SETrans to get AI and Nural

SETrans development is going grate. Pramod and I have been successful in improving SETrans translation accuracy. More and more it gets evident that we need to improve our approach to get get better results.

We have been using meta information of the language to detect patterns. We store patterns in one language and how it translates to the other (kind of re arrangement) and then do a dictionary based translation.

One problem we have is that some patterns from the source language, maps with the destination language, differently based on the context. Making things more complicated, the two languages pack information differently, meaning some times we don’t just have sufficient information to make a proper translation.

Solving this problem is our main focus for the next few months.

The plan is to have a daemon to monitor the context of the translation to try patch missing information. These assumptions made will be made available to the user who will then make a final decision.

Then the tackle the uncertainty in patterns, we plan to implement a neural based approach. This will take in to account individual words and build it self up from user corrections.