University of Helsinki
2015–2016 / 2016–2017 / 2017–2018 / 2018–2019 / 2019-2020
General Information
This is the time table of the general research seminar in language technology. Presentations will be given primarily by doctoral students but also other researchers in language technology including external guests
NEWS – NEWS – NEWS – NEWS
All seminars from March onwards are postponed until further notice due to the spread of the corona virus and recommendations by the University.
Place and Time:
Thursdays 14:15-16, Metsätalo lecture room 24
(group meetings and reading groups are in Metsätalo B610)
Registration and Announcements:
Please subscribe to our e-mail list lt-research by sending a message to majordomo@helsinki.fi with the following text in the body of your e-mail:
subscribe lt-research your.email@helsinki.fi
Tentative Schedule – Spring 2020
- January 16: Janine Siewert, Kaisla Kajava and Anna Dmitrieva
- 3 short presentations about new members in our team, their background and what they will be working on
- January 30: Umut Sulubacak: The Case for Context in Machine Translation for Subtitling: A Public Dissection
- February 6: Aina Garí Soler: Teaching BERT the meaning of words
- Since its appearance about a year ago, BERT has revolutionized the field of NLP. Its good performance on a wide range of NLP tasks has motivated numerous works aiming at better understanding its inner workings or at improving the learned representations.
Capturing word usage in context, contextualized representations have attracted a lot of interest in the field of computational semantic analysis as well. However, a closer inspection of word similarity estimations provided by BERT reveals that they are not as semantic as one would expect, but are instead heavily influenced by surface structure-related properties.
In this talk, I will present our ongoing work where we propose a number of strategies for improving BERT’s meaning representation of words in context. We intervene directly at the input and classification head levels of the model, or perform an additional fine-tuning step. An evaluation of our models on three semantics-related datasets shows an occasional moderate improvement over the standard pre-trained BERT. Finally, I will discuss a variety of directions that we believe promising to pursue in future work, as well as the main challenges faced during our experiments.
- Since its appearance about a year ago, BERT has revolutionized the field of NLP. Its good performance on a wide range of NLP tasks has motivated numerous works aiming at better understanding its inner workings or at improving the learned representations.
- February 13: Emily Öhman: Collaboration with the University of Tokyo & Waseda – Experiences from a 3 months research stay in Japan
- (March 5 – no seminar because of reading week?)
- March 12: Hande Celikkanat: Looking for Interlingua in Correlated Representations
March 19: Lidia Pivovarova: Detection of word meaning shifts (tbc)March 26: Updates from MeMAD (tbc)April 2: Tatu YlönenApril 9April 16: Andrey Indukaev- April 24 (POSTPONED): Workshop on NLU and language grounding
- (May 14: LREC)
May 28: Elaine Zosa: multilingual and diachronic topic models (tbc)June 4: Mikko KiveläJune 11June 18
