moos/wordpos

getNouns() is not accurate

dimavolo opened this issue · 3 comments

Thanks for this useful library! I noticed the getNouns() function is not accurate.

For example:

Please give this note to the man in the blue hat.

Returns the following nouns:

["give","note","man","blue","hat"]

It's true the word 'give' can sometimes be a noun, such as when a material has 'some give', but in this sentence, and 99% of the time, it's a verb.

Here's another simple example:

Can you ask her what time it is?

WordPos returns the following as nouns:

["Can","time"]

But of course a sentence starting with 'Can' does not refer to a metal container. And it skips the pronoun.

Would it be possible to improve the accuracy of getNouns()? I realize the answer might be no, as NLP like this is very hard.

Thanks!

moos commented

Hi -- I get this question a lot. I think I need to make the disclaimer in the README more prominent:

This has no relation to correct grammar of given sentence, where here only 'bear' and 'squirrel' would be considered nouns.

WordNet is just a lexicographic database of words -- there is no NLP here. The results you get from wordpos are the same as those from WordNet online.

All plural nouns will return as false. Tested with "cars" and "waves"

moos commented

@AusPrinzip wordpos is an interface only for WordNet. To get other features use natural's full library, e.g. http://naturalnode.github.io/natural/inflectors.html#nouns