Natural Language Embedded Programs for Hybrid Language Symbolic Reasoning

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Zhang, Tianhua, Ge, Jiaxin, Luo, Hongyin, Chuang, Yung-Sung, Gao, Mingye, Gong, Yuan, Wu, Xixin, Kim, Yoon, Meng, Helen, Glass, James
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917624946884608
author Zhang, Tianhua
Ge, Jiaxin
Luo, Hongyin
Chuang, Yung-Sung
Gao, Mingye
Gong, Yuan
Wu, Xixin
Kim, Yoon
Meng, Helen
Glass, James
author_facet Zhang, Tianhua
Ge, Jiaxin
Luo, Hongyin
Chuang, Yung-Sung
Gao, Mingye
Gong, Yuan
Wu, Xixin
Kim, Yoon
Meng, Helen
Glass, James
contents How can we perform computations over natural language representations to solve tasks that require symbolic and numeric reasoning? We propose natural language embedded programs (NLEP) as a unifying framework for addressing math/symbolic reasoning, natural language understanding, and instruction following tasks. Our approach prompts a language model to generate full Python programs that define functions over data structures which contain natural language representations of structured knowledge. A Python interpreter then executes the generated code and prints the output. Despite using a task-general prompt, we find that this approach can improve upon strong baselines across a range of different tasks including math and symbolic reasoning, text classification, question answering, and instruction following. We found that the generated programs are interpretable since they outline the exact reasoning process followed by the program interpreter.
format Preprint
id arxiv_https___arxiv_org_abs_2309_10814
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Natural Language Embedded Programs for Hybrid Language Symbolic Reasoning
Zhang, Tianhua
Ge, Jiaxin
Luo, Hongyin
Chuang, Yung-Sung
Gao, Mingye
Gong, Yuan
Wu, Xixin
Kim, Yoon
Meng, Helen
Glass, James
Computation and Language
How can we perform computations over natural language representations to solve tasks that require symbolic and numeric reasoning? We propose natural language embedded programs (NLEP) as a unifying framework for addressing math/symbolic reasoning, natural language understanding, and instruction following tasks. Our approach prompts a language model to generate full Python programs that define functions over data structures which contain natural language representations of structured knowledge. A Python interpreter then executes the generated code and prints the output. Despite using a task-general prompt, we find that this approach can improve upon strong baselines across a range of different tasks including math and symbolic reasoning, text classification, question answering, and instruction following. We found that the generated programs are interpretable since they outline the exact reasoning process followed by the program interpreter.
title Natural Language Embedded Programs for Hybrid Language Symbolic Reasoning
topic Computation and Language
url https://arxiv.org/abs/2309.10814