AMTA 2006 - Workshop:  Automated Post-Editing Techniques and Applications

Machine translation (MT) system developers have traditionally focused their efforts on improving the quality of a particular engine's output text through the modification of either its input recognition algorithm or its output generation algorithm. MT developers tend to give attention to the intelligibility of the output very late in the development process - focusing instead on lexical coverage and language coverage. By the same token, intelligibility is frequently cited as a reason for the lack of acceptance of MT output.  We ask the question then, can MT intelligibility be improved solely by automatically post-editing the output text.  In the absence of non-subjective reference or gold standard texts, post-editing of MT output represents a three part problem:  error detection, error categorization, and error remediation.  All three of these are active areas of research in the natural language processing (NLP) community, particularly in the area of second language learning.

 

Workshop Program

 

09:00 - 09:15   Greetings and Introductions

 

09:15 - 10:15   Invited Talk:  Post-Editing of MT Output in a Production Setting

                        Julia Aymerich and Hermes Camelo, Pan American Health Organization

 

10:15 - 10:45   Break

 

10:45 - 11:15   Automated MT Improvement Through Post-Editing Techniques:

                        Developing an MT Error Taxonomy

                        Rachael Allbritten, The MITRE Corporation

 

11:15 - 11:45   Automating Post-Editing to Improve MT Systems

                        Ariadna Font Llitjos and Jaime Carbonell, Carnegie Mellon University

 

11:45 - 12:15   Post-Editing in a Multi-Evidence Evaluation Paradigm:  The PLATO Syntax Assessment

                        Michelle Vanni, Army Research Laboratory

 

12:15 - 1:45     Lunch

 

1:45 - 2:15       Human Post-Editing of MT Output

                        Marty Roberts, Linguistic Systems, Inc.

 

2:15 - 2:45       Post-Editing in the GALE Program I

                        Greg Sanders, National Institute of Standards and Technology

 

2:45 - 3:15       Post-Editing in the GALE Program II

                        Stephanie Strassel, Linguistic Data Consortium

 

3:15 - 3:45       Break

 

3:45 - 4:45       Panel:  Does Statistical Machine Translation Need Post-Editing?

 

4:45 - 5:00       Wrap-Up

 

Date:

August 12, 2006

 

Contacts:

Jennifer Doyon (jdoyon@mitre.org),

Florence Reeder (freeder@mitre.org) and

Keith Miller (keith@mitre.org)