Courses in previous years: [ 2004 ]

Replaces the course:
T-61.186 Special course in Language Technology V P 2-5 cr
T-61.186 Kieliteknologian erikoiskurssi V L 2-5 ov

T-61.6090 Special course in Language Technology V P
T-61.6090 Kieliteknologian erikoiskurssi V L

Statistical Machine Translation, Autumn 2005

Organizer	Timo Honkela
Assistants	Tiina Lindh-Knuutila, Jaakko Väyrynen
Researchers	Krista Lagus, Ann Russell
Credit points	7
Semester	Autumn 2005 and to some extent spring 2006 (during periods I, II and III)
Place	virtual discussions in the Fle3 environment, actual meetings at the computer science building, Konemiehentie 2, Otaniemi, Espoo
Time	weekly discussions with new topics in the Fle3 environment
Language	English
Web	http://www.cis.hut.fi/Opinnot/T-61.6090/
E-mail	t616090@cis.hut.fi

• News
• Seminar course description
• Course structure
• Timetable
• Prerequisites for attending
• Requirements for passing the course
• Signing up for the course
• Literature
• More information

News [back to top]

• November 17th Draft article deadline moved to December 15th

Seminar course description [back to top]

Machine Translation (MT) refers to automatic translation from one natural language to another. To this date, even state-of-the-art systems are not able to provide high quality translations, except in some limited domains, such as weather reports etc. Moreover, the human effort needed to build a traditional MT system based on rules, transformations, or some kinds of knowledge representation formalisms is also substantial.

The availability of monolingual and parallel corpora enables the use of statistical methods for machine translation. A parallel corpus contains the same text in different languages. Parallel corpora are interesting because of the opportunity offered to align the original and the translated text and thus gain insights into the nature of translation. However, useful information for MT can be also gained by analyzing statistically monolingual corpora.

Significant improvements to translation system performance may be achieved using statistical approaches. Potential benefits of the statistical approaches include:

• Less human labor, enables development of translation systems also for minority languages and even for dialects and so-called sublanguages.
• As it is based on actual language use, it leads to better performance on real world data.

The following list outlines some topics that will or may be covered during the seminar. Some topics will be covered in more detail and some other themes will emerge during the seminar.

• Basics
  • The translation problem
    • The statistical translation problem
    • Analysis, Generation
    • Different transfer models (syntactic, semantic, interlingua...)
    • Different translation tasks:
      • Prose (long texts)
      • Discourse (interactive, or short texts)
      • Summarization
      • Expert to layperson within language
  • The Alignment problem
    • Is alignment always necessary?
    • Different approaches to alignment
    • Other applications of alignment
• Focus
  • State-of-the-art in statistical machine translation
    • Research systems
    • Systems in industrial use
  • Methods for evaluation
  • How is structure represented and translated?
    • Generative grammars
    • N-gram models
    • Tree models
    • Other models
  • Cognitive aspects
    • Can human cognition be used or considered as an interlingua?
    • Approaches to semantic representation
      • Formal logic in linguistics
      • Conceptual spaces
      • Image-schemas
    • How do cognitive scientists approach machine translation?

Course structure [back to top]

The seminar is namely organized as a knowledge building or progressive inquiry learning activity. In order to facilitate this process and to provide a possibility to attend to the course even for those in other universities than in Helsinki area, a web-based learning environment will be used. The course is thus given as a virtual course by using a specific web-based learning environment (Fle3) on the Internet. The course is organized in a workshop kind of fashion: The participants will form groups and participate to discussions over the internet. Each group will work on their selected topic, present their work orally in a final session, and write a scientific article.

There will be two meetings where physical attendance is required: At the beginning of the course there will be an introduction lecture and a guest lecture on the theme by Prof. Lauri Carlson. In addition, information on the basic arrangements and on how to use the learning environment will be given. At the end of the course, intermediate project works are presented orally in a full-day seminar.

Initial timetable [back to top]

Important dates:

12 September 2005	Introduction lecture + guest lecture by Prof. Lauri Carlson 14-16, room T4
15 December 2005	Deadline for draft papers
23 January 2006	Oral presentations + guest lecture
28 February 2006	Deadline for final papers

Year	2005																	2006
Month	September				October				November					December				January					February				March
Period		I							II											III
Week	36	37	38	39	40	41	42	43	44	45	46	47	48	49	50	51	52	1	2	3	4	5	6	7	8	9	10
Actual meetings
Virtual meethings
Introduction lecture
Group formation
Discussion
Research plan
Group research
Draft article DL
Oral presentations
Guest lecture
Article DL

Prerequisites for attending [back to top]

This special course is intended mainly for graduate students, however, advanced undergraduate students may also be admitted (please contact the course organizers). Each participant is expected to have reasonably advanced understanding of some discipline neighboring the domain of interest (machine translation, statistical methods, cognitive science, language philosophy).

Students, e.g., from the following disciplines and areas are welcome:

• Language Technology major (Kieliteknologian pää-/sivuaine), Computer and Information Science, and Information Processing Science at Helsinki University of Technology,
• KIT (Kieliteknologian opetuksen verkosto, network of language technology education in Finland),
• Cognitive science, language technology, general linguistics and computer science at University of Helsinki and elsewhere, and
• Media Lab of University of Art and Design.

Successful participation does not require specific mathematical modeling or programming skills. However, basic familiarity with computational and/or mathematical modeling is useful.

Requirements for passing the course [back to top]

The course will be graded as accepted / failed. Requirements for passing the course:

• Active participation in the discussions in the learning environment,
• Oral presentation of group work, and
• Finalization of a scientific article based on the group work.

Signing up for the course [back to top]

HUT students:	In Webtopi
Other universities:	Send email to the organizers

Or sign up at the introduction lecture.

Literature [back to top]

Some possibly relevant literature can be found from this .bib file. There is a collection of links for Machine Translation on a separate link page.

More information [back to top]

For more information, please send email to t616090@cis.hut.fi.

Welcome,
Timo Honkela, Tiina Lindh-Knuutila, Jaakko Väyrynen