Say I wanted to learn a new language and I had a data set of images w/ the English labels and the same images w/ labels in the new language.
Would it make sense to use a neural network to back out a dictionary of words in the new language with the English translation?
For example, say I have newspapers from around the world with the same picture captioned in different languages. I could prepare this data set by using a model to determine if the images are all of the same thing – or maybe it’s really the same image.