Programme
Wednesday November 23: Swedish Dialogue Workshop 2022 (NB! separate registration)
- 13.00-13.15
- Welcome
- 13.15-13.35
- Dialogue systems and EdTech
- 13.15-13.35
- Challenges of Applying Large Language Models to Open-Domain Dialogue with the Elderly
- 13.35-13.55
- Socially Aware Dialogue Systems
- 14.15-14.35
- Predictive Turn-taking with Voice Activity Projection Models
- 14.35-15.05
- Break
- 15.05-15.25
- Maintenance of a large scale conversational AI solution in the Electrolux contact center
- 15.25-15.45
- Presenting with adaptive robotic agents
- 15.45-16.05
- Tips and tricks for treating transcribed talking (dialogues) as text
- 16.05-16.25
- Conversational speech synthesis
- 16.25-16.55
- Break
- 16.55-17.15
- A system for pedestrian wayfinding
- 17.15-17.35
- How open are conversations with "open-domain" chatbots?
- 17.35-17.55
- Embodied dialogue datasets
- 17.55-18.15
- Experiences of creating an office-bot prototype for employee well being using the Furhat SDK
Thursday November 24: SLTC 2022 day 1 (F11, Ground floor, Lindstedsvägen 22 - to the right of TMH's entrance)
- 9.00-9.30
- Coffee
- 9.30-10.30
-
Studying unscripted informal conversation may be more important than you think
Jussi Karlgren is a principal research scientist at Spotify, where he works on analysis of podcasts. His background is in language technology, and he has worked on various aspects of understanding information access use cases, on semantic knowledge representations, and on stylistic variation in language, which in spite of sounding as if they might be different things actually converge quite nicely.
- 10.30-10.40
- Break
- 10.40-11.00
- Dialogue strategies for... hur säger man ordförrådsträning? (PDF)
- 11.00-11.20
- Automatically generating question-answer pairs for assessing basic reading comprehension in Swedish (PDF)
- 11.20-11.40
- Using BERT to measure coordination in bi-lingual dialogue: An information-theoretic approach (PDF)
- 11.40-13.00
- Lunch
- 13.00-13.20
- Annotated Job Ads with Named Entity Recognition (PDF)
- 13.20-13.40
- Hallucinate or ground: how general or specific are object descriptions generated by a vision-and-language transformer? (PDF)
- 13.40-14.00
- Fast corpus queries with unary and binary indexes (PDF)
- 14:00-14:20
- MEET: a corpus for studies of meeting characteristics (PDF)
- 14.20-14.50
- Break
- 14.50-15.10
- Improving Swedish Part-of-Speech Tagging for hen (PDF)
- 15.10-15.30
- Towards a computational model of reference and re-reference in visual scenes (PDF)
- 15.30-15.50
- Automatic Classification of Budget Allocation Conditions (PDF)
- 15.50-16.10
- Detecting Gender Bias in Course Evaluations (PDF)
- 16.10-16.40
- Break
- 16.40-17.00
- Rook – A New Tool for Visualising Word Frequency Changes (PDF)
- 17.00-17.20
- Towards A Word Complexity Score for Individuals with Dyslexia (PDF)
- 17.20-17.40
- Methods for increasing cohesion in automatically extracted summaries of Swedish news articles (PDF)
- 17.40-18.00
- Nordic language ambassadors - Teaching speakers of small Nordic languages about language technology (PDF)
- 18.00-18.30
- Break
- 18.30-21.00
- Reception
Friday November 25: SLTC 2022 day 2 (Fantum, 5th floor, Lindstedtsvägen 24 - TMH's facilities)
- 9.00-9.30
- Coffee
- 9.30-10.30
-
GPT-SW3: building the first large generative language model for the Nordic languages
This talk gives an overview over the process of building the first large generative language model for the Nordic languages. We cover the motivation for building the model, as well as challenges and opportunities with data and compute. We also give examples of applications of the model, and discuss future directions for building and deploying large language models for smaller languages.
- 10.30-10.40
- Break
- 10.40-11.00
- The Diachrony of Political Terror: Tracing Terror and Terrorism in Swedish Parliamentary Data 1867-1970 (PDF)
- 11.00-11.20
- Colloquialization in the Narrative and Dialogue of Swedish Fictional Prose (PDF)
- 11.20-11.40
- What Causes Unemployment? Unsupervised Causality Mining from Swedish Governmental Reports (PDF)
- 11.40-13.00
- Lunch
- 13.00-14.00
-
Practical Ethics of AI and Music
My research for the past six years has focused on the application of artificial intelligence to music. Reactions from some stakeholders have compelled me to ask the dangerously murky questions of how my work benefits the world, how it harms the world, and how I know. This talk summarizes my journey and hopefully contributes some insights that are useful more broadly.Bob Sturm is Associate Professor of Computer Science at the KTH Royal Institute of Technology, Stockholm, Sweden. He has degrees in physics, music, multimedia, and engineering, and specializes in signal processing and machine learning applied to music data. He currently leads the MUSAiC project funded by the European Research Council (https://musaiclab.wordpress.com), and is probably most known for his work on horses (https://ieeexplore.ieee.org/document/6847693), the GTZAN dataset (https://arxiv.org/abs/1306.1461), and playing AI generated folk music on his accordion (https://tunesfromtheaifrontiers.wordpress.com).
- 14.00-14.10
- Break
- 14.10-14.30
- Towards a Swedish test set for speech-oriented text normalization (PDF)
- 14.30-14.50
- Really good grammatical error correction, and how to evaluate it (PDF)
- 14.50-15.00
- End