Natural Language Processing

I need a NLP software that rewrites text. It can be written in any language (PHP,Python and Java are preffered) as long as it can run both on Linux and Windows XP and newer. If you use any libraries/modules/toolkits let me know which they are and their license. Basically free software is prefered and licenses that can be used for commercial projects for free (ideally without having to share the code back to the open source community - LGPL vs GPL). I am not familiar with NLP (but I am a programmer) so this description and the ideas I wrote here might not be the best ones. I expect you to help me understand what exactly can be achieved from what I want and how you will achieve it. Please read through this brief description and if you think you can develop what I need, message me and we can continue discussing the details. We can also use the chat feature on vWorker to talk, just send me a message to let me know that you're in, because I won't stay logged in there all the time. Read below more details about what I want. ## Deliverables What I want: Basically the user inputs an article and the software adds synonym words and phrases to the existing ones. Ideally it should be able to generate as many rewrites that use different words and maybe even paraphrase, while keeping the resulting text grammatically correct and maybe even make sense. Simple example: Input: John is going to drive his car on the highway. Output: John {is going to|will} drive his car on the highway. This is a very simple example that illustrates synonym addition while maintaining the meaning of the sentence very close to the original. The program however can use synonyms that would change the accuracy of the resulting sentence compared to the original but still makes perfect sense. For example: Output: John {is going to|will} drive {his|some|a} car on the {highway|street|road|boulevard}. Ideally, the program should be able to generate completely different sentences with the same meaning, replacing not just words but phrases or even build equivalent sentences. Example: Input: As a matter of fact, my new monitor is quite big. Output: {The thing is|As a matter of fact}, my {new|newly bought|recently acquired} monitor {is|happens to be} {quite|pretty|fairly} big. Then, I will feed this resulting text with curly braces into a program that will choose just one word from the braces basically generating one version of all possible sentences. I am not familiar with NLP but I believe what I need can be implemented. At least that is what I consider after playing a bit with the Natural Language Toolkit for Python and reading through a NLP book. I expect you to tell me how smart the system you will build is going to be, its strengths and limitations and suggest ideas about the implementation. I can provide a database of text (text corpus) and ideally the program should be able to recalibrate itself to a new text corpus for each project. For example if I want to rewrite articles about airplanes I can train it with several hundred articles about airplanes. If I want articles about pets I can train it with several hundred articles about pets. That way the accuracy of the program will be higher. The bottom line is that it is more important to generate a high number of rewrites than it is to maintain logical accuracy. However, ideally the text should be readable and make sense. If that is not possible, it has to be at least grammatically correct. Ideally there should be a way to instruct the program to give more weight to accuracy in the detriment of total number of rewrites that can be generated, or the opposite. Aside from the database of texts that can be used to train the program for a particular topic, it should use a database of synonyms like WordNet and also a web service that I will provide, which will return synonyms not just for words but also for multi-word phrases. These synonyms and phrases were extracted from pieces of text that were manually spun (actual people took an article and manually wrote it in the format with curly braces in a way that makes sense). In the worst case scenario the program should couple the database of synonyms with a Part of Speech tagger or something like that so that when adding synonyms, only the ones that make sense from a grammatical point of view are used. Like for example if I have the expression "computer boots up fast" I don't get something like "computer shoes up fast". The program could also maintain a database of N-grams (I believe that is what they are called) of word+PoS pairs so that it can add synonyms taking into consideration the previous and next word(s). These are just some ideas that may or may not make sense to be used, but it is the best I was able to come up with considering I am not familiar with NLP. You should also take a look at www*article~queen*com (remove ~ and replace * with .). In the worst case scenario, the program you will build for me has to do at least that.