When you read the name “Paris Hilton” just now, did you think immediately of the socialite or the city of Paris; possibly the hotel was a connection into the city?
Your answer will be based on whichever has had the most impact on you. Have you heard more about Conrad’s great granddaughter (good or bad) which enabled you to form an opinion? Perhaps you’re a traveller and you have internally formed a stronger association to the name Hilton (of hotels) and or the location of Paris to logically derive your thought.
So what exactly is the Paris Hilton effect?
It occurs when a collection of words are used together, which without context, can cause referential ambiguity in a cognitive structure.
Yes that sounds altogether too complicated. So in simple terms, the COG gets confused, just like you and me.
This happens quite often in human communications and is for the most part not recognized at the point of occurrence. We give off little signals in our body language to suggest we need more details (more context), or sometimes we overtly state that we don’t understand.
Another example is if I said “Where is the Armadillo?”
Without context, you may assume that I am looking for quite possibly the weirdest pet in the world, or another alternative could be that there is a restaurant down the road, named "The Armadillo".
As you will see in the rest of this post, the challenges are stacked. Resolving an understanding and disambiguation of a sentence can have additional road blocks, that make this more than a problem that can be figured out with pure linguistic analysis.
As an aside, I named this effect after Paris Hilton, as I didn’t think “The Armadillo Effect” had the same ring to it?
Allow me to create two new questions and assume your answers. Read the questions out allowed to yourself.
Q1: How old is Paris Hilton?
I am assuming here that you’re considering me to be asking about the person, regardless of if you know the name or not. You could ask the same question again with an entirely made up name and you will almost always certainly consider the subject matter to be that of a person (or living thing).
Q2. How old is the Paris Hilton?
I am assuming here that in your mind, I am asking about a building or specifically the Hilton hotel in Paris. This is a little trickier. You almost always have to have some version of the object in your mind to relate this to. “Paris” (and probably the image of the Eiffel tower) may just be enough in this scenario.
It’s a simple change that creates a complex difference in logical reasoning.
NLP is short for Natural Language Processing. In the field of AI, you could consider this to be “how we work with human language”. Not entirely to do with the meaning of words, but more precisely to do with the understanding of the structure of language and linguistic concepts.
Below you will see the structural breakdown in linguistic terms of both these questions which does a good job at showing just what effect the “the” has in the sentence.
As you can see, the composition of the sentence is incredibly important. We consider each "part" (Tokens) of the sentence individually as well as collectively.
You will note that “the” has a little box tagged with “DT” above it. This means that it is a Determiner. In linguistic terms, a Determiner (or sometimes a Determinative) is used to express the relationship between objects within the context of a noun phrase (NP) (you will note the little “NP” box connecting the sentence together above the “DT”).
Now because we are talking about nouns here (a person or thing), the assessment of the noun phrases (circled in red) shows us two different results:
- Paris Hilton
- the Paris Hilton
There are other determiners such as “a” or “an” that can have a similar effect on a sentence, to change entirely the context and or the subject assumption. But rather than taking a strictly engineering approach to try and derive the context of the sentence in isolation, this is where we need to consider the process that we use when we speak to others as people.
This is part of a process that I call "Conscious Deconstruction".
The context chain.
Firstly, the context of our entire conversation is a factor. Each input and output from both the human and the COG* are links in a chain. Without creating this context chain, we would be quite lost in trying to understand the subject, so the COG must contain a viable context chaining mechanic.
Secondly, if an accurate assessment of the chain can’t be achieved in the given context, a clarifying question needs to be asked. Similar to our own process where we should ask “do you mean…?”
I concede that sometimes we (as humans) don’t always ask the question(s), but that can be assigned to other processes regarding shyness and confidence (another blog post and discussion).
Inside the COG, we maintain references to Entities that have definitions and sub-definitions, each segmented again by Context. These Entities can contain a type definition for both people and buildings (and cities and more).
The point being that the COG can get into a confused state with or without the assistance of the context chain and so the solution is to follow our own human process and ask the question.
Intelligence is knowing when to ask a qualifying or clarifying question, not in the ability to deconstruct a sentence.
This is a key difference with cognitive intelligence based reasoning, rather than pattern based derivation most simple non-intelligent systems employ.
Even though NLP shows us the differences in the structure of what is being asked, this is not just a simple NLP problem/solution. It goes much further than that by demonstrating the complex relationship between words, conceptual representations of Context, and Entities.
Go ahead and try the two original example questions in Google searches. The results will pretty much be the same because Google is looking at word linking in the search term, not what the actual words mean (individually or collectively) and certainly not how you are using the words to infer if the subject matter is a person or a building.
That is not a berating of Google – it is to demonstrate the difference in my approach to elicit an accurate and appropriate response from the COG, going beyond NLP.
*COG refers to Cognitive Artificial Intelligence