2nd Place Winner in Section 01: Spring 2019
by Team T-Posers
We are training a model using Keras as a high-level API to interface with TensorFlow. We started with a dataset mapping words to the language of their immediate etymological parents. The input consists of the vectorized characters of those words (process described below), and the output consists of OneHotEncoded language tags. Our goal is to produce a neural net which can predict the etymological parent of a word based on its internal orthographic structure. For example, our model might recognize that the consonant cluster "sch" is likely to have derived from German .
Commentaires