Saved in:
Bibliographic Details
Main Authors: Ma, Ziqi, Nguyen, Sao Mai, Xu, Philippe
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.24259
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917047937531904
author Ma, Ziqi
Nguyen, Sao Mai
Xu, Philippe
author_facet Ma, Ziqi
Nguyen, Sao Mai
Xu, Philippe
contents Emergent symbolic representations are critical for enabling developmental learning agents to plan and generalize across tasks. In this work, we investigate whether large language models (LLMs) can translate human natural language instructions into the internal symbolic representations that emerge during hierarchical reinforcement learning. We apply a structured evaluation framework to measure the translation performance of commonly seen LLMs -- GPT, Claude, Deepseek and Grok -- across different internal symbolic partitions generated by a hierarchical reinforcement learning algorithm in the Ant Maze and Ant Fall environments. Our findings reveal that although LLMs demonstrate some ability to translate natural language into a symbolic representation of the environment dynamics, their performance is highly sensitive to partition granularity and task complexity. The results expose limitations in current LLMs capacity for representation alignment, highlighting the need for further research on robust alignment between language and internal agent representations.
format Preprint
id arxiv_https___arxiv_org_abs_2510_24259
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Can LLMs Translate Human Instructions into a Reinforcement Learning Agent's Internal Emergent Symbolic Representation?
Ma, Ziqi
Nguyen, Sao Mai
Xu, Philippe
Computation and Language
Robotics
Emergent symbolic representations are critical for enabling developmental learning agents to plan and generalize across tasks. In this work, we investigate whether large language models (LLMs) can translate human natural language instructions into the internal symbolic representations that emerge during hierarchical reinforcement learning. We apply a structured evaluation framework to measure the translation performance of commonly seen LLMs -- GPT, Claude, Deepseek and Grok -- across different internal symbolic partitions generated by a hierarchical reinforcement learning algorithm in the Ant Maze and Ant Fall environments. Our findings reveal that although LLMs demonstrate some ability to translate natural language into a symbolic representation of the environment dynamics, their performance is highly sensitive to partition granularity and task complexity. The results expose limitations in current LLMs capacity for representation alignment, highlighting the need for further research on robust alignment between language and internal agent representations.
title Can LLMs Translate Human Instructions into a Reinforcement Learning Agent's Internal Emergent Symbolic Representation?
topic Computation and Language
Robotics
url https://arxiv.org/abs/2510.24259