What do you mean with granularity ?

Granularity is a theory introduced by Hobbs 1985 see this site

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.74.7822

It simply seeks a multi resolution or multi-scale representation of information.

For example in a paper, you can represent a section by one word, like introduction, and then you can expand it into several words or granules, like problem definition, previous work and paper over view. Yet you may expand the modules untill you count words and then count letters. Granularity is important for machine learning of high information content such as images or text based documents. I am searching metrics used to identify it in text or other domain.

On topical granularity, you could try segmenting, see e.g. http://www.jlcl.org/2012_Heft1/jlcl2012-1-3.pdf

Thanks Chris.

I want to apply the concept of granularity to reach evoloving intelligence from the collection of small set of rules. For example we can associate a general meaning or trend for a group of words. When this is done in heirarchical manner, we should be able to reach highly grannular automatic text understanding, and should help for mining large scale quantity of text such as a sceintific article. The same concept for behavioral robotics theory which decomposes the robot complex behaviors into smaller basic behavior. I think of grannularity the same way. A grannule can be a simple rule of assignment, and collection of grannules in the hierarchy will give highr evolving intelligence.

In the view of set theory, granularity is the size of granule, and a granule is regarded as a element of power set of set.

Suppose you are buying a pair of shoes in a shop. During this, your girl friend are waiting for you in a 'café'. You are late because the sender  is an old friend of you. She calls. What do you answer ? : "I just end buying a pair of shoes" ? or "I am speaking with my old friend Paul, you know, the shoes seller". Probably with the first sentence (at least, I advise you to do so). Conclusion : granularity depends on the action you are claiming to do (and not necessarily you are doing ; all depends on the event you study), i.e. on the hierarchy it belongs and granularity of an analyse depends on your theoretical hypothesis. So there is no categorical answer to your question. It's always relative to your goals.

Thanks Trognon, 

I agree totally with your nice example.

My question is about the system of representing the knowledge and the work done to relate it to the goals.