Help with NLG project!

My name is Carl Strathearn, a postdoc working on an NLG project and I was wondering if someone would not mind giving me a bit of advice on the best way to generate text from objects and their common storage locations. I am fairly new to NLG (I started in Sept 2020) as I come from Human Robot Interaction, but I want to create a masked model for household tasks in Colab i.e: The Kettle belongs in the [MASK]: 0.434 Kitchen, 0.214 Living Room, 0.123 Bathroom. The target would be kitchen, and the next probable Living room ect. The idea of the project is to create a system that can be embedded in an assistive/care robot to help people with preventative disabilities locate objects in their home. Does anyone know of any model or methods that I can use to achieve this? Your advice would be greatly appreciated.

I haven’t explored this but BERT (Bidirectional Encoder Representation of Transformer) might be of use to you.

Here’s a demo of how BERT might work (I added Kitchen or Living Room or Bathroom at the end in this example as otherwise it suggests a separator for the MASK value). You will have to look up how to change MASK’s association to left-only or right-only instead of keeping it bidirectional.

I found this Google Colab Implementation and related paper. Hopefully this can help you get started at the least.

1 Like

Thank you !! I really appreciate your help.

1 Like