Date sent:      	Mon, 22 Nov 1999 19:08:02 +0000
Subject:        	passives
From:           	Dick Hudson <[log in to unmask]>
To:             	[log in to unmask]
Send reply to:  	[log in to unmask]

> Dear WG,
> Maybe Chet and Jasper are right. I said in an earlier message that I
> thought English passives were formed by derivation (as well as inflection)
> because inflection alone wouldn't allow the passive linking pattern to
> override the active one. This argument rested on the assumption that the
> active pattern would be fixed lexically, i.e. lexeme by lexeme, but of
> course the problem would disappear if active linking was fixed by default
> rules for all verbs, as a lot of people assume (including some people on
> this list). Since Passive (the inflection) isa Verb, its rules would
> automatically override those which apply to Verb. In fact you could use
> the existence of passive as an inflection as evidence for a high-level
> linking rule.

Quite. Passive is a construction, in the sense that we use in that 
paper. Just as with (English!) Middle, Resultative etc, the lexical 
representation of the construction is arranged around something 
that isa VERB: PASSIVE is complement ofa BE (or ofa GET, but 
that's another construction); object of PASSIVE = its subject, etc.

> But suppose this is true. How does the active rule work? And how does it
> get the right linkages for all the problem cases we know of (like/please,
> spray-load alternations, and so on)? The rule will have to map the
> referent of the subject to X, that of the direct object to Y and that of
> the indirect object to Z; and then the passive rule will change these
> mappings so that the subject maps to Z or Y and the by phrase to X. But
> what are X, Y and Z? They sound to me remarkably like our old friend
> 'argument structure', which some of us have tried so hard to do without
> because we couldn't decide whether it belongs in syntax or in semantics. 

Actually, I don't think we need a default active specification. Active 
is just another construction (perhaps lexically associated with 
VERB rather than a subcategory, in which case it *is* a default 
after all!): ref of subject of ACTIVE is er of its sense, ref of o of A is 
ee of its sense... (what's a bit worrying is that this second property 
*is* overriden by the model for passives, though if ACTIVE is a 
subcatgeory of verbs there's no problem).

As for like/please etc, we kind of do have argument structure 
already: er and ee. The meanings of LIKE and PLEASE are pretty 
close (though they are by no means identical: Liking is a state, 
Pleasing an activity). They form part of the same conceptual 
schema. However, just as Buying and Selling profile different parts 
of their relevant schemas, so do Liking and Pleasing. It is difficult 
to decide which argument of Liking/Pleasing is most like an er and 
which most like an ee. This is why some languages have one 
arrangment and some another, and some have both. Similarly for 
spray/load: the thing that is most affected by the action is made 
into the ee, since the other properties of ee's are pretty evenly 
spread between the two non-er arguments.

This isn't quite the same as 'argument structure', though it comes 
close. For purity it would be better to say that 'er' is simply 
shortand for 'argument that has the most of the properties typically 
associated with the referents of subjects' and 'ee' is.... '...objects'. 
Then they could be taken out of semantics altogether (once we've 
identified the properties typically associated with the referents of 
subjects and objects).
> 
> Maybe argument structure is ok after all? I.e. maybe there is a
> language-oriented level of cognitive structure, as I think Jasper and I
> decided in our Re-cycling paper. But if there is, there's a lot of
> redundancy if I'm also right about the need for empty (PRO) subjects
> because this provides a relatively semantics-oriented syntactic structure
> which interfaces rather easily with the syntax-oriented semantics of
> argument-structure.

I think that the semantics of lexemes must be 'syntax oriented', 
because it must tell you which semantic arguments link with which 
syntactic ones (the linking generlaities I alluded to above are, after 
all, just generalisations over observed lexical characteristics). This 
was also the point of the comparison we did of the English Dutch 
and German words for driving, riding etc. No-one's going to say that 
English and German speakers conceptualise the act of cycling 
differently, but it is obvious (look at the facts!) that the senses of 
the verbs RIDE and REITEN are different. At the point where it is 
closest to syntax, semantic structure is (can be) quite closely 
controlled by its requirements.


> 
> Dick
> 
> Richard (= Dick) Hudson
> 
> Phonetics and Linguistics, University College London, 
> Gower Street, London WC1E  6BT.
> +44(0)171 419 3152; fax +44(0)171 383 4108;
> http://www.phon.ucl.ac.uk/home/dick/home.htm




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%