Deep Learning as tool to distinguish words for sharp and round objects in natural languages

Sound Symbolism is a well studied psychological phenomena of the relation between sound and meaning in natural language. Though the phenomena has been studied by psychologists and linguists, the phenomena has not been put to use natural language processing or modeled by machine learning. In this work we select words for round and sharp objects from various natural languages. We attempted to see if a machine learning algorithm could perform better than Chance in distinguishing words for round and sharp objects in natural languages. We performed a psychophysics experiment to see if human subjects will associate words for sharp objects with a round object and round object with sharp figure. We show that human subjects are more likely than chance to associate words for sharp objects with sharp figure and vice versa. We propose that the algorithms can be improved by using training sets consisting of words whose sound symbolic properties are labelled by psychophysics experiments.