Mari Ostendorf
What language technology looks like when you've been responsible for it as deployable engineering for thirty years before LLMs arrived to redo the field.
Ostendorf spent decades on the speech side of language technology — speech recognition, prosody, dialogue systems — in a tradition that took engineering-grade reliability for granted because the systems had to run on real audio for real users. That posture is increasingly relevant to LLM evaluation: as language models move from text-only demos into voice-enabled products, the engineering questions speech researchers solved for thirty years matter again. Her current work, combined with the systems-engineering tradition she comes from, makes her a useful read for anyone trying to evaluate language tech as something with deployment consequences.