Do Audio Language Models Hear and Read Distinctive Features Alike?

2026-09-24

AI Summary

This study investigates whether speech and text streams in audio language models encode phonological distinctive features along consistent representational directions. To address this, the authors propose an evaluation framework based on randomly paired reference systems, which extracts minimal pair offset vectors and computes their cosine similarities to systematically compare cross-modal geometric alignment across multiple models and languages. The findings reveal that model family architecture, rather than parameter scale, governs the degree of feature alignment. Notably, only Qwen2.5-Omni significantly surpasses the random baseline on the voicing feature, demonstrating that auditory and textual representations in most current models remain fundamentally misaligned.

Abstract

Audio language models pass speech and text through a single decoder. We ask whether that decoder represents a distinctive feature in the same direction when a phoneme is heard and when it is read. For minimal pairs of phonemes differing in one feature, we take the offset between the two members' mean representations. Averaging those offsets gives a direction for each stream, and we measure the cosine between the two. Because the two streams already agree about arbitrary phoneme pairs, we compare every measure against a reference built from random pairings rather than against zero. We apply this to 6 models, 7 features and 15 languages from 11 families. Only voicing in the two Qwen2.5-Omni models exceeds that reference after correction for multiple testing, and the reference varies by a factor of seven between models. In three of the six models, voicing has one direction in audio across the 14 languages with enough minimal pairs to measure it, and every language pair agrees in two of them. The model family, not the model size, predicts which stream represents a feature.

Problem

• audio language models
• distinctive features
• speech-text alignment
• phoneme representation
• multimodal decoding

Innovation

• Audio Language Models
• Distinctive Features
• Cross-modal Alignment
• Minimal Pairs
• Representation Geometry