Dan Roth

Learning to detect objects in images via a sparse, part-based representation

We study the problem of detecting objects in still, gray-scale images. Our primary focus is the development of a learning-based approach to the problem that makes use of a sparse, part-based representation. A vocabulary of distinctive object parts is automatically constructed from a set of sample images of the object class of interest; images are then represented using parts from this vocabulary, together with spatial relations observed among the parts. Based on this representation, a learning algorithm is used to automatically learn to detect instances of the object class in new images. The approach can be applied to any object with distinguishable parts in a relatively fixed spatial configuration; it is evaluated here on difficult sets of real-world images containing side views of cars, and is seen to successfully detect objects in varying conditions amidst background clutter and mild occlusion. In evaluating object detection approaches, several important methodological issues arise that have not been satisfactorily addressed in the previous work. A secondary focus of this paper is to highlight these issues, and to develop rigorous evaluation standards for the object detection problem. A critical evaluation of our approach under the proposed standards is presented.

Design Challenges and Misconceptions in Named Entity Recognition

We analyze some of the fundamental design challenges and misconceptions that underlie the development of an efficient and robust NER system. In particular, we address issues such as the representation of text chunks, the inference approach needed to combine local NER decisions, the sources of prior knowledge and how to use them within an NER system. In the process of comparing several solutions to these challenges we reach some surprising conclusions, as well as develop an NER system that achieves 90.8 F1 score on the CoNLL-2003 NER shared task, the best reported result for this dataset.

Learning question classifiers

In order to respond correctly to a free form factual question given a large collection of texts, one needs to understand the question to a level that allows determining some of the constraints the question imposes on a possible answer. These constraints may include a semantic classification of the sought after answer and may even suggest using different strategies when looking for and verifying a candidate answer.This paper presents a machine learning approach to question classification. We learn a hierarchical classifier that is guided by a layered semantic hierarchy of answer types, and eventually classifies questions into fine-grained classes. We show accurate results on a large collection of free-form questions used in TREC 10.

Learning a Sparse Representation for Object Detection

We present an approach for learning to detect objects in still gray images, that is based on a sparse, part-based representation of objects. A vocabulary of information-rich object parts is automatically constructed from a set of sample images of the object class of interest. Images are then represented using parts from this vocabulary, along with spatial relations observed among them. Based on this representation, a feature-efficient learning algorithm is used to learn to detect instances of the object class. The framework developed can be applied to any object with distinguishable parts in a relatively fixed spatial configuration. We report experiments on images of side views of cars. Our experiments show that the method achieves high detection accuracy on a difficult test set of real-world images, and is highly robust to partial occlusion and background variation.In addition, we discuss and offer solutions to several methodological issues that are significant for the research community to be able to evaluate object detection approaches.

Emotions from Text: Machine Learning for Text-based Emotion Prediction

In addition to information, text contains attitudinal, and more specifically, emotional content. This paper explores the text-based emotion prediction problem empirically, using supervised machine learning with the SNoW learning architecture. The goal is to classify the emotional affinity of sentences in the narrative domain of childrens fairy tales, for subsequent usage in appropriate expressive rendering of text-to-speech synthesis. Initial experiments on a preliminary data set of 22 fairy tales show encouraging results over a naive baseline and BOW approach for classification of emotional versus non-emotional contents, with some dependency on parameter tuning. We also discuss results for a tripartite model which covers emotional valence, as well as feature set alternations. In addition, we present plans for a more cognitively sound sequential model, taking into consideration a larger set of basic emotions.

A Winnow-Based Approach to Context-Sensitive Spelling Correction

A large class of machine-learning problems in natural language require the characterization of linguistic context. Two characteristic properties of such problems are that their feature space is of very high dimensionality, and their target concepts depend on only a small subset of the features in the space. Under such conditions, multiplicative weight-update algorithms such as Winnow have been shown to have exceptionally good theoretical properties. In the work reported here, we present an algorithm combining variants of Winnow and weighted-majority voting, and apply it to a problem in the aforementioned class: context-sensitive spelling correction. This is the task of fixing spelling errors that happen to result in valid words, such as substituting to for too, casual for causal, and so on. We evaluate our algorithm, WinSpell, by comparing it against BaySpell, a statistics-based method representing the state of the art for this task. We find: (1) When run with a full (unpruned) set of features, WinSpell achieves accuracies significantly higher than BaySpell was able to achieve in either the pruned or unpruned condition; (2) When compared with other systems in the literature, WinSpell exhibits the highest performance; (3) While several aspects of WinSpells architecture contribute to its superiority over BaySpell, the primary factor is that it is able to learn a better linear separator than BaySpell learns; (4) When run on a test set drawn from a different corpus than the training set was drawn from, WinSpell is better able than BaySpell to adapt, using a strategy we will present that combines supervised learning on the training set with unsupervised learning on the (noisy) test set.

The importance of syntactic parsing and inference in semantic role labeling

We present a general framework for semantic role labeling. The framework combines a machine-learning technique with an integer linear programming-based inference procedure, which incorporates linguistic and structural constraints into a global decision process. Within this framework, we study the role of syntactic parsing information in semantic role labeling. We show that full syntactic parsing information is, by far, most relevant in identifying the argument, especially, in the very first stagethe pruning stage. Surprisingly, the quality of the pruning stage cannot be solely determined based on its recall and precision. Instead, it depends on the characteristics of the output candidates that determine the difficulty of the downstream problems. Motivated by this observation, we propose an effective and simple approach of combining different semantic role labeling systems through joint inference, which significantly improves its performance. Our system has been evaluated in the CoNLL-2005 shared task on semantic role labeling, and achieves the highest F1 score among 19 participants.

Understanding the Value of Features for Coreference Resolution

In recent years there has been substantial work on the important problem of coreference resolution, most of which has concentrated on the development of new models and algorithmic techniques. These works often show that complex models improve over a weak pairwise baseline. However, less attention has been given to the importance of selecting strong features to support learning a coreference model. This paper describes a rather simple pairwise classification model for coreference resolution, developed with a well-designed set of features. We show that this produces a state-of-the-art system that outperforms systems built with complex models. We suggest that our system can be used as a baseline for the development of more complex models -- which may have less impact when a more robust set of features is used. The paper also presents an ablation study and discusses the relative contributions of various features.

Learning question classifiers: the role of semantic information

To respond correctly to a free form factual question given a large collection of text data, one needs to understand the question to a level that allows determining some of the constraints the question imposes on a possible answer. These constraints may include a semantic classification of the sought after answer and may even suggest using different strategies when looking for and verifying a candidate answer. This work presents a machine learning approach to question classification. Guided by a layered semantic hierarchy of answer types, we develop a hierarchical classifier that classifies questions into fine-grained classes. This work also performs a systematic study of the use of semantic information sources in natural language classification tasks. It is shown that, in the context of question classification, augmenting the input of the classifier with appropriate semantic category information results in significant improvements to classification accuracy. We show accurate results on a large collection of free-form questions used in TREC 10 and 11.

Integer linear programming inference for conditional random fields

Inference in Conditional Random Fields and Hidden Markov Models is done using the Viterbi algorithm, an efficient dynamic programming algorithm. In many cases, general (non-local and non-sequential) constraints may exist over the output sequence, but cannot be incorporated and exploited in a natural way by this inference procedure. This paper proposes a novel inference procedure based on integer linear programming (ILP) and extends CRF models to naturally and efficiently support general constraint structures. For sequential constraints, this procedure reduces to simple linear programming as the inference process. Experimental evidence is supplied in the context of an important NLP problem, semantic role labeling.