Text summarization aims to condense lengthy documents into concise, coherent summaries. Although recent abstractive models achieve strong performance, they often struggle to accurately identify the most salient information, leading to summaries that lack faithfulness and coherence. Our goal is to demonstrate how summarization can benefit from discourse parsing to enhance the faithfulness of generated summaries. To achieve this, we propose RST Nuclei for Discourse-Aware Abstractive Summarization (RST-NDAS), a framework that leverages Rhetorical Structure Theory (RST) to improve salience identification, summary coherence, and overall quality. Our method segments documents into Elementary Discourse Units (EDUs), identifies nucleus components, and uses this discourse structure to guide summarization models. Evaluations on the RST-DT and CNN/DailyMail datasets demonstrate that RST-NDAS outperforms strong baselines, showing that integrating discourse structure as importance signals improves salience identification, coherence, and summary faithfulness.

Full natural language understanding requires identifying and analyzing the meanings of metaphors, which are ubiquitous in both text and speech. Over the last thirty years, linguistic metaphors have been shown to be based on more general conceptual metaphors, partial semantic mappings between disparate conceptual domains. Though some achievements have been made in identifying linguistic metaphors over the last decade or so, little work has been done to date on automatically identifying conceptual metaphors. This paper describes research on identifying conceptual metaphors based on corpus data. Our method uses as little background knowledge as possible, to ease transfer to new languages and to mini- mize any bias introduced by the knowledge base construction process. The method relies on general heuristics for identifying linguistic metaphors and statistical clustering (guided by Wordnet) to form conceptual metaphor candidates. Human experiments show the system effectively finds meaningful conceptual metaphors.

A central concern in Evidence Based Medicine (EBM) is how to convey research results effectively to practitioners. One important idea is to summarize results by key summary statistics that describe the effectiveness (or lack thereof) of a given intervention, specifically the absolute risk reduction (ARR) and number needed to treat (NNT). Manual summarization is slow and expensive, thus, with the exponential growth of the biomedical research literature, automated solutions are needed. In this paper, we present a novel method for automatically creating EBM-oriented summaries from research abstracts of randomly-controlled trials (RCTs). The system extracts descriptions of the treatment groups and outcomes, as well as various associated quantities, and then calculates summary statistics. Results on a hand-annotated corpus of research abstracts show promising, and potentially useful, results.

Abstract Statistical authorship attribution has a long history, culminating in the use of modern machine learning classification methods. Nevertheless, most of this work suffers from the limitation of assuming a small closed set of candidate authors and essentially unlimited training text for each. Real‐life authorship attribution problems, however, typically fall short of this ideal. Thus, following detailed discussion of previous work, three scenarios are considered here for which solutions to the basic attribution problem are inadequate. In the first variant, the profiling problem, there is no candidate set at all; in this case, the challenge is to provide as much demographic or psychological information as possible about the author. In the second variant, the needle‐in‐a‐haystack problem, there are many thousands of candidates for each of whom we might have a very limited writing sample. In the third variant, the verification problem, there is no closed candidate set but there is one suspect; in this case, the challenge is to determine if the suspect is or is not the author. For each variant, it is shown how machine learning methods can be adapted to handle the special challenges of that variant.

In this paper, we address a relatively new and interesting text categorization problem: classify a political blog as either liberal or conservative, based on its political leaning. Our subjectivity analysis based method is twofold: 1) we identify subjective sentences that contain at least two strong subjective clues based on the General Inquirer dictionary; 2) from subjective sentences identified, we extract opinion expressions and other features to build political leaning classifiers. Experimental results with a political blog corpus we built show that by using features from subjective sentences can significantly improve the classification performance. In addition, by extracting opinion expressions from subjective sentences, we are able to reveal opinions that are characteristic of a specific political leaning to some extent.

