Catálogo de publicaciones - libros

Natural language interaction between a student and a tutoring or an assistance system for mathematics is a new multi-disciplinary challenge that requires the interaction of (i) advanced natural language processing, (ii) flexible tutorial dialog strategies including hints, and (iii) mathematical domain reasoning. This paper provides an overview on the current research in the multi-disciplinary research project , whose goal is to build a prototype dialog-enabled system for teaching to do mathematical proofs. We present the crucial sub-systems in our architecture: the input understanding component and the domain reasoner. We present an interpretation method for mixed-language input consisting of informal and imprecise verbalization of mathematical content, and a proof manager that supports assertion-level automated theorem proving that is a crucial part of our domain reasoning module. Finally, we briefly report on an implementation of a demo system.

We report on two tests we performed in Hong Kong and Shanghai to verify the hypothesis that one can learn better when being given access to visualizations beyond the standard verbal explanations in a classroom. The outcome of the first test at HKUST was inconclusive, while the second test at Fudan University showed a clear advantage for those students who had access to visualizations.

Knowledge, cognition and intelligence are three tightly connected concepts. The Turing test is widely accepted as a test stone for machine intelligence. This paper analyzes experiences obtained in a research project on a Turing test for children and discusses its meaning with respect to some knowledge and cognition issues.

The definition, assembly and manipulation of learning objects is becoming more and more popular in learning environments. But despite standardization efforts their appropriate markup and practical usage still faces many difficulties, such as retrieval, true interoperability and cognitively adequate selection and presentation. This paper describes current work of the authors tackling some of these challenges.

Information Retrieval (IR) has played a very important role in our modern life. However, the results of search engines are not satisfactory for human intelligent activities. The platform proposed in this paper tried to solve the problems from three aspects: One is to provide domain specific IR, which consists of task oriented searching and topic related filtering. The second is to provide open-domain, concept-based retrieval to reduce irrelevant pages and overcome ambiguous keywords. The third is to provide the exact answer by question answering. The intelligent platform will facilitate searching ability on the Web. It will be easier for users to locate the interest information by keywords or questions. Researches described in the paper are developed at Shanghai Jiaotong universities and Fudan university. The experiments have showed a promising result from each aspect. The integration of these three aspects is one of the challenges of IR in the near future.

, a peer-to-peer (P2P) text retrieval and search prototype system is introduced in this paper. Compared with existing P2P systems, has three novel features: 1) The text content of the shared documents is searchable. 2) The system is open for extensions. 3) Our search and query processing techniques are implemented in the system. These techniques are designed for achieving high efficiency and scalability. The presentation of the system includes the design strategies of the system and the technologies that are implemented. We also discuss the on-going research and development work related to .

COLLATE is a project dedicated to building up a German authority center for language technology in Saarbrücken. Under this project, a computational model with three-stage pipeline architecture for Chinese information extraction has been proposed. In this paper, we concentrate on the presentation for the third stage, viz., the identification of named entity relations (NERs). A learning and identification approach for NERs called positive and negative case-based learning and identification is described in detail. It pursues the improvement of the identification performance for NERs through simultaneously learning two opposite cases, automatically selecting effective multi-level linguistic features for each NER and non-NER, and optimally achieving an identification tradeoff etc. The experimental results have shown that the overall average recall, precision, and F-measure for 14 NERs are 78.50%, 63.92% and 70.46% respectively. In addition, the above F-measure has been enhanced from 63.61% to 70.46% due to adoption of both positive and negative cases.

With the rapid growing amount of information available to us, the situations that a user needs to use a retrieval system to perform querying a multilingual document collection are becoming increasingly emerging and common. Thus an important problem is formed, to match the user queries specified in one language against documents in another different language, i.e. Cross-Language Information Retrieval (CLIR). Based on the work in CLIR evaluation task in the 9th Text Retrieval Conference (TREC-9), we have constructed an English-Chinese bi-directional CLIR system. In this system, we adopt English-Chinese bi-directional query translation as the dominant strategy, use English and Chinese queries as translation objects, and utilize English and Chinese machine readable dictionaries as the important knowledge source to acquire correct translations. By combining English and Chinese monolingual IR systems constructed by us, the complete English-Chinese bi-directional CLIR process can be implemented successfully.

We introduce in this article a blobs perspective for understanding image texture, and the subsequent motivation to characterize texture through analyzing the blobs in the textured image. Three texture description schemes arising from this motivation are discussed and their performance is experimentally evaluated. The experiment results show that a 94.9% correct classification rate on the entire Brodatz set of 112 different types of texture is achieved, which is the highest classification performance to date among the published methods according to the literature survey carried out by the authors of this article.

Enabling everybody to access content in a very simple and user friendly way becomes an increasingly important topic as content availability still increases exponentially. In this paper we will first address the problem of content annotation. The core part of the paper is a description of the LISy system which analyses TV content and allows the user to access it using a natural language dialogue system. Finally we will give a brief overview of the SmartWeb system which extends the capabilities of LISy to the access to the semantic web.