This paper is a tentative note investigating the potential utility of using a representation of actions developed for an animation project as a general representation of text meaning (an interlingua). The proposal is flawed at the outset because it deals only with actions of change of state, change of location and exertion of force. Outside the task of specifying what animation must be done, the utility of such an approach in practical terms is nil for obvious reasons of coverage.
Theoretically, there is nothing strikingly novel or otherwise interesting in the proposal (it is clear from the text that no experiments have been carried out, in fact, that there is no system: otherwise, a number of difficulties and shortcomings of the method, as described, would have become apparent to the authors, and they might have considered discussing such issues). The parameterized action representation formalism is a bread-and-butter frame or TFS representation, with appropriately vague (or, which is the same, application-driven) semantics for the properties involved.
The emphasis in the discussion of issues concerning language is on treating divergences. While the authors correctly assume that the interlingual representation should be the same irrespective of the way a language expresses a particular meaning, they still believe that the realization differences among the languages are a special feature worth a special treatment, thus displaying a presupposition that this state of affairs is somehow unexpected or that it plays any important role in interlingual MT. In reality, of course, each source (and each target) language in an interlingual MT environment must be treated individually. Juxtaposition against other languages is simply not needed or useful, unless one is thinking about transfer-oriented MT systems, as the authors mention in comparison with other work.
In sum, the most charitable statement about the proposal is that it is not possible from this paper to judge whether the proposed formalism is useful as an interlingua, as the formalism will have to undergo a great deal of elaboration and enhancement before any realistic interlingua gets developed. And, of course, the real problems with interlingual MT, namely, problems with semantic analysis could be addressed by the authors only after other major development steps occur, namely, the development of a dedicated semantic analyzer based on an available detailed interlingua as well as the development a dedicated semantic lexicon based on a general ontology. (Parsers and lexica developed predominantly for syntactic analysis have been shown to be grossly insufficient as bases for semantic analysis.)
Until all these prerequisites are met (and they do not need to be developed from scratch by the researchers, there are resources available, at a price of learning and accepting, if temporarily, a different approach and formalism), the best that the authors can do are paper experiments.
To SIG-IL Workshop Series Home Page
Copyright 2000 Computing Research Lab.