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..

Are definitions of words inherently weak?
 
 

Hi,
I’ve been pondering this question for some time. I read Jan’s post tonight along with his white paper and thought I would bring this topic up.

The reason I ask is that the brain is able to solve problems by substituting one object with another object. E.g. use a chair instead of a ladder (if not available) to climb up and grab something. 

Here are two definitions to illustrate my point.

Ladder - steps consisting of two parallel members connected by rungs; for climbing up or down
Chair   - a seat for one person, with a support for the back

The best chat bot would most likely fail to see that a chair and ladder, in the context of extending ones reach, serve the same purpose.  A chair may be substituted for a ladder.

Other objects that may be substituted but have definitions that are not similar.
* glass and bucket - moving water from point A to B
* tree branch or golf club - to whack a bad guy in self-defense
* concrete blocks or 4x4 pieces of lumber - for raising something off the ground

The list goes on. 

I believe definitions are inherently weak when used by themselves. I believe it is necessary to ‘supplement’ definitions in order to get a chat bot to ‘seem’ more human.

I’m curious what thoughts you all have on this topic.

Regards,
Chuck

 

 
  [ # 1 ]

That’s why, in my bot, I have several “properties” for each word.

glass, bucket, tree, block - they all have property of “pos”, with a value of “noun”.
BUT, glass, and bucket could have property of something like “can-hold” with value of “water”, where as block wouldn’t really have that property.  “tree” could have perhaps a property of “living = true” where glass, bucket wouldn’t have.  But all these DO have pos=noun.

 

 
  [ # 2 ]

Interesting statement. I’d agree with Victor, to do this using properties + doing a useful search on the database of known objects.
With respect to a weak definition, I don’t think it’s really ‘weak’ as you say, but perhaps more ‘multi faceted’.
What’s also interesting is this: you can have a chair in general, or a specific chair (my chair), which can have different properties.

 

 
  [ # 3 ]

Yes, or the chair of a meeting… or to chair a meeting (verb, well infinitive phrase)

 

 
  [ # 4 ]

Hi,

I suppose its a big enough challenge determining the intended meaning of the word ‘chair’ in a sentence….like the example usages you both mention above.

I agree with the notion that an object (e.g. bucket) is multifaceted in the sense that an English definition is generally inadequate. It only describes one facet.  The property (“can-hold”) is a good idea in that it provides an additional facet of information unrelated to the definition.  And each object can have more than one property.

So, a scoop, trashcan, cup, tank would all share this property “can-hold”.  So if the human expresses the need to store something and needs a suggestion, the bot could iterate through all the properties and suggest something ‘unconventional’.  How many of us use old coffee cans to store various bits in the garage or basement?

I ramble on…

Regards,
Chuck

 

 
  [ # 5 ]

Yes, and this is exactly how my bot understands words… I have all kinds of ‘weird’ property names… whatever does the trick.

yes, a lot of words share the same properties.  So the real “meaning” of a word is its unique SET of properties.

Words that mean pretty much exactly the same thing, will pretty much share all the same properties (and values of those properties).

if X and Y are synonyms, then the set of properties that X has will also be shared with Y

And yes, I agree, word definition are incomplete… they assume all the experiences we collect over the years - which is not a problem for humans, but is for computers!!

 

 
  login or register to react