André Blessing

Acquisition and Presentation of Diverse Spatial Context Data for Blind Navigation

In order to allow blind people independent navigation in unknown areas, we have developed a navigation system that seamlessly integrates both static maps and dynamic location-based textual information from a variety of sources. Each information source requires a different kind of acquisition technique. The acquired information is integrated by a context management platform and then presented to the user on a tactile-acoustical map depending on the sources available for his current position. Positioning is achieved by a combination of an inertial tracking system, RFID technology and GPS and the user is guided to a desired destination by speech output and a hap tic cane. The resulting system is the first of its kind to integrate a variety of maps and other accumulated location-based information on a unified interface for blind people.

On a generic uncertainty model for position information

Position information of moving as well as stationary objects is generally subject to uncertainties due to inherent measuring errors of positioning technologies, explicit tolerances of position update protocols, and approximations by interpolation algorithms. There exist a variety of approaches for specifying these uncertainties by mathematical uncertainty models such as tolerance regions or the Dilution of Precision (DOP) values of GPS. In this paper we propose a principled generic uncertainty model that integrates the different approaches and derive a comprehensive query interface for processing spatial queries on uncertain position information of different sources based on this model. Finally, we show how to implement our approach with prevalent existing uncertainty models.

Language-derived information and context models

There are a number of possible sources for information about the environment when creating or updating a context model, including sensorial input, databases, and explicit modeling by the system designer. Another source is natural language, either in the form of electronic text (e.g., the World Wide Web) or speech. In this paper, we investigate the implications for context models when some of their information is derived linguistically with an emphasis on the issues of hybrid models and mapping between entities in language and context model. We present a prototype that tests some of our ideas

Towards a context model driven german geo-tagging system

In this paper, we present a new approach for recognition and grounding of geographic proper names for German. Named Entity Recognition (NER) in German is more difficult than in English because not only proper names, but all nouns start with capital letters, which results in a large pool of potential ambiguous entities. Our approach makes critical use of a geographic knowledge base that is more detailed (down to the level of streets) and more structured than most knowledge bases used before. We have designed a three-stepmodel (spotting, typing, referencing) that specifies the sources of information that are necessary for geo-tagging and their dependency relationships. Basic aspects of the model were implemented and evaluated in a proof of concept. The model can be applied to other NER tasks by simply substituting the appropriate knowledge base for the one used here and retraining the model.

Automatic acquisition of vernacular places

This paper delineates our approach to augment geospatial datasets by named regions which are not typically handled by surveyors. Such regions are very important in vernacular speech and play a significant role in context-aware systems. Three different region types can be distinguished: (i) functional regions which are named after their purpose (e.g., financial district, business quarter), (ii) regions which have vernacular names (e.g. Bohnenviertel) and (iii) regions which are named after spatial attributes (Stuttgart-Süd). To acquire such regions we use a web-based approach. Our first implementation is used as a proof of concept and provides promising results. We describe several improvements which will be implemented in the future. Finally a possible scenario for a rigorous evaluation is introduced.

Crosslingual distant supervision for extracting relations of different complexity

We propose crosslingual distant supervision (crosslingual DS) for relation extraction, an approach that automatically extracts labels from a pivot language for labeling one or more target languages. The approach has two benefits compared to standard DS: (i) increased coverage if target language labels are not available; and (ii) higher accuracy of automatically generated labels because noisy labels are eliminated in crosslingual filtering. An evaluation for two relations of different complexity shows that crosslingual DS increases the accuracy of relation extraction. Our approach is language independent; we successfully apply it to four different languages: Chinese, English, French and German.

Towards a text analysis system for political debates

Social scientists and journalists nowadays have to deal with an increasingly large amount of data. It usually requires expensive searching and annotation effort to find insight in a sea of information. Our goal is to build a discourse analysis system which can be applied to large text collections. This system can help social scientists and journalists to analyze data and validate their research theories by providing them with tailored machine learning methods to alleviate the annotation effort and exploratory facilities and visualization tools. We report initial experimental results in a case study related to discourse analysis in political debates.

An End-to-end Environment for Research Question-Driven Entity Extraction and Network Analysis

This paper presents an approach to extract co-occurrence networks from literary texts. It is a deliberate decision not to aim for a fully automatic pipeline, as the literary research questions need to guide both the definition of the nature of the things that co-occur as well as how to decide co-occurrence. We showcase the approach on a Middle High German romance, Parzival. Manual inspection and discussion shows the huge impact various choices have.