Future Work
- The first thing I need to do is get CABot3 working.
- From the parsers perspective, I need to put something together
so that instances are better used. This means I don't have to
reset the STP synapses for variable binding.
- I also need to stick in some new sentences.
- Beyond that, the parser has a lot to improve. I'd like to
fix up phrase combination rules so that I don't need a rule per
instance pair.
- I could also increase the coverage by engineering rules, but
I think the really interesting thing is to learn rules.
- We've already submitted a follow on grant, CABot5, to EPSRC.
- In that, we've got a segment on learning grammar rules from input.
- This should combine variable binding into rule learning.
- This is a shot at Universal Grammar. (We just say we'll try
in CABot5.)
- Here we're not just engineering a solution to learn rules; they
emerge through interaction with the environment including language.
- It's proposed that CABot4 and CABot5 are conversational agents,
so they can produce text.
- This isn't the only type of learning we're looking at for CABot5,
and learning is the main point of that project.
- I'd also like to look at mechanisms for backtracking, and
more psycholinguistic modelling (not part of CABot5).
- This slide could run for pages, attention, improved vision,
speech recognition, spelling correction......
- In the long run, I'd like to take a shot at the Turing test.
