The Language Systems
- The language parsing bit was how I got into Cell Assemblies.
- The EPSRC project is actually called Natural Language Parsing
with Cell Assemblies: comutational linguistics with attractor nets.
- I probably spent the most time on parsing in both CABot1 and 2.
- In CABot1 I parsed using a stack. See
Parsing with fLIF Neurons
- Word CAs were activated, and links to them got stuck on
- This requires variable binding. I do it, I think uniquely,
via Short-term potentiation. This is a paper I'm still trying
to get accepted by Connection Science.
- Verb frames were used to store the dynamic structre.
- Erasing happened automatically over time (STP) so the
mechanism could be reused.
- Unfortunately this erasing took a long time.
- So, for CABot2 I built a stackless parser.
Pyscholinguistic Model of Natural Language Parsing Implemented
in Simulated Neurons.
- This took advantage of memory decay
to replace the stack
- It turns out that this parser attaches PPs via semantics,
generates semantics, and behaves in human like times.
- It's also pretty fast.
- I'm only making minor (grammar and lexicon) changes to the
parser for CABot3.