Abstract Most text analysis and retrieval work to date has focused on the topic of a text; that is, what it is about. However, a text also contains much useful information in its style, or how it is written. This includes information about its author, its purpose, feelings it is meant to evoke, and more. This article develops a new type of lexical feature for use in stylistic text classification, based on taxonomies of various semantic functions of certain choice words or phrases. We demonstrate the usefulness of such features for the stylistic text classification tasks of determining author identity and nationality, the gender of literary characters, a text's sentiment (positive/negative evaluation), and the rhetorical character of scientific journal articles. We further show how the use of functional features aids in gaining insight about stylistic differences among different kinds of texts.

We developed a prototype for integrated retrieval and aggregation of diverse information contained in scanned paper documents. Such complex document information processing combines several forms of image processing together with textual/linguistic processing to enable effective analysis of complex document collections, a necessity for a wide range of applications. This is the first system to attempt integrated retrieval from complex documents; we report its current capabilities.

This paper focuses on a method for the stylistic segmentation of text documents. Our technique involves mapping the change in a feature throughout a text. We use the linguistic features of conjunction and modality, through taxonomies from Systemic Functional Linguistics. This segmentation has applications in automated summarization, particularly of large documents.

Much interesting text n the web consists largely of opinionated or evaluative text, as opposed to directly informative text. The new field of 'sentiment analysis' seeks to characterize such aspects of natural language text, as opposed to just the bare facts. We suggest that 'appraisal expression extraction' should be viewed as a fundamental task for sentiment analysis. We define an 'appraisal expression' to be a piece of text expressing some evaluative stance towards a particular object. The task is to find these elements and characterize the type and orientation (positive or negative) of the evaluative stance, as well as its target and possibly its source. Potential applications of these methods include new approaches to the now-traditional tasks of sentiment classification and pinion mining, as well as possibly for adversarial textual analysis and intention detection for intelligence applications.

In this paper, we use a blog corpus to demonstrate that we can often identify the author of an anonymous text even where there are many thousands of candidate authors. Our approach combines standard information retrieval methods with a text categorization meta-learning scheme that determines when to even venture a guess.

Research and development of information access technology for scanned paper documents has been hampered by the lack of public test collections of realistic scope and complexity. As part of a project to create a prototype system for search and mining of masses of document images, we are assembling a 1.5 terabyte dataset to support evaluation of both end-to-end complex document information processing (CDIP) tasks (e.g., text retrieval and data mining) as well as component technologies such as optical character recognition (OCR), document structure analysis, signature matching, and authorship attribution.

Recently, interest is growing in non-topical text classification tasks such as genre classification, sentiment analysis, and authorship profiling. We study to what extent OCR errors affect stylistic text classification from scanned documents. We find that even a relatively high level of errors in the OCRed documents does not substantially affect stylistic classification accuracy.

Little work to date in sentiment analysis (classifying texts by `positive' or `negative' orientation) has attempted to use fine-grained semantic distinctions in features used for classification. We present a new method for sentiment classification based on extracting and analyzing appraisal groups such as ``very good'' or ``not terribly funny''. An appraisal group is represented as a set of attribute values in several task-independent semantic taxonomies, based on Appraisal Theory. Semi-automated methods were used to build a lexicon of appraising adjectives and their modifiers. We classify movie reviews using features based upon these taxonomies combined with standard ``bag-of-words'' features, and report state-of-the-art accuracy of 90.2%. In addition, we find that some types of appraisal appear to be more significant for sentiment classification than others.

This paper considers the use of computational stylistics for performing authorship attribution of electronic messages, addressing categorization problems with as many as 20 different classes (authors). Effective stylistic characterization of text is potentially useful for a variety of tasks, as language style contains cues regarding the authorship, purpose, and mood of the text, all of which would be useful adjuncts to information retrieval or knowledge-management tasks. We focus here on the problem of determining the author of an anonymous message, based only on the message text. Several multiclass variants of the Winnow algorithm were applied to a vector representation of the message texts to learn models for discriminating different authors. We present results comparing the classification accuracy of the different approaches. The results show that stylistic models can be accurately learned to determine an author's identity.

