Well put !  I agree that probably both approaches could end up being as successful as each other!  I highly agree with your statements that fuzzy/pattern matching suffers more from false positive problems than grammar based.  And yes, grammar based can suffer more false negatives (NOT saying something when you should have).  The approach then, is a matter of taste I guess.  I wonder though, which will require more work, making a ‘crisp’ system more fuzzy or making a fuzzy system more ‘crisp’?

And by ‘more work’ I mean both human coding & data entry, and work for the machine.  Since human labour costs more than CPU time, I suspect we all agree that we want the machine to do the work, self teaching system.

What I will probably do to make CLUES less strict, is to have the “did you mean ____?” functionality.

Example, if it generates for a given user input, <str>, 3 parse trees :

pt1
pt2
pt3

Now, for those 3 parse trees, if there is no semantic match, it finds a given parse tree, where:

pt1 - 2 nodes are different than a suggested parse tree
pt2 - 5 nodes are different than a suggested parse tree
pt3 - 8 nodes are different than a suggested parse tree

The ideal parse tree with proper grammar differs by only 2 nodes than the parse tree of user input, then I can convert pt1 into str_1, and respond “did you mean <str_1>?”

In my opinion, a bot should always ask “did you mean ____”, especially in a production system, instead of just assuming, and taking some (potentially costly) action !, which could be disasters in real life !!

Since human labour costs more than CPU time, I suspect we all agree that we want the machine to do the work, self teaching system.

Agreed. Although in Bot land, “self teaching” may not be the same or as desirable as “supervised learning”.

In my opinion, a bot should always ask “did you mean ____”, especially in a production system, instead of just assuming, and taking some (potentially costly) action !, which could be disasters in real life !!

That depends on the application. In entertainment based conversational systems the “did you mean ____” response can inhibit the conversational flow. You also might find that a “quality score” threshold could be used to give it flexibility to use something other than an absolute match. Of course, in mission critical apps, the request for clarification prompt is a necessity.

Merlin - Jan 31, 2011:

In my opinion, a bot should always ask “did you mean ____”, especially in a production system, instead of just assuming, and taking some (potentially costly) action !, which could be disasters in real life !!

That depends on the application. In entertainment based conversational systems the “did you mean ____” response can inhibit the conversational flow. You also might find that a “quality score” threshold could be used to give it flexibility to use something other than an absolute match. Of course, in mission critical apps, the request for clarification prompt is a necessity.

That is true.  In fact, one could really show the power of a bot that way.  What I mean is, if it responds one way, and the user responds with “no, no, I meant….....” which it then says “Ah, I see, so then…......”.  That will be amazing functionality in a bot, to re-process based on feedback.

The name CLUES I chose for my engine has 2 reasons, the acronym and the actual word clues - it uses clues, hints, and the more that contribute to a certain conclusion (including conversational context), the more likely it choses that option.  And a subsequent input of “no, no, I meant…..” will affect future processing.

Merlin - Jan 31, 2011:

This also lets some of us throw grammar completely out the window and attempt to identify meaning from the input in a “fuzzy” fashion.

I wish you luck in that ‘attempt’.  Is the solution forthcoming ?  I mean do you have any ideas on how to go from fuzzy to crisp? 

So far, the more crisp solution seems to be showing a LOT of promise, and there are many ways I can make it ‘fuzzy’ later.

But, for efficiency, Grace will first try proper grammar, then various degrees of ‘relaxing’ the rules by selectively by-passing certain ones).  Most people, most I know anyway, speak and write with proper grammar.

Especially for shorter, simple sentences, I think most people ‘get it right’ and there is no need to be fuzzy for most cases.  Also, keep in mind, that the core CLUES engine doesn’t care or differentiate between proper and improper grammar, they are simply different set of production rules.  Later, Grace will have production rules (parse tree generators) which will assist in her ability to learn other languages.

Hans Peter Willems - Jan 27, 2011:

@Erwin: thanks for the welcome. If ever you are around Zeeland (Terneuzen) somewhere be sure to drop by for a coffee

Victor Shulist - Jan 27, 2011:

Yes, I am putting extreme focus on the understanding functionality.  I want a bot that can comprehend complex natural language constructs, and be able to differentiate all the subtleties.  I want full understanding.

This is one thing where I totally agree with you; the bot needs ‘understanding. However, I might not (completely) agree with your implementation of this ‘understanding’. When I look at human communication, only a few have real understanding of language constructs but most humans communicate fine without that knowledge. Also I think we are too much hung up on proper grammar, syntax, etc.

