Ask me about CFG generators

**trigotron** · 07-14-2007 05:18 AM

Ok, if you're intrigued enough by the obscure title to actually click on it, here's what this is all about:
CFG - Context Free Grammar

Basically i worked with a guy on a computer program a while ago (the name of it is classified for personal/legal reasons) that writes essays without any prior knowledge of the subject matter. To put it simply, he created an artificial intelligence that could write essays, its use is very limited and specialized, but it is very effective when used correctly. Ask me anything you like about the program, the output, the research involved, whatever.
http://automaticessaywriter.150m.com/index.html

There is a sample of one of the essays here:
http://automaticessaywriter.150m.com/Complete.html

The only information entered was the name of the novel, and the name of the author to write that text. This actually won me and my associate the title of "most likely to bullshit".

**Capoeirista** · 07-15-2007 01:15 PM

How do i get this to do my homework?

**pokilty** · 07-19-2007 04:40 PM

I'm very impressed that you two were able to construct such a powerful CFG! I think I wouldn't have even attempted such a huge project. I have some questions that help me understand how you turned this into a reality.

Are the leaves of the CFG derived from words in an English dictionary?
What data was the grammar initially constructed off of?
What programming language and/or optimizations do you use to generate the essay?
Can the program be used to input an essay and determine whether it exists in the grammar you defined, possibly to be extended as an English grammar checker?

I recently took a course on Finite State Machines, but we didn't explore any examples more complex than a CFG for palindromes

**trigotron** · 07-19-2007 07:16 PM

Originally Posted by pokilty

I'm very impressed that you two were able to construct such a powerful CFG! I think I wouldn't have even attempted such a huge project. I have some questions that help me understand how you turned this into a reality.

Are the leaves of the CFG derived from words in an English dictionary?
What data was the grammar initially constructed off of?
What programming language and/or optimizations do you use to generate the essay?
Can the program be used to input an essay and determine whether it exists in the grammar you defined, possibly to be extended as an English grammar checker?

I recently took a course on Finite State Machines, but we didn't explore any examples more complex than a CFG for palindromes

-Yes, it takes words from different sections of a word bank of about 10,000 words arranged into different parts of speech and word types (adjective, verb, noun, etc).
-The only data that needs to be inputted at the beginning is the name of the author and the name of the book. You have the option to input more, like a thesis, or symbolic devices the author uses and stuff like that, but for the most part, it comes out better if you let the computer do all the randomizing. If you mean what data does the program use to construct grammar, it's just a large bank of sentence structures that i got off of another program.
-I used Java API, and it's not very well optimized to say the least =P
-I toyed around with the last idea, of inputting premade essays into the program, analyzing them and taking the structure of them and putting it into a structure bank of sorts, however this proved to be very difficult and i abandoned the project. I also toyed around with the idea of the program generating its own sentences by generating a random sentence with a randomized structure, then asking a user how grammatically correct it was, taking the user input into a function, then kindof narrowing down the grammar until the program could generate coherant sentences on its own, however this fell through also.

**pokilty** · 07-19-2007 08:19 PM

Originally Posted by trigotron

If you mean what data does the program use to construct grammar, it's just a large bank of sentence structures that i got off of another program.

Yeah, this is what I meant, thanks!

Originally Posted by trigotron

I toyed around with the last idea, of inputting premade essays into the program, analyzing them and taking the structure of them and putting it into a structure bank of sorts, however this proved to be very difficult and i abandoned the project. I also toyed around with the idea of the program generating its own sentences by generating a random sentence with a randomized structure, then asking a user how grammatically correct it was, taking the user input into a function, then kindof narrowing down the grammar until the program could generate coherant sentences on its own, however this fell through also.

I'm faced with problems like these all the times in my programs - a lot of times I think it's just infeasible without a strong background in statistics. Time to pay Google a visit!

Anyway, very nice job

I hope the experience you gained helps you achieve more in the future!