Word Dictate - keeps typing 'what'

A bit of a stuff around, but good advice, nonetheless. Thanks.

When I called Dragon support and suggested I might be able to help, via testing and evaluation, to prevent them from marginalizing a group of autistic customers, they were very, very disinterested.  It's an interesting situation. 

.

That is unfortunate. Many developers are resistant to letting individuals help them. But if you are still interested in trying, see if you can interest a recognized group to make the offer to Dragon.  I don't know is there is anyone who represents / helps autistic people in general. The specific example I can suggest is the CNIB, Canadian National Institute for the Blind. Is there an equivalent group for people with autism?

Dictate is one of MS's "A.I." (Artificial Intelligence / Idiot) SaaS. So for the most part it is not a question of programming lines of code.  This sort of AI depends a lot on "learning" from test data sets.  The programmer feeds the test data, voice recordings in this case, to the AI. The AI makes its best guess on the content. The programmer corrects errors. Then they move on to the next test data.

Even if they used one set of samples to provide the core English translation, I would expect that they would use regionally specific samples to fine tune the AI.  Hopefully after MS gets the core functionality developped they will provide a way for users and developers to "fine tune" it. Like the "custom dictionary" file and feature in spell checking. 

So the resulting "quality" of the product depends entirely on test data. If your voice inflection or accent is substantially different than the data used to develop the AI, it won't work for you. 

Hmmm. I suspect Dragon is doing the same thing. I never had good luck with Dragon, until I did an OS-10 update which somehow caused Dragon to work super well for a short time, but crash. The AI algorithms must be more tuned toward those characteristics that are regional and individual, as opposed to the core language. Perhaps, we could think of words as being layered, like an onion. I'm guessing the AI algorithms are learning to look primarily at the outer layer and less at the inner, which is where the core language exists.  But, people like me don't have much of an outer layer. Over the years, I've purchased four versions of Dragon and the latest is the worst, for me. I suspect the changes in AI emphasis coupled with those things missing from my autistic voice leave the processor with too little to work with. When I called Dragon support and suggested I might be able to help, via testing and evaluation, to prevent them from marginalizing a group of autistic customers, they were very, very disinterested.  It's an interesting situation. 

Your observations about differences between English dialects is interesting. I'm an American and a high-functioning autistic. As an autistic, I lack many of the intonations other have, regardless of language and dialect.

Windows is really contradictory to the new software-as-a-service cloud-based paradigm.

Dictate is one of MS's "A.I." (Artificial Intelligence / Idiot) SaaS. So for the most part it is not a question of programming lines of code.  This sort of AI depends a lot on "learning" from test data sets.  The programmer feeds the test data, voice recordings in this case, to the AI. The AI makes its best guess on the content. The programmer corrects errors. Then they move on to the next test data.

.

Even if they used one set of samples to provide the core English translation, I would expect that they would use regionally specific samples to fine tune the AI.   Hopefully after MS gets the core functionality developped they will provide a way for users and developers to "fine tune" it. Like the "custom dictionary" file and feature in spell checking.

.

So the resulting "quality" of the product depends entirely on test data. If your voice inflection or accent is substantially different than the data used to develop the AI, it won't work for you.