Watson’s Linguistic Struggles

When you talk to experts in linguistics about teaching language to a computer, they stress all of the context that humans use to make sense of sentences. Most of the time we know to put words into the right boxes.The “Lincoln” that’s jammed at rush hour is a tunnel. The “Lincoln” wearing the stove-pipe hat was a president. And no, that wasn’t a real stove pipe on his head. What are you, an idiot? (Watson, don’t answer that question.)

Humans also read each other’s gestures. We know which friends are more likely to speak sarcastically, and we understand why. Watson doesn’t know such things. We interpret sentences by the accent on certain words. By stressing each different word, you get seven different meanings from this one sentence: “I didn’t say she stole my money.” That would mystify Watson (and the computer is fortunate that Jeopardy clues are written).

But even with our mastery of context, intonation, and gestures, we don’t understand each other very well. I was thinking of all of the expressions we use to get a better grip on what is being said. Just consister:
Huh?
Are you serious?
Get out of here!
What?
You’re kidding!
I mean…
Seriously.
What are you saying!?
Yeah, right!
I think you misunderstood…
Excuse me?
Spit it out.

Our dialogs, if you look at them from a computer’s point of view, are a back and forth with constant fixes and updates about what we’re trying to communicate. Getting meaning out of our own heads and into someone else’s is a monumental endeavor. And that’s even when the other person shares the same language and culture–and sometimes even the same house and bed. Watson is and will always be an outsider to language. The fact that it makes sense of Jeopardy clues still amazes me.

More Read

Derailing Your Supply Chain BI Project

Does Data Quality Matter in Social Media?

Applying BIM to Design of Sites and Structures

Delivering Quality – Where it Counts, When it Counts

Good Data – An Essential Part of the Marketing Recipe