Uppsala universitet  

Machine Translation, 7.5 HP / 5 HP, Autumn 2021

Course codes : 5LN711 (master; 7.5 hp) 5LN718 (master; 5hp)
Course coordinator: Gongbo Tang
Course examiner: Sara Stymne
Teachers: Sara Stymne, Gongbo Tang

Syllabus: 5LN711, 5LN718

The course information for 2021 is currently being updated.

News

  • 210908: Update: The schedule for the final seminars have been updated, both the date and the classrooms!
  • 210816: Course page under construction


More Information on

Literature
Activities
Examination for Masters (7.5 hp)
Examination for Masters (5 hp)


Schedule

Date
   Time   
Room
Type
 Content Teacher
Reading / Assignments
2021-08-30 15-17 6-k1031 Lecture Introduction Gongbo Tang Koehn Ch.1; JM 25.1-2; HS
2021-08-31 10-12 2-k1028 Lecture MT evaluation Gongbo Tang Koehn Ch.8; JM 25.9
2021-09-02 9-12 Chomsky, Turing Lab Assignment 1: MT Evaluation Gongbo Tang
2021-09-07 10-12 16-0043 Lecture Introduction to SMT Sara Stymne Koehn Ch. 2-4, 7
2021-09-08 10-12 16-0043 Lecture Phrase-based SMT Sara Stymne Koehn Ch. 5-6
2021-09-09 13-16 Chomsky, Turing Lab Assignment 2: Moses Sara Stymne
2021-09-13 15-17 2-0076 Lecture Introduction to NMT Gongbo Tang Koehn13 Ch.1-4; GB Ch.9
2021-09-14 13-15 2-1077 Lecture Sequence to sequence models Gongbo Tang Koehn13 Ch.4-5, 7; Luong Ch.2.3
2021-09-15 9-12 Chomsky, Turing Lab Assignment 3: Neural Language Models Gongbo Tang
2021-09-15 13-15 2-0076 Lecture Advanced NMT Gongbo Tang Koehn13 Ch.5; Luong Ch.4-5
2021-09-16 09-12 Chomsky, Turing Lab Assignment 4: seq2seq Gongbo Tang
2021-09-21 09-12 Chomsky, Turing Lab Lab: Seq2seq model with attention (Only 5LN711) Gongbo Tang Code and reports
2021-09-22 10-12 16-2043 Lecture Advances and Challenges Gongbo Tang Koehn13 Ch.8; Tsinghua MT List
2021-09-29 13-16 Chomsky, Turing Lab Assignment 5: Training NMT Models Gongbo Tang
2021-10-04 13-15 Chomsky Project supervision Gongbo
2021-10-11 15-17 Chomsky Project supervision Gongbo
2021-10-19 13-15 Chomsky Project supervision Gongbo
2021-10-28 Updated 9-12 Blåsenhus 11:131 Seminar Student presentations Sara, Gongbo Exact time subject to change
2021-10-28 Updated 13-16 Blåsenhus 13:028, 13:029 Seminar Student presentations Sara, Gongbo Exact time subject to change

Please observe that attendance is mandatory for the assignment examination sessions and for the two final seminars (student presentations).
If you cannot attend, please inform your teacher beforehand and we will find an alternative solution for you.

Teaching mode, Covid-related information

We will do campus teaching following the recommendations, but we might change to a hybrid or online teaching mode, depending on the development of the pandemic.

Please respect the current regulations and stay home if you are not feeling well, and maintain social distancing! Note that this also applies to teachers, so any Campus activities may be moved entirely online on short notice. Please always check the Studium or your email before going to Campus!

This information will be continually updated throughout the term.

Literature

The course is based on Koehn's introduction to statistical MT, parts of Yoav Goldbergs book on neural networks for NLP, and other material as specified below. You will also be expected to find and read additional research articles online.

Additional reading:

Activities

The following activities are part of the examination
  • Assignments are lab-style exercises, which will be examined orally during class time. Attendance for assignment session discussions is mandatory. There are 5 assignments.
  • Labs are larger assignments for which a written lab report should be handed in. Supervision for labs are given during scheduled (non-mandatory) lab hours. There will be 1 lab (only for 5LN711, not for 5LN718).
  • A practical group project focusing on a low-resource language. The group project will be presented both orally during mandatory seminars and in a written report.
  • An individual report discussing your project and research literature for some specific topic, or an implementation.
  • An individual theoretical presentation based on your individual reports (only 5LN711).
Both labs and assignments can be done individually or in pairs of two students. It is not necessary to work in the same pairs for all different assignments and labs. It is up to students who wants to work in pairs to find a lab/assignment partner.

