On the Entropy of Language Models in Getting Semantic from Tokens
Keywords: LLM evaluation
TL;DR: We designed a new metric to quantify the entropy reduction between semantic level and token level to represent the capability of capturing semantic from tokens
Abstract: Large language models (LLMs) are widely recognized for their exceptional capacity to capture semantic meaning. Yet, there remains no established metric to quantify this capability. In this work, we introduce a quantitative metric, Information Emergence (IE), designed to measure LLMs’ ability to extract semantics from input tokens. We formalize “semantics” as the meaningful information abstracted from a sequence of tokens and, leveraging information theory, quantify this through comparing the reduction in entropy observed for a sequence of tokens (macro-level) and individual tokens (micro-level). To achieve this, we design a light-weight estimator to compute the mutual information at both micro and macro levels for each transformer layer, which is agnostic to different tasks and language model architectures. We apply IE in both synthetic in-context learning (ICL) scenarios and natural sentence contexts. Experiments show a high-level informativeness of our metric reflected in semantic faithfulness, sensitivity, and connection with emergence. In addition, we highlight some interesting findings: 1) IE explains why ICL offers clearer semantics and benefits compared to natural text through changes
in entropy. 2) We could associate certain hallucination phenomenon with increased variance in IE. 3) IE can effectively differentiate between human-written and LLM generated text, proving especially useful for extremely large and closed-source language models. Our codes are available at: https://anonymous.4open.science/r/Emergence/.
Primary Area: foundation or frontier models, including LLMs
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2108
Loading