A Writer's Notebook, Day Nine-Hundred-And-Fifty-Four
I am still struggling with getting this project off the ground, but I am learning more and more each day. I have a good sense of what to do, but their are many little things that I don't understand enough about yet. It seems significant for me to do this without having to go far beyond just looking around the internet, as a part of the intellectual conceit of this endeavor. That is to say, a major part of why this is a thing I think I should do is that I feel it is very near to happening, whoever takes the step. If it were a project I felt would be impossible without years of study, I don't believe it would be worthwhile in the same way. That is not to say I won't spend more time on this, overall, but that if I can't get a basic example of my intended concept functioning in some rudimentary sense without such an education, the point is somewhat lost.
I may need to step back a tiny bit from some of the specifics of my goal, maybe doing a few less complicated things for a bit to get more acquainted with the work I am doing. I have a few ideas that I might be able to play around with, things that are connected to larger aspects of this project farther down the line, but which I think I might be able to begin exploring separately, but which may require less than the kind of language processing I hope to do.
Some of this is connected to a possible thought on training a system to recognize and compose metaphors. The idea would be to train a model on comparisons outside of language first, possibly. I am considering the idea of visual pairing as a task, where the items share a visible similarity even though they are different in other ways. The goal would be to create a network that can draw certain kinds of visual comparisons, then to apply a linguistic model on top of that, with the idea of comparison already established, even if in a limited sense. It is still a very loose conception, obviously, at least beyond the initial step, but I think I might be able to play around in this space, as I think visual recognition has different challenges than language, and many of the tools for that seem to exist already in a lot of the software I have installed now. Thus, I might be able to get a bit of this started immediately, using it as a chance to learn more about the overall processes and frameworks involved.
Comments
Post a Comment