Guten Tag

Gerne lade ich euch diesen Mittwoch und Donnerstag, 24. und 25. März 2021 zum virtuellen NLP Hackathon der Uni Bern ein. Auf dieser Website und unten im Email sind die 4 spannenden Challenges aufgeführt, die bis jetzt eingereicht wurden:

https://www.cnd.philnat.unibe.ch/ueber_uns/aktivitaeten/nlp_hackathon/

Der Ablauf des Hackathons ist wie folgt:

Kickoff am Mittwoch, 24. März 2021, 9:00 - 10:00 Uhr
- Begrüssung und Einführung
- Vorstellung der Challenges
- Team-Building

Präsentation der Resultate am Donnerstag, 25. März 2021, 15:00 - 16:00 Uhr
- Präsentation der Ergebnisse
- ab 16h virtuelles Abschlussbier

Meeting auf BigBlueButton: https://bbb.ch-open.ch/b/mat-f4n-qtn

Kommunikation per Slack: https://nlphackathon.slack.com

Aktuell sind rund 25 Personen angemeldet. Wer ebenfalls teilnehmen möchte, kann sich per Email an dh@wbkolleg.unibe.ch anmelden.

Danke auch fürs Weiterleiten dieser Nachricht an weitere interessierte Personen!

Wir freuen uns auf spannende zwei Tage NLP-Hacking!

Herzliche Grüsse,

Matthias Stürmer

Challenges

Folgende vier Challenges sind aktuell eingereicht:

1. Forschungsstelle Digitale Nachhaltigkeit Uni Bern: Kompetitive Challenge "Klassifikation von Schweizer Gerichtsurteilen"
   The legal language is very special in many regards compared to regular natural language. It is highly structured, rather complicated, contains its own special terms and uses certain words differently than they are used in regular text. Text classification is simple to define but has a myriad of possible applications and good systems can provide immense value. Common general applications of text classification include for example spam filtering, email priority rating, or topic classification. And in the legal domain text classification includes legal judgement prediction (predict outcome of a case based on description of case's facts) or legal area prediction. So in this challenge, you will predict the chamber based on the text of a court decision. The chamber is structured in the form of {federal level}_{court}_{chamber number} (e.g. SG_KG_002 => St. Gallen, Kantonsgericht, 002).
2. Statistisches Amt Kanton Zürich: Kreative Challenge "STATBOT.CH" (English Documentation on GitHub)
   If you are searching for some form of statistical information, it is not always easy to find it in the shortest time possible. Particularly in Switzerland, the data and information are not only spread vertically over different federal levels. They are also spread within these federal levels horizontally over different offices and even there sometimes over different sites/channels with different formats. Looking for the needle in the haystack looks comparably easy next to that. Further, even search engines are only of limited help, as they follow an indexing logic that excludes information stored in databases or files. The background of a more difficult search for facts, is also a risk for democratic processes: The harder it is for the average citizen to find truthful information, the easier it is to spread fake news. Therefore, the Statistical Office of the Canton of Zurich, together with other organizations, would like to develop a Swiss Statistical Bot (STATBOT), which would provide data and statistical information directly and quickly across all organizations.
3. Digital Humanities Uni Bern: Kreative Challenge "NER for Historical Documents"
   Developments towards NER solutions have shown significant outcome in the past few years already. Nevertheless, applications for sparse language data are still a challenge, specially when dealing with data from pre-modern times. In this challenge, we focus on language data from the 16th to the 18th century from the Bernese Turmbücher (legal documents protocolled in the Tower of Bern, Switzerland). These documents are currently hosted in the State Archives of Bern. Language models are not provided.
4. Digital Humanities Uni Bern: Visualization of Language Models
   Language models (e.g. character embeddings) are essential to succeed in NLP tasks. Especially when it comes to Part-of-Speech and Named Entity Recognition, tasks result in more precise models if supported by adequate language models already. Since the advent of word2vec and large transformer-based language models (such as BERT or GPT-3) a variety of specialized and fine-tuned language models is currently available. Despite the widespread use and the necessity when it comes to specific model training (e.g. for language entities with only sparse data), our understanding of the models themselves is limited at best. In order to strengthen our understanding of language models and to start the process of reflecting them, this challenge asks for creative ways of visualizing language models. We envision 3D-visualizations based on dimension reduction to identify the positioning of e.g. synonym/homonyms in vector spaces or listing of semantic fields (neighboring vector values). For context insensitive approaches (e.g. word2vec or GloVe) we imagine to use the fixed vectors and represent calculations in grids.