This thesis will focus on argument mining for Norwegian texts.
The goal of argument mining is the automatic extraction and identification of argumentative structures from natural language text, in order to provide structured data for example for reasoning engines. An argument can be defined as a set of statements consisting of three main elements: a set of premises, a conclusion, and an inference that goes from the premises to the conclusion. Conclusions are often seen as claims, while premises are usually defined as the evidence or reasons, whereas the inference is seen as the argument itself (Lippi and Torroni, 2016). The relations between these components are usually identified as support and attack. Such that premises can for example be used to support or attack a given claim.
In this project, we will use a corpus of Norwegian reviews. The structure of review documents usually contain evidence and reasons on why a product is good or bad, as well as a conclusion identified as the rating given to the product. A previous master student has made a huge effort in annotating a small testset of these reviews. Here, both components and relations are annotated. However, since the size of the annotated dataset is rather small, a seemingly unavoidable solution to this task is to employ machine learning techniques capable of dealing with unannotated data. Deep learning techniques seems to be one of the most interesting choices in this direction. It is however, if preferred, possible for the candidate(s) to annotate more data for this task.
The precise details and scope of the thesis will be further decided in agreement between the supervisors and the candidate(s), and the thesis can be well suited for two students who are comfortable working together.
The project presupposes a good balance of technical and linguistic expertise. Good programming skills, experience with machine learning and a solid background in NLP are relevant qualifications. Please contact the supervisors to discuss further details.