How can NLP tell the word "stool" is closely related to the word "dirty?" When the word relationship is based on a corpus, most likely the word "stool" is related to some medical terms such as "bloody" "pale" "solid," etc. But the defining character should be "dirty" in people's mind. I wonder how NLP handles that? i.e., given the word "stool," return a high score of relatedness to the word "dirty."

Thank you.

  • For relatedness, see third point here. For "character" or associations, much harder, because these are vague, often individual and, most importantly, mental conceptions that are not directly encoded in the language (of course any meaning is somewhat purely conceptual, but lexical relatedness manifests stronglier in language than associations and emotions), you'd need a lot of world knowledge in order to automatically detect such connotations which is much harder to do than checking for semantic similarity (like co-hyponyms). – lemontree Jul 19 '16 at 20:17
  • That's you again, Lemontree, thank you very much for replying! It's intuitive to me that hoping the NLP to figure out emotions is a stretch. That was not my intent of the question. My original question is more like whether there is an NLP technique to recognize a piece of common sense, e.g., "stool" is "dirty" This relationship is not vague, not unique to individual. – bizbuzz Jul 19 '16 at 21:16
  • So the only reason for this to be difficult, let me paraphrase your words a little bit, is that a common sense like this will just be in the back of people's mind, and it will unlikely to be explicitly said "stool is dirty" many times in daily communication, and therefore the evidence of this fact will not exist in any corpus (meaning "not directly encoded in the language"). Is that a correct understanding of your reply? Thank you. – bizbuzz Jul 19 '16 at 21:18
  • Do you know what "stool" means? – Gaston Ümlaut Jul 19 '16 at 22:48
  • stool in the sense of "feces" – bizbuzz Jul 20 '16 at 1:29

The closest answer to your question I can think of would be Word2Vec. The language models produced by it tend to capture semantic relationships between words very well. They can answer questions such as "woman is to queen as man is to X" or calculate the "semantic distance" (whatever that means exactly) between two words. Among others, there are implementations in C and Python. The latter is introduced here.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.