The first workshop on representation learning from multilingual data
Invited keynotes by:
Željko Agić, IT University of Copenhagen(unfortunately canceled)
- Kyunghyun Cho, NYU
- Manaal Faruqui, Google/New York
- André Martins, Unbabel
- Ivan Vulić, University of Cambridge
The goal of this workshop is to bring together researchers that are interested in learning meaning representations from natural languages using multilingual data sets. The main focus is on the use of translations as “semantic mirrors”, allowing to infer meaning from translational grounding. With this, FoTran 2018 also constitutes the start of an ERC project with the same name and we hope that this event will set up a growing network that we will make use of during the project. Topics of special interest of this kickoff-event are not restricted to the objectives of the ERC project but include:
- sentence representation learning in general
- multilingual neural machine translation
- cross-lingual NLP and transfer models
- approaches to natural language inference
- interpretation and analyses of neural networks used for NLP
As mentioned above, participation is free but requires registration (space limits may apply). There is also a pre-workshop dinner on Thursday evening for people who signed up for that event. This dinner will start at 7pm at Bryggeri Helsinki (Sofiankatu 2). Here is a link for directions.
09:15 Coffee & Welcome
- 09:30 Invited talk: André Martins: Beyond Softmax: Sparsity, Constraints, Latent Structure – All End-to-End Differentiable! [slides]
- 10:50 Jörg Tiedemann: FoTran, MeMAD and all that Jazz [slides] (Invited talk by Željko Agić is canceled)
- 11:10 Invited talk: Ivan Vulić: Multilingual NLP via Cross-Lingual Word Embeddings [slides]
12:00 Lunch (catering in front of the lecture hall)
- 13:30 Invited talk: Manaal Faruqui: Understanding Structure in Language through Wikipedia Edits
- 14:20 – 15:20 Short Presentations:
- Alessandro Raganato: An Analysis of Encoder Representations in Transformer-Based MT
- Aarne Talman: Natural Language Inference with Hierarchical BiLSTM’s
- Yves Scherrer: Translational Grounding: Using Paraphrase Recognition and Generation to Demonstrate Semantic Abstraction Abilities of MultiLingual NMT
- Eetu Sjöblom, Mathias Creutz and Mikko Aulamo: The paraphrase project
- 15:50 Invited talk: Kyunghyun Cho: Three recent directions in neural machine translation [slides]
- 16:40 – 18:00 Short Presentations and Posters:
- Jose María Hoya Quecedo: Revita: a language-learning platform at the intersection of ITS and CALL (co-authors: Anisia Katinskaia, Javad Nouri, Roman Yangarber) [slides]
- Philipp Dufter: Embedding Learning Through Multilingual Concept Induction [poster]
- Robert Östling and Murathan Kurfali: Better word vectors for the next thousand languages
- Amir Bakarov: Representativeness of cross-language word embeddings on a lexical level: could it be general, or is it always task-specific?
- Stig-Arne Grönroos: Cognate-aware morphological segmentation for multilingual neural translation
- Niko Partanen: Multilingual Dependency Parsing for Low-Resource Languages via Multilingual Word Embeddings [poster]
- Hotel Arthur is located right across the street from the department.
- Hotel Cumulus Kaisaniemi and GLO Hotel Kluuvi are nearly as close.
- Omena Hotels have two somewhat cheaper hotels in central Helsinki, though a bit further away than the options above.
- Eurohostel offer even cheaper accommodation, also within walking distance of the venue.
- More options are available at the Visit Helsinki accommodation page.
Arriving by air
The Helsinki-Vantaa Airport is located some 17 km from Helsinki city centre and is easily reachable by train. The airport is of moderate size and quite efficiently run, but do allow 1-2 hours before departure for check-in and security. On arrival, you can expect to be out of the airport in 20 minutes after landing.
The easiest connection to the city center of Helsinki is by train. You can also look at other options of public transportation (search for the route from Lentoasema, Vantaa to Unioninkatu 40 if you want to go to the University City Campus; the trip should take about half an hour and cost €5.50) or by taxi which takes a similar amount of time and costs about €50. More airport transportation options can be found on the Finnair website. The schedule for the Finnair city bus is available here.
You can use the journey planner of HSL to find your way along the public transportation system of Helsinki. The department’s central location also allows walking to many destinations in the city.
All public transportation in the Helsinki metropolitan area uses the same ticket system. Single tickets can be purchased with your cell-phone (just download the app) or from ticket machines found on some of the stops.
Single tickets inside Helsinki cost 2,20 € when bought through the mobile app, 2,50 € when bought from a ticket machine and 3 € when bought from the bus driver (no tickets can be purchased in trams or trains). The ticket is valid for one hour from purchase and you can transfer freely from one line to another for that time. If you plan on using the public transport extensively, there are also day tickets available for periods of 1–7 days.
Local Organisation Team
The event is funded by FoTran, an ERC grant from the European Union’s Research Framework Programme, Horizon 2020. The workshop is also supported by the Faculty of Arts in Helsinki. Thank you!