Apple's voice recognition software completely ignores context

I note this because I find it a very curious decision on the part of the engineers who built it.

An example: a minute ago, I just spoke the words "an ant colony." The software inserted the words "an and colony." That does not even make syntactical, let alone semantic sense. But the software did offer me two alternatives: "Aunt" and "an." An "Aunt colony" is an amusing suggestion: I pictured a group of spinsters gathered together on an island, knitting and serving each other tea. But again, "an an colony" makes no sense at all.

Surely, far and away the most common word, which sounds like these choices, to go in the blank in the phrase "an ____ colony" is "ant," right? But that was not even offered as an alternative.

Does anyone know why the designers decided against taking any clues from context in deciding what was spoken?

No comments:

Post a Comment