Figurative Language Classification of English Manner of Motion Verbs for a Machine Translation System

Being able to differentiate between figurative and literal words and phrases is useful for a variety of natural language processing tasks, including machine translation. In this project, I implemented a linear classifier with a logistic loss function to distinguish between figurative and literal uses of English manner of motion verbs. The feature inputs to this models were the lemma of the manner of motion verb, the types of syntactic dependents attached to it, and the semantic hypernyms of the verb’s syntactic dependents, extracted with WordNet. The linear classifier achieved 86% accuracy on a set of test sentences, compared to 77.5% baseline accuracy of a bagof-words classifier. Finally, I test the effect of adding figurative/literal annotations to English manner of motion verbs on the performance of the Moses machine translation system. I find an increase in BLEU score for annotated inputs, but discuss issues with the experiment that make this result difficult to interpret. Note: Some of the work for this project was done in conjunction with my final project for CS221: Artificial Intelligence: Principles and Techniques. The work for CS221 mainly comprises the implementation and analysis of a linear classifier model. The work for CS224N comprises the feature design and extraction, the in-depth error analysis of the classifier’s performance, and the experiment with the Moses machine translation system.