This document presents a method to predict whether words are used literally or sarcastically based on word embeddings. It collects target words and their contexts to learn embeddings. It then uses the embeddings to disambiguate literal and sarcastic senses of words using distributional and classification approaches. The distributional approach calculates cosine similarity between word embeddings, while the classification approach trains models on sentence embeddings. Evaluation shows the approaches perform well in predicting literal and sarcastic meanings of words. The authors conclude and discuss directions for future work.