In this paper, we address the tradeoff between exploration and exploitation for agents which need to learn more about the structure of their environment in order to perform more effectively. For example, a software agent operating on the World Wide Web may need to learn which sites on the net are most useful, and the most efficient routes to those sites. We compare exploration strategies for a repeated task, where the agent is given some particular task to perform some number of times. Tasks are modeled as navigation on a partially known (deterministic) graph. This paper describes a new utility-based exploration algorithm for repeated tasks which interleaves exploration with task performance. The method takes into account both the costs and the potential benefits (for future task repetitions) of different exploratory actions. Exploration is performed in a greedy fashion, with the locally optimal exploratory action performed during repetition of each task. We experimentally evaluated our utility-based interleaved exploration algorithm against a heuristic search algorithm for exploration before task performance (a priori exploration) as well as a randomized interleaved exploration algorithm. We found that for a single repeated task, utility-based interleaved exploration consistently outperforms the alternatives, unless the number of task repetitions is very high. In addition, we extended the algorithms for the case of multiple repeated tasks, where the agent has a different, randomly-chosen task (from a known subset of possible tasks) to perform each time. Here too, we found that utility-based interleaved exploration is clear in most cases.

In many real-world learning tasks, it is expensive to acquire a sufficient number of labeled examples for training. This paper investigates methods for reducing annotation cost by `sample selection'. In this approach, during training the learning program examines many unlabeled examples and selects for labeling only those that are most informative at each stage. This avoids redundantly labeling examples that contribute little new information. Our work follows on previous research on Query By Committee, extending the committee-based paradigm to the context of probabilistic classification. We describe a family of empirical methods for committee-based sample selection in probabilistic classification models, which evaluate the informativeness of an example by measuring the degree of disagreement between several model variants. These variants (the committee) are drawn randomly from a probability distribution conditioned by the training set labeled so far. The method was applied to the real-world natural language processing task of stochastic part-of-speech tagging. We find that all variants of the method achieve a significant reduction in annotation cost, although their computational efficiency differs. In particular, the simplest variant, a two member committee with no parameters to tune, gives excellent results. We also show that sample selection yields a significant reduction in the size of the model used by the tagger.

In this paper we consider the problem of `theory patching', in which we are given a domain theory, some of whose components are indicated to be possibly flawed, and a set of labeled training examples for the domain concept. The theory patching problem is to revise only the indicated components of the theory, such that the resulting theory correctly classifies all the training examples. Theory patching is thus a type of theory revision in which revisions are made to individual components of the theory. Our concern in this paper is to determine for which classes of logical domain theories the theory patching problem is tractable. We consider both propositional and first-order domain theories, and show that the theory patching problem is equivalent to that of determining what information contained in a theory is `stable' regardless of what revisions might be performed to the theory. We show that determining stability is tractable if the input theory satisfies two conditions: that revisions to each theory component have monotonic effects on the classification of examples, and that theory components act independently in the classification of examples in the theory. We also show how the concepts introduced can be used to determine the soundness and completeness of particular theory patching algorithms.

Corpus-based methods for natural language processing often use supervised training, requiring expensive manual annotation of training corpora. This paper investigates methods for reducing annotation cost by sample selection. In this approach, during training the learning program examines many unlabeled examples and selects for labeling (annotation) only those that are most informative at each stage. This avoids redundantly annotating examples that contribute little new information. This paper extends our previous work on committee-based sample selection for probabilistic classifiers. We describe a family of methods for committee-based sample selection, and report experimental results for the task of stochastic part-of-speech tagging. We find that all variants achieve a significant reduction in annotation cost, though their computational efficiency differs. In particular, the simplest method, which has no parameters to tune, gives excellent results. We also show that sample selection yields a significant reduction in the size of the model used by the tagger.

An issue that must be addressed in map-learning systems is that of error accumulation. The primary emphasis in the literature has been on reducing errors entering the map. The authors suggest that this methodology must reach a point of diminishing returns, and hence focus on explicit error detection and correction. By identifying the possible types of mapping errors, structural constraints can be exploited to detect and diagnose mapping errors. Such robust mapping requires little overhead beyond that needed for nonrobust mapping. A mapping system was implemented based on those ideas. Extensive testing in simulation demonstrated the effectiveness of the proposed error-correction strategies.>