This repository contains the code that I used for a research in the capita selecta ai course at radboud university. In this research I wanted to see if I could learn a word to word translation of two languages without the use of any examples. Instead I assume that languages share certain patterns in which certain concepts or words would occur. And thus by matching these patterns we could see which words are likely to be in

The scripts in the code folder. This scripts can be run stand alone(as a batch) and imported in other code. In the notebooks I developed the code and provide a more detailed explanation.

All my recurrent networks are implemented in Chainer.

My code consists of 3 parts:

• Wiki dataset: Where I obtain my datasets by extracting text from a wikidump, using the chainer helper functions
• Language model(lm): Where I train a simple language model that predicts the next word in a sequence. Using the dataset obtained from wikpedia and the pbt tutorial from chainer.
• Unsupervised word 2 word(uw2w): Where I use the network parameters of one language to predict the words of the other language. While doing so I learn a linear transformation matrix that maps words from language a to b.