4 Non-Trivial Questions to Ask before Committing to Production Document Capture
In late 2009 and I got a call from the brother of a good friend. He was a researcher at IBM's Watson Labs - soon to became famous for the "Watson" artificial intelligence engine that spectacularly beat the top humans on the trivia game-show, Jeopardy!
My friend was trying to solve a problem and thought that my company, Datacap (the acquisition of Datacap by IBM was not even on the horizon at this point), could help, since we specialized in optical character recognition (OCR) and related document capture technologies.
I said, "great, let me ask you 3 or 4 questions about what you are trying to do:
1) What is the volume of documents/pages/images you need to process per day, week, month, or year?
2) What data do you need to extract from those pages, any special considerations to take into account?
3) Are the pages consistent in format, variable, something in between?"
He said he had 5000 pages. Clearly to him that was a big number, but he was a bit deflated when I asked, "is that per day?" In the production document capture business, it is definitely common that a volume like that may be literally processed "before breakfast."
But 5000 pages were all he had. Not every day or week, or even every month, just once. I was a little skeptical, but I wanted to learn more.
He needed to extract information from an English language pronunciation guide. He wanted to read the word to be pronounced, and then the linguistically precise definition of the pronunciation, including diacritical marks (accents) commonly used in those definitions. In other words, this was not just straight English language OCR. My skepticism increased.
I wasn't surprised when I next learned that the pages were not at all consistent, that the definitions for a specific word could wrap from one page to the next, or that the pages to be scanned were in bound books...
That was it. Did he really expect to use a production capture product to process - one time - 5000 pages with specialized text and words on them and no fixed format? Well, yes, he did. He had a real challenge and his expectation was not unreasonable... it just is not what production document capture is about.
Those three questions can help anyone quickly assess a document capture problem. In this case, the answer was simple, but perhaps wrong. I advised him that it would not be economically feasible for him to invest in production document capture, but in giving that answer I missed a great opportunity.
Turns out I should have asked a 4th question, "why do you need to read a pronunciation guide?"
I learned later that my friend was working on a major artificial intelligence project, one that would need a computer capable of blurting out words under extreme time pressure. He was, in fact, working on giving "Watson" a voice. It was that voice, having been trained to enunciate thousands of words, that went on prime time to beat the best human players at a live game of Jeopardy!
He eventually used a desktop OCR program and a lot of patience to translate the pronunciation guide from paper to something Watson could understand. Although my 3 questions helped me quickly assess the value of the opportunity, by skipping the 4th question, I missed the opportunity to brag how Datacap helped to give Watson a voice!
Is production document capture and imaging right for you? Click here to learn more on using capture solutions.