Uppsala universitet  

Individual work in Machine Translation
5LN711 (7.5 credits)

Overview

For your individual work there are two options: (1) theoretical work focused around a topic and (2) an implementation project.

Theoretical work

The main option is to base your indivdual work on a theoretical topic. You will read research articles and maybe also text books related to your topic, write a report and give an individual presenation of it.

Each student will choose a topic to work on. A list of potential topics can be found below, but you are also welcome to propose your own topic. Once you have decided on a topic, email Gongbo with your choice. Maximum two persons can work on each topic, you will see below when someone has been assinged to a topic. The topics will be distributed on a first come, first served basis. If you choose your own topic, it needs to approved before you start working on it.

  • Lingusitics in NMT
  • Document-level MT [Eirini]
  • Domain adaptation [Emma]
  • Zero-shot translation [Yifan]
  • Multilingual translation [James]
  • Back translation (for NMT) [Oreen][Jae Eun]
  • Unsupervised MT [Siyi]
  • Word order / reordering (mainly for SMT)
  • Syntactical/tree-based SMT
  • Speech translation [K├Ątriin]
  • Multi-modal translation [Martina]
  • Automatic post-editing of MT [Chen]
  • Quality estimation
  • Test suites for MT evaluation
  • MT for professional translators
  • Pre-training for MT
  • Your own choice:

For each topic you should use the course literature and/or research articles to get introduced to the topic. Then you should also read additional research articles about your topic. We recommend to use articles available from the journals and conferences listed in the ACL Anthology. The Tsinghua MT Reading List is also a good resource. Note that the majority of articles should have been properly published at a conference or in a journal. You should limit the use of articles only appearing on arXiv, for example. You may choose articles on your own, but make sure they are really concerned with your topic! If you feel unsure about the choice, you can discuss it with Gongbo.

In the report you should give an introduction to the topic, present important research on it, and summarize important challenges. If applicable it is good to use a text book, tutorial or similar as a basis for the introduction to the topic, but it depends on your topic if this is applicable. You should then also select a number of research articles related to your topic. You can choose to read a lower number of research articles in detail (3-4 depending if you also use a text book), or a higher number of articles in less detail, but to give a broader overview of research related to your topic.

In your report you should then discuss your topic in general and give an overview of research activities related to your topic. If you choose a smaller number of articles you can summarize specific studies, if you choose a larger number you will likely rather give a more high-level overview of strategies being used. We do not just want to see summaries of individual articles, but a more coherent description, discussion and analysis of your topic and research related to it. This part of the report should be maximum 3 A4 pages, and not much shorter.

Implementation project

You may also choose to do an implementation project. You should then choose to implement a component in an MT system, or implement an evaluation metric. To do this porject you need to read up on relevant literature, and come up with a concrete plan for what you want to implement. You may choose to reimplement a component already described, but all code should be your own, you may not use any already existing code.

If you want to do an implementation project, you should present a brief plan for your project before starting. This can be handed in by email.

You should hand in the code for your project, some kind of results based on running your code (exactly what depends on what you implement, and can either be integrated in your report, or specified in some other way), and a report. The report should describe literature realted to your implementation project, and motivate and describe your implementation project. This part of the report could be around 1.5 pages.

Choose a project

You need to choose a project before October 14. If you want to do a theoretical project, you just need to decide on a topic to work on, and email it to Gongbo. Maximum two students can work on each topic, and the anmes will be listed on this page. If your preferred topic is already full, you will be asked to choose another one.

If you choose to do an implementation project, we want a short specification of your plan by email.

Report

Your individual report should consist of two separate parts.

You should discuss research on your topic, or your implementation project, as detailed above. For an implementation project you should also hand in code.

In addition you should include a separate part in the report (around 0.5 A4 page) where you discuss your own role in the project work, describing which parts of the project you were active on (either on your own, or in colaboration with others).

The deadline for handing in the report is October 29. It should be written in English, and handed in as a pdf in studium.

Presentation

You should also do an individual oral presentation of your topic project during one of the final seminars on October 28. The length of each presentation is 8 minutes plus some time for questions. The final schedule will be presented later. It is compulsary to present and to attend the full seminars.