Workshops

Tuesday, October 8, 2002

W1. Jessie Pinkham, Deborah Coughlin, Bill Dolan: Customization Strategies for MT

This workshop has regretably been cancelled. Those who registered for it should contact the AMTA Registrar to see about a reimbursement.

W2. SIG-IL: Interlingua Reliability

W2. SIG-IL: Interlingua Reliability

Interlingua representations are usually very rich and extremely knowledge intensive. One of the benefits of such a system is that meaning representation can be relatively uniform across multiple source languages. This richness, however, leads to wide potential variation among different producers (human or machine) of interlingual representations. Given an interlingua language specification and a specific document, how similar will the interlinguas for that document be if produced by different linguists from different sites? from the same site? the same linguist on different days? by different semantic analyzers/parsers? 

The goal of the workshop is to investigate issues of reliability of interlingua representation. Papers are invited on:

• inter-annotator agreement in manually producing interlingual representations
• discussions on whether interlingual representations need to be canonical, and, if so, if there is any hope that they ever will be; and whether observations about the semantic structure of languages can inform the design of such a canonical representation
• methodology or technology for ensuring common understanding of interlingual representations
• discussions of the impact of differences/errors in interligual representations on the overall NLP system
• methodology or technology for ensuring reproducibility of interlingual representations
• measures of semantic similarity of interlingual expressions within an interlingual system
• measures and methods for inter-translatability of interlingual representations in different interlingual systems
• interaction of monolingual or parallel proposition bank/predicate-argument bank efforts (or other broad but shallow semantic resources) with interlingual reliability

The workshop will consist of paper presentations, as well as one or more experiments in inter-annotator agreement. Pre-registered participants will receive a specification of an interlingual language, plus some text to annotate in advance. The experiments during the workshop will involve discussions of annotation differences among the participants, measures of agreement, etc.

Instructions for submitting abstracts

Length: 500 to 1500 words Submission
deadline: July 14, 2002
Send to: Lori Levin, lsl@cs.cmu.edu
Format: One of the following: ascii, ps, psf, doc

Instructions for Participation in Interlingua Coding Experiment

If you want to participate in the interlingua coding experiments, send an email containing your name and institution to Lori Levin, lsl@cs.cmu.edu. You have until September 24, 2002 to submit your results. Click here for more information.

Time Line

Program Committee

The workshop is sponsored in part by the Special Interest Group on Interlinguas of the AMTA. For further information about this series of workshops see http://crl.nmsu.edu/Events/FWOI/index.html.

This workshop has regretably been cancelled. Those who registered for it should contact the AMTA Registrar to see about a reimbursement.

Machine translation systems must have customization capabilities in order to claim success as a commercial product or a research prototype. Minimally, these include the ability to add translations for new words and phrases, but may also include more sophisticated functionality such as adapting to new syntactic structures or writing styles, and may even be the means of acquiring all the system's translation knowledge (e.g., in statistical systems), We propose to bring MT developers and researchers together to discuss the customization capabilities of their systems, with an emphasis on using common data for the discussion.

This workshop is intended to cover all types of customization strategies, although preference in selection will be given to novel strategies. We encourage participation from MT developers of commercial systems, and researchers working on all types of MT systems (traditional transfer, interlingua, example-based, or statistical).

To make the discussion more interesting, we request that participants demonstrate the customization capabilities of their system using data freely available to everyone, such as Hansard data for French-English or other data available from ELRA or LDC. Microsoft has also agreed to make technical manual data available for several language pairs (English and any of these: French, Spanish, German, Japanese) for purposes of research related to this workshop. (For access to this data please send email to CustomWS@microsoft.com.)

We request that interested parties submit a two page abstract with the following information:

• overview of your MT system
• description of customization capabilities
• comparison of this strategy to other known strategies
• data that will be used to test capability
• proposed evaluation to determine the effectiveness of the customization
• estimate of time that would be required to customize for the chose domain, based on your sample run.

Dates

Instructions for submission

All submissions should be in English, and it is recommended that they be submitted in one of the following three formats: PDF (preferred); PostScript; Microsoft Word. All submissions will be received and processed using the Conference Management Toolkit (CMT) located at http://cmt.research.microsoft.com/CustomWS.

Authors should follow the instructions at the CMT web site to register, enter information about themselves and their abstract, and upload a copy of their abstract in one of the acceptable formats by the submission deadline. Report any problems with the website to CustomWS@microsoft.com.

For information on obtaining Microsoft Data to participate in the workshop, please contact CustomWS@microsoft.com

webmaster@amtaweb.org
Last updated: Tue Sep 24 15:25:11 PDT 2002