Graduate School of Language Technology in Finland
Kieliteknologian valtakunnallinen tutkijakoulu - Språkteknologiska forskarskolan i Finland
Department of General Linguistics
Siltavuorenpenger 20 A
22 - 26 August 2005
Much previous work in computational morphology has centered on using human experts either to directly create rule systems or to annotate training data which can be used by supervised machine learning algorithms. The performance of these supervised methods has been very good; however, this manual construction/annotation can not be transferred between languages. This means that human experts will be required every time morphological tools are desired in a new target language.
This course will focus on the ways in which the morphology of a language can be learned with little or no human expertise. We will begin by examining what parts of the morphology can be inferred without direct supervision. We will show how features such as orthography and word occurrence frequency can be used to bootstrap a supervised morphological learner.
We will then explore the parts of the morphology which are not easily learnable without some supervision. We'll look at what kinds of data sources we could use to obtain the necessary information and how we could use this information most effectively. Importantly, we will explore how we can gather small amounts of information from non-experts and from those who do not even speak the target language to aid us in constructing effective morphological tools.
The class will divided into a morning session, held as a combination of lectures and discussions, and an afternoon session which will involve exploring the ideas we discussed in the morning in a hands-on lab setting.
Back to the Courses Page