Back up to the PAPPI homepage

Notes on VP Internal Subject Hypothesis and PAPPI

David Lebeaux and Sandiway Fong

Contents

Changes:

Problems:

Changes

X-Bar

Added the [Spec,VP] position in file xbar2.pl as follows:
	spec(v2,np).
Needed to make the new position an A-position:
rhs [np(NP),v2(V2)] ordered specInitial app_goals [aPos(NP)]. 

New Theta Configuration

Eliminated [Spec,IP] as a theta-position by commenting out:
thetaConfig(IP,Role,Subject) :-			% [IP Subject [I1 IO VP]]
	cat(IP,i2),
	VP complement_of IP,
	Subject specifier_of IP,
	externalRole(VP,Role).
We added [Spec,VP] and [Spec,AP] as the replacement theta configuration:
thetaConfig(VP,Role,Subject):-               %thetaConfig (VP... is new code
	(cat(VP,vp) ; cat(VP,ap)),                            
	\+ adjoined(VP),
	Subject specifier_of VP,
	externalRole(VP,Role).
Added new code for thetaRelevant(X). Category that is theta relevant is extended to VP.
thetaRelevant(X):- cat(X,VP).          %new code

Theta Role Assignment

External theta role was not propagated out to the auxiliary to assign to the [Spec,VP] position in the following tree:

Reason was that subcategorizes_for/2 doesn't work properly when it is presented with a [VP Spec [V' V AP]] structure. Previously it was only given something like [VP V AP]]. It's job is to pick out the AP phrase.

In file principles14.pl:

inheritsFeature(X,F) :-	
	X has_feature F
	-> true
%	;  (X subcategorizes_for Y
	;  (Y complement_of X
	   -> inheritsFeature(Y,F)
	   ;   adjoined(X,_,LowerX),
	       inheritsFeature(LowerX,F)).
Who called inheritsFeature/2:
externalRole(XP,Role) :-
	\+ inheritsFeature(XP,blockTheta),
	inheritsFeature(XP,grid([Role],_)).	  % external Role

Passive be

Changed lexical entry so that feature blockTheta is no longer a feature of the passive verb, but a feature that is assigned downwards to the verb it subcategorizes for.

In file lexiconE.pl, we have:

lex(be,v,[morph(be,[]),aux,
	  oneof(_,[
			...
		   [subcat(vp$[morph(_,ed(2)),grid([_],[_|_]),
			       not(noCasemark(+))],[blockTheta,noCasemark(+)])],
		   ...])]).
instead of:
		   [subcat(vp$[morph(_,ed(2)),grid([_],[_|_]),
			       not(noCasemark(+))],[noCasemark(+)]),
		    blockTheta],

Idea is that the main verb should not assign the theta role (normally assigned to the subject position). The blockTheta feature tells the main verb not to do this:

thetaConfig(VP,Role,Subject):-               %thetaConfig (VP... is new code
	cat(VP,vp),                             
	\+ adjoined(VP),
	Subject specifier_of VP,
	externalRole(VP,Role).
See definition of externalRole/2 above

Expletive Linking

Altered expletiveLinking so the VP internal subject links to the object instead of the IP subject.

expletiveLinking in_all_configurations CF
	where linkConfig(CF,Subject,XP) then expletiveLink(Subject,XP).

linkConfig(VP,Subj,XP) :-
	expletiveSubjChain(VP,Subj,C),
	linkPhrase(VP,C,XP).
expletiveSubjChain(VP,Subj,C) :-
	specVP(VP,X),
	(partOfChain(X)
	-> baseTrace(X),
	   headOfChain(Subj,X),
	   expletive(Subj,C)
        ;  expletive(X,C),
	   Subj = X).

specVP(VP,Spec) :-
	\+ adjoined(VP),
	cat(VP,vp),
	Spec specifier_of VP.

Note here that the else-clause in expletiveSubjChain/2 is never used in English. Previously, the subject was base-generated in [Spec,IP], so the then-clause was only used for long-distance expletive linking, e.g. there seems to be a man in the room. Now, the then-clause is always used.

Problem: Massive overgeneration for there seems to be a man in the room. See:

Second problem with expletive linking: Expletive linking not adjoined: can link to the subject of any category.

Some changes in Expletive Linking:
added specMax predicate in expletiveSubjChain(XP,Subj,C)

expletiveSubjChain (XP,Subj,C):-
     specMax(XP,X),
     (partOfChain (X)
     -> baseTrace (X),
        headOfChain (Subj,X),
        expletive (Subj,C)
        ; expletive (X,C(,
        Subj = X).

% should work for VP, AP...   
% need to eliminate the \+ adjoined/1 restriction for
% [AP [AP Spec [A1 tough] [C2..]]]
% consequence: we consider all segments now => possible multiple linking

specMax(XP,Spec):-
%    \+ adjoined(XP),               commented out adjoined code
     cat (XP,C),
     max (C),
     Spec specifier_of XP.

External Theta Role

Needed to modify externalRole/2 otherwise matrix auxiliary inherits role pHolder(location) all the way down the subcategorization chain:