Agentivity and intentionality

[dan everett]

     Tom Givon says:


     >>RE: AGENT: I think what Scott alludes to is that you cannot reduce
     all instances of 'agent' to 'intentional'. But if AGENT is a natural
     human cognitive category -- i.e. constructed as a PROTOTYPE with
     multiple features--you need not assume that ALL instances of AGENT
     will always display the feature of intentionality, but only that a
     large majority will (say at least 90%, my guess?). Natural categories
     are not exceptionless, they just represent an strong statistical trend
     (preponderance of evidence). So, I suspect, if a usage-based
     assessment will be made of supposed 'agentive' or 'active' verbs in
     text, you'll find a very robust association of AGENT with
     intentionality. As well as with 'control', 'acting', 'responsibility',
     and 'blame-worthiness'. So one would expect, as in other natural
     categories a-la E. Rosch, to have strong but not absolute FEATURE
     ASSOCIATION. If someone wishes to falsify this hypothesis, it's
     relatively easy to do so, with a large enough text (say 50 pp.). But
     until one did such quantification, it is not clear that we have
     sufficient ground for treating AGENT differently that all other
     natural cognitive categories. They are all frequency-driven. Cheers,
     TG>>
     ************************


     But let's try to make this talk of agents empirical in the following
     way: is the intentional vs. nonintentional actor distinction causally
     implicated in any interesting set of generalizations
     crosslinguistically or within a single language? Or is this merely a
     conceptual distinction, useful perhaps for human psychology, but not
     for human language? If there are such generalizations, then we need
     both kinds. If there are none, then we do not.

     Now, a wide range of linguists, from Beth Levin to Bob Van Valin, have
     concluded that the syntax does not, in general, need to appeal to
     separate classes of actors based on intentionality. Tom suggests the
     same in his concept of prototype. The question in regard to prototypes
     is whether there are *linguistically* significant generalizations to
     be gained by introducing such an entity into *linguistics* at all
     (whether it is necessary in psychology or not is irrelevant). Role and
     Reference Grammar (and other models, from Chomskyan theory to
     Tagmemics) has/ve concluded that in fact prototypes like this lead to
     no syntactic ends, merely obfuscating results. The best
     generalizations, by and large, so RRG contends, are in terms of the
     Macroroles Actor and Undergoer.

     At one level there are intentional vs. nonintentional actors which we
     can all recognize, e.g. in examples likethe one Scott presented. And
     some languages, say, perhaps, Acehnese, may indeed use such notions in
     its syntax. But other languages may not. Generally the answer is that
     this distinction isn't made much of crosslinguistically, but there can
     be many exceptions to this.

     We need to be careful to allow for flexibility in what we assume to be
     relevant crosslinguistically, lest we be mistaken for proponents of
     Universal Grammar, rather than what I consider to be the much more
     interesting (empirically) proposal of Boas, i.e. that we are looking
     not for UG but for 'patterns'.

     Dan Everett