What Do Multilingual Self-Supervised Speech Encoders Know About Sentence Meaning? A study posted to arXiv (2611.17686v1) probing sentence-level semantics in massively multilingual self-supervised speech models across 76 languages found that sentence meaning is linearly accessible from frozen speech representations but highly uneven across languages and resource tiers. Within the OmniASR wav2vec family, scaling from 300M to 7B parameters disproportionately benefited high-resource languages, and ASR tuning sharply improved semantic decodability while making representations more language-specific. Cross-lingual probes matched language-specific probes on languages never seen in probe training, but per-model monolingual decodability did not predict transfer, and the authors report semantic content is entangled with acoustics, inflating apparent performance on single-speaker corpora. arXiv:2611.17686v1 Announce Type: new Abstract: Self-supervised speech models S3Ms have been scaled up to thousands of languages and millions of hours of audio, yet little is known about how these models structure semantic knowledge across languages. In this work, we probe sentence-level semantics in massively multilingual S3Ms across 76 languages, using linear maps from frozen speech representations into a frozen text embedding space, evaluated by cross-modal retrieval. Sentence meaning is linearly accessible but highly uneven across languages and resource tiers. Cross-lingual probes can match language-specific probes on languages never seen in probe training, but per-model monolingual decodability does not predict transfer. Within the OmniASR wav2vec family, scaling from 300M--7B disproportionately benefits high-resource languages, and ASR tuning sharply improves semantic decodability while making representations more language-specific. However, semantic content is entangled with acoustics, inflating apparent performance on single-speaker corpora.