I highly disagree.  Is there any statistics evidence to support the claim?  (that “only a few have real understanding of language constructs”  ?  I think it simply seems that way since our minds have gotten so proficient at analyzing auditory language and written language.  No, I would argue that we fully understand language constructs, and that the ability to do this in a fuzzy fashion only comes from first understanding it clearly.  Crisp understanding should come first, then fuzzy.  Crisp rules can be more easily relaxed, you know what your target is, going from fuzzy to crisp, good luck; you don’t know your destination, you’ll have a very difficult time getting there.  When I listen to people speak to each other, even in a very complex lengthly sentence, the other party can easily see the difference in semantics, even if it is as simple as modifying a verb with an auxiliary verb like ‘should’.

May I add something here?

I noticed that English natives have a lot of knowledge (not only on this forum) about language construct, i.e. terminology to address language construct. So they not not able to say: ‘that’s a verb, and that’s a noun’, but also all the other constructs. In the Netherlands, higher educated people have knowledge about grammar classification, but the average man in the street doesn’t. They have learned something about it in primary school, but as they need to learn other language as well, there is not an emphasis to go in depth with their own language. Understanding, text comprehension, and good writing is good enough.

I’ve noticed that people from the UK tend to used much longer sentences than non-English speakers use in their own language.

Furthermore, the UK culture, the origin of English, can be classified a ‘highly contextual’ cultural in language. In conversations, a lot is assumed to be known by the other party. Therefore ‘people can speak to each other in very complex lengthy sentences’. It’s an English characteristic though.

From that perspective, once you’ve solved the semantics in English, it should be quite easy in other languages as well.

Unfortunately, there are many more other perspectives

Erwin Van Lun - Feb 3, 2011:

They have learned something about it in primary school, but as they need to learn other language as well, there is not an emphasis to go in depth with their own language. Understanding, text comprehension, and good writing is good enough.

This is the funny thing; they know it, but don’t know they know it.

People DO understand grammar.  Now, they may not know what a preposition means or prepositional phrase or adverb, but they know how it is used in a sentence.

“The dog jumped over the cat”

We know what was doing the jumping, we know what the dog did, etc.  The mind uses just as much world knowledge as it does grammar , to understand text.  Not knowing the names of those things doesn’t mean we don’t know them.

Otherwise, “The dog jumped over the cat” is simply a list of words.    Even if we knew what a cat was, a dog was, and what jumped meant, we wouldn’t know the meaning if we didn’t know that, in English, one of the ways to encode a sentence is to have a subject followed by a predicate.  We know when we see a subject followed by a predicate, even though many people may not know the terms subject and predicate.  Similar situation, I can’t ask Grace “What’s a noun?” right now, but she knows how to generate any parse tree given almost unlimited number of free-form inputs, parse trees containing nouns.  Same situation with people.

Victor Shulist - Feb 4, 2011:

People DO understand grammar.  Now, they may not know what a preposition means or prepositional phrase or adverb, but they know how it is used in a sentence.

We agree. The only thing that remains is that English natives tend to use more complex sentences in their language, also when they speak.

I’m from Holland.

I use simply sentences.

Easy

Funny you should say that, I know some people that have English as second language, and they would agree; English is one of the more complex languages; they even suggested that I create a bot in another, easier language first !!

So is the continuum:
* data (user inputs only if you really believe that computers are incapable of experiencing a world which bothers me a bit because I know I can find weather forecasts about snow storms from a device that has never felt a single snowflake);
* information (that stuff you find after applying grammar and semantic rules or fuzzy matching patterns, etc.);
* knowledge (perhaps something which appears produced by reasoning or maybe strong AI);
* wisdom (who knows if any machine has this kind of stuff, or is it embedded in social networks or forums like chatbots.org)?
  Learning is the capture of data. Maybe that data is of a higher processing order like grammar rules or the instance of the application of those rules. Learning is using information. Could you say you learned a thing because you can explain why (whoops, there goes physics which only explains the what)? Learning is gaining knowledge. Self aware means we know we learned something? It is easy to say the bot has to learn by itself, but what are we talking about here? If it learns the grammar rules, I think we are getting pretty fuzzy.
  BTW, all computer programming is philosophy. You can’t escape it or pretend to avoid it. You have to say, “this is how I believe it is done.” Period.
  Unless you can get at least to analogies and storytelling, all your crunching of text will fall short of any human inspiration. You will always have just a machine with a glorified interface to an encyclopedia of facts.  I believe IBM is marketing one such machine called Watson!

