In this paper we present a human-based evaluation of surface realisation alternatives. We examine the relative rankings of naturally occurring corpus sentences and automatically generated strings chosen by statistical models (language model, loglinear model), as well as the naturalness of the strings chosen by the log-linear model. We also investigate to what extent preceding context has an eff...

Categorical data—frequency data, and discrete data—are most often presented in tables, and analyses using loglinear models and logistic regression are most often presented in terms of parameter estimates. Over the past decade, I and others have developed novel visualization methods for categorical data, designed to provide exploratory and confirmatory graphic displays analogous to those used re...

F2F is a C++ library for large-scale machine learning. It contains a CPU optimized implementation of the Fastfood algorithm in Le et al. (2013), that allows the computation of approximated kernel expansions in loglinear time. The algorithm requires to compute the product of Walsh-Hadamard Transform (WHT) matrices. A cache friendly SIMD Fast Walsh-Hadamard Transform (FWHT) that achieves compelli...

Observations which seem to deviate strongly from the main part of the data may occur in every statistical analysis. These observations, usually labelled as outliers, may cause completely misleading results when using standard methods and may also contain information about special events or dependencies. Therefore it is of interest to identify them. We discuss outliers in situations where a gene...

Language models can be formalized as loglinear regression models where the input features represent previously observed contexts up to a certain length m. The complexity of existing algorithms to learn the parameters by maximum likelihood scale linearly in nd, where n is the length of the training corpus and d is the number of observed features. We present a model that grows logarithmically in ...

Statistical methods for categorical data, such as loglinear models and logistic regression, represent discrete analogs of the analysis of variance and regression methods for continuous response variables. However. while graphical display techniques are common adjuncts to analysis of variance and regression, methods for plotting contingency table data are not as widely used. This paper provides ...

The F&B-index is used to speed up pattern matching in tree and graph data, and is based on the maximum F&B-bisimulation, which can be computed in loglinear time for graphs. It has been shown that the maximum F-bisimulation can be computed in linear time for DAGs. We build on this result, and introduce a linear algorithm for computing the maximum F&B-bisimulation for tree data. It first computes...

Statistical methods for categorical data. such as loglinear models and logistic regression, represent discrete analogs of the analysis of variance and regression methods for 'ontinuous response variables. However. while graphical display techniques are common adjuncts to analysis of ~~-variance and regression. methods for plotting contingency table data are not as widely used. This paper provid...

We present a new approach to stochastic modeling of constraintbased grammars that is based on loglinear models and uses EM for estimation from unannotated data. The techniques are applied to an LFG grammar for German. Evaluation on an exact match task yields 86% precision for an ambiguity rate of 5.4, and 90% precision on a subcat frame match for an ambiguity rate of 25. Experimental comparison...