Prevailing approaches do not have a computational lexicon

[Andrew Carnie]

Hi Carl,

I speak here only for myself, and not for all P&P practictioners, or even
all those that don't believe in a generative lexicon.

Of course, the answer to your question raises all *sorts* of issues, and
I'm really not the best person to answer (I'm a case person, not a lexicon
person), but I'll give it a shot. What I'm about to say is at least
partly incompatible with Chomsky 95.

First off, we might assume that it isn't feature checking/unification per
se, but feature value satisfaction. That is, Item A has a feature [A] that
needs a value, Item B has a similar (or identical) feature [A'] with the
value <a>, value <a> is then assigned to feature A in item A as part of
the operation MERGE (Move is simply a kind of merge, that takes a
pre-existing term in the set, and copies it and remerges it).
This treatment should sound kind of familiar <ahem!>. Chomsky has a
whole lot of discussion about "labels" where only the head provides
syntactic features to the phrase. I think this is clearly wrong, and
what you have is some kind of intersection of the features (don't take
"intersection" too literally) that functions as the phrase "label". There
is a whole lot of ink spilled on this "label" business. See Chris Collin's
work for a detailed discussion, also there is a paper in the Specifiers
volume edited by Adger et al. (Sorry can't remember the author and I'm at
home).

Ok, so with the idea that you don't have "checking" per se, we can at
least partly underspecify the lexical items. The work of the neo-
generative-semantics types hold that there are a relatively limited number
of syntactic primitives. The formal features that drive their combination
result in a set of possible structures. The "Encyclopedia" (ie. the list
of form meaning pairs) then comes over and fills in the blanks. The idea
here is that asfar as the syntax is concerned the difference between
"cat", "dog" and "banana slug" is irrelevant. The sentences:

	The cat bit the dog
	The dog bit the cat
	The cat bit the banana slug
	The cat chased the dog
	The dog hit the cat
	The banana slug poisoned the cat

have essentially the same syntactic structures (Ok, I'm oversimplifying
again. Obviously there are featural difference between the objects of hit
and the objects of poisoned, but the underlying point should be obvious).

Ok, now to the question of the difference between say the noun
father and the verb father. I said previously this kind of difference was
underspecified. One usual way of doing this is to claim that there are
functional categories (n) and (v) that select for (i.e. have features that
require values) underspecified roots. The root FATHER can be merged with
either n or v and result in a "noun" or a "verb". The neo-GS way of
marking event structure is also done this way. Case features are "checked"
by way of requiring a DP with a particular set of case feature values,
presumably marked on the D.

Was that even remotely comprehensible?

Andrew


On Fri, 4 Oct 2002, Carl Pollard wrote:

> Hi Andrew,
>
> Hope you don't mind another naive question. I know you said it is a
> maytter of some debate, but in your own view, how does feature checking
> happen while the structure is being built, pre-morphology, if the
> features are still underspecified?
>
> Thanks,
>
> Carl
>
> >
> Hi Carl,
>
> Yes, the debate is roughly as you've characterized it. "Classical" MP
> (if I may use the term "classical" loosly) holds that lexical items come
> fully formed into the syntax, complete with feature structures (albeit not
> ones as rigourously defined as those in HPSG). The words are then combined
> via merge, into phrase markers, where all the relevant features are
> "checked" (very roughly equivalent to unified -- which is why some of us
> lurk on these lists). The alternative view (a la Distributed morphology)
> is that major class lexical items such as verbs and nouns come into the
> syntax partly underspecified (including underspecification for syntactic
> category). For example, a root such as "DIE" comes in only partly
> specified, it is by virtue of combining it with functional categories such
> as v (a verbalizer), aspect, tense, etc. it becomes either the verb die or
> the noun death. The actual lexical item (technically vocabulary item) is
> inserted at the end of the derivation. There is a fair amount of debate
> about such things as the role of various semantic primitives (such as
> "Cause" and "become" and the extent to which these are present in the
> grammar. How checking works in this structure building system is also a
> matter of some debate.
> >>
>

------------------------------------------------------------------------------
     O  O  O  O  O  		Andrew Carnie, Ph.D.
    <|\/|\/|\/|\/|>  		Asst. Professor of Linguistics
     =  =  =  =  =  		Department of Linguistics
    << << << << << 		Douglass 200E, University of Arizona
    ~~ ~~ ~~ ~~ ~~ 		Tucson, AZ 85721

				Tel: (520) 621 2802
				Cell: (520) 971 1166
				http://linguistlist.org/~carnie