ACL, EMNLP, and NAACL are the top conferences. They're all about equally selective, and they cover all areas of NLP. The same people attend them and the format is the same.
So how do you choose between them? They have different deadlines, that's all. If you don't get your paper done in time for one conference, you submit to the next one. (You might also consider whether the location and date are convenient for you.)
ACL might be slightly more selective than the others, and it traditionally has the largest audience, with many audience members who show up only to listen. At EMNLP, most of the attendees are also authors, which makes the coffee breaks more fun.
If you publish a paper in the TACL journal, you get to give a talk about it at the next conference. TACL papers are like ACL papers but can be slightly longer and go through a journal-style reviewing process.
Good papers also appear in CoNLL, EACL, COLING, and IJCNLP, although I think some people would avoid submitting their very best work there, other things equal. CoNLL is particularly focused on learning issues, and also runs a significant "shared task" competition each year. EACL may have more papers on topics that are popular in Europe, including formal grammars and linguistic issues.
Note: I have served as area chair for most of the above conferences (usually more than once) and am an action editor for TACL.
LREC is a specialized conference on linguistic resources and evaluation. You publish an LREC paper to inform the community about new useful resources that required a lot of effort, care, and judgment, but may not be as technically novel. It has a high acceptance rate (by design).
Various special-topics workshops also exist, usually held at the conferences. While workshop papers are generally not considered archival, some of the workshops can be relatively large events with strong papers. This seems especially true lately in MT and semantics. CoNLL started out as a workshop.
In some cases it may be appropriate to publish at a machine learning conference; just don't expect everyone there to know about NLP.
You can get some statistics here: CL conference acceptance rates.