conference on Empirical Methods in Natural Language Processing, pp.3763 - 3775 | - |
dc.identifier.uri | http://hdl.handle.net/10203/316862 | - |
dc.description.abstract | Compositional reasoning across texts has been a long-standing challenge in natural language processing. With large language models like GPT-4 taking over the field, prompting techniques such as chain-of-thought (CoT) were proposed to unlock compositional, multi-step reasoning capabilities of LLMs. Despite their success, the prompts demand significant human effort to discover and validate them. Our work draws attention to the idea of transferring task-specific inductive biases from finetuned models to prompts, as a way of improving GPT-4’s compositional reasoning capabilities. To leverage these inductive biases, we formulate prompt templates to ease the transfer of inductive biases. The experimental results on multi-hop question answering and numerical reasoning over text show that our proposed prompt scheme shows competitive zero-shot and few-shot performances compared to existing prompts on complicated reasoning tasks, highlighting the importance of adopting the validated biases of the previous paradigm. | - |
dc.language | English | - |
dc.publisher | Association for Computational Linguistics | - |
dc.title | FinePrompt: Unveiling the Role of Finetuned Inductive Bias on Compositional Reasoning in GPT-4 | - |
dc.type | Conference | - |
dc.type.rims | CONF | - |
dc.citation.beginningpage | 3763 | - |
dc.citation.endingpage | 3775 | - |
dc.citation.publicationname | The 2023 Conference on Empirical Methods in Natural Language Processing | - |
dc.identifier.conferencecountry | SI | - |
dc.identifier.conferencelocation | Resorts World Convention Centre, Singapore | - |
dc.identifier.doi | 10.18653/v1/2023.findings-emnlp.245 | - |
dc.contributor.localauthor | Myaeng, Sung-Hyon | - |
dc.contributor.localauthor | Whang, Joyce Jiyoung | - |
dc.contributor.nonIdAuthor | Kim, Jeonghwan | - |
dc.contributor.nonIdAuthor | Hong, Giwon | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.