I'm not super deep into the world to NLP, but from what I have kept up with, it seems that the state-of-the-art is almost constantly open-sourced. I understand certain companies have personal relationships already built which may be what MS is after more so than the IP, but isn't there a "canonical" way of building a, for example, speech to text system for your phone?

Is there really a lot of NLP IP hidden behind corporate walls at this point? I just assumed Alexa, Siri, Google Assistant, etc were all using the same model architectures. Genuine question, can anyone shed some light?