Give me a demo for NLP

What I am trying to do is in a given paragraph I am trying to separate out the information which is given and the information which is sought. This information which is sought is not always truncated with a question mark. For example consider the following paragraph: --------- More than 5000 students have cleared the test. They are mostly in the age group of 10-14. The toppers are in the range of 10-11 years. We have no idea behind this strange pattern in the outcome of the examination. --------- My software should output me the following text : Given Data :1- 5000 students have cleared the test, 2- Age group of 10-14 3- Toppers are in the range of 10-11 years To find: Why this strange pattern coming up in the outcome of the examination ? .......... You can see that I get the output in two categories: Given data and what is to be found or what is being sought. How do we do this using automation using the NLP such as by Watson ? You will have to give me proof-of-concept, with just 60% accuracy. You will explain how you achieved it and what tools you used. If you can help me achieve this through IBM watson NLU. that will be great! Will be posting a series of projects once this poc is done! Budget is under USD 50 only since this is only a demo with low accuracy and efficiency. More details: We shall enter 50 paragraphs one by one. Each para that we shall enter will have two components or section: Given , and Sought. We humans can identify it by looking at it (even with and without question mark). we want to see if it can be done by machines. For example in the given paragraph the information which is being seek is not terminated by question mark. "We have no idea behind this strange pattern in the outcome of the examination." This is the information that should come under the category of" Sought". Other info should automatically go in : Given category I am expecting that we enter around 50 paragraph one-by-one manually, results wit categorization will appear in the tool itself , we take the screenshot of each one, and estimate accuracy by human expert, then report it as percentage of acceptable outputs. We need a minimal poc just for a demo. No coding is required if you are working manually with the help of a tool.