GF meets SMT

1 Nov 2010
5 Nov 2010
Europe/Vaduz

November 1-5, 2010, Chalmers University of Technology

The UPC team experts in Statistical Machine Translation are visiting the GF group to work on the MOLTO project's task of making GF more robust by SMT methods. The first days of this visit are devoted to introductory lectures, suitable to a general audience and are therefore open to the public.

The goal is that all participants will

  • learn to use standard SMT tools (Giza, Moses)
  • understand the basic principles and algorithms behind SMT
  • build an SMT system of their own as the course assignment (worth 1.5 European credits for those interested)

Since some software depends on the operating system it is advisable to install them before the session. For the tutorial we will use the following packages on a linux machine:

GF is mentioned in the title because the ultimate goal is to acquire understanding of SMT techniques in the GF group - but there is no planned GF tutorial. If there are participants who want a crash course, it can be provided. The advanced topics and group work will study the ways in which GF and SMT can be combined.

Day 1-2

Place: EDIT Room, 3rd Floor, ED house, Chalmers. Except Monday till 11.00: room 6128, 6th floor.

1 Nov. 2010 title speaker
9:30 - 10:00 Machine Translation in MOLTO Aarne
10:15 - 12:00 SMT Tutorial 1: Basics Cristina España
12:00 - 14:00 Lunch
14:00 - 15:30 SMT Tutorial 2: Hands-on Cristina España
15:30 - 16:00 Coffee break
16:00 - 17:00 SMT Tutorial 3: Evaluation & Hands-on Cristina España
2 Nov. 2010 title speaker
10:15 - 12:00 GF grammars, probabilities, statistics, alignments Ramona & al
12:00 - 14:00 Lunch
14:00 - 14:30 Manual Evaluation Methods Maarit
14:30 - 15:00 Large-scale grammar writing: first experiences from the patent corpus Aarne & al
15:00 - 15:30 Coffee break
15:30 - 17:00 SMT/GF combination. Alignments and translation tables Working Session

Day 3-4

Place: rooms 5128 (Wed 9-17, Thu 9-13), 4128 (Thu 13-17), ED house, Chalmers. Or private offices.

General theme: ST-GF hybrid baselines: GF-based alignments; GF fragments hard/soft combination.

Own work or group work

4 Nov. 2010 title speaker
10:15 - 11:15 A TAG formalism for Parsing and Translation (seminar) Xavier Carreras
11:15 - 12:15 On discriminative GF models for Parsing and Translation (brainstorming) Xavier Carreras
12:15 - 14:00 Lunch
14:00 - 15:30 Statistical models for GF syntax Working Session

Day 5 (advanced topics)

5 Nov. 2010 title speaker
10:15 - 11:15 MT evaluation Lluís Màrquez
11:15 - 12:00 Soft integration SMT/GF, GF driven (brainstorming) Lluís Màrquez, Cristina España
12:00 - 14:00 Lunch
14:00 - 14:45 Working Session X
14:45 - 15:15 Coffee break
15:15 - 17:00 Working Session X

Lodging

A few rooms have been reserved at Quality Hotel Panorama. Contact Aarne if you want to get one of these.

Signups closed for this Event

AttachmentSize
tutorialSMT.pdf1.63 MB
mt-in-molto.pdf216.35 KB
TAG-formalism-for-parsing-and-translation.pdf565.69 KB
discriminative-GF-grammars.pdf194.62 KB
MT-evaluation-seminar.pdf1.54 MB
MOLTO-hybrid.pdf187.13 KB
GF_SMT_manualeval.pdf126.42 KB