It's not a hard and fast rule, but Ns tend to prefer language that is abstract. They think in terms of concepts rather than concrete things. So conversation, interests, and thoughts tend to converge along those dimensions.
Keirsey's argument was the N and S cause the biggest barriers between people.
Since Ns tend to deal with abstractions and concepts, he saw the biggest division between them in how they dealt with and organized their inner world (T vs F).
But Ss concentrate more on their physical surroundings and environment, and he thought the prime distinction between them was in how they organized and interacted with the outside world (P vs. J).
That was my understanding, anyway. It comes from Keirsey, who focuses more on temperaments and rejects the Jungian psychological functions. IMO, he has an advantage because he emphasizes that type is more than "just the sum of four letters" and comes up with observations about communication style, interaction, motivation, etc.
You can't wait for inspiration. You have to go after it with a club. - Jack London
Indeed, it comes from Keirsey, and his four temperaments. He explains how the concepts of 4 temperaments has existed ever since the Antiquity, in a rather consistent way. But in order to fit the 16 types into those pre-existing 4 temperaments, he had to pick NT and NF, and SP and SJ. SF and ST would not have worked.
Berens' Interaction Styles seem more based on how people behave on the surface while Keirsey's temperaments are based more on how people think. (Notice, for example, that the Get-Things-Going Interaction Style is compiled of the "flakes" of the MBTI.)
In a person's behavior, I'd say that E/I and J/P are the most important functions, but a person's way of thinking, I think, is much more important for psychology, and that is governed by S/N and T/F.
On the other hand, a person's way of thinking could reflect their behavior, and it seems to be mostly in the S/N dichotomy, but I think that it has more to do with the fact that there are more S's than N's, making the N type more likely to be considered socially "abnormal."
Then what is about behaviour that splits the N's (by their attitude function) and unites the S's, and what is it about thinking that splits the S's and unites the N's?
Well, what I mean is that the Interaction Styles are compiled of the type lookalikes. I'm a little confused about where you're getting at here. But one of the four types of each temperament is grouped into each of the Interaction Styles.