Re: Word Order in typology

>>On the other hand, Argument roles I would argue do have meaning that
>>doesn't change from language to language,
>>
>>
>
>Not so.  A brief look at Navaho morphosyntax suggests that agentivity
>is a gradient phenomenon as well, highly dependent on animacy hierarchies.
>
>
>
 I concede this. I've been thinking more about such things recently: I
think agentivity etc are somewhat similar to colour terms, in that there
are peaks (centers) that tend to occur as the prototypical example of
colour terms in languages, but away from the peaks the boundaries can
shift a little from language to language. But I would argue that the
prototypical Agent is a notion that all people share. Note I'm not
claiming that all languages build a lot of grammar around a case/GR/etc
in which prototypical Agents fall, but I do believe that all people
share this notion, even if their language only expresses it in a minimal
way.
 Also, I don't have problems with statistical methodology really, except
I still feel that (for the sake of clarity) you should try to find less
language dependent terms to express your statistical universals in.
Maybe Agent isn't perfect, but its better than talking about Subjects
which is clearly a far from universal notion. And if there's something
better than Agents, Patients etc then I'd be happy to adopt that. :)