Semantics

• One of the big problems, is that the scientific community does not really understand semantics or meaning.
• Note that most parsers work largely (or entirely) on the part of speech of each word.
• It's not entirely clear what each word means, much less a sentence, or a paragraph.
• One resource that is useful is WordNet. It's a dictionary with a semantic net placing words in a semantic hierarchy.
• Bag of word techniques like Latent Semantic Analysis are also useful.
• You take a corpora, and you build a matrix of words by documents.
• You then mark the cells where words occur in documents.
• The meaning of the document is the vector of the words, and the meaning of the words is the vector of documents.
• LSA reduces this to two much smaller matrices, and thus compresses the vector.
• While these are useful, they don't really give the meaning of a word.