Abstract. In this paper, we propose the use of the Maximum Entropy approach for the task of automatic image annotation. Given labeled training data, Maximum Entropy is a statistical technique which allows one to predict the probability of a label given test data. The techniques allow for relationships between features to be effectively captured. and has been successfully applied to a number of language tasks including machine translation. In our case, we view the image annotation task as one where a training data set of images labeled with keywords is provided and we need to automatically label the test images with keywords. To do this, we first represent the image using a language of visterms and then predict the probability of seeing an English word given the set of visterms forming the image. Maximum Entropy allows us to compute the probability and in addition allows for the relationships between visterms to be incorporated. The experimental results show that Maximum Entropy outperforms one of the classical translation models that has been applied to this task and the Cross Media Relevance Model. Since the Maximum Entropy model allows for the use of a large number of predicates to possibly increase performance even further, Maximum Entropy model is a promising model for the task of automatic image annotation. 1

We present a heuristic meta-learning search method for finding a set of optimized algorithmic parameters for a range of machine learning algorithms. The method, wrapped progressive sampling, is a combination of classifier wrapping and progressive sampling of training data. A series of experiments on UCI benchmark data sets with nominal features, and five machine learning algorithms to which simple wrapping and wrapped progressive sampling is applied, yields results that show little improvement for the algorithm which offers few parameter variations, but marked improvements for the algorithms offering many possible testable parameter combinations, yielding up to 32.2 % error reduction with the winnow learning algorithm. 1

Data-driven grammatical function tag assignment has been studied for English using the Penn-II Treebank data. In this paper we address the question of whether such methods can be applied successfully to other languages and treebank resources. In addition to tag assignment accuracy and f-scores we also present results of a task-based evaluation. We use three machine-learning methods to assign Cast3LB function tags to sentences parsed with Bikel’s parser trained on the Cast3LB treebank. The best performing method, SVM, achieves an f-score of 86.87 % on gold-standard trees and 66.67 % on parser output- a statistically significant improvement of 6.74 % over the baseline. In a task-based evaluation we generate LFG functional-structures from the functiontag-enriched trees. On this task we achive an f-score of 75.67%, a statistically significant 3.4 % improvement over the baseline. 1

In this paper we investigate the use of machine learning techniques to classify a wide range of non-sentential utterance types in dialogue, a necessary first step in the interpretation of such fragments. We train different learners on a set of contextual features that can be extracted from PoS information. Our results achieve an 87 % weighted f-score—a 25 % improvement over a simple rule-based algorithm baseline. Keywords Non-sentential utterances, machine learning, corpus analysis 1

resolution

This paper describes a method for automatically extending the lexicon of wide-coverage parsers. The method is an extension to the automatic detection of coverage problems of natural language parsers, based on large amounts of raw text (van Noord 2004). The goal is to extend grammar coverage, focusing in particular on the acquisition of lexical information for missing and incomplete lexicon entries (including subcategorization frames). In order to assign lexical entries for unknown words, or for words for which the lexicon only contains a subset of its possible lexical categories, we propose to apply a parser to a set of unannotated sentences containing the unknown word, or to a set of unannotated sentences (found by error mining) in which the word apparently was used with a missing lexical category. The parser will assign all universal lexical categories to the problematic word. Once the parser has found a result for the sentence, it can output the lexical category that was eventually used in its best parse. If this process is repeated for a large enough sample of sentences, it is expected that either a single or a small number of lexical categories can then be identified which are to be taken as the correct lexical categories of this word. A maximum entropy classifier is trained to select the correct lexical categories.

This dissertation addresses the task of detecting instances of object categories in photographs. We propose modeling an object category as a collection of object parts linked together in a deformable configuration. We propose two different approaches to model the appearance of object parts that provide robustness to intra-class variations and viewpoint change. The first approach models object parts as locally rigid assemblies of dense feature points and part detection proceeds by incrementally matching the feature points between the model image and the test image. The second approach employs a discriminative classifier (Support Vector Machine) based on a descriptor that consists of a combination of a sparse visual word histogram pyramid and a dense gradient and edge histogram pyramid. We also propose two different approaches for modeling the inter-part relations and algorithms for efficiently learning the model parameters. The first approach uses a generative model that models the joint probability distribution over the locations and visibility of all the object parts. The second approach employs a discriminative Conditional Random Field based model to encode the relative geometry and co-occurrence constraints.

Abstract not found

at SemEval 2007. The system is based on two supervised

Abstract not found