Alice in...WordLand?

I attempted impersonating Alice for the past four weeks, falling through rabbit holes of unending questions and a strong determination to wholly organize the influx of information guiding me through a marvelous world of...messy data, bidirectional recurrent neural networks and matrices twisting every which way.

The rabbit I ran after was my Original Work Project. After creating part of a dynamically changing plan, I was able to tumble into learning - no assignments, no deadlines, nothing but curiosity. The biggest disadvantage I faced, apart from not having a complete map of the Wonderland I found myself lost in, was having walked into the uncharted territory of Unsupervised Learning and Neural Networks that tend to travel forwards and backwards, computing output based on input from the future. I do not know if Alice had motion-sickness, but reading that last phrase made my stomach drop at the same time my excitement soared at having found yet another puzzling riddle.

This future-past input-output network was what I learned was Encoder-Decoder architecture. Though my clarity on the concept remains no doubt fuzzy, its application in machine translation opened my eyes to what a complex brain we have. Our natural neural processes concerning linguistics - listening to someone speak, encoding it into meaning, storing the information, forming phrases that are grammatically, semantically and contextually meaningful, and converting these sentences to facial twitches for speech - amazed me. I had never consciously thought about the steps involved, information gleaned and wrapped and reorganized almost instantaneously, different processes occurring simultaneously. However, learning about the Encoder-Decoder made me acutely aware of how I view the world, and I realized I could never be grateful enough.

In a few words, the experience was equivalent to Alice missing home, every inch of Wonderland worthy of consciously thinking about home, realizing a small fraction of a phenomenon not less than a miracle, regardless of what Charles Darwin would argue. For such a long time, we have spoken to computers in their language - zeroes and ones. As we try to bridge the gap (since we clearly do not speak in numbers, and certainly not in two numbers only), teaching computers our language through their own, I realized, required understanding how we do it ourselves. As we teach computers, we learn about ourselves. I suppose describing Wonderland or rather, WordLand, to others requires us to explore first.