Question

I am having this problem with word boundary identification. I removed all the markup of the wikipedia document, now I want to get a list of entities.(meaningful terms). I am planning to take bi-grams, tri-grams of the document and check if it exists in dictionary(wordnet). Is there a better way to achieve this.

Below is the sample text. I want to identify entities(shown as surrounded by double quotes)

Vulcans are a humanoid species in the fictional "Star Trek" universe who evolved on the planet Vulcan and are noted for their attempt to live by reason and logic with no interference from
emotion They were the first extraterrestrial species officially to make first contact with Humans and later became one of the founding members of the "United Federation of Planets"

Answer 1

I think what you’re talking about is really still a subject of burgeoning research rather than a simple matter of applying well-established algorithms.

I can’t give you a simple “do this” answer, but here are some pointers off the top of my head:

• I think using WordNet could work (not sure where bigrams/trigrams come into it though), but you should view WordNet lookups as part of a hybrid system, not the be-all and end-all to spotting named entities
• then, start by applying some simple, common-sense criteria (sequences of capitalised words; try and accommodate frequent lower-case function words like ‘of’ into these; sequences consisting of “known title” plus capitalisd word(s));
• look for sequences of words that statistically you wouldn’t expect to appear next to one another by chance as candidates for entities;
• can you build in dynamic web lookup? (your system spots the capitalised sequence “IBM” and sees if it finds e.g. a wikipedia entry with the text pattern “IBM is … [organisation|company|…]”.
• see if anything here and in the “information extraction” literature in general gives you some ideas: http://www-nlpir.nist.gov/related_projects/muc/proceedings/muc_7_toc.html

The truth is that when you look at what literature there is out there, it doesn’t seem like people are using terribly sophisticated, well-established algorithms. So I think there’s a lot of room for looking at your data, exploration and seeing what you can come up with… Good luck!