If you are building a system which makes use of syntactic trees, but you do not want to use a parser which was trained on annotated trees, this parser could be helpful for you.
This parser was trained from part-of-speech tags (gold standard) from the Penn Treebank, and then wrapped in a perl script for parsing natural language sentences. First, the natural language sentences are tagged using treetagger, and then they are parsed using the unsupervised POS parser. The output format is CoNLL's format (dependency trees).
If you are interested in training the parser from scratch, see DAGEEM. This page also includes links to the papers that describe the main technical ideas behind this unsupervised parser.
Click here to download the parser. There is a README file included which explains how to use it. The parser was tested on Linux machines only. If you port it to another platform, please let me know about it.