Assignments

There are 5 assignments during the course, see below for the content of each assignment. The assignments will be performed and examined during a 3-hour session each. To pass an assignment, it is mandatory to be present during the full session, to actively perform the tasks given, and to be prepared to discuss your results and experiences in class. Please read through the instructions for each assignment before the session, in order to use the classroom time in the best way possible! The assignments can be performed in pairs of students or individually. You do not have to work in the same pair for each assignment.

If someone fails to attend an assignment examination session, you should make up for this by presenting the assignment to the teacher during one of the project supervision sessions (given that the teacher has time), or prepare a short written report. The deadline for handing in such a report is October 22. Such a report or presentation could be done individually or in pairs.

Assignment Theme Date
Assignment 1 Evaluation September 2
Assignment 2 SMT with Moses September 9
Assignment 3 Neural Language models September 15
Assignment 4 Sequence to sequence models September 16
Assignment 5 Transformer and Toolkits September 29

Labs

There will be one lab on seq2seq model with attention mechanisms, for which you are required to hand in a written lab report. This lab is only for 5LN711 (master 7.5hp), not for 5LN718 (master 5hp).

The lab can be performed individually or in pairs.

Project

 There is a practical group project for all students, where you will perform MT experiments in practice. You will work in groups of 3-4 students, working on one language pairs focusing on translation for a low-resource language (translating from or into English). The languages will be announced in Studium, and you will be able to sign up in a group working on a specific language. You may not pick a langauge that you know, or where you know a (closely) related language.

Each group should perform a project where you train and evaluate several MT systems for your language pair, linking your work to previous research on MT. The project should result in a joint project report and a joint seminar presentation.

Detailed information about the project is available here!

The project is graded. There will be a joint grade (VG/G/U) for all members of the group, based on the written group report and the presentation. Your grades for the full course will also depend on your personal contribution to the group work, see below.

The report should be written in English and are due October 29 via Studium in pdf format.

Individual report/presentation

All students should write an individual report consisting of two parts: (1) a description and discussion of your individual contribution to the group project and (2) a discussion of one (5LN718) or more (5LN711) research articles.

Students in 5LN711 should also do a theoretical presentation, based on the articles discussed in your individual report.

Detailed information about the individual work is available here: 5LN711, 5LN718

You individual report (taken together with the individual presentation for 5LN711) will be graded (VG/G/U).

Deadlines

Activity First deadline Second deadline
Sign up for a project TBD -
Assignment 1 September 2 *
Assignment 2 September 9 *
Assignment 3 September 15 *
Assignment 4 September 16 *
Assignment 5 September 29 *
Lab 1 (5LN711) October 1 October 29
Individual topic/project choice (5LN711) October 12 --
Project presentation October 29 By agreement
Project report October 29 November 27
Theoretical presentation (5LN711) October 28 By agreement
Individual report October 29 November 27

* For assignments the first deadline is for participation in the oral examination. If you miss the session for an assignment, you need to do the assignment on your own, and either present it to the teacher during a project supervision session (given time), or hand in a written report at the latest October 22.

Examination for Masters (7.5 hp; 5LN711)

In order to pass the course, you have to:
  • Do the lab and pass the lab report. Lab reports will not be graded.
  • Do all assignments and pass the oral examination (or written report in case you fail to attend a session). Assignments will not be graded.
  • Perform an MT project in a group of 2-4 students. This includes practical group work resulting in a written group report and a joint seminar presentation.
  • Write an individual report and give a theoretical seminar presentation.

Grading

Passing all requiremnts in the course will earn you a G grade. In order to get a VG grade you need at least one VG on your individual work, or on the group work combined with a strong individual contribution to the group work (i.e. not doing a fair share of the project work will not grant you a VG, even if your group earns a VG).

Examination for Masters (5 hp; 5LN718)

 In order to pass the course, you have to:
  • Do all assignments and pass the oral examination (or written report in case you fail to attend a session). Assignments will not be graded.
  • Perform an MT project in a group of 2-4 students. This includes practical group work resulting in a written group report and a joint seminar presentation.
  • Write an individual report.

Grading

Passing all requirments in the course will earn you a G grade. To earn a higher grade you will either need to have a VG on the project, combined with a strong individual contribution to the group work (i.e. not doing a fair share of the project work will not grant you a VG, even if your group earns a VG), or have a strong individual contribution on a G project together with a VG on your individual report.