Is SPA suitable for Nature Language Processing(NLP)?

Semantic Pointer Architecture
1

Dear All,

I have a general open question. Is SPA suitable for NLP applications such as question-answer/reasoning ?

I read several tutorials. It seems the SPA suits for representing complex structures with great scaling performances(with chained memory). Also, with high dimensions the system has a large space to encode conceptual semantic points. Therefore it suits for reasoning and Q-A tasks.

Anyone can comments on this?

Thanks a lot.

Best,
Jun

2

Hi @Ryan,

The SPA is a vector symbolic architecture, so it is capable in addressing any problems that other vector symbolic architectures have be applied to. This does in include question answering and reasoning, as well as general NLP.

While he was at the CNRG lab at the University of Waterloo, Peter Blouw spearheaded research into using the SPA for NLP tasks. His PhD thesis might be of particular interest to you.

Outside of NLP, the SPA has been used to do tasks like question answering and reasoning. We even have a question answering example on the NengoSPA documentation page. While this example is fairly simplistic, it does demonstrate the SPA capability of facilitating such tasks. The SPA was also used to construct the Spaun model which is able to perform reasoning tasks, a video of which is here. This video presents a good overview of the Spaun model.

1 Like
3

Hi Xchoo,

Thanks a lot for your help and information! It is very useful. I will have a look at details.

Best,
Ryan