Dear L,

There are these new breed of computer software, called neural networks that can classify images; For example, labelling an image of a dog as “dog”, or a cat photo as “cat”. The natural question to ask would be, “So does the neural network know what a cat or dog is?”. The answer, would be complicated.

The only reason you and I can communicate anything associated with dogs is because we both agree on what a dog is as an abstraction; A abstract representation of a physical entity that we both learned over the course of our existence. Our languages have enabled us to handle abstract symbolic and aural representations of said notions, that enables us to communicate. So if I ask you, “Do you know what a dog is?”, depending on your familiarity with the word “dog” as an entity, you would probably answer “yes”, or “no”. But the enabler of the question is our agreement on existence of a block of knowledge; In this case, knowledge of dogs. No matter how we represent an idea, there’s always some sort of agreement between us about its existence. And then we initiate exchange of information as the answer. But, this sort of agreement doesn’t exist in case of a neural network. If you open up the guts of a neural network, you would only see numbers, and connections, a grand jumble of fine tuned equations. So how do you even begin to answer the question whether the software knows what a dog is. How would you initiate an conceptual agreement to exchange information upon? I don’t think possible. The best we can do is to say that the neural network is arbitrarily good at determining a dog’s image compared to everything else.

This agreement is indirectly skimmed over in Douglas Hofstadter’s masterpiece, “Gödel, Escher, Bach”. While trying to explain how seeminlgy meaningless structures give rise to meaning out of almost nothing, Dr. Hofstadter argues that lot of our language constructs has an underlying structure, a sort of stack that the listener and the speaker instinctively keeps track of while speaking or listening. Recursive Transition Network as these constructs are called, a mean to dissect language and syntax, into context free schemas. Without going all nerdy and mentally frail with all existentially horrifying details in search for the meaning of meaning, I noticed something in there on an unusual moonlit winter night that I wanted to share with you.

The author, points out that there’s a distinct difference of placement of verbs in German compared to English; Positioning the verb at the end of the sentence. In almost-a-footnote, he implies that it’s an efficient mean to communicate since it’s average “Recursive Transition Network” never grows very large in sentences, in turn the user of the language never has to keep track too many chunks of information strewn around sentences. He even writes,

But in normal spoken German, such deep stacks almost never occur — in fact, native speakers of German often unconsciously violate certain conventions which force the verb to go to the end, in order to avoid the mental effort of keeping track of the stack.

That chapter, compelled me to think about my own writing. My sentences sometimes long, with layers stacked on top of each other. For example, take the sentence I just wrote,

Without going all nerdy and mentally frail with all existentially horrifying details in search for the meaning of meaning, I noticed something in there on an unusual moonlit winter night that I wanted to share with you.

While I am already referring to the contents I wrote just before, I am trying to cram a few things in this single sentence,

  1. I was reading the book at night.
    1. It was a moonlit night.
    2. It’s unusual, because where I live it’s usually cloudy. Moonlit winter nights are uncommon.
  2. The book, that I previously mentioned elaborates on the very nature of meaning.
  3. Studies on linguistic structures is complicated.
    1. I don’t want to get very nerdy.
    2. I think these sort of discussions are tiring.
    3. I don’t go want into the details.
  4. But I still want to share something with you.

And if you had to read that sentence, you probably had kept track of all those squished layers. I don’t know if that was neccessary in order to get my point accross, that’s a matter for another time. But when my Mother asked me to share these open letters with her, in Bengali, my native tounge; The first thing came to my mind, was that I can’t write my thoughts in Bengali the same way I write them in English. At the risk of sounding like a pretentious expert in language studies, I would say that’s because I can’t really figure out a way to construct long winded sentences in Bengali, without making them sound unnatural; I can’t create large “stacks” in Bengali, like I can do in English. So, naturally I ended up asking myself if I really need lather my sentences with layers of complexity just to get my point accross.

I didn’t really answer that, I cheated. I started recording a audio clip, like a podcast; Speaking out, like I would on a dinner table, or over groggy morning tea that my Mother brews.

And who knows... maybe I’ll start speaking English too.

Fantastic Features We Don’t Have In The English Language

Why the Chess Computer Deep Blue Played Like a Human - Issue 18: Genius - Nautilus

Umberto Eco And His Legacy In Open-World Games

Google’s AI won the game Go by defying millennia of basic human instinct