It’s been a rough week, but I wanted to get an update out if anyone actually reads this or cares.

I’m still experimenting with different methods, but I think I found something that fits my needs.

It would actually be similar to a translator in that it converts text into a binary representation.  But the difference is that the binary representation is not the black magic of a deep neural network, but rather an structured data array.

The data array represents a “concept”.  It is not a word-for-word map of the input, but rather a conceptual map of the meaning of the sentance.

The biggest barrier to this approach is the lack of extensive training data.  I’m having to write it all by hand.

The goal will be to allow the system to supplement that concept with a “common sense” knowledge base.  For example: an old abandoned hut may have a damp, musty smell.

The next phase would be having the system produce text containing the amalgamation of the defined scenario with the knowledgebase.  This will allow our “translator” to take simple descriptions and “translate” them into rich descriptive narrations.

This also lays the groundwork for converting player input into an intent concept.

 

 

Categories: Machine Learning

Leave a Reply

Your email address will not be published.