C R Hunt - Jun 12, 2013:

an NLP-based system should be able to figure out what “it” is. And answer questions like “Does Susie have a dog?”, “Does Susie’s dog have a collar?”, “Can collars be blue?”, etc. One could imagine extracting this kind of information using AIML, but then you’re simply using AIML for NLP! Why not have a more general parsing system designed for the task and save yourself the trouble?

The reason not to use a more general parsing system is that none have proven to be able to handle a simple conversational example, and the overhead has yet to justify their use.

I think this is the crux of the conversation, I think Victor and I once had a conversation where we concluded the NLP/POS people and the pattern matching people were just trying to do the same thing from 2 different directions.

In this exchange, there is a lot of data that the bot could be extracting (although the bot picks out the key lines in the conversation,  it doesn’t further digest the information). Ultimately you would like the bot to know after this conversation that:
‘Susie is a girl who has a dog with a blue collar. The dog does not like wearing the collar.’ You could also surmise that:
Susie likes dogs (although that man not truly be the case).
The dog does not like wearing the collar (versus the fact that the collar is blue or that Suzie gave the dog the colar).
I will really consider my bot a success when it can extract this data automatically.

It’s great how this conversation developed further, learning a lot

@Bruce: Thanks for correcting, NLP could include NL generation, but I meant actually just the understanding part to then match to an answer. Basically using NLP to enhance the understanding of the enquiry and match to a prewritten answer.

@Mike: Do you have somewhere defined this rising scale / maturity curve of NLP? Would be helpful to measure different systems against each other. The Gartner maturity level for virtual agents was a bit to fluffy for my needs.

@C R Hunt: I like your thinking. I have to say I see your view of NLP going further into the machine learning area. I’d be happy if bots/search engines would use a bit more than just keyword matching to understand my enquiries - learning from my enquiries is a whole big next step.

@Merlin: Good luck.

Bora - Jun 12, 2013:

@Mike: Do you have somewhere defined this rising scale / maturity curve of NLP? Would be helpful to measure different systems against each other. The Gartner maturity level for virtual agents was a bit to fluffy for my needs.

No, sorry. I just made up those examples for that post.

C R Hunt - Jun 12, 2013:

I wish I could just take a year off from my research and dedicate my full energy to this project! The problem with phd studies is that they’re supposed to absorb all your time as it is. Sometimes I have to set everything aside for months, or just put in an hour here and there. It’s frustrating, but I’m sure we’ve all been there at one point.

Maybe you are wasting your talents in the wrong field CR, but if that’s the case, take heart from the thought that many a physicist has gone on to do great things in computer science. You’re making a very good start, that’s for sure.

There are numerous parallels between the current topic of conversation and the evolution of science and technology over the ages. As I’ve mentioned elsewhere, one of my favourites is to be found in aviation. For centuries at least some physicists and engineers had the solution figured out (the aerofoil), but couldn’t make it work until other related technologies advanced far enough (the internal combustion engine and then the jet engine).

Meanwhile there was no shortage of other “experts” making a tidy living proclaiming how and why it could never be done, while yet others (let’s call them the “results oriented” folk shall we) were sustained by big bags of hot air or died jumping from high places while desperately flapping their arms.

There is still a lot of work to be done in NLP. While some of you are fooling around with stop-gap measures like pattern matching and statistical analysis, linguists are still arguing over how language actually works. It’s not even certain yet if there is such a thing as a Universal Grammar. There must be something innate (and therefore reproducible) or we wouldn’t be having this discussion.

We just have to keep working until we find it.

Andrew Smith - Jun 13, 2013:

It’s not even certain yet if there is such a thing as a Universal Grammar. There must be something innate (and therefore reproducible) or we wouldn’t be having this discussion.

Yes, a pretty big issue with English, and not only with regard to AI. I would say though that it is certain there isn’t a universal grammar as far as English is concerned. I think this originates from having those that believe we should follow original latin-based rules (such as not splitting infinitives) and those who see the evolution of the language as determining the grammar.

This should be an advantage for bot masters of other languages (for example, Spanish in Spain) where there is a universal grammar, having a single accepted national organisation that makes such decisions (much like UK has the Oxford English dictionary for vocabulary).

C R Hunt - Jun 12, 2013:

My goal now is to get a version of my bot online by the end of the year. At the moment it’s just not robust enough for general use. (Certain types of learning can’t be trusted to occur correctly in a free-flow conversation environment. Yet.) And I’m changing around the knowledge base structure fairly regularly still. But maybe by the end of the year it’ll at least be stable enough to let trusted users play around with it.

I made a mistake a few years ago when I spent ages putting together a Verbot-based bot that had the aim of breaking down input into grammar sections. I never actually put it online for testing. What I missed out on was all the potential useful feedback, and became too involved in something that stayed too theoretical. I now see the immense value in having others test, even if that means highlighting flaws in the system (which for me probably would have made me change direction for the better).