Disclosure or it didn't happen — and Google's grant US11769011B2 (“Universal language segment representations learning with conditional masked language model,” issued 2023-09-26) is the disclosure. Assigned to GOOGLE LLC and classified CPC G06F 40/284 and G06N 3/04, it covers learning universal language representations using a conditional masked language model.
The business requirement is global reach. A language AI that only works well in English is a fraction of a product; serving many languages well is both a market expander and a cost — more data, more training, more evaluation. Universal cross-lingual representations are how you get broad coverage without training a separate model per language.
Google routes its AI revenue through Cloud and its core products, and its filings discuss AI investment in aggregate rather than crediting specific architectures. The grant is the technique-level record under that aggregate: dated 2023, owned, aimed at the multilingual capability its products depend on.
House discipline: a grant proves invention and ownership, not a revenue figure, and we attribute none. The patent also doesn't claim a particular product ships this method. What it establishes is dated, owned IP behind a capability — multilingual AI — that is genuinely load-bearing for a global platform.
For the markets reader, the value is grounding a soft claim in a hard document. “Our AI is multilingual” is the kind of statement that's easy to make and hard to verify; a 2023 Google grant on cross-lingual representation learning is at least primary evidence that the capability has real, dated IP behind it.