A Writer's Notebook, Day Nine-Hundred-And-Fifty-One
As mentioned last night, I am endeavoring to create a computer program, bot of sorts, that is modelled from my writing. I spent a good deal of time today on learning more about computer learning, and considering my goals with this project. It is not something that is short term, of course, as I do not think that I can even conceive of an end state, but I can think of many steps along the way. My initial intent is just to get something functioning to write poetry that replicates the qualities of my writing, even if not fully sensible. It is possible, I know, to do such things, and I have seen programs to emulate various writers, though the results are not usually meaningful. In most cases, the network has discerned a pattern of language, but does not have a real sense of what that language is intended to do. It is only stringing together words based on what language it has seen, but with no reference point for the language outside that framework. It does not know what any of the words are about, and so it only can emulate a style and structuring of the language itself. Still, it is an impressive thing, I think, and if I can get that done first, I will be quite pleased.
Of course, once I have achieved something of that sort, which I expect shall be rather simple, as I have already collected a number of examples doing similar with corpus from around the internet. In truth, I am having most of my difficulty in getting the programming language set up. I need to download various libraries and install them, and doing it all has proven somewhat of a challenge. It is just a matter of spending a bit of time, though, and then I will be able to get it started.
Assuming that goes well, I am not certain the next exact step, but I am thinking that it might be possible to use the trained network to generate text for a content driven task, such as using it for description or summarization, thus providing some level of meaning. It may be difficult to do such a thing, I am not certain, but I expect some version of this would be plausible. I am also wondering if it could be used in creating a translation tool, where I can put in other English and have it come out transformed through the language model, in essence crafting a tool that just replicates meaning in a new style. It may be that I can combine some of these tools to make the output more realistic.
That, of course, asks the real question at the deepest heart of this: can such a program ever be brought to a point of genuinely producing anything real, with actual meaning, or is it always going to be an illusion? I do not know, and I am not certain I could ever know, no matter what it writes, but I am going to try. There is a great deal behind this, but I think a further discussion shall need to wait. For now, I need to focus on getting things up and running, if only at the most basic level.
Comments
Post a Comment