Catálogo de publicaciones - libros

A new task is identified in the ongoing analysis of opinions: finding propositional opinions, sentential complement clauses of verbs such as “believe” or “claim” that express opinions, and the holders of these opinions. An extension of semantic parsing techniques is proposed that, coupled with additional lexical and syntactic features, can extract these propositional opinions and their opinion holders. A small corpus of 5,139 sentences is annotated with propositional opinion information, and is used for training and evaluation. While our results are still quite preliminary (precisions of 43–51% and recalls of 58–68%), we feel that our focus on opinion clauses, and in general the use of rich syntactic features, helps point to an important new direction in opinion detection.

The tagging of discourse is important not only for natural language processing research, but for many applications in the social sciences as well. This chapter describes an evaluation of a range of different tagging techniques to automatically determine the attitude of speakers in transcribed psychiatric dialogues. It presents results in a marriage-counseling domain that classifies the attitude and emotional commitment of the participants to a particular topic of discussion. It also gives results from the Switchboard Corpus to facilitate comparison for future work. Finally, it describes a new Java tool that learns attitude classifications using our techniques and provides a flexible, easy to use platform for tagging of texts.

We address the problem of automatically classifying academic citations in scientific articles according to author affect. There are many ways how a citation might fit into the overall argumentation of the article: as part of the solution, as rival approach or as flawed approach that justifies the current research. Our motivation for this work is to improve citation indexing. The method we use for this task is machine learning from indicators of affect (such as , or and of presentation of ownership of ideas (such as , or ). Some of these features are borrowed from Argumentative Zoning (Teufel and Moens, 2002), a technique for determining the rhetorical status of each sentence in a scientific article. These features include the type of subject of the sentence, the citation type, the semantic class of main verb, and a list of indicator phrases. Evaluation will be both intrinsic and extrinsic, involving the measurement of human agreement on the task and a comparison of human and automatic evaluation, as well as a comparison of task-performance with our system versus task performance with a standard citation indexer (CiteSeer, Lawrence et al., 1999).

In this chapter, two empirical pilot studies on the role of politeness in dialogue summarization are described. In these studies, a collection of four dialogues was used. Each dialogue was automatically generated by the NECA system and the politeness of the dialogue participants was systematically manipulated. Subjects were divided into groups who had to summarize the dialogues from a particular dialogue participant’s point of view or the point of view of an impartial observer. In the first study, there were no other constraints. In the second study, the summarizers were restricted to summaries whose length did not exceed 10% of the number of words in the dialogue that was being summarized.

Amongst other things, it was found that the politeness of the interaction is included more often in summaries of dialogues that deviate from what would be considered normal or unmarked. A comparison of the results of the two studies suggests that the extent to which politeness is reported is not affected by how long a summary is allowed to be. It was also found that the point of view of the summarizer influences which information is included in the summary and how it is presented. This finding did not seem to be affected by the constraint in our second study on the summary length.

We present experiments on modifying the semantic orientation of the near-synonyms in a text. We analyze a text into an interlingual representation and a set of attitudinal nuances, with particular focus on its near-synonyms. Then we use our text generator to produce a text with the same meaning but changed semantic orientation (more positive or more negative) by replacing, wherever possible, words with near-synonyms that differ in their expressed attitude.

This chapter uses Systemic Functional Linguistic (SFL) theory as a basis for extracting semantic features of documents. We focus on the pronominal and determination system and the role it plays in constructing interpersonal distance. By using a hierarchical system model that represents the author’s language choices, it is possible to construct a richer and more informative feature representation with superior computational efficiency than the usual bag-of-words approach. Experiments within the context of financial scam classification show that these systemic features can create clear separation between registers with different interpersonal distance. This approach is generalizable to other aspects of attitude and affect that have been modelled within the systemic functional linguistic theory.

This chapter studies the use of textual features based on systemic functional linguistics, for genre-based text categorization. We describe feature sets that represent different types of conjunctions and modal assessment, which together can partially indicate how different genres structure text and may prefer certain classes of attitudes towards propositions in the text. This enables analysis of large-scale rhetorical differences between genres by examining which features are important for classification. The specific domain we studied comprises scientific articles in historical and experimental sciences (paleontology and physical chemistry, respectively). We applied the SMO learning algorithm, which with our feature set achieved over 83% accuracy for classifying articles according to field, though no field-specific terms were used as features. The most highly-weighted features for each were consistent with hypothesized methodological differences between historical and experimental sciences, thus lending empirical evidence to the recent philosophical claim of multiple scientific methods.

We present a system that applies Argumentative Zoning (AZ) (Teufel and Moens, 2002), a method of determining argumentative structure in texts, to the task of advising novice graduate writers on their writing. For this task, it is important to automatically determine the rhetorical/argumentative status of a given sentence in the text. On the basis of this information, users can be advised that a different sentence order might be more advantageous or that certain argumentative moves are missing. In implementing such a system, we had to port AZ from English to Portuguese, as our system is designed to help the writing of Brazilian PhD theses in Computer Science. In this chapter, we report on the overall system, named SciPo, the porting exercise, including a human annotation experiment to verify the reproducibility of our annotation scheme, and the intrinsic and extrinsic evaluation of the AZ module of the system.

Citations in scientific writing fulfil an important role in creating relationships among mutually relevant articles within a research field. These inter-article relationships reinforce the argumentation structure intrinsic to all scientific writing. Therefore, determining the nature of the exact relationship between a citing and cited paper requires an understanding of the rhetorical relations within the argumentative context in which a citation is placed. To determine these relations automatically, we have suggested that various stylistic and rhetorical cues will be significant. One such cue that we are studying is the use of hedging to modify the affect of a scientific claim. We provide evidence that hedging occurs more frequently in citation contexts than in the text as a whole. With this information we conjecture that hedging is a significant aspect of the rhetorical structure of citation contexts and that the pragmatics of hedges may help in determining the rhetorical purpose of citations. A citation indexing tool for biomedical literature analysis is introduced.

This chapter describes an automated system for detecting polar expressions about a specified topic. The two elementary components of this approach are a shallow NLP polar language extraction system and a machine learning based topic classifier. These components are composed together by making a simple but accurate collocation assumption: if a topical sentence contains polar language, the polarity is associated with the topic. We evaluate our system, components and assumption on a corpus of online consumer messages.

Based on these components, we discuss how to measure the overall sentiment about a particular topic as expressed in online messages authored by many different people. We propose to use the fundamentals of Bayesian statistics to form an aggregate authorial opinion metric. This metric would propagate uncertainties introduced by the polarity and topic modules to facilitate statistically valid comparisons of opinion across multiple topics.