Language Writ Large: LLMs, ChatGPT, Grounding, Meaning and Understanding

2024-02-03

arXiv.org

AI Summary

This paper addresses the foundational question of whether large language models (LLMs) possess genuine linguistic understanding, arguing that their lack of sensorimotor experience results in semantically ungrounded representations. Method: Drawing on cognitive linguistics, neural category learning, computational semantics, and Chomskyan thought experiments, the study proposes six convergent bias hypotheses—first identifying how intrinsic linguistic properties (e.g., definitional circularity, production–comprehension symmetry, propositional iconicity) induce structural biases in ultra-large-scale statistical learning—and employs dialogic reasoning and conceptual modeling. Contribution/Results: The work demonstrates that LLMs instantiate only structured linguistic constraints—not semantic understanding—thereby bridging the theoretical gap between statistical pattern recognition and meaning grounding. It provides a novel explanatory framework for grounded language learning, recharacterizing LLM “understanding” as emergent from domain-general statistical convergence rather than embodied cognition.

Technology Category

Natural Language Processing: (Large) Language ModelsMachine Learning: Large Multimodal Models (LMMs)Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Apart from what (little) OpenAI may be concealing from us, we all know (roughly) how ChatGPT works (its huge text database, its statistics, its vector representations, and their huge number of parameters, its next-word training, and so on). But none of us can say (hand on heart) that we are not surprised by what ChatGPT has proved to be able to do with these resources. This has even driven some of us to conclude that ChatGPT actually understands. It is not true that it understands. But it is also not true that we understand how it can do what it can do. I will suggest some hunches about benign biases: convergent constraints that emerge at LLM scale that may be helping ChatGPT do so much better than we would have expected. These biases are inherent in the nature of language itself, at LLM scale, and they are closely linked to what it is that ChatGPT lacks, which is direct sensorimotor grounding to connect its words to their referents and its propositions to their meanings. These convergent biases are related to (1) the parasitism of indirect verbal grounding on direct sensorimotor grounding, (2) the circularity of verbal definition, (3) the mirroring of language production and comprehension, (4) iconicity in propositions at LLM scale, (5) computational counterparts of human categorical perception in category learning by neural nets, and perhaps also (6) a conjecture by Chomsky about the laws of thought. The exposition will be in the form of a dialogue with ChatGPT-4.

Problem

• LLMs' surprising capabilities and understanding
• Lack of direct sensorimotor grounding in ChatGPT
• Convergent biases aiding ChatGPT's performance

Innovation

• LLM scale biases enhance performance
• Lack sensorimotor grounding for meaning
• Computational counterparts in neural nets