Experiments extracting semantic information from the WordNet

pip install -r requirements.txt

Using the semantic relationships between entries in the wordnet to to extract semantic relationships between synset entries. This work is meant to serve as a proof of concept of how strengthening the wordnet in terms of accuracy and vastness can be beneficial to provide researchers/developers with more information about simple semantic relationships between objects in simple phrases.

As with all Knowledge Bases, the inference of this system is limited by the totality of it's own enumeration. In other words, I cannot make inferences about terms that are not in the WordNet. In addition, all the inferences made my program are not necessarily correct semantically.

I am using the wordnet to detect simple examples of figurative speech.

How it works:

Grammar: NP1 + conj('is') + NP2

A car is a motor vehicle -> fact
A vehicle is a car -> a false overgeneralization Love is war -> figurative speech

Fact

If NP2 is a direct Hypernym of NP1

Falsehood

If NP1 is a direct Hypernym of NP2

Generalization

An indirect hypernym and fact or falsehood

Figurative

Two Wordnet entries with non-common roots

-The accuracy of Pattern's POS Tagger

i.e. (love is a nutrient)

Parse tree: [Sentence('Love/NN/B-NP/O is/VBZ/B-VP/O a/DT/O/O nutrient/JJ/B-ADJP/O')]

-limited by entries in WordNet

i.e. (entries not in WordNet)

-No support for pronouns/people

i.e. The gender problem (she was George Washington... is figurative?)

-deep philosophical questions

Your brain is a computer -> figurative speech (two entries, with no roots)

-using recursive roots of the word net to make inferences

The kids were monkeys on the jungle gym -> is a verifiable falsehood

Pattern 2.6

Syntax: python program, sentences to check, name of ouput file

In Terminal:

python figur_detection.py common_metaphors.txt common_meta_test

Develop web interface for user friendly processing