Re: [ontolog-forum] Improved Elizae being used in Call Centers

["Rich Cooper" <rich@xxxxxxxxxxxxxxxxxxxxxx>]

Dear John,

I've been mulling over your description of Elizae, along with Doug's, Pat's, Ronald's and Leo's contributions in view of the IPsoft usage.  My comments are below,

-Rich

Sincerely,

Rich Cooper

EnglishLogicKernel.com

Rich AT EnglishLogicKernel DOT com

9 4 9 \ 5 2 5 - 5 7 1 2

John Wrote:

On 2/27/2013 5:37 PM, Obrst, Leo J. wrote:

> Anyone who uses Eliza-based NLP templates, I say: pity the fools.

An "Eliza-based template" is a regular _expression_.  Such expressions

have been widely used for searching and information extraction for

decades.

I've been thinking about such odd sentences as:

(1.1) The rat the cat the dog chased caught escaped. 

Although comp linguists like to model such sentences as recursive grammars, it occurs to me that I have never uttered, read, or written such a sentence to convey real meaning.  An FSA would simply have to encode that in a state sequence, three terms deep, to process it as well as an infinitely recursive grammar.  The FSA just couldn't keep up the depth of the nesting arbitrarily deep like a recursive grammar could.  But people, like me, don't utter, read or write such a sentence anyway. 

A few years ago, there was an MIT seminar entitled something like "Natural Language: Have We Gotten It All Wrong?"  The point was that for all our mathy NLP representations, we haven't made MUCH progress past Elizae. 

So I started thinking about FSA parsers, perhaps using a Turing-like stack, but only for the semantic vaguenesses rather than the syntactic ones. 

But Eliza had more than just a set of templates.  Each template was used as a pattern that would be matched to the current input stream and would trigger some rule.  In effect, Eliza implemented production rules with just one-step forward chaining.

Later Eliza-like systems used production rules that allowed multiple-step forward chaining *and* the option of a "context" switch.  Each context had its own set of patterns and rules.

The set of contexts and the rules that made transitions from one context to another could be represented by a finite-state machine.

The above two paragraphs are the basis for many, if not most of the annoying phone answering systems you get when you call

an airline, a bank, or any company from which you would like some useful service.

I agree that they are annoying, and they're not state of the art in NLP.  But at least they're a step up from the useless help facilities based on keywords.

John

 

Also, I noticed that the brain is not arbitrarily recursive either.  Nobody uses the form of (1.1) because it seems unnatural, and certainly nobody could keep it up for very long before the entire meaning of the sentence could be lost.  For example, the "cat" could be the subject of "escaped" instead of "caught" in some interpretation, and the dog could have "caught" the cat instead of just "chased" it, letting the rat do the "escaped" part. 

We know that word order, in English, is important for certain semantic renderings.

(1.2) "The dog brought the truck home"

implies that the truck is a small toy, but

(1.3) "The truck brought the dog home"

implies that the truck is big enough to carry a dog. 

But the standard reaction to such word ordering significance has consistently been to support the contention that syntax is important, while a FSA implementation could distinguish between the two orderings also.  Just not infinitely deeply. 

Perhaps those of us interested in language processing should pay more attention to the FSA versions, perhaps with a semantic stack of objects and phrases, rather than a purely syntactic stack as we have consistently used over the last few decades since Chomsky's suggestions re syntactic interpretations. 

JMHO,

-Rich

_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (01)