Gary Dubuque - Feb 4, 2011:

  BTW, all computer programming is philosophy. You can’t escape it or pretend to avoid it. You have to say, “this is how I believe it is done.” Period.

I highly agree with this.

Gary Dubuque - Feb 4, 2011:

  Unless you can get at least to analogies and storytelling, all your crunching of text will fall short of any human inspiration.

Past what I am trying to achieve, (only so many hours in a life-time, after work week and family) but probably true.  But I think if we can get a machine to really understand and follow a conversation, and learn and reason via natural language, strong AI or not, it will be a huge step forward.  If, for true AI we need the thing to walk on water, read our minds, time travel and teleport, it won’t happen, not in our life time, perhaps not ever.  Let’s get real

Gary Dubuque - Feb 4, 2011:

You will always have just a machine with a glorified interface to an encyclopedia of facts.  I believe IBM is marketing one such machine called Watson!

Hum, yes, Watson of course is impressive, and I am looking forward to 14th of Feb to watch the Jeopardy challenge.  Still, more work I think needs to be done though for it to carry a conversation, learn via a conversation.  I’m not sure if it truly understands the sentences it reads.

Gary Dubuque - Feb 4, 2011:

You will always have just a machine with a glorified interface to an encyclopedia of facts

I like “just a”, never mind how incredible that actually is. 

This is the AI effect, ( http://en.wikipedia.org/wiki/AI_effect ) and the reason why “strong AI” will absolutely never ever be realized.

For strong AI, we need computer do to X.
Computer can do X.
Oh, that is just a computer that can do X, actually for strong AI we need Y.
Computer can do Y
Oh, that is just a computer that can do Y, actually for strong AI we need Z.

and on and on we go.

Until the computer can do everything.

Well.. it is not doing any MORE functionality this year than it did last year, so therefore, another excuse, its not strong AI.

its even more powerful than human, but, hum.. its the WAY it does it, it does it too “mechanically” or something, so not strong AI.

Victor Shulist - Feb 3, 2011:

I highly disagree.  Is there any statistics evidence to support the claim?  (that “only a few have real understanding of language constructs”  ?  I think it simply seems that way since our minds have gotten so proficient at analyzing auditory language and written language.

I don’t have any statistics, I’m not even sure if there is actual research being done in this field. However, what I do have is my own experience with people that communicate with a very limited vocabulary and a clear lack of grammatical rules. Most youth I come across have totally abandoned any idea of proper grammar in conversation, both written and spoken. I guess the ‘modern’ world of texting and instant-messaging are one reason for this development. Maybe I should have said ‘only a few have real understanding of CORRECT language constructs’.

But that was not the point I was trying to make. My point is that if I say something like ‘me drinking like in the evening a whisky glass of’... you still understand what I meant to say. There is no grammar involved in that understanding, it is based on weighing the combination of words and their possible meaning in relation to each other. So to sum it up: to ‘understand’ the meaning of a sentence, that sentence does not need to be grammatically correct, but it does need enough information (words?) to get the meaning across. Ergo, grammar is less important then content.

Hans Peter Willems - Feb 4, 2011:

me drinking like in the evening a whisky glass of…

I disagree.  That sentence DOES have grammar, not exactly “proper grammar”, but it DOES have grammar. 

me drinking like in the evening a whisky glass of…..

me (subject)

drinking like in the evening a whisky glass of.  (predicate)

drinking (predicate verb)

like in the evening (prepositional phrase (with 2 prepositions, which is not the best, but still, I can allow a “bad grammar” rule like that)

evening (the object of the prepositional phrase (“like in the evening”)

“the” - adjective describing WHICH “evening” , so we know from this that it is any evening, instead of “that evening”.

“glass” - object of verb “drinking” (‘drinking’ being predicate verb)

“a” - modifies the object ‘glass’ - so, ok, we mean any glass of whiskey, not *THAT* glass or MY glass, or your glass

“whiskey”  - ok, what kind of glass,  it is a whiskey glass, not water glass, or juice glass.  We know that, many times, 2 nouns, like “<noun1> <noun2>” we often use <noun1> like an adjective to modify <noun2>—hay, a grammar rule that we all know.

**
now, extreme example… “the of in house when of they car since over who” . .. that pretty much has no grammar,  not sure what that would means!!

Another example, “She hit me”

It is a “convention” (or call it grammar, whatever you like), that in this case the subject comes first (in most cases, but there is inverter order yes), followed by predicate.  Here we see subject then predicate. 

But if we don’t care about that, we just have

She
hit
me

if we don’t care, then, I guess we have a 50/50 chance of knowing if she hit me, or if I hit her.