Pattern-spotting

This week's readings have been my favorite of the quarter so far. I particularly enjoyed "Building on a corpus" reading because it made me think about machine learning in contrast to natural learning.

I think that humans are naturally pattern-spotters, and that that's what makes us so intelligent as a species. I think we are able to notice patterns in word usage that help us create language in turn. For example, the sentence "the cell lights shuddered" conveys a particular meaning to me because 'shuddered' is an action that I associate to animate objects who can internally cause events. When I read that sentence, I infer a personification or a use of poetic language -- I believe that's partly been a consequence of the particular way I've been exposed to that word and its near-synonyms.

I agree with the authors that these patterns we notice inform our lexicon and therefore our word selection. It's interesting that this is knowledge acquired and not learned -- we gain intuitions about words and concepts by being exposed to language. These intuitions may be partially true or partially consistent with the general usage: sometimes we might use a word in an unorthodox way because our pattern wasn't detailed enough. This might be written off as mistake most of the time, but sometimes we might even push our language in a new direction.

The fact that computers could go through a similar process of "training" by analyzing a large corpus of electronic writing intrigued me from the beginning of the reading. It then became clearer that computers were just an aid for human intuition.We really haven't found a suitable proxy for this intuition -- the process by which we interpret the patterns and modify our mental maps accordingly. When we get to the point where a computer might notice on its own why "the house shuddered" sounds poetic, we will have taken a huge step towards conversational AI.