The Pragmatic Programmer says, "Learn a new language every year". This is great advice, not just because it puts new tools into your mental toolbox that you can pull out on various occasions to get a job done, but also because it opens your mind to new ideas and new concepts that will filter their way into your code even without explicit language support. For example, suppose you've looked at (J/Iron)Ruby or Groovy, and come to like the "internal iterator" approach as a way of simplifying moving across a collection of objects in a uniform way; for political and cultural reasons, though, you can't write code in anything but Java. You're frustrated, because local anonymous functions (also commonly--and, I think, mistakenly--called closures) are not a first-class concept in Java. Then, you later look at Haskell/ML/Scala/F#, which makes heavy use of what Java programmers would call "static methods" to carry out operations, and realize that this could, in fact, be adapted to Java to give you the "internal iteration" concept over the Java Collections:

 1: package com.tedneward.util;

 2:  

 3: import java.util.*;

 4:  

 5: public interface Acceptor

 6: {

 7: public void each(Object
obj);

 8: }

 9:  

 10: public class Collection

 11: {

 12: public static void each(List
list, Acceptor acc)

 13: {

 14: for (Object
o : list)

 15: acc.each(o);

 16: }

 17: }

Where using it would look like this:

 1: import com.tedneward.util.*;

 2:  

 3: List
personList = ...;

 4: Collection.each(new Accpetor()
{

 5: public void each(Object
person) {

 6: System.out.println("Found
person " + person + ", isn't that nice?");

 7: }

 8: });

Is it quite as nice or as clean as using it from a language that has first-class support for anonymous local functions? No, but slowly migrating over to this style has a couple of definitive effects, most notably that you will start grooming the rest of your team (who may be reluctant to pick up these new languages) towards the new ideas that will be present in Groovy, and when they finally do see them (as they will, eventually, unless they hide under rocks on a daily basis), they will realize what's going on here that much more quickly, and start adding their voices to the call to start using (J/Iron)Ruby/Groovy for certain things in the codebase you support.

(By the way, this is so much easier to do in C# 2.0, thanks to generics, static classes and anonymous delegates...

 1: namespace TedNeward.Util

 2: {

 3: public delegate void EachProc<T>(T
obj);

 4: public static class Collection

 5: {

 6: public static void each(ArrayList
list, EachProc proc)

 7: {

 8: foreach (Object
o in list)

 9: proc(o);

 10: }

 11: }

 12: }

 13:  

 14: //
...

 15:  

 16: ArrayList
personList = ...;

 17: Collection.each(list, delegate(Object
person) {

 18: System.Console.WriteLine("Found
" + person + ", isn't that nice?");

 19: });

... though the collection classes in the .NET FCL are nowhere near as nicely designed as those in the Java Collections library, IMHO. C# programmers take note: spend at least a week studying the Java Collections API.)

This, then, opens the much harder question of, "Which language?" Without trying to infer any sort of order or importance, here's a list of languages to consider, with URLs where applicable; I invite your own suggestions, by the way, as I'm sure there's a lot of languages I don't know about, and quite frankly, would love to. The "current hotness" is to learn the languages marked in bold, so if you want to be daring and different, try one of those that isn't. (I've provided some links, but honestly it's kind of tiring to put all of them in; just remember that Google is your friend, and you should be OK. :-) )

• Visual Basic. Yes, as in Visual Basic--if you haven't played with dynamic languages before, try turning "Option Strict Off", write some code, and see how interacting with the .NET FCL suddenly changes into a duck-typed scenario. If you're really curious, have a look at the generated code in Reflector or ILDasm, and notice how the generated code looks a lot like the generated JVM code from other dynamic languages on an execution environment, a la Groovy.
• Ruby (JRuby, IronRuby):
• Groovy: Some call this "javac 2.0"; I'm not sure it merits that title, or the assumption of the mantle of "King of the JVM" that would seem to go with that title, but the fact is, Groovy's a useful language.
• Scala: A "SCAlable LAnguage" for the JVM (and CLR, though that feature has been left to the community to support), incorporating both object-oriented and functional concepts, plus a few new ideas, into a single package. I'm obviously bullish on Scala, given the talks and articles I've done on it.
• F#: Originally OCaml-on-the-CLR, now F# is starting to take on a personality of its own as Microsoft productizes it. Like Scala and Erlang, F# will be immediately applicable in concurrency scenarios, I think. I'm obviously bullish on F#, given the talks, articles, and book I'm doing on it.
• Erlang: Functional language with a strong emphasis on parallel processing, scalability, and concurrency.
• Perl: People will perhaps be surprised I say this, given my public dislike of Perl's syntax, but I think every programmer should learn Perl, and decide for themselves what's right and what's wrong about Perl. Besides, there's clearly no argument that Perl is one of the power tools in every *nix sysadmin's toolbox.
• Python: Again, given my dislike of Python's significant whitespace, my suggestion to learn it here may surprise some, but Python seems to be stepping into Perl's shoes as the sysadmin language tool of choice, and frankly, lots of people like the significant whitespace, since that's how they format their code anyway.
• C++: The grandaddy of them all, in some ways; if you've never looked at C++ before, you should, particularly what they're doing with templates in the Boost library. As Scott Meyers once put it, "We're a long way from Stack<T>!"
• D: Walter Bright's native-compiling garbage-collected successor to C++/Java/C#.
• Objective-C (part of gcc): Great "other" object-oriented C-based language that never gathered the kind of attention C++ did, yet ended up making its mark on the industry thanks to Steve Jobs' love of the language and its incorporation into the NeXT (and later, Mac OS X) toolchain. Obj-C is a message-passing object language, which has some interesting implications in its own right.
• Common Lisp (Steel Bank Common Lisp): What happens when you create a language that holds as a core principle that the language should hold no clear delineation between "code" and "data"? Or that the syntactic expression of the language should be accessible from within that langauge? You get Lisp, and if you're not sure what I'm talking about, pick up a Lisp or a Scheme implementation and start experimenting.
• Scheme (PLT Scheme, SISC): Scheme is one of the earliest dialects of Lisp, and much of the same syntactic flexibility and power of Lisp is in Scheme, as well. While the syntaxes are usually not directly interchangeable, they're close enough that learning one is usually enough.
• Clojure: Rich Hickey (who also built "dotLisp" for the CLR) has done an amazing job of bringing Lisp to the JVM, including a few new ideas, such as some functional concepts and an implementation of software transactional memory, among other things.
• ECMAScript (E4X, Rhino, ES4): If you've never looking at JavaScript outside of the browser, you're in for a surprise--as Glenn Vanderburg put it during one of his NFJS talks, "There's a real programming language in there!". I'm particularly fond of E4X, which integrates XML as a native primitive type, and the Rhino implementation fully supports it, which makes it attractive to use as an XML services implementation language.
• Haskell (Jaskell): One of the original functional languages. Learning this will give a programmer a leg up on the functional concepts that are creeping into other environments. Jaskell is an implementation of Haskell on the JVM, and they've taken the concept of functional further, creating a build system ("Neptune") on top of Jaskell + Ant, to yield a syntax that's... well... more Haskellian... for building Java projects. (Whether it's better/cleaner than Ant is debatable, but it certainly makes clear the functional nature of build scripts.)
• ML: Another of the original functional languages. Probably don't need to learn this if you learn Haskell, but hey, it can't hurt.
• Heron: Heron is interesting because it looks to take on more of the modeling aspects of programming directly into the language, such as state transitions, which is definitely a novel idea. I'm eagerly looking forward to future drops. (I'm not so interested in the graphical design mode, or the idea of "executable UML", but I think there's a vein of interesting ideas here that could be mined for other languages that aren't quite so lofty in scope.)
• HaXe: A functional language that compiles to three different target platforms: its own (Neko), Flash, and/or Javascript (for use in Web DOMs).
• CAL: A JVM-based statically-typed language from the folks who bring you Crystal Reports.
• E: An interesting tack on distributed systems and security. Not sure if it's production-ready, but it's definitely an eye-opener to look at.
• Prolog: A language built around the idea of logic and logical inference. Would love to see this in play as a "rules engine" in a production system.
• Nemerle: A CLR-based language with functional syntax and semantics, and semantic macros, similar to what we see in Lisp/Scheme.
• Nice: A JVM-based language that permits multi-dispatch methods, sometimes known as multimethods.
• OCaml: An object-functional fusion that was the immediate predecessor of F#. The HaXe and MTASC compilers are both built in OCaml, and frankly, it's in a startlingly small number of lines of code, highlighting how appropriate functional languages are for building compilers and interpreters.
• Smalltalk (Squeak, VisualWorks, Strongtalk): Smalltalk was widely-known as "the O-O language that all the C guys turned to in order to learn how to build object-oriented programs", but very few people at the time understood that Smalltalk was wildly different because of its message-passing and loosely/un-typed semantics. Now we know better (I hope). Have a look.
• TCL (Jacl): Tool Command Language, a procedural scripting language that has some nice embedding capabilities. I'd be curious to try putting a TCL-based language in the hands of end users to see if it was a good DSL base. The Jacl implementation is built on top of the JVM.
• Forth: The original (near as I can tell) stack-based language, in which all execution happens on an execution stack, not unlike what we see in the JVM or CLR. Given how much Lisp has made out of the "atoms and lists" concept, I'm curious if Forth's stack-based approach yields a similar payoff.
• Lua: Dynamically-typed language that lives to be embedded; known for its biggest embedder's popularity: World of Warcraft, along with several other games/game engines. A great demonstration of the power of embedding a language into an engine/environment to allow users to create emergent behavior.
• Fan: Another language that seeks to incorporate both static and dynamic typing, running on top of both the JVM or the CLR.
• Factor: I'm curious about Factor because it's another stack-based language, with a lot of inspiration from some of the other languages on this list.
• Boo: A Python-inspired CLR language that Ayende likes for domain-specific languages.
• Cobra: A Python-inspired language that seeks to encompass both static and dynamic typing into one language. Fascinating stuff.
• Slate: A "prototype-based object-oriented programming language based on Self, CLOS, and Smalltalk-80." Apparently on hold due to loss of interest from the founder, last release was 0.3.5 in August of 2005.
• Joy: Factor's primary inspiration, another stack-based language.
• Raven: A scripting language that "rips off" from Python, Forth, Perl, and the creator's own head.
• Onyx: "Onyx is a powerful stack-based, multi-threaded, interpreted, general purpose programming language similar to PostScript. It can be embedded as an extension language similarly to ficl (Forth), guile (scheme), librep (lisp dialect), s-lang, Lua, and Tcl."
• LOLCode: No, you won't use LOLcode on a project any time soon, but LOLCode has had so many different implementations of it built, it's a great practice tool towards building your own languages, a la DSLs. LOLcode has all the basic components a language would use, so if you can build a parser, AST and execution engine (either interpreter or compiler) for LOLcode, then you've got the basic skills in place to build an external DSL.

There's more, of course, but hopefully there's something in this list to keep you busy for a while. Remember to send me your favorite new-language links, and I'll add them to the list as seems appropriate.

Happy hacking!