Natural Language Processing


    Our Chair will start with a full teaching offer from the Winter Semester 2022/2023. In the running Summer Semester, we offer a Seminar and a Praktikum, both covering hot topics in (Deep Learning-based) Natural Language Processing (primarily in the program Master Informatik, but students of all other study programs are more than welcome to register): 

    BSc/MSc Seminar in Natural Language Processing (08 151300, I-SEMx-1S) 
    2 SWS, 08 151260, biweekly meetings, block session (close to end of the semester)

    • TopicDeep(er) and Sustainable? Efficient Natural Language Processing
    • Format: Each student chooses between one of three concrete sub-topics related to computational efficiency of modern (deep) NLP models. For each topic, one or two seminal papers will be provided as the starting point for the seminar exploration
    • Deliverables: The students will organize their findings and present them (1) in front of the teaching staff and colleagues (15 minute presentations in the block session close to the end of the semester, exact date to be announced) and (2) in the form of a discussion paper (6-8 pages) that critically summarizes the state of the art in the subarea/topic of efficient NLP they investigated      

    MSc Praktikum in Natural Language Processing (08 140750, I=PRAK-1P)  
    6 SWS, (bi)weekly meetings, block session at the end of the semester

    • Topic: Geographically Specialized Language Models    
    • Description: The students will work on implementing geographically specialized large pretrained language models that can capture specific properties of dialects and closely related languages (e.g., variants of English -- UK, USA, India, Australia, ...; Norwegian, Swedish, Danish; Croatian, Serbian, Bosnian, Montenegrin). The project will include implementation of a web application (i.e., interface) for making predictions on geolocations (in supervised and zero-shot fashion) given dialectal language as input.
    • Deliverables: The project is expected to result with a platform for exploring geographically specialized language models, which is to be described in a system demonstration paper to be submitted for a top-tier NLP conference.