The syntax is just weird, not only in a paradigm-way (pattern matching is not huge in most languages, but, hey, that's the way of doing stuff in functional programming), but on strange places ("read" lines ending on dot, semicolon takes time, it does not share any common syntax definitions with the languages used by 99% of the programmers). That sets a high bar in approaching the language, so it's difficult to "play around" with it (at least compared with other languages)
While I like some of the advantages of Erlang, the lack of general support for a lot of common operations (and yes, string manipulation is a huge deal) and the fact that it is designed with a very very particular problem in mind makes it "a silver bullet". Not in the usual meaning, but in the way that's only useful for killing a werewolf. For every other task is too expensive and just not the proper tool. I was involved in a project that used Erlang for something not well suited for it, and it was absolutely awful, you have to wrestle with it to perform common stuff that in other languages is done by the standard library. Again, you win something, but only in a very very VERY specific problem.
(I've used it in another project when it was the proper tool, and, in that case, it's still not the most pleasant experience, but you're getting a clear win)