In this story, I want to talk about the abstract concept of a word. The meaning of it in different contexts. How the current Machine Learning algorithms understand it and why is it hard to achieve a general, human-like knowledge. I would like to illustrate my thoughts with the word _elephant. _I will donate the revenue of this post to save the species.*
The elephant in the dictionary
This is the definition of the elephant in the Merriam-webster dictionary [1]:
_1a: _a thickset, usually extremely large, nearly hairless, herbivorous mammal (family Elephantidae, the elephant family) that has a snout elongated into a muscular trunk and two incisors in the upper jaw developed especially in the male into long ivory tusks:
_(1): _a tall, large-eared mammal (Loxodonta africana) of tropical Africa that is sometimes considered to comprise two separate species (L. africana of sub-Saharan savannas and L. cyclotis of central and western rain forests)
— called also African elephant
_(2): _a relatively small-eared mammal (Elephas maximus) of forests of southeastern Asia
— called also Asian elephant, Indian elephant
_b: __any of various extinct relatives of the elephant — see MAMMOTH, _MASTODON
_2: _one that is uncommonly large or hard to manage
This entry can illustrate some hard challenges of understanding the concept of a word. Firstly, words might have many entirely separate meanings. Here, the word elephant has two meanings and the only connection between them is that both refer to something large. Secondly, 1a, 1b and 1a subtypes show that in a different context the elephant word represents different things, larger or smaller groups of animals. For a kid learning about animals, the difference between African and Asian elephant is less important than it is for a biologist (see a complete list of elephant species on Wikipedia). Lastly, reading and understanding this definition requires knowledge of words used in it (for example mammal, _Asia _and Africa).
#computer-vision #artificial-intelligence #machine-learning #nlp
