Many statistical learning problems in NLP call for local model search methods.But accuracy tends to suffer with current techniques, which often explore either too narrowly or too broadly: hill-climbers can get stuck in local optima, whereas samplers may be inefficient.We propose to arrange individual local optimizers into organized networks.Our building blocks are operators of two types: (i) transform, which suggests new places to search, via non-random restarts from already-found local optima; and (ii) join, which merges candidate solutions to find better optima.Experiments on grammar induction show that pursuing different transforms (e.g., discarding parts of a learned model or ignoring portions of training data) results in improvements.Groups of locally-optimal solutions can be further perturbed jointly, by constructing mixtures.Using these tools, we designed several modular dependency grammar induction networks of increasing complexity.Our complete system achieves 48.6% accuracy (directed dependency macro-average over all 19 languages in the 2006/7 CoNLL data) -more than 5% higher than the previous state-of-the-art.

This paper describes a fast and fully automated dictionary based approach to gene annotation and exon prediction. Two dictionaries are constructed, one from the nonredundant protein OWL database and the other from the dbEST database. These dictionaries are used to obtain O(1) time lookups of tuples in the dictionaries (4 tuples for the OWL database and 11 tuples for the \ndbEST database). These tuples can be used to rapidly find the longest matches at every position in an input sequence to the database sequences. Such matches provide very useful information pertaining to locating common segments between exons, alternative splice sites, and frequency data of long tuples for statistical purposes. These dictionaries also provide the basis for both homology determination, and statistical approaches to exon prediction. For instance, using the OWL protein database on a benchmark test set of 130 genes, and after removing sequences from the database with exact amino acid homology to genes in our test set, we find 88% of coding nucleotides, and 99% of our predictions of coding nucleotides are correct. Also, 81% of coding exons are predicted exactly, while 82% of our predictions of exons agree exactly with the published annotation of their genes.