Suppose two crossword puzzle compilers start with the same key word.
1. The word has to be divided and embedded into phrases.
2. Puzzle designers share the same sense of “taste”: It is considered elegant to divide the key word into as many separate words as possible.
3. Puzzles have to have a certain shape (“180 degree rotational symmetry”).
So, if two or three puzzle designers start off with the same keyword they are highly likely to come up with very similar crosswords as there are only so many solutions given the constraints.
Interestingly, while same problem can occur in economics research, I believe it is less likely than in design of crossword puzzles. For example, I attended at NBER conference on relational contracting. This area of contract theory studies how incentives between a firm and its supplier might be aligned as they interact repeatedly (eg early “cheating” might be punished by a worsening of the relationship later on).
Many researchers at the conference had the same motivation: Why does Toyota deal with a core group of suppliers while GM acquires parts via competitive bidding? So, there are two constraints: same motivation and same theoretical approach (ie relational contracting). And yet the papers were quite different.
The universe of potential models is infinite, unlike natural language, and hence accidental and near identical replication is less likely. To enjoy the infinite, the human brain must know no bounds. Some say this is the case though their claims are controversial.
(Hat tip: Matt Gaffney at Slate)