T-61.182 Robustness in Language and Speech Processing
Special course on Information Technology
II (4 ov, L)
Lecturer: Prof. Mikko Kurimo
Assistant: DI Teemu Hirsimäki
Semester: Spring 2003
Credit points: 2-4 cr
Place: Seminar room T4
Time: Thursdays 14-16, starting from February 6
Language: English (or Finnish)
Seminar course description
As speech and language technology innovations enter the everyday use,
a research issue with growing importance is to develop methods that are robust
to different forms of noise and changes in their using environment.
Although some automatic speech recognition and natural language understanding
systems do already a pretty good job in certain well-defined applications and conditions,
their performance fails dramatically compared human speech and language processing abilities
when the conditions change abruptly or the input is somehow disturbed.
The problem in estimation of statistical models and learning representations from data
is that it is difficult to collect enough training data to be able to properly model
even the most likely types of noise and conditions.
For adaptive methods the challenge is to make as general adaptations as possible using
only a very limited amount of available adaptation data.
This course addresses robustness issues at the speech recognition and natural language
parsing levels, with a focus on feature extraction and noise robust recognition,
adaptive systems, language modeling, parsing, and natural language understanding.
The purpose is to give a clear overview of the main technologies used in language
and speech processing and discuss their robustness.
It also brings together speech and language technologies often considered separately.
The course is based on the recent book:
Robustness in Language and Speech Processing, edited by Jean-Claude Junqua and Gertjan van Noord, Kluwer Academic Publishers (2001).
The basic knowledge of pattern recognition, speech processing,
speech recognition, and natural language processing methods is helpful.
Requirements for passing the course
NOTE: To cover well this broad topic
we decided that each participant should give two presentations.
To balance the work load there will be no home exercises.
To pass the course (4cr), you now have to
To pass with distinction, the seminar talk, the handouts, and the
project work must each be very good.
Without the project work the course can be completed to obtain two
- participate actively,
- give two seminar talks on book chapters or articles selected for the course
- carry out a small project work, that is, a small-scale research project
or a literature study on a topic related to the course.
Relationship to other studies
At TKK the course is suited for the Language Technology major
(Kieliteknologian pää/sivuaine) and for studies in Information
Also students and staff from the KIT
(Kieliteknologian opetuksen verkosto) are welcome.
Mikko.Kurimo@hut.fi (tel. 451 5388)
Wednesday, 19-Mar-2003 11:49:15 EET