When one is about to accept (store) a proposition as true it is appropriate to reject the proposition if an opposing proposition is already known. Below is an example of this type of reasoning.

Given: Jane is a human.
Given: Humans are animals.
Given: Animals are either male or female.
Given: Jane is female.
Q:  Jane is male.
A: No, Jane is not male because Jane is female.

I would argue that simply saying “Jane is female” is enough to answer the question. The first 3 statements are moot.

without knowledge of the first three then the result would be:

Given: Jane is female.
Q:  Jane is male.
A: Jane is male and female.

True. However, I would assume all bots know someone or something can’t be both male and female.

If the bot already knows something, then that something needs to be stated as “given”. This allows the test to be validated.

Steve Worswick - Apr 9, 2013:

I would argue that simply saying “Jane is female” is enough to answer the question. The first 3 statements are moot.

Nova is female, yet she isn’t human! ‘La Boîte’ is female, but not even an animal,....

Toborman - Apr 10, 2013:

without knowledge of the first three then the result would be:

Given: Jane is female.
Q:  Jane is male.
A: Jane is male and female.

If you make ‘sex’ a singleton, the system can now that you can’t be male and female, that would make it a hermaphrodite.

Steve Worswick - Apr 10, 2013:

True. However, I would assume all bots know someone or something can’t be both male and female.

Isn’t the question, how does the bot know something can’t be male and female? Hard-coding it in is one way. But if your goal is to have the bot learn world knowledge through natural language, then the preceding sentences are necessary to provide this information.

In general, I think being able to draw conclusions using natural language relationships between objects is crucial to giving the appearance of understanding, especially in long or more in-depth conversations. This goes beyond simple statements of identity (A is B, B is not C, etc.). Any sort of sentence should be able to be treated as a relationship (A catches the faintest whiff of B when the night is C and the D can E in the F.). These relationships may not be unique—there are many ways to state the same information, after all.

This kind of data manipulation is where NLP should shine, in principle. Being able to break down a statement into its component parts allows you to map those parts onto equivalent statements or use them to perform logic operations in a generalized way.

I agree with you Hunt.

When it comes to knowledge I don’t think anything should be taken as definite. The knowledge might be accurate one moment but is always subject to change. I think Jan brought up a good point, what if the person is a hermaphrodite? What do you consider them then if a person can only be considered male or female and not both? What if Jane was a female and had some sort of sex change operation and does not consider herself/himself to be male anymore? Would “Jane is male.” not be an accurate statement then?

Right on target, C. R.

By giving this type of test to children we can establish an “age” at which this test is passed. Then “special needs” individuals can be tested to determine their “mental acuity age”. This testing could also apply to artificial lifeforms (i.e., bots) to evaluate its “reasoning” ability.

In this type of test we are not asking “do you know the answer”, we are asking “with the information provided can you figure out the answer”.

A more sophisticated form of this test might be in conversation. Tom is the tester and Harry is the Responder.

Tom: Jane is male.
Harry: Jane is not male.
Tom: Why?
Harry: Because Jane is female.
Tom: How did you determine that Jane is not male?
Harry: When you said “Jane is male” I challenged the validity of the statement.Harry: I found the following:
Harry: Jane is a human.
Harry: Humans are animals.
Harry: Animals are either male or female.
Harry: Jane is female.
Harry: This let me figure out that “Jane is male” is not true.

Chad, you are right,of course.

We are not determining the truth of the statements, only that the reasoning is valid.

Efficiency can be a good reason to store the negation of a statement, or more generally: efficiency considerations can be a good reason to store the partial results of anything.

Some examples:
(1)
It is easier to search for default properties of “male” rather than default properties of “not(female)”, if only because each matching process would require one extra calculation (the “not”) at each step otherwise.
(2)
An inferencing system that stored (= learned) the fact that A causes B could greatly reduce its search time in looking for a diagnosis of what caused C, since the search tree depth would be reduced so that only B might need to be searched, rather than considering both A and B independently.
(3)
An associative memory search looking for visual events where objects that were “red and moving downward and that happened about 1 hour ago” would be faster if the concepts of “red and moving downward” were already stored together as a single concept.

Mark, while your idea may be initially true (in some cases), storing (learning) “conglomerate concepts” will have diminishing returns, as each one will add to the body of information to search through each time, and will eventually not only fail to be “more efficient”, but, in fact, would become LESS efficient than the original method, over time. It’s often faster and more efficient to perform two queries against a moderately sized database than it is to perform a single query against a massive database. Of course, this greatly depends on a number of external factors that are too numerous to mention, and as always, “Your Mileage May Vary”.

Looking at this from a different perspective, from a child, you are told that people can be only male or female.

So, if someone was to tell you, that someone was male AND female, the truth of that original learned idea would come in to play, and you would question it….quite rightly, as in nearly ALL circumstances, people either are male, or female.

Upon being told the statement that “JohnJane is a male AND a female” you would question said statement to be false, at which point, the govening party would explain to you why, with a CONDITIONAL.

“A person can be male and female only if they have both sex organs”

If you apply that conditional, with the already learned rule of “a person is male or female”, you can then handle both scenarios, no need for singletons, or any other fancy jazz.

This is how Caesar handles situations such as these, providing that there is a previous learned rule which will call into question the statement, Caesar can apply any conditionals applicable to that rule when something out of the ordinary comes along to determine its truth. 

Caesar always seeks verification when a conditional is encountered within a strong truthed “be” rule, as would you, in disbelief of such a claim, which is exaclty what happens within Caesar.  He doesn’t believe what you have said, even though he understands (if the conditional is learned) that it’s possible in some circumstances.

Dave Morton - Apr 10, 2013:

in fact, would become LESS efficient than the original method, over time.

Good point. I should have remembered and pointed out that there is always an inherent space-time tradeoff: with my suggestion you’d be gaining time at the expense of losing memory space. Which of the two is better depends on the intended application.

Mark Atkins

Efficiency can be a good reason to store the negation of a statement, or more generally: efficiency considerations can be a good reason to store the partial results of anything.

As Mark points out, efficiency is a consideration. When Harry challenges a proposition, he tries to refute first (i.e., checks for a negation) then looks for opposition, because its more efficient to check for a negative than for opposition.

Creating and resolving a dilemma or disjunction is the other side of the opposition situation. The question is, do we report or respond to the discovery of the dilemma or just remember it until more information is provided to resolve the issue?  Here are some possible conversations.  T is the tester and R is the responder.

Test 1: report dilemma
Given: humans are male or female.
T: Jane is human.
R: Jane is male or female.

Test 2: ask for resolution of dilemma
Given: humans are male or female.
T: Jane is human.
R: Is Jane is male or female?
T: Jane is female.
R: Jane is female.
T: describe Jane.
R: Jane is a female human.

Test 3: recall dilemma
Given: humans are male or female.
T: Jane is human.
R: Jane is human.
T: describe Jane.
R: Jane is a male or female human.

Test 4: resolve dilemma with positive
Given: humans are male or female.
T: Jane is human.
R: Jane is human.
T: Jane is female.
R: Jane is female.
T: describe Jane.
R: Jane is a female human.

Test 5: resolve dilemma with negative
Given: humans are male or female.
T: Jane is human.
R: Jane is human.
T: Jane is not male.
R: Jane is not male.
T: describe Jane.
R: Jane is a female human.