Date: Tuesday, October 28, 1997 (preceding the MT-Summit 97 conference)
Place: San Diego, California
The SIG-IL of the AMTA is organizing a workshop to be held the day before the opening of the MT-Summit 97. This workshop is a follow-on to last year's pre-workshop on Interlinguas at AMTA-96 in Montreal.
This workshop will focus on two issues of significance that arose at the preworkshop: the nature of interlingual representation, and the content of that representation.
Two clearly different understandings of interlinguas were evident at the pre-workshop. For some, the interlingual representation is an attempt to represent in a non-language-specific way the meaning of the source language text. As such, its foundations lie in logical representations of the meaning of natural language. This IL representation is supported by a complex ontology of world-knowledge, suitable for applying contextual information to the language data to produce the meaning representation (definitional ontology).
The other approach seeks to populate the interlingual ontology and the IL representation of the source language text with entities that encode correspondences across languages. This approach can be regarded as a generalization of the transfer approach to machine translation, except that multilingual correspondences are encoded rather than bilingual ones. This approach to IL representations is often supported by an ontology with a larger set of entities (word senses) with less internal structure, organized in a generally shallower hierarchy than the definitional ontology (terminological ontology).
These approaches tend to differ also in acquisition modes (hand-crafting with assistance from automatic and statistical approaches vs. semi-automatic or statistical approaches combined with subsequent manual cleanup). They also tend to differ in depth of knowledge acquisition.
Important questions in this area are whether these two approaches are antithetical or mutually supportive; what kinds of situations/applications are best suited for each approach; and whether these approaches converge at some point or whether they represent fundamentally different approaches to translation.
The second issue deals specifically with the amount of language-specific information that must be included in the IL representation and how it is best represented. It has been claimed that an interlingual representation should contain no language-specific information. Experience shows that at least some language-specific information needs to be present in order to produce an adequate translation--for instance, whether or not a particular type of construction (e.g., passive) was used or whether or not a particular type of figure of speech (metaphor, metonymy) was used in the source language text. Other claims have been made that an adequate translation must depend not simply on a plausible semantic representation of the input text, but on an _interpretation_ (a coherent explanation) of that semantic representation in terms of the communicative purposes of the author of the text. Finally, others have argued that some _semantic_ information (semantic information necessary for classifier choice, the semantic information necessary to distinguish "pork" from "pig") need not be present in the IL representation, but can be determined during generation.
Questions surrounding this issue include whether the language-specific information necessary can be determined in advance; whether it can be completely represented in a non-language specific way; whether it is possible to describe methods for deciding which information should be discarded and which should be retained for the interlingual representation.
Extended abstracts are invited on either of these topics. More than one paper is acceptable from one author. Submissions are expected to be between 3 and 8 pages in length, but there are no actual requirements beyond being clear and concise. Using specific systems to exemplify a position would be welcome and even encouraged. Abstracts should be in camera ready format.
Extended abstracts must be received by August 20, 1997. Submission may be in printed or electronic form (LaTex, FrameMaker) and should be sent to:
Stephen Helmreich
Computing Research Laboratory
New Mexico State University
PO Box 30001/3CRL
Las Cruces, NM 88003 (USA)
phone: (505) 646-2141
fax: (505) 646-6218
e-mail: shelmrei@crl.nmsu.edu
Submitters will be notified by September 5 of acceptance. Accepted abstracts will be compiled into a Technical Report/Proceedings available at the workshop. We look for approximately 20 presentations with a total workshop attendance of 40. Non-presenters will be accepted on a first-come, first served basis. A registration fee of not more than $50 is expected.
Important dates:
Submission deadline: August 20, 1997
Notification: September 5, 1997
Revision deadline: October 1, 1997
Workshop: October 28, 1997
Copyright 1997 Computing Research Lab.