Talk:Link grammar
From Wikipedia, the free encyclopedia
An explanation of the grammar syntax would be helpful. E.g. what does {}, +, -, etc mean?
A number of points about link grammars seem to have gotten lost here.
The example shows the links between words, that's good, but it doesn't show that the links can form cycles. What's shown is a tree and that misleads the reader into thinking this is forming parse trees. You can form parse trees of constituents using the links as input but that's quite secondary to the idea.
It should be emphasized that words and links for the atomic units in this approach. Given the words in your language you declare _for each word_ one or more patterns of the links that would satisfy (meet the requires of) that word. For example you state that the word "red" can be satisfied by being linked to a noun on it's right. You do this for each word, one at a time!
In answer to the question above all the syntax used in the grammar examples are syntactic sugar to help write out these link patterns quickly and concisely. As such they are a bit misleading; because you don't declare a class of words. They are only introducing a macros which are then used to help define large number of words quickly.
A successful parse in this framework must uses these atoms and conform to three rules. The unsurprising first two rules are that the words in a sentence must be linked together, and each word has to have one of it's required linkage patterns satisfied.
Most interesting, and the feature observed in the wild upon which this approach is based, is that the resulting link graph is planar; links never cross.
English has a few dozen different kinds of links; a half dozen examples would be a big help. That punctuation, idioms, and sentence boundaries are handled in the same model is one of the delightful aspects of this approach. Bhyde 15:10, 1 June 2007 (UTC)
[edit] Sourceforge Project
I am planning on starting a sourceforge project whose goal is to continue maintenance of the CMU Link Grammar Parser.
I am already carrying a dozen or so patches, with everything from cosmetic fixes, to a Makefile that builds shared libraries, to a couple of bug fixes in the code and the dictionaries, and also a Java JNI binding. It would certainly be appropriate to include the OCaml, perl and ruby bindings as a part of this distribution.
I got tired of carrying around these patches, and would like to post them publicly. To do so, I envision creating a sourceforge project. I invite you to participate, and/or tell me that I'm going in the wrong direction, and/or tell me that there already is such a project somewhere else.
Please let me know your thoughts. Please forward this email to anyone else you feel would be interested in, or affected by, this project.
Dr. Linas Vepstas linas (talk) 22:58, 25 December 2007 (UTC)
- That sounds great. Hopefully this will bring together work done by the Debian Link Grammar package mantainers. –jonsafari (talk) 21:33, 26 December 2007 (UTC)
[edit] Vulnerability
Since there has been some reverting going on, we should discuss this. My personal opinion is that it's not very appropriate to put a vulnerability notice in an encyclopedia article. We wouldn't expect a vulnerability notice for other software articles like, eg. Firefox, OpenCCG, or exim. –jonsafari (talk) 03:50, 17 January 2008 (UTC)
- Ok, I removed this bit from the article: "A serious vulnerability in this program was reported in November 2007 by Alin Rad Pop from Secunia Research,[1] this vulnerability was quickly patched and is fixed in version 4.2.5.[2]" I suppose if the description of the program was more verbose it may be relevant to include the vulnerability…? — fnielsen (talk) 11:29, 17 January 2008 (UTC)
