Ahead of the Glyph: Predictive Intelligence for Multilingual Text Engines
Every layer of today’s internationalization stack waits to be told what to do. Developers pick an ICU data configuration at build time. The engine fails, then fetches, when it misses a data file. Software guesses a text’s encoding only after the bytes are already garbled. Language is detected one call at a time. The result is an i18n stack that is reactive, manually configured, and increasingly strained by the messy, mixed-script, AI-generated text users now paste into every application. This session presents a unifying approach: a single intelligence layer that sits in front of ICU and Adobe’s Globalization Library and reads the incoming text once to get ahead of every downstream Unicode decision. From that one pass it predicts the language and script of each run — including code-switches within a sentence; prefetches exactly the ICU data those scripts need, just ahead of use; resolves the most likely source encoding from context to prevent mojibake; and flags text that is technically valid but visually broken — denormalized sequences, orphaned combining marks, mixed-script homoglyphs, and stray bidi controls — before it ever reaches a glyph. This connects directly to “Unicode in the World”: it makes correct, multilingual text handling automatic and deployable everywhere — for every language, on every platform, and for the AI-generated content now reshaping how text is produced. The approach is grounded in published work on neural prefetching, compressed language identification, and learned caching, and built on production internationalization primitives. Attendees will leave with a concrete architecture for a predictive i18n layer, the trade-offs it introduces, and a set of open problems to collaborate on.