Emergent Semantics Beyond Token Embeddings: Transformer LMs with Frozen Visual Unicode Representations

TL;DR

This paper demonstrates that large language models can achieve high performance using frozen Unicode visual embeddings instead of trainable semantic embeddings, highlighting the emergent nature of semantics.

Contribution

It introduces a method of using frozen Unicode visual embeddings in Transformer models, challenging the traditional view of trainable semantic input embeddings.

Findings

Models with frozen Unicode embeddings outperform those with trainable embeddings on reasoning tasks.

High-level semantics emerge from the model architecture and data, not from trainable input embeddings.

The approach is compatible with any tokenizer, including a new Unicode-centric tokenizer.

Abstract

Understanding the locus of semantic representation in large language models (LLMs) is crucial for interpretability and architectural innovation. The dominant paradigm posits that trainable input embeddings serve as foundational "meaning vectors." This paper challenges that view. We construct Transformer models where the embedding layer is entirely frozen, with vectors derived not from data, but from the visual structure of Unicode glyphs. These non-semantic, precomputed visual embeddings are fixed throughout training. Our method is compatible with any tokenizer, including a novel Unicode-centric tokenizer we introduce to ensure universal text coverage. Despite the absence of trainable, semantically initialized embeddings, our models converge, generate coherent text, and, critically, outperform architecturally identical models with trainable embeddings on the MMLU reasoning benchmark. We attribute this to "representational interference" in conventional models, where the embedding layer is burdened with learning both structural and semantic features. Our results indicate that high-level semantics are not inherent to input embeddings but are an emergent property of the Transformer's compositional architecture and data scale. This reframes the role of embeddings from meaning containers to structural primitives. We release all code and models to foster further research.