Suppose we want to have a phrase/task that prints out statements of the form ?a gave ?b the ?c.  We could do this with a task with three parameters:

# Try: [Give “Mary” “Jill” “a red rose”]
Give ?giver ?receiver ?item:
   ?giver gave ?receiver ?item.
[end]

If you then say:

[Give “Mary” “Jill” “a nice raise”]

it responds:

Mary gave Jill a nice raise

But if we say:

[Give “Mr. Boss” “Mr. Boss” “a nice raise”]

it responds:

Mr. Boss gave Mr. Boss a nice raise.

Which isn’t fluent.  How could we fix this?

Detecting self-dealing

We can fix it by having another method that is only used when someone gives something to themself:

# Try: [Give “Mr. Boss” “Mr. Boss” “a nice raise”]
Give ?giver ?giver ?item: ?giver gave themself ?item
Give ?giver ?receiver ?item: ?giver gave ?receiver ?item

Let's diagram the choices here.  Again, the rounded boxes are calls and the square green boxes are final results.  But now the arrows are labeled not only with the method that's chosen, but with the values it assigns to the variables:

graph TB
   g("[Give “Mr. Boss” “Mr. Boss” “a nice raise”]") -- "<b>Give ?giver ?giver ?item: ?giver gave themself ?item</b><br>?giver = 'Mr. Boss'<br>?item = 'a nice raise'" --> r1["Mr. Boss gave themself a nice raise"]
   g -- "<b>Give ?giver ?receiver ?item: ?giver gave ?receiver ?item</b><br>?giver = 'Mr. Boss'<br>?receiver = 'Mr. Boss'<br>?item = 'a nice raise'" --> r2["Mr. Boss gave Mr. Boss a nice raise"]
   style r1 fill:green
   style r2 fill:green

If we had tagged Give with [randomly], then it would randomly choose between the two outputs shown in green.  Since we didn't randomize Give, Step will always prefer to choose methods that come earlier in the program over ones that come later.  That means Step will always choose the leftmost solution path in the diagram.¹  In this case, that means it generates:

Mr. Boss gave themself a nice raise

But in the Mary/Jill example, the parameters can’t match to that first method because ?giver would have to somehow be both Mary and Jill at the same time.  So we say that that method "fails":

graph TB
   g("[Give “Mary” “Jill” “a nice raise”]") -- "<b>Give ?giver ?giver ?item: ?giver gave themself ?item </b>" --> r1[fail]
   g -- "<b>Give ?giver ?receiver ?item: ?giver gave ?receiver ?item </b> <br> ?giver = 'Mary' <br> ?receiver = 'Jill' <br> ?item = 'a nice raise'" --> r2["Mary gave Jill a nice raise"]
   style r1 fill:red
   style r2 fill:green

Since the first method would fail, it has to choose the second, resulting in:

Mary gave Jill a nice raise

Notes