M.Sc. Sebastian Germesin


  • M.Sc. Sebastian Germesin
  • Researcher
  • DFKI GmbH
  • ℅ DFKI GmbH
    Office: C0.13
    Stuhlsatzenhausweg 3
    66123 Saarbrücken
  • Phone:(+49) (0)681.85775.5079
    Fax:(+49) (0)681.85775.5021
Sebastian Germesin

Curriculum Vitae

Sebastian Germesin was born in Merzig, Germany and started his studies of computer science at the Saarland University in 2002.

He finished his studies in May 2008 with a Master's degree and after that, he started working in the position of a Researcher at the DFKI - the German Research Center for Artificial Intelligence - at Saarbrücken in the AMIDA project which is a multi-disciplinary research project whose aims are to develop technology to support human interaction in meetings and to provide better structure to the way meetings are run and documented. His main research topics were the classification of dialogue acts and furthermore the detection and classification of speech production errors (so called disfluencies) which occur in spoken language.

With the end of AMIDA in 2009, he joined the EU-funded IKS project. IKS' (Interactive Knowledge Stack) main objective is to provide open source technologies for semantically enhanced content management systems CMS. Sebastian Germesin's key task is the definition and development of semantic Interaction Patterns to model the interaction of a user with semantically enriched content. During this work, he also joined the Semantic Editor group and initiated the Semantic Interaction webpage.

Feel free to download the full CV.


  • Semantic Interaction Patterns
  • Web Development - jQuery, jQuery UI, HTML5, CSS3, PHP
  • Discourse Modeling
  • Machine Learning
  • Disfluency Detection & Correction
  • Agreement/Disagreement Detection
  • Dialog Act Recognition

Recent Publications


pdfVishal Juneja and Sebastian Germesin and Thomas Kleinbauer, "A Learning-based Sampling Approach to Extractive Summarization", In: External LinkStudent Research Workshop of NAACL-HLT 2010, Los Angeles, USA, 2010


pdf Sebastian Germesin and Theresa Wilson, "Agreement Detection in Multiparty Conversation", In: Proceedings of External LinkICMI-MLMI 09, Cambridge, USA, 2009, starOutstanding Student Paper Award
pdf Thomas Kleinbauer and Sebastian Germesin, "ARKTiS - A Fast Tag Recommender System Based On Heuristics", In: External LinkECML PKDD (Discovery Challenge) 09, Bled, Slovenia, 2009
pdf Andrei Popescu-Belis, Peter Poller, Jonathan Kilgour, Mike Flynn, Sebastian Germesin, Alexandre Nanchen and Majid Yazdani, "Comparing Meeting Browsers using a Task-based Evaluation Method", Research Report no. Idiap-RR-38-09, External LinkIdiap Research Institute, Martigny, Switzerland, December, 2009


pdf Sebastian Germesin, Tilman Becker and Peter Poller, "Domain-specific Classification Methods for Disfluency Detection", In: Proceedings of External LinkInterspeech 08, Brisbane, Australia, 2008
pdf Sebastian Germesin, Tilman Becker and Peter Poller, "Determining Latency for Online Dialog Act Classification", In: Proceedings of External LinkMLMI 08, Utrecht, The Netherlands, 2008
pdf Sebastian Germesin, Tilman Becker and Peter Poller, "Hybrid Multi-Step Disfluency Detection", In: Proceedings of External LinkMLMI 08, Utrecht, The Netherlands, 2008
pdf Sebastian Germesin, "Disfluency Classification and Correction with a Hybrid Machine Learning and Rule-based Approach", Master's Thesis, External LinkUniversität des Saarlands, Saarbrücken, Germany, 2008


pdf Sebastian Germesin, "Spracherkennung mit dynamisch geladenen, spezifischen Akustikmodellen", Bachelor's Thesis, External LinkUniversität des Saarlands, Saarbrücken, Germany, 2006

Deliverables and Reports

pdf Massimo Romanelli, Sebastian Germesin, Alessandro Adamou, Violeta Damjanovic, Sabine Janzen, Andreas Filler, Alex Conconi, Tobias Kowatsch, Tilman Becker , "IKS Deliverable - D3.1 Model of Knowledge Based Interaction",Public Deliverable of External LinkIKS Project, 2010
Sabine Janzen, Wolfgang Maass, Tobias Kowatsch, Andreas Filler, Massimo Romanelli, Sebastian Germesin, Tilman Becker, Gokce B. Laleci, Ozgur Kilic, Cagdas Ocalan, Erdem Alpay & Asuman Dogac, "IKS Deliverable - D2.1: AMI Case Requirements Specification",Restricted Deliverable of External LinkIKS Project, 2010

Invited Talks

  • Hybrid Multi-Step Disfluency Detection
    HAIL/SALS-SIG Seminar, ICT Centre, CSIRO, September 2008, Sydney, Australia

  • Hybrid Multi-Step Disfluency Detection
    UMNL Research Group, Monash University, September 2008, Melbourne, Australia

Tutoring activities

  • pdf Hanne Kosinowski, "Modular Grammars for Speech Recognition in Ontology-Based Dialogue Systems", Bachelor's Thesis, External LinkUniversität des Saarlands, Saarbrücken, Germany, 2010


  • 01.2010 - today:
    • IKS - Interactive Knowledge Stack
    • IKS is an integrating project targeted at the hundreds of small and medium enterprises (SMEs) in Europe, which are providing technology platforms for content and knowledge management to thousands of end user organisations. Downstream, hundred-thousands of corporate end users and millions of content consumers are affected by the quality of service provided through these platforms. The majority of these platforms lack the capability for making use of semantic web enabled, intelligent content, and therefore, lack the capacity for users to interact with the content at the user's knowledge level. The major technological result of the project will be the "Interactive Knowledge Stack", a layered set of software components and specifications which will make traditional content management platforms capable of dealing with the future "Semantic Web".
  • 04.2008 - 12.2009:
    • AMIDA - Augmented Multiparty Interaction with Distant Access
    • The AMIDA project was an EU-funded project, concerned with the development of multimodal technologies to support human interaction before, during and after business meetings. Meeting rooms at IDIAP (Martigny, Switzerland), TNO and the University of Edinburgh were equipped with an extensive setup of microphone arrays, camera's and further capturing devices. In these meeting rooms the interactions between people during meetings were recorded and stored in the publicly available AMI Meeting Corpus. The development of tools for the multimodal annotation of the recorded meetings (e.g., speech transcription, emotion, gestures, dialogue acts, etc.), as well as the subsequent usage of this data as a basis for a variety of research directions were the key objectives of the project.


Fork me on GitHub