From robin@xxxxxxx.xxx.xxx Sun Apr 4 06:23:48 1999 X-Digest-Num: 104 Message-ID: <44114.104.573.959273824@eGroups.com> Date: Sun, 04 Apr 1999 16:23:48 +0300 From: Robin Turner > One solution would be to adopt a feature-based analysis > > of the gismu involved, using features which are, as far > > as possible, consistent across cultures. > > > An alternative ... use the Natural Semantic Model ... which > > aims to define terms using a limited number of universally > > accepted words (I think the current total is 90). > > In the end, aren't these two methods basically the same? > In the final analysis, don't they both say ... > > There is a small, enumerated list of points in semantic > space which have the following properties: > > 1. Every language has a word at that location > 2. Between them they cover *all* of semantic space. > > ?? Well, yes, more or less, which is what I find questionable about both, especially the second claim. OK, maybe you could produce a minimal set of words to cover all semantic space, but it would end up with some hopelessly vague categories, in the same way that Klingon covers all of grammatical space by classifying words as "nouns, verbs and everything else". However, if we leave the more exaggerated claims of both methods aside, and concentrate on what they can do in practice, I think it's largely a matter of taste as to which one you use. In my own work I use a variable weighted feature approach, since that I find that is the most convenient way of describing the structure of a category, but NSM may be more useful for definition-writing. For example ninmu x1 [is female] AND {[is human] OR [is like a human]} Obviously with 5-place gismu it gets a bit more complicated. Also, in compiling a master definitions list, you don't have to break every entry down into your minimal set of words/features - you can define the more common compounds at the start. co'o mi'e robin.