I'm starting this blog as an experiment. My intention is that writing here will help me during my research, however, I'm not sure if this will work in any way. The content may change over time, same as my thoughts. Here, I will try to structure and explain all the ideas that came to my mind --often they do it in non-linear way, so writing them is hard--, for they not stay abandoned in my book of notes.
My research, which should lead to a doctoral dissertation for me to get a PhD in Computer Sciences, is about natural language, more precisely about natural language understanding for robots; so this blog will be (mostly) all about it. The reader, if interested, may find the tale in one of the blog entries.
First thing my readers --such a thing will ever exist-- must know is that, even though I'm writing in it, English is not my native language, but thats something the reader have already guessed. It is not because I'm expecting to be read, I'm humble enough to know I'm just another PhD student with delusions of greatness whose ideas, problems, and self are completely irrelevant to the world and its people. This is written in English for mere skill improvement, but on that I will write later.
I would recommend the reader to check first the tale to get a better understanding of what I try to explain in the following lines: the problem my research addresses. Though it has grown and shrink over time, the ultimate goal to give machines the ability to understand language as humans do, and by understand I mean a really understanding of what is being told, more or less in the same sense of what Jimmy understands when Jimmy's mom asks him "make some fusili pesto for dinner". To Jimmy is completely unknown what fusili pesto, how is cocked and the exact time for dinner, even the word "cook" is not mentioned, but Jimmy knows (1) he has to cock something, (2) needs to investigate what and (3) figure out how. The research is about trying to find (1) and if possible also solve (2), while how to accomplish tasks is completely out of scope.
All sources I've found to this point somehow bypasses the problem with no clear lead to a solution. The machine knows already all the words will be dealing with (or their categories which relations are fixed when the machine can learn new terms), which is not the case for Jimmy, or it uses a super powerful blackbox that computes the probability of a sentence and may require quite some time to learn new lexicon which also is neither Jimmy's case. Nowadays the second scope has proven extremely efficient and is used by huge companies like Google, Microsoft or Apple to power up their assistants. In this regard, I line up with Noam Chomsky since, even if those methods are an impressive achievement, also are completely irrelevant to science since there is no advance in the understanding of the cognitive process of language.
My research suggest that, in order to solve (1) the machine must be given the ability to learn by itself a language. What we know is that humans born ignorant of any language, but they have the ability to learn them so I believe (this far is a belief, a theory, I've not enough human brains at my disposal to be 100% sure and I'm not sure it will help due to the differences between living tissue and silicon circuits), as it does Chomsky, that there is a machine in the human brain that renders it capable of learning languages. If modeled, even when the model was imperfect, this machine could be artificially replicated or at least provide new insights to understand the language acquisition phenomenon.
However, the behavior of such machine cannot be observed directly. For this reason, and considering that this machine is designed to produce another machine for language recognition and generation, it might be only possible to try to guess its behavior by looking at the patterns and similitudes in the machines it produces.
Find a model, that's the problem.
No comments:
Post a Comment