Skip to main content
conference lunch move company map contacts lindholmen lindholmen 2 travel info

NLP Wednesdays

Every second Wednesday is all about NLP (Natural Language Processing) at the AI Sweden Stockholm office.

What are the NLP Wednesdays?

The NLP webinar series is a bi-weekly webinar series for people who work with or are interested in NLP. The main focus of the webinars is NLP development in Sweden and for the Swedish language, but international breakthroughs may also be discussed. 

Now, we are turning the NLP webinars into hybrid seminars with the possibility to attend in person at the AI Sweden Office in Gallerian in Stockholm!

Join us for the webinar at 13.45 or take the opportunity to work from the AI Sweden office alongside our NLP team for the entire day. We promise great discussions on language technology, coffee, and maybe some table tennis! 

The webinars are organized by RISE NLU group and AI Sweden, and moderated by Magnus Sahlgren, Head of the Natural Language Understanding Group at RISE Research Institutes of Sweden.
Time: Wednesdays odd weeks 14:00-15:30

Upcoming webinars

Here you will find upcoming webinars

Previous webinars


  • November 24: Isaiah Hull Riksbanken
    Presentation recording (Youtube)
  • Oktober 13: Martin Malmsten och Robin Kurtz, KB-labb
    Martin Malmsten and Robin Kurtz from KB lab will present their work on Swedish speech synthesis: How to build a Swedish wav2vec?

    Presentation recording (Youtube)

  •  September 29: Improving Multilingual Models for the Swedish Language: Exploring Cross­Lingual Transferability and Stereotypical Biases

    Styliani Katsarou, Peltarion: We explore whether a multilingual model can be improved for Swedish, through the lens of zero-shot cross-lingual transferability. Taking into account the responsibilities that arise from the increased use of such models in real-world applications, we also focus on bias detection with respect to gender.

    Presentation recording (Youtube)

  •  May 26: SuperLim: a Swedish SuperGLUE (General Language Understanding Benchmark)
    Magnus Sahlgren at RISE; Aleksandrs Berdicevskis, Yvonne Adesam, Gerlof Bouma, Dana Dannélls at Språkbanken, Gothenburg University
    This seminar presents the outcome of the SuperLim project, which provides the first General Language Understanding Benchmark (GLUE) for Swedish. The test collection covers various aspects of language understanding, including reading comprehension, inference and entailment, coreference, sentiment analysis, lexical semantics, and word sense disambiguation. The presentation provides details on the various test sets, and we also discuss future development of the benchmark.

    Presentation recording (Youtube)

  • May 12: CLIP: Connecting Text and Images
    - Ariel Ekgren, RISE
    Diving deep into CLIP: Connecting Text and Images. Explaining the CLIP network, general text/image classification and extensions to it.

    Presentation recording (Youtube)


  • April 28: Gender Bias in NLP 
    - Hannah Devinney, Umeå University

    This seminar will focus on gendered aspects of "bias": exploring what it is, how it manifests in NLP, the harms it causes, and what we as NLPers can do to combat these harms.

    Presentation recording (Youtube)

  • April 14: NLP and Language Modeling for Norwegian
    - Erik Velldal and Andrey Kutuzov
    , Olso university.
    In this presentation we survey the current state of NLP resources for Norwegian, focusing on data sets as well as ongoing efforts on large-scale language modelling.

    Presentation recording (Youtube)

  •  March 31: Speech recognition for Swedish
    - Jens Edlund, KTH

    A significant proportion of showcase applications of AI puts spoken language at its core: intelligent assistants, deep fakes, speech search, and automatic transcription, to mention a few. This seminar will highlight some of the factors that often go missing when speech is treated as e.g. a simple encoder-decoder problem, and why they are important. The seminar will also attempt to show why speech-centric research historically is truly interdisciplinary, and why it should remain that way.

    Presentation recording (Youtube)


  • March 17: Machine Translation
    - Jörg Tiedemann, Helsinki University

    Presentation recording (Youtube)
    Presentation slides (PDF)

  • March 3: NLP and disinformation: Detection of computer-generated text
    Fredrik Johansson, FOI (Swedish Defence Research Agency)

    Presentation slides (PDF)

  • February 17: Climate change and large-scale language models: a panel discussion
    In this week's edition of the Swedish NLP webinar series, we will host a panel discussion on climate change and large-scale models. More information on the topic and confirmed panellists to be released shortly.

    Presentation recording (Youtube)

  • February 3: Best practices for operationalizing the Language models for Swedish authorities project
    Anders Arpteg, Peltarion

    This seminar will discuss and give a proposal for an additional way to share the knowledge and best practices from the Language models for Swedish authorities project. Hopefully, this can help empower as many as possible to operationalize the results, i.e., find value in their own organizations.

    Presentation recording (Youtube)

    Presentation slides (PDF)

  • January 20: What is a Transformer and what is the difference between the current models?
    - Love Börjeson, National Library of Sweden, and Evangelia Gogoulou and Amaru Cuba Gyllensten, RISE NLU Group

    This seminar provides an overview of the Transformer architecture for language models, and discusses the differences between the currently most popular models (e.g. BERT, GPT, RoBERTa, ALBERT, XLNet, Electra, etc.). We also present the current Swedish language models, and introduce the latest, and to date biggest, Swedish language model based on the Electra architecture. 

    Transformers crash course (Youtube) 
    Data and models from KBLab (Youtube)


  • From needs to solutions: Working with NLP at the Swedish Tax Agency
    - Gintare Grigonyte, Oscar Kalldal, Jakob Wedman; Skatteverket

    In this presentation, we aim to give an overview of what role NLP plays at the Swedish Tax Agency (SKV) in the year of 2020. We will talk about some of the different perspectives that we need to consider in our internal NLP projects, focusing on:
    - What motivates and what is a precondition for various NLP services at SKV
    - What possibilities and limitations there are in using SKV's data; how our projects are affected by laws and/or regulations on an internal, national, and EU level; and what ethical considerations we make
    - How we absorb scientific advances into conceptual solutions, and how we bring conceptual solutions to services

    Presentation slides (PDF)

  • Bootstrap annotation for Named Entity Recognition
    - Felix Stollenwerk, Arbetsförmedlingen AI centre

    The annotation of job ads enables a better understanding of the labour market and helps to improve Arbetsförmedlingen’s services. In this presentation, we will describe our ongoing efforts to annotate job ads using Swedish language models and named entity recognition. Lack of training data poses a challenge that we try to overcome with bootstrapping techniques and active learning.

    Presentation slides (PDF)


  • A new state of the art for unsupervised sentence embeddings
    - Fredrik Carlsson, RISE

    Presentation slides (PDF)

  • Data readiness for NLP
    - Fredrik Olsson, RISE

    Presentation slides (PDF)

A high-level summary of each topic from 2020 can be found in this blog post: Method for efficient annotation, NLP from an organisational perspective, and more: Summing up the first four Swedish NLP webinars.