Catálogo de publicaciones - libros

Abstract argumentation systems are formalisms for defeasible reasoning where some components remain unspecified, the structure of arguments being the main abstraction. In the dialectical process carried out to identify accepted arguments in the system some controversial situations may appear. These relate to the reintroduction of arguments into the process which cause the onset of circularity. This must be avoided in order to prevent an infinite analysis. Some systems apply the sole restriction of not allowing the introduction of previously considered arguments in an argumentation line. However, repeating an argument is not the only possible cause for the risk mentioned. A more specific restriction needs to be applied considering the existence of subarguments. In this work, we introduce an extended argumentation framework where two kinds of defeat relation are present, and a definition for .

Honorific agreement is one of the main properties of languages like Korean or Japanese, playing an important role in appropriate communication. This makes the deep processing of honorific information crucial in various computational applications such as spoken language translation and generation. We argue that, contrary to the previous literature, an adequate analysis of Korean honorification involves a system that has access not only to morpho-syntax but to semantics and pragmatics as well. Along these lines, we have developed a typed feature structure grammar of Korean (based on the framework of HPSG), and implemented it in the Linguistic Knowledge Builder (LKB). The results of parsing our experimental test suites show that our grammar provides us with enriched grammatical information that can lead to the development of a robust dialogue system for the language.

In this paper, we propose a named-entity recognition (NER) system that addresses two major limitations frequently discussed in the field. First, the system requires no human intervention such as manually labeling training data or creating gazetteers. Second, the system can handle more than the three classical named-entity types (person, location, and organization). We describe the system’s architecture and compare its performance with a supervised system. We experimentally evaluate the system on a standard corpus, with the three classical named-entity types, and also on a new corpus, with a new named-entity type (car brands).

Semantic knowledge is important in many areas of natural language processing. We propose a new unsupervised learning algorithm to annotate groups of nouns with hypernym labels. Several variations of the algorithm are presented, including a method that utilizes semantic information from WordNet. The algorithm’s results are compared against an independently-developed labeling method. The evaluation is performed using labels assigned to noun clusters by several participants of a specially designed human study.

The paper shows how to construct language patterns that signal influence strategies and tactical moves corresponding to such strategies. We apply corpus analysis methods to the extraction of certain multi-word patterns from the text data of electronic negotiations. The patterns thus acquired become features in the task of classifying those texts. A series of machine learning experiments predicts the negotiation outcome from the texts associated with first halves of negotiations. We compare the results with the classification of complete negotiations.

Over the last decades, more and more visually handicapped students have attempted post-secondary studies. This situation has created many new challenges. One of them is the need to study text and electronic documents in depth and in a reasonable time. Blind students cannot flip through the pages of a book, skim through the text or use a highlighter. In this paper, we propose a solution in the form of an experimental prototype and show how natural language processing techniques can profitably assist blind students in meeting their academic objectives. The techniques used include the automatic creation of indices, passage retrieval and the use of WordNet for query rewriting. The paper presents a technology application of a practically usable software.

The system was evaluated quantitatively and qualitatively. The evaluation is very encouraging and supports further investigation.

We present a method for text compression, which relies on pruning of a syntactic tree. The syntactic pruning applies to a complete analysis of sentences, performed by a French dependency grammar. Sub-trees in the syntactic analysis are pruned when they are labelled with targeted relations. Evaluation is performed on a corpus of sentences which have been manually compressed. The reduction ratio of extracted sentences averages around 70%, while retaining grammaticality or readability in a proportion of over 74%. Given these results on a limited set of syntactic relations, this shows promise for any application which requires compression of texts, including text summarization.

Sentence selection shares some but not all the characteristics of Automatic Text Categorization. Therefore some but not all the same techniques should be used. In this paper we study a syntactic and semantic enriched text representation for the sentence selection task in a genomics corpus. We show that using technical dictionaries and syntactic relations is beneficial for our problem when using state of the art machine learning algorithms. Furthermore, the syntactic relations can be used by a first order rule learner to obtain even better performance.

We present a sentiment tagging system which is based on multiple bootstrapping runs on WordNet synsets and glosses using different non-intersecting seed lists of manually annotated words. The system is further enhanced by the addition of a module for partial sense disambiguation of sentiment-bearing adjectives using combinatorial patterns. This (1) enables sentiment annotation at the sense, rather than whole word level, and (2) provides an effective tool for the automatic cleaning of the lists of sentiment-annotated words. The resulting cleaned list of 2907 English sentiment-bearing adjectives achieved a performance comparable to that of human annotation, as evaluated by the agreement rate between two manually annotated lists of sentiment-marked adjectives. The issues of sentiment tag extraction, evaluation and precision/recall tradeoffs are discussed.

Adaptive Benford’s Law [1] is a digital analysis technique that specifies the probabilistic distribution of digits for many commonly occurring phenomena, even for incomplete data records. We combine this digital analysis technique with a reinforcement learning technique to create a new fraud discovery approach. When applied to records of naturally occurring phenomena, our adaptive fraud detection method uses deviations from the expected Benford’s Law distributions as an indicators of anomalous behaviour that are strong indicators of fraud. Through the exploration component of our reinforcement learning method we search for the underlying attributes producing the anomalous behaviour. In a blind test of our approach, using real health and auto insurance data, our Adaptive Fraud Detection method successfully identified actual fraudsters among the test data.