Fall 2026 AAC Roundup: What the Big Labs Are Bringing to Communication Tech 👩🔬🧠📱
AAC has always moved slower than the rest of consumer tech, for good reason — vocabulary systems, motor planning, and clinical trust take years to earn. But this year, the gap between "frontier AI lab" and "AAC device" has been closing fast. Here's what's new and what's coming as we head into fall 2026.
Google DeepMind and Android: sign language finally gets a text bridge
The headline release this season is SL2T (sign-language-to-text), a model from Google DeepMind and Android that translates ASL directly into text in real time. It's now live in Gboard and Live Transcribe, starting on Pixel 11, with more devices and sign languages planned. It's not marketed as an AAC app, but functionally it's doing AAC-adjacent work: turning an expressive modality that most devices can't parse into text that any app, keyboard, or captioning system can use. Google has said explicitly that full parity with spoken and written languages is the long-term goal, and that sign language generation — not just recognition — is next. That's the piece AAC teams should watch closely, since a model that can generate sign output could eventually let AAC systems talk back in sign, not just text-to-speech.
Project Euphonia's open-sourcing moment
Google's longer-running effort, Project Euphonia, has spent years collecting speech samples from people with ALS, cerebral palsy, and other conditions that produce atypical speech, in order to train better speech recognition. This year the project shifted from closed data collection to an open-source GitHub release, handing its research and tooling to developers who want to build personalized ASR models for non-standard speech. Paired with the Speech Accessibility Project (a University of Illinois–led effort Google backs alongside Apple, Amazon, and Microsoft), this is a meaningful move: the training data and techniques for understanding disordered speech are no longer locked inside one company's research org. Expect smaller AAC and speech-recognition vendors to start building on this over the next year.
The incumbents are quietly adding AI, not replacing themselves
The established AAC players — Proloquo (AssistiveWare), TD Snap (Tobii Dynavox), TouchChat, CoughDrop, Grid — haven't pivoted to chatbot interfaces, and that's mostly by design. But AI is showing up in other ways:
Just-in-time vocabulary generation. Academic work (presented at ASSETS and CHI this year) has tested using large multimodal models to auto-populate visual scene displays with contextually relevant vocabulary on the fly, evaluated directly against SLP-created boards. Early results are promising… but we already knew this. 🙂
AI-generated symbols. A framework published this year explores using text-to-image generation to create personalized, culturally specific pictograms — useful for the long tail of concepts that commercial symbol sets don't cover well.
Ultra-personalized phrase prediction. CHI 2026 also featured an autoethnographic study of a fine-tuned personal language model for AAC phrase suggestions — and it surfaced the hard part of this trend: personalization improves speed, but it raises real questions about agency, identity, and privacy when a model starts finishing your sentences based on your own communication history.
What to watch this fall
Whether SL2T's roadmap toward sign language generation starts showing up in developer previews.
Whether any major AAC vendor formally integrates a generative "board-builder" feature rather than leaving it in research prototypes.
More open datasets and models emerging from Project Euphonia's GitHub release, and whether they get adopted outside Google's own products.
The theme underneath all of this: the frontier labs are increasingly treating communication access as a first-class AI problem, not a side project — and we all stand to benefit from the AAC space taking a giant AI leap forward.