I Have No Mouth, and I Must Rhyme: Uncovering Internal Phonetic Representations in LLaMA 3.2

TL;DR

This paper explores how LLaMA 3.2 internally encodes phonetic information, revealing a structured phoneme representation and a specialized head that facilitates rhyming tasks without explicit phonetic training.

Contribution

It uncovers the internal phonetic representations in LLaMA 3.2, including a phoneme organization and a dedicated 'phoneme mover head' that aids in phonetic tasks.

Findings

LLaMA encodes a rich internal model of phonemes.

High-level organization of phoneme representations in latent space.

Identification of a 'phoneme mover head' that promotes phonetic information.

Abstract

Large language models demonstrate proficiency on phonetic tasks, such as rhyming, without explicit phonetic or auditory grounding. In this work, we investigate how \verb|Llama-3.2-1B-Instruct| represents token-level phonetic information. Our results suggest that Llama uses a rich internal model of phonemes to complete phonetic tasks. We provide evidence for high-level organization of phoneme representations in its latent space. In doing so, we also identify a ``phoneme mover head" which promotes phonetic information during rhyming tasks. We visualize the output space of this head and find that, while notable differences exist, Llama learns a model of vowels similar to the standard IPA vowel chart for humans, despite receiving no direct supervision to do so.