July 20th-25th, 2014
The seminar will be held in the BiCi, which is located in the small medieval hilltop town of Bertinoro, Italy, about 50km east of Bologna. The town is picturesque. Meetings are held in an archiepiscopal castle that has been converted into a modern conference center.
Large amounts of text are added to the Web daily from social media, web-based commerce, scientific papers, eGovernment consultations, etc. Such texts are used to make decisions in the sense that people read the texts, carry out some informal analysis, and then (in the best case) make a decision; for example, a consumer might read the comments on an Amazon website about a camera before deciding what camera to buy. The problem is that the information is distributed, unstructured, and not cumulative. In addition, the argument structure – justifications for a claim and criticisms – might be implicit or explicit within some document, but harder to discern across documents. The sheer volume of information overwhelms users. Given all these problems, reasoning about arguments on the web is currently infeasible.
A solution to these problems would be to develop tools to aggregate, synthesize, structure, summarize, and reason about arguments in texts. Such tools would enable users to search for particular topics and their justifications, trace through the argument (justifications for justifications and so on), as well as to systematically and formally reason about the graph of arguments. By doing so, a user would have a better, more systematic basis for making a decision. However, deep, manual analysis of texts is time-consuming, knowledge intensive, and thus unscalable. To acquire, generate, and transmit the arguments, we need scalable machine-based or machine-supported approaches to extract arguments. The application of tools to mine arguments would be very broad and deep given the variety of contexts where arguments appear and the purposes they are put to.
On the one hand, text analysis is a promising approach to identify and extract arguments from text, receiving attention from the natural language processing community. For example, there are approaches on argumentation mining of legal documents, on-line debates, product reviews, newspaper articles, court cases, scientific articles, and other areas. On the other hand, computational models of argumentation have made substantial progress in providing abstract, formal models to represent and reason over complex argumentation graphs. The literature covers alternative models, a range of semantics, complexity, and formal dialogues. Yet, there needs to be progress not only within each domain, but in bridging between textual and abstract representations of argument so as to enable reasoning from source text.
To make progress and realize automated argumentation, a range of interdisciplinary approaches, skills, and collaborations are required, covering natural language processing technology, linguistic theories of syntax, semantics, pragmatics and discourse, domain knowledge such as law and science, computer science techniques in artificial intelligence, argumentation theory, and computational models of argumentation.
Objectives and Outcomes
The objective of the seminar is to gather an interdisciplinary group of scholars together for an extended, collaborative discussion about the various aspects of connecting argumentation and natural language processing. The intended outcome of the seminar is a roadmap that outlines the state-of-the art, identifies key problems and issues, and suggests approaches to addressing them. More precisely, theseminar is conceived for the writing of a monograph “A Prospective View of Argumentation Theory and Natural Language Processing” that should become a standard reference in the field and should provide guidelines for future research by putting that activity in focus and identify the most significant research issues in combining these two research fields. This roadmap will have several sections authored by the participants at the seminar and edited by the seminar organizers.
Format and Process
The seminar will adopt a structure, where personal interaction and open discussion are prominent, emphasizing discussion of results, ideas, sketches, works in progress, and open problems. Participants will be requested to prepare individual contributions around specific topics (see a tentative list below) so that the outcome of the workshop will constitute a roadmap for the area to be published in the near future. The allocation of topics as well as the mechanism for compiling and elaborating contributions into a coherent draft — that will form the working document for the workshop — will be made known in a future communication to those individuals who accept to participate in this workshop.
Currently we have identified the following areas of research to be presented for discussion at the workshop (and we welcome suggestions about additional topics):
- Automatic identification of argument elements and relationships between arguments in a document;
- Argumentation and negation & contrariness;
- Argumentation and discourse;
- Argumentation and dialogue;
- Approaches combining NLP methods and argumentation frameworks;
- Creation/evaluation of high quality annotated natural language corpora to prove argumentative models on naturally occurring data, or to train automatic systems on tasks related to argumentation (e.g. arguments detection).
- Applications of argumentation mining: summarization, extraction, visualization, retrieval;
INRIA Sophia-Antipolis Mediterranee, France
INRIA Sophia-Antipolis Mediterranee, France
University of Aberdeen, Scotland
Structure of Position Paper Submissions
Participants will be expected to submit position papers (with references) prior to the seminar. Submissions details will be discussed over the course of the seminar. The seminar organizers will facilitate a fruitful exchange of ideas and information in order to integrate the discussion topics.
Position papers should follow the two-column format of ACL 2014 proceedings without exceeding eight (6) pages of content plus two extra pages for references. We strongly recommend the use of ACL LaTeX style files. Submissions must conform to the official style guidelines, which are contained in the ACL style files, and they must be in PDF.
Subsequent to the seminar, draft roadmap documents will be circulated amongst the participants for further discussion and prior to submission for publication. We plan to publish the roadmap in a volume of the CEUR workshop proceedings series. In addition, we have a journal that has agreed to publish a special issue based on expanded and revised versions of the material presented at the workshop.
The total registration fees for each person for the whole stay (arrival Sunday evening – departure Friday after lunch) are 600 Euro. Participants pay their own costs; however, organizers are seeking funding to defray the expenses. We will update as information becomes available. Fees include seminar registration, accommodation, WiFi and meals (included an excursion and the social dinner).