Re:Ask HN: Are there any NLP projects that use context?
by wodenokoto on Sunday April 19, 2015 @11:19PM
(#9405980)
There are a lot of ways of creating context in NLP. When a normal person thinks of contex, they probably consider the meaning of the sentence and from there they'll figure out the POS.
For a simple POS-tagger you would today probably use an HMM. What this means in simple terms in you look at how likely is the next word (regardless of what word there may be) to be a given category. Let's say the next word has a 50/50 chance of being a verb or a noun. Then we look at the actual word "project" which has a 20/80 chance of being a verb or a noun.
In another context the POS might have a 90% chance of being a verb, and then project would probably end up being. Tagged as such.
You can also start looking at bi, and trigrams (series of two or three words) and so on.
So what you describe is pretty much an actual thing.
For a simple POS-tagger you would today probably use an HMM. What this means in simple terms in you look at how likely is the next word (regardless of what word there may be) to be a given category. Let's say the next word has a 50/50 chance of being a verb or a noun. Then we look at the actual word "project" which has a 20/80 chance of being a verb or a noun.
In another context the POS might have a 90% chance of being a verb, and then project would probably end up being. Tagged as such.
You can also start looking at bi, and trigrams (series of two or three words) and so on.
So what you describe is pretty much an actual thing.