DRAGON
Baker built a system in which every level of knowledge about language is one probabilistic model, and recognition is a search for the most likely path.
Why it matters
Instead of a chain of heuristic rules there was one model, and that bet displaced every other approach a decade later.
Other participants assembled systems from separate modules, each with its own rules. Baker instead described acoustics, phonetics, lexicon and syntax as a single Markov process and searched it for the best path. Accuracy trailed the competition, but the approach scaled. Baker and his wife Janet founded Dragon Systems in 1982 and carried the line through to a consumer product.