Walther von Hahn

Functional Validation of a Machine Interpretation System: Verbmobil

Evaluation of NLP systems is on its way to a deep and detailed standardization. Methods and techniques are developed, but only for the evaluation of ready-to-sell products; the evaluation of a system that is still under development is not standardized and not even ad hoc tools are available for this purpose. The evaluation of Verbmobil required the development of an adequate evaluation technique and a tool that could deal both with the need to validate the system as a quasi-product and to produce useful feedback to the developers for further improvement of the system. This paper explains the methodological and technical choices that led to the implementation of a graphic evaluation tool (GET), discusses the GET and shows the results that have been gathered by its use. The paper includes a discussion of the complex problem of evaluating translations.

Multilingual Verbmobil-Dialogs: Experiments, Data Collection and Data Analysis

In this article we describe the collection and analysis of multilingual dialogs with a human or machine interpreter within the Verbmobil framework. As the dialogs represent very rare speech data with high acoustic quality, analysis is still in progress and further research is ongoing.

Von der Verknüpfung zur Integration: Kontrollstrategie oder kognitive Architektur

Die integrierte Bearbeitung von Sprachsignalen und linguistischer Struktur setzt generelle Uberlegungen uber die Architektur einer solchen Integration voraus. Die konzeptuellen Grundlagen der Arbeit im Projekt ASL1, namlich deterministische, inkrementelle und zeitsynchrone Verarbeitung, werden diskutiert, und eine Software-Losung mittels Modulinteraktion ohne zentrales Regime wird vorgestellt.

The gain of failures: Using side-effects of anaphora resolution for term consistency checks

The paper discusses a prototype module for on-line checking of term consistency in a workbench for knowledge-based Machine Aided Human Translation (MAHT). We present the linguistic resources and the knowledge base (KB) of the system as well as their place in the processes. To discover missing or misleading translations, the checker relies on the lexicon information and the hierarchy in the KB. To detect comprehension difficulties, the module checks ambiguities in the target text by resolving anaphora. The module design is based on the assumption that in aligned parallel texts some types of difficulties to resolve anaphora could be used to find wrong, missing and/or cognitively difficult translations. The paper explains the algorithms for checking term consistency. An evaluation of the approach and prospects for further work conclude the paper.

System Architectures for Speech Understanding and Language Processing

Processing natural language dialogues obviously is much more than simply processing a sequence of more than one sentence. Similarly, processing spoken dialogues is much more than handling just another type of input.

Making historical texts accessible to everybody

In this paper we discuss the degree of readability of historical texts for a broad public. We argue that text simplification methods can improve significantly this aspect and bring an added value to historical texts. We present a specific example, a genuine multilingual historical texts, which should be available at least to r e s a r h e r r m d i f r n t f i e l s a n d p r p o s a mechanism for simplifying the text.

The contribution of artificial intelligence to the human factors of application software

This paper deals with some mechanical, cognitive and social aspects of ergonomy in the following fields of Artificial Intelligence: Vision, robotics, theorem proving, speech recognition, natural language processing. A few benefits that are now visible are shown. To evaluate the use of natural language AI-systems, research (1) must consider the natural (language) environment of the problem-solving process with its heterogeneous types of information, (2) must appreciate highly restricted language utterances as linguistically appropriate to specific task environments. The benefits of future natural language systems depend on three conditions: dominance of cognitive and communicative abilities over linguistic ones, realization of an elementary, dynamic partner model, transparency, as provided by an explanation component.

On the annotation of vague expressions: a case study on Romanian historical texts

Current approaches in Digital .Humanities tend to ignore a central as-pect of any hermeneutic introspection: the intrinsic vagueness of analyzed texts. Especially when dealing with his-torical documents neglecting vague-ness has important implications on the interpretation of the results. In this pa-per we present current limitation of an-notation approaches and describe a current methodology for annotating vagueness for historical Romanian texts.

Spoken Language Systems: Technological Goals and Integration Issues

The vision of the talking, listening computer is by now, twenty-five years after HAL in Arthur Clarke’s 2001, a common-place. When people are asked to participate in a ‘Wizard of Oz’ experiment and interrogate what they are told is a computer over the telephone, it apparently never occurs to them to be surprised that a computer can understand what they have to say. But in fact as we know all too well, such an interaction is still far beyond the capabilities of the technology, outside carefully structured and constrained contexts. What will it take to move us forward toward that millenial plus one goal?

LOKI A Logic Oriented Approach to Knowledge and Data Base Supporting a Natural User Interaction

Das Projektziel von LOKI ist, ein flexibles, komplexes Zugangssystem aufzubauen, das eine naturliche Interaktion zwischen dem Benutzer und Hintergrundsystemen (vorzugsweise einer relationalen Datenbank) erlaubt.