When Text Became Number Crunching
by Dwayne Phillips
The number crunchers now rule the world. How did that happen?
Many years ago I was a number cruncher. I did then what people still call “digital signal processing.” We took analog signals, magically made them numbers in computers via gadgets called analog-to-digital converters or A/D converters, and happily applied digital approximations of mathematics. Wow. That was great fun. And, to a large extent, it all worked.
We used supercomputers (which were far advanced but are now puny compared to an iPhone), programmed in FORTRAN (another story in technology management for another day), and it all worked well. We wanted x87 co-processors in our PCs. These were special number-crunching processors that boosted the performance of the x86 CPUs. They were the predecessors of today’s GPUs, which perform the same function.
Then there were “the other guys.” “They” processed text. They considered how to teach a computer to understand subject-verb-object. They created lists of unique words. They did “stemming” and other things that linguists and English majors understood. Those things might be interesting to some persons, but really? We were crunching numbers.
What would real programmers want to do, CRUNCH or parse?
One day I woke to find that the text processors were crunching numbers. Let’s apply machine learning to text processing. Machine learning was the term the newspapers used for pattern recognition and machine intelligence (PAMI) and all types of supervised and unsupervised learning and long equations that were digital approximations of mathematics — wait, we are using the same terms used earlier. These machine learning algorithms turned text into digital approximations of linear algebra. Matrix multiply. “DO LOOPS” (an ancient term sometimes quoted by old FORTRAN programmers) and all that. The levels of the loops were deep, but they were just loops and IF and THEN and ELSE. The same thing us old “number crunchers” did.
All of a sudden, those linguists and English majors wanted to talk to “us” not “them.” And, some of “them” wanted to become “us.”
Time and technology march on. Now we are all crunching numbers to a greater extent than ever before. MegaFLOPS has become some greek-prefix-FLOPS that I’ve never heard before. Enough ranting for an old man.
Now, where can I find one of those new Nvidia GPUs that has an air conditioner built in to keep the computer from bursting into flames so I can pop it into my mega PC and crunch some numbers?