Hi Sam,
You mean the textual patterns of relations?
For each NE NELL has literal strings like: “New York”, “NY”, “New York City”, etc. And also textual patterns, like: “_is a city”, “_is located in US”, etc. In the same way, for relations NELL has textual patterns like: “_is a city located in_”, “_has offices at_”, etc.
In few words, and speaking more specifically about CPL, NELL learns categories and relations.
So, in the categories learning, NELL uses NEs (e.g. “Washington”) to learn new textual patterns (e.g.: “_is a big city”). And also using textual patterns NELL learns new NEs.
For relation learning is the same, but here NELL uses pairs of NE (“Seattle” and “USA”) to learn textual patterns of relations (“_is a big city of_”). And also from textual patterns of relations, NELL learns new pairs of NE.
Let ne know if you have any doubts.