Views
3 years ago

The MATE Workbench - Architecture

The MATE Workbench - Architecture

The MATE Workbench -

The MATE WorkbenchLaila Dybkjær and Niels Ole BernsenNatural Interactive Systems Laboratory, University of Southern DenmarkScience Park 10, 5230 Odense M, Denmark{laila, nob}@nis.sdu.dkAbstractThe growing commercialisation and sophistication of spoken dialogue systems has increased the need for methods and tools in supportof annotating and extracting information from spoken dialogue resources, unimodal as well as multimodal. This paper describes resultsand future prospects of the MATE project which investigated spoken dialogue annotation and built a workbench in support ofannotation and information extraction. This workbench may have the potential for being generalised to providing tools support formultimodal dialogue annotation.1. IntroductionThe growing commercialisation and sophisticationof spoken dialogue systems has generated a strong needfor spoken dialogue corpus annotation and re-use of theannotated resources. The next generation of dialoguesystems most of which are still in the research labs, willnot be speech-only systems but will include othermodalities as well, such as gestural input and graphicsoutput. This forces extension of the focus of corpusannotation to multimodal spoken dialogue resources.During the next few years, we will probably be seeing aseries of initiatives in tools support, annotation schemesand standards for multimodal dialogue corpora.This paper describes results and future prospects ofthe MATE project (http://mate.nis.sdu.dk) which endedby the end of 1999. MATE has successfully addressedsome core issues in spoken dialogue annotation andtools support and may have the potential for beinggeneralised to providing tools support for multimodaldialogue annotation. This is now being investigated inthe ISLE project which started in 2000.2. MATEMATE was launched in March 1998 in response tothe increasing need for standards and tools in support ofcreating, annotating, evaluating and exploiting spokenlanguage resources. Corpus annotation is time- andcost-intensive, and re-use of annotated data would seemvery attractive. So far, however, re-use has usually beendifficult and time-consuming if not downright unattractivedue to the lack of standards and widely used tools.MATE aimed to facilitate re-use of spoken languageresources by addressing theoretical issues as well as thepractical implementation of solutions.MATE reviewed more than 60 existing annotationschemes relating to the annotation levels addressed inthe project, i.e. prosody, (morpho-)syntax, co-reference,dialogue acts, communication problems, and cross-levelissues. The resulting report, (Klein et al. 1998) providesdetails on their coding book, number of annotators whohave worked with it, number of annotated dialogues/-segments/utterances, evaluation results, underlying task,list of annotated phenomena, and markup languageused. Annotation examples are also provided. Theamount of pre-existing work varies enormously fromlevel to level, which causes very different state-of-theartproblems at the individual levels. The quality ofdescriptions of the individual coding schemes analysed,varies considerably which made it extremely difficult tocompare schemes even within the same annotationlevel. Furthermore, with such non-standardised descriptionsit would be impossible to create generally reusabletools within, as well as across, levels.The collected information and joint experience inthe consortium formed the basis for the development ofthe MATE markup framework for spoken dialogue corpusannotation at multiple levels, including those mentionedabove (Dybkjær et al. 1998). The core concept ofthe framework is the coding module which extends andformalises the concept of a coding scheme. Roughlyspeaking, a coding module describes everything that isneeded in order to perform a certain kind of markup of aparticular spoken language corpus. A coding moduleprescribes what constitutes a coding, including markuprepresentation and relations to other codings.The MATE markup framework ensures a commonapproach and uniform description across levels. Foreach annotation level, one or more state-of-the-art codingschemes were selected to form the basis of the bestpractice coding schemes proposed by MATE (Mengelet al. 2000). Common to the selected coding schemes isthat these are among the most widely used schemes fortheir level, which means that they have been used byseveral annotators and for the annotation of manydialogues. Examples of coding schemes selected for theMATE workbench are SAMPA, ToBI, MapTask,Verbmobil and MUC-7. All MATE best practice codingschemes are expressed in terms of coding modules.A range of existing tools for annotating spokendialogues were reviewed early in the project to provideinput to the specification of the MATE workbench(Isard et al. 1998). Examples of reviewed tools are theAlembic workbench, AnnoTag, DAT and Nb.Building on the specification, the MATE markupframework, and the best practice coding schemes, abasic set of coding functionalities has been implementedin the Java-based MATE workbench. Implementationdetails are described in (Isard et al. 2000).Workbench usability is supported by the MATE markupframework which facilitates use of the same set ofsoftware tools and the same interface look-and-feelindependently of annotation level, and serves as anintermediate layer ensuring that the user interface need

Practical part 5DD40 - Video Coding and Architectures Research ...
Open Architecture for Condition Based Maintenance - Engineering ...
Methods for Complex Single-Mind Architecture Designs - MIT Media ...
Compact, Open Architecture, Security and ... - Tejas Networks
MATE Deliverable D2.1 MATE Dialogue ... - Andreas Mengel
Introduction to the Systems Biology Workbench
A workbench for Biomedical Text Mining - Universidade do Minho
Cqi tutorial - The IMS Open Corpus Workbench
Using OpenOCD server and IAR Embedded Workbench
TABC41 ABAP Development Workbench Basics 1/2 TABC41 ... - Yimg
Ontology Architecture - i2b2
Contractual date of delivery - Architecture
Advanced Distributed Architectures
An Architecture for Linguistic and Semantic Analysis on the ... - Kwarc
A Distributed Architecture for Interactive Parse Annotation
SAAM: A Method for Analyzing the Properties of Software Architectures
Architectural Evolution - The 2nd International Conference on ...
An Approach to the Automated Evaluation of Pipeline Architectures ...
1 Instrument description and architecture
Equipment Management and Control Architecture
Architecture and module interfaces
Advanced Distributed Architectures
An Architecture for Extensible Click'n Prove Interfaces
paper pdf - Video Coding and Architectures Research Group
FPGA based generalized architecture for Modulation and ...
Qnet: A Modular Architecture for Reconfigurable Computing
Service Component Architecture Client and Implementation Model ...