Making intelligent decisions - ArgumenText uses the Internet’s wealth of argumentative knowledge to help in the balanced search for arguments. The diverse media coverage of the TU Darmstadt project reflects the considerable interest in and potential of this innovative access to knowledge.
The ArgumenText validation project, which is being developed at the Ubiquitous Knowledge Processing (UKP) department of the TU Darmstadt, provides methods for integrating previously unused knowledge into decision-making processes. It enables its users to filter out concrete pro and con arguments from the vast data pool of the Internet for any topic in German and English, especially for controversial ones in which strong arguments are used. ArgumenText presents an intuitive, clear summary of the pro and cont arguments for the topics and additionally links the sources from which the arguments originate. This ensures that the results remain transparent and that arguments can be traced back to the original text. The validation project particularly aims to assess the economic and social impact and future exploitation of the technology.
The resulting potentials for utilization are also of increasing importance for the interdisciplinary application in the field of digital humanities. In order to harness this profitably, the team is collaborating with the Department of Social and Historical Sciences at the TU Darmstadt on the Kritis project, for example.
The project has already been received very positively in national and international media thanks to its potential for various areas of application. The news portal IDW (Informationsdienst Wissenschaft) reports on the opportunities that ArgumenText offers for decision-makers in business who have to weigh up whether it is worth entering into an innovation (Staub, idw 2018). According to the c't (Magazin für Computertechnik), ArgumenText can automate key activities of decision-makers in business and support them in evaluating innovations (Grävemeyer, c't 2018).
In other areas, too, the application can make processes easier and more efficient. Phys.org highlights how journalists can benefit from automatic argument search when they need to get to grips with a topic quickly and thoroughly (Phys.org, 2018). The project is constantly changing and further features are planned. As the Austrian industry magazine FACTORY also points out, a real-time search as well as an aggregated presentation, i.e. according to topics, will soon be possible (Weber, FACTORY 2018). In addition to these reports, ArgumenText was also made available to a wider audience at the Hannover Fair 2018, among other events. Visitors had the opportunity to try out the demo version and ask questions about the tool. A corresponding report can be found in the campus newspaper of the TU Darmstadt hoch3 (Paradowski, hoch3 2018).
ArgumenText project offers many social and scientific opportunities. The ArgumenText team is looking forward to the valuable feedback of active users to further improve the application. A demo version of ArgumenText can be tested on the project's website.
For a deeper insight into the project - its background and technical basics- the following publications are recommended:
Stahlhut, C. (2018) Searching Arguments in German with ArgumenText. Proceedings of the First Biennial Conference on Design of Experimental Search & Information Retrieval Systems, Volume 2167
Stab, C., Daxenberger, J., Stahlhut, C., Miller, T., Schiller, B., Tauchmann, C., Eger, S. and Gurevych, I. (2018) ArgumenText: Searching for Arguments in Heterogeneous Sources. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: System Demonstrations. p. 21-25
● Phys.org (2018) Searching through noise for pros and cons.
● Weber, R. (2018) ArgumenText: Das „neue Google“ für Ingenieure?
● Staub, C. (2018) TU Darmstadt entwickelt Argumentsuche für Internet-Texte.
● Paradowski, S. (2018) Pro und Kontra aus dem Rauschen.
● Grävemeyer, A. (2018) Suchmaschine der Argumente.
● Witte, J. (2016) System filters the arguments in texts and checks their quality.
In December, ArgumenText's Johannes Daxenberger joined a company mission to San Francisco organised by Hessen Trade & Invest (HTAI). ArgumenText extracts pro and con arguments from Big Data (e.g. websites) through Deep Learning. The technology enables innovation assessment and trend recognition in various sectors. Various startups and transfer-oriented research projects from Hesse participated in the delegation. Getting to know the American market, especially the tech scene in Silicon Valley, was the top priority for the delegetaion. The agenda included visits to big and smaller tech companies in and around San Francisco (LinkedIn and SAP, among others), pitch meetings, networking events and a visit to the Stanford University. The American market – and in particular, Silicon Valley – is an important opportunity for German startups and tech companies, not just because of the higher amount of venture capital. ArgumenText aims to establish contact with the American market but also to get to know innovative business models in the software sector.
ArgumenText's Iryna Gurevych spoke on "Natural Language Processing for Automated Fact-Checking" at the Fake News and Other AI Challenges Conference in Vienna on November 30th. More details at Language Intelligence. In another invited talk, Iryna spoke about "Let's Debate: Natural Language Processing at a Large Scale" at Novartis in Basel on November 26th.
We will present ArgumenText at the 2nd Hessian Innovation Congress in Frankfurt on Thursday November 15th. Looking forward to see you there.
We will present ArgumenText at the Startup & Innovation Day in Darmstadt, organized by HIGHEST/TU Darmstadt on October 22nd. The event is in its third year already and connects players from industry, politics and research.
ArgumenText's Iryna Gurevych gave Invited Lectures on "Cross-Topic Argument Mining" in July 2018 at the School of Computing and Information Systems at University of Melbourne, as well as at the School of Information Technology and Electrical Engineering at University of Queensland, both Australia. In August 2018, she gave another talk about "Cross-Topic Argument Mining" at the Machine Learning Group at the University of Waikato, New Zealand.
On June 12th, ArgumenText's Iryna Gurevych gave a keynote speech on "Disentangling the Thoughts: Latest News in Computational Argumentation" at the 3rd Swiss Text Analytics Conference (SwissText 2018) in Winterthur, Switzerland. SwissText is a major yearly NLP event in Switzerland, aiming to bring together practitioners and researchers from the domain of text analytics.
We will be presenting ArgumenText at the "Wissenschaftstag" hellwach! on Sunday June 10th. The public event at TU Darmstadt's main campus (downtown) is open between 11AM and 5PM.
Several short and long papers with contributions from ArgumenText have been accepted at the 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL) in New Orleans and the 27th International Conference on Computational Linguistics (COLING) in Santa Fe, both USA. At NAACL, we'll be presenting our latest work on multi-task learning for Argument Mining and a system description of our argument search engine, at COLING we'll show how to leverage language adaptation techniques for multi-lingual Argumentation Mining.
The German public broadcasting radio station Deutschlandfunk held an interview with ArgumenText's Johannes Daxenberger at Hannover Fair. You can listen to the program as part of the "Campus & Karriere" show here (the interview starts at min. 22, in German).
HANNOVER MESSE (Hanover Fair) is one of the world's largest trade fairs. ArgumenText will be presenting our latest demo, research insights and use cases. You will find us in Hall 2, Stand 25 at the joint exposition area of Hessian research institutions - we would be very happy to meet you there. More details here:
On March 1st, Christian Stab presented ArgumenText at the Bitkom AI Summit 2018 in Hanau. In his talk, he showed how ArgumenText can be used to assess the latest innovations and technologies by mining supporting and opposing arguments from the Web. In this context, he also introduced our latest argument search demonstrator, which provides a new paradigm to search large document collections for valuable information. Further information about his talk can be found at:
On October 11, Christian Stab presented our project to an international audience at the Language Technology Summit 2017 in Brussels. His talk, "Improving Decision Making with Argument Mining", presented our latest research results on how decision-making processes can be supported with Deep Learning.
On October 11, Johannes Daxenberger gave an invited 90-minutes talk at the Innovation Summit of Alexander Thamm GmbH in Munich (non-public event). He gave an introduction to text classification and deep learning for language technology, followed by an overview of the latest research results in argumentation mining and a project summary of ArgumenText, including demo session.
On September 10, Johannes Daxenberger and co-authors presented latest research finding about detecting claims (statements that should be supported with reasons) at the annual Conference on Empirical Methods in Natural Language Processing. Our goal in this work was to find and learn about claims in different textual domains including online discourse, legal text, and student essays. The extensive experiments carried out in this research showed that simple lexical clues are most helpful to detect claims across domains. The paper can be found here:
On September 10, our colleague Andreas Rücklé presented our latest work on detecting complex argumentative structures at the main event of the Natural Language Processing research community, the annual meeting of the Association for Computational Linguistics (ACL). We showed that an end-to-end approach (a deep learning system which models several tasks jointly) on detecting arguments in student essays has superior performance as compared to non-neural pipeline approaches. We also showed that modeling argument mining as a sequence tagging problem achieves state-of-the-art performance on this task. The paper can be found here: