AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Natural Language Generation
 
 

A colleague of mine is interested in developing a process that would allow for the “automated” creation of hundreds of bots that can answer a predefined set of questions. He wondered if ChatScript (or another software platform) could generate grammatically correct responses using user assigned variables.

So to the question “Where is your pain?”

bot one would respond “My pain is on the right side”
bot two would respond “The pain I am having is mostly in my head”
bot three would respond “Well doctor, my pain is pretty diffuse”

I know we can create variables or facts for the pain location for each bot so that part seems fairly straightforward.

We could also manually write several variations of these forms and have CS randomly choose, but he wanted the responses to be automatically generated using rules of sentence structure, grammar, syntax, etc.

Is that possible?

Thanks,

Doug

 

 
  [ # 1 ]

I don’t see why not. Generating output from rules is simply having appropriate rules and initial data. CS can certainly be used to do it.

 

 
  [ # 2 ]

I’ve written a good bit on Quora about Natural Language Generation.  NLG seems to be most commonly achieved using templating systems, of one kind or another.  I’ve also done quite a bit of work with automated knowledgebase creation, but in AIML.  There is a field of study dedicated to natural language interfaces to databases, variously known as NLIDB or NLDB.  (At the end of the day, you’re talking to a database.)  Then there is the task of question generation (see questiongeneration.org).  Beyond question generation there is even hypothesis ranking.

Here is where the Turing test comes into play.  For instance, early work with NLG was attempted using Markov models essentially to generate random noise.  In other words, with NLG there are varying degrees of believability, from wrong answer, to right answer, to human like right answer, then to questions about (potential) answers, etc. 

 

 
  [ # 3 ]
Doug Danforth - Sep 3, 2015:

A colleague of mine is interested in developing a process that would allow for the “automated” creation of hundreds of bots that can answer a predefined set of questions. He wondered if ChatScript (or another software platform) could generate grammatically correct responses using user assigned variables.

Is that possible?

Doug,
This is pretty simple with JAIL. Natural language generation is a core part of the system and is used to generate stories, small talk and press releases.

In fact, I coded your example and added it with a couple additional variations to Skynet-AI. Just launch the bot and ask:

Where is your pain

It will generate a variety of responses each time. Of course, if you say the same thing 4 times in a row it will object, just say something different and you can continue.

 

 

 
  [ # 4 ]

Had a few minutes to expand the sample I did last night.

Transcript:
Usertell me about your pain
Bot
I think my pain is on the right side.
User: do you have any pain
Bot
I think where it hurts is right between my eyes.
UserWhat is your pain?
BotWhere it hurts is in the muscles of my right arm.
User: do you have any pain?
BotMy pain is pretty diffuse.
Userwhere is your pain?
BotThe pain is in the joints of my right arm.
Userwhat pain do you have?
BotWellwhere it hurts is on the left side.
UserWhere is you pain?
BotWell doctorwhere it hurts is on the right side

This simple example is done with 1 neuron via JAIL (Javascript Artificial Intelligence Language).

 

 
  login or register to react