Times are given in Irish Standard Time (IST), i.e., either UTC+0 or UTC+1 (Daylight Saving).
- This event has passed.
Cardamom Seminar Series #3
June 28 @ 5:00 pm – 6:00 pm IST
The Unit for Linguistic Data at the Data Science Institute, National University of Ireland Galway is delighted to welcome Remo Nitschke of the Linguistics department at the University of Arizona as our next speaker in the Cardamom Seminar Series. Register here.
In this talk, we will introduce a method for vocabulary reconstruction/prediction via a neural machine translation (NMT) setup. This method relies on adapting a translation/sequence prediction task to predict cognates of a target language from cognates of related languages. It shows that by using a many-to-one mapping in the input (mapping multiple language items to one target), we can reduce the need for large datasets significantly. This has to potential to allow us to use a neural-net based approach even in some low-resource contexts.
Further, we will show that this is possible via a proof-of-concept in the Romance language family. We will then discuss the current efforts to conduct an analog experiment in the Uto-Aztecan language family, an effort which is currently underway.”
About the Speaker:
Remo Nitschke is a PhD-candidate at the University of Arizona linguistics department. His research areas are Syntax and NLP for low-resource languages. He is currently working on the Syntax of German adjectives and ways to utilize neural machine translation for low-resource languages.