Lexy Cal Am Big You Itty Rezo Loo Shin.
A computer/science illiterate attempts to understand the intelligence of artificial intelligence.

The most frequently used words in English are highly ambiguous: Webster’s Ninth New Collegiate Dictionary lists 94 meanings for the word “run” as a verb alone.

The word "bat" can describe an animal, a sports apparatus, the use of a sports apparatus, the blink of an eye, and more.

And yet computers are learning all the “runs”.
And all the “bats”.
And all the other words.
And all the other combinations of words.
And all the other combinations of words—in all their combinations.
Or are they?

How can a computer program be made to understand so many words in an almost infinite number of contexts --especially when those contexts are compounded by all the other mitigating conundrums of language like idiom and syntax?

If I knew the answer I’d be working at a LAR Lab.

LAR is Lexical Ambiguity Resolution ( appearing phonetically in the headline of this post) and it’s the central problem in natural language and computational semantics research. And it’s what’s behind all that voice recognition stuff that you get when you call customer service lines, and what Apple built into the new iPhone that let’s you find the nearest Spicy Salmon Roll with a simple query like “I’m in the mood for Sushi.”

Sidebar: When I looked up LAR I found a site that listed LAR as the acronym for 52 things, none of which was Lexical Ambiguity Resolution. Talk about context?

I guess if they can teach a computer to play Chess, they can make it learn words. Yet, somehow I think Chess and Language is apples and oranges. Hey, I don’t know what the hell I’m talking about since I don’t know the first thing about programming computers, yet I do have some appreciation of the complexity of the English language and a fair understanding of how tricky it can be (which is why I love wordplay).

But programming a computer to “understand” context in language blows my mind. What would they make of:

“They passed the port as dawn rose”

Was that port by the river, or from the wine rack?
Did they pass it by or pass it around?
And did Dawn get a sip?

Or

“Visiting distant relatives can be inconvenient.”

Unless, of course, it's the other way around.

Or

“The rabbi married my sister who was looking for a match ”

Oy Vey.

So I did some reading and here’s what I learned.

“The problem is specifying the nature of context and how it interacts with the rest of an “understanding” system. “

“Advanced computer speech does not need all words to be programmed, it only needs all transitions between any pair of sounds (phonemes) in a language.”

Aha, so it’s sounds not words. It’s not really the language that has to be interpreted, it’s the sound of the components of the language.

Which to my mind begs the question:
Who’s making the sounds? A drawling pig farmer from Arkansas? A nasal soccer mom from Long Gi-land? A Cockney lad from Liverpool? George Bush? Me?

I had a voice recognition system installed in my desktop by the geniuses at IBM back in the mid-90’s when I worked at Prodigy -- remember Prodigy? It was the ill-fated online service that was 10 years too early and ran out of money and vision just when AOL and internet Browsers hit the scene. All I had to do was say “open Mail” and my e-mail screen would appear. Or I could speak any number of other commands and the computer would magically comply. Until one day I came to work with a bad cold and my nasal drip and phlegmy throat altered my voice and rendered the system incapable of carrying out my congested wishes.

“The most complicated part of the programming is turning "human readable" text into phonemes. Sounds. Mistakes in this process are what account for the often hilarious results in GPS pronunciation.”

The GPS in our car (we call her Ethel) is a real comedienne. She tells us to: “Take The Van Why Sick” (for the Van Wyck Expressway) I can only imagine the fun in LA or in the South-West with all the Mexican names: “Now arriving at I’ve a need a Cab Brillo? ” Our landline at home has caller ID voice that always says “Call from Verizon Were-Eels. “ (Verizon Wireless)

(Most phonetics in the GPS-device market are produced by a company called the Acapela Group…gotta love that)

Good Luck LAR folk, but don’t mess with Ethel . She’s our portable dependable and entertaining Mrs. Malaprop and we like her just the way she are.