Version 1.4 of cascadilla.cls, my LaTeX class for typesetting articles according to the Cascadilla Proceedings Project’s stylesheet (used by WCCFL and other linguistic conferences) is now available.
Jason Riggle, Morgan Sonderegger and I have posted a paper proving that the VC dimension of Harmonic Grammar is k+1 for k constraints, which is the same value as has been shown for Optimality Theory.
In this paper we analyze the complexity of Harmonic Grammar (HG), a linguistic model in which licit underlying form to surface form mappings are determined by optimization over weighted constraints. We show that the Vapnik-Chervonenkis Dimension of HG grammars with k constraints is k-1. This is the same as the VC dimension of Optimality Theory (OT), which is similar to HG, but uses ranked rather than weighted constraints in optimization. The parity of the VC dimension in these models is surprising because OT defines finite classes of languages — there are at most k! ways to rank k constraints — while HG defines infinite classes of languages. The linear growth of the VC dimension with the number of constraints has broad positive ramifications for the learnability of HG grammars.
I’m happy to report that Jason Riggle and I have had our submission accepted to the SIGMORPHON 2008 Workshop, which takes place concurrently with the annual meeting of the ACL at The Ohio State University this June 19th.
The citation of the paper is as follows:
Bane, Max and Jason Riggle. To appear. Three correlates of the typological frequency of quantity-insensitive stress systems.
A copy of the paper is available on ROA as document number 966.
Here is the abstract of the paper:
We examine the typology of quantity-insensitive (QI) stress systems and ask to
what extent an existing optimality theoretic model of QI stress can predict
the observed typological frequencies of stress patterns. We find three
significant correlates of pattern attestation and frequency: the trigram
entropy of a pattern, the degree to which it is “confusable” with other
patterns predicted by the model, and the number of constraint rankings that
specify the pattern.
I recently went through the odyssey of typesetting a paper for WCCFL 26 in LaTeX. WCCFL, along with a number of other conferences, publishes its proceedings through the Cascadilla Proceedings Project, which unfortunately provides no LaTeX package for authors to implement its style sheet. Some of the requirements were sufficiently tricky to implement (particularly the copyright notice) that it seems worthwhile for me to release my solution as a reusable LaTeX document class. Grab it, along with a documented example paper, here.
I gave a talk yesterday on “Multilingual Learning as Parameter Co-occurrence Clustering”, as part of the CAS Language and Cognition workshop series. The handout is available here.
My colleagues Jason Riggle, James Kirby, John Sylak and I have had our submission “Distinguishing Grammars in Multilingual Learning Using Parameter Co-occurrence” accepted to the January 3-6, 2008 annual meeting of the LSA as a 20-minute presentation. We’ll be speaking January 5th at 11:30, in a session yet to be named. Be there or be convex!
Well, my new Wordpress-based site is up and running. Unfortunately most of the content from my previous site has been lost to a catastrophic hard drive failure. So it goes…