Andjective-Noun composition learning to detect metaphoricity

Using this tool, one can train an adjective-noun composition function which can then be used to predict its metaphoricity.

First, one needs to provide a list of positive and negative examples on each line to train the model. When this process is complete, one can test the model using different adjective-noun phrases.

We recommend using at least 100 sample phrases, both metaphoric and literal. The insert button will load 500 samples randomly from this corpus. Note that this corpus only has compositions of the following adjectives with nouns which do not require extra context to be considered metaphoric:

heated, clean, strong, dense, clear, warm, soft, sour, brilliant, murky, solid, weak, icy, rough, bitter, smooth, cold, heavy, shallow, bright, deep, dim, sweet

In this demo, we are using these pretrained word embeddings. We only have limited word-vectors for this online demo. In order to load several words-vectors you might need to disable the ad-blocker in your browser.

This tool is using ConvNetJS to train the neural network model.