Natural Language

Scribblenauts
Liz England: Scribblenauts objects definitely had systems and inheritance.

They were placed in a 2-tier hierarchy. Ex: "Mammal -> Water Type -> Dolphin"

Some of the top level categories were things like "Tools" "People" "Food" "Buildings" "Plants" "Miscellaneous" (ridiculously useful)

There were some broad strokes things that could be inherited. Like, I dunno, "Mammal -> Water" all had a checkbox for swim underwater

And you could do things like if an object was X by Y pixels, automatically set it's weight to Z.

In a perfect world these systems would inherit across categories and you'd have to do very little hand-authoring to tweak outliers.

In practice when you are dealing with cartoon objects - "stereotypes" - the most important thing is how they are different from each other

For example, all humans were scared of category "monsters" and ate from category "food". This is why the "vegetarian" ate "hamburgers".

But when you look at individual objects, the rules don't make sense. [...]

There was an interview where someone said we read scoured dictionaries for words. This wasn't ENTIRELY accurate. It was actually wikipedia.

It didn't actually matter if an object was accurate, What mattered is that it did what people expected. Wikipedia was best source for that.

Wikipedia also conveniently categorizes stuff, and includes lots of synonyms of other things people might call a thing from various cultures [...]

In Super Scribblenauts I did download a public domain dictionary and stripped out everything but adjectives. Left with ~80k words.

Which had to be sorted. By hand. Into synonyms and categories. Most were deleted (no time, or too specific - like medical terminology) [...]

https://scribblenauts.fandom.com/wiki/Category:Words