Jim Henry, On 09/07/2008 03:12:
> On Tue, Jul 8, 2008 at 8:36 PM, And Rosta <[log in to unmask]> wrote:
>> Rick Harrison, On 08/07/2008 21:53:
>> In working on my own conlang
>> I have found that it is very useful to include argument structure as a
>> criterion. By 'argument structure' I mean what Lojban calls 'place
>> structure' or 'sumti(-place) structure'. In a language in which grammatical
>> behaviour mirrors logical properties, members of the same semantic category
>> behave alike if and only if they have the same argument structure, and it's
>> these that form natural rather than arbitrary categories.
> Do you mean distributional categories -- sets of words that are
> liable to occur in the same context as each other?  

After I sent that message I realized that what I was describing is probably applicabe far more to predicate-based languages like Lojban or Livagian than to more naturalistic ones. In Livagian there is just one part of speech.

I meant predicates involving the same set of participant roles (in a system in which the inventory of possible participant roles is not finite, and participant roles can be as idiosyncratic as predicates are). Normally a predicate's syntax-semantics has two parts: the set of participant roles, and the residue. I find it kind of takes a conceptual weight off my mind -- by making things simpler & tidier -- to group together predicates involving the same set of participant roles and that differ only in terms of the residue; the set of colour predicates would be an example.

> Yes, I think
> members of true distributional categories would necessarily
> have both the same argument structure and some aspect of
> meaning in common with each other.  But does it make more
> sense, in making a hierarchy of distributional categories,
> to put the argument-structure categories at the top of the
> hierarchy, and then subdivide those argument-structure
> categories into subcategories based on their semantics,
> or vice versa?  That is, would it make more sense to have
> top-level categories like animate noun, inanimate noun,
> intransitive verb, transitive verb, ditransitive verb, preposition,
> etc., and then subdivide the animate nouns into spirit, human,
> animal and the intransitive verbs into physical state,
> involuntary process, voluntary process, etc., --- or have top-level
> categories like "the mental world", "the physical world",
> "relationships", etc., and then subdivide "the mental world"
> into categories for e.g. thinking verbs, speaking verbs,
> nouns for mental states, adjectives for mental states, etc....?

The grouping together of predicates that differ by residue alone would I think precede other taxonomizing, so that the taxonomies serve to classify sets of predicates that differ only by residue.

Taxonomies are of limited use, because it is hard to classify polyadic predicates in a taxonomy. For example, take a predicate meaning "X drinks milk Y produced by lactator Z":  it's hard to see where the predicate as a whole would go in the hierarchy (except as a daughter of some very general node such as Event); it's easier to see how X, Y and Z could separately be classified, but that would work for a thesaurus but not as a vocabulary organization method (given the reasonable assumption that each vocable should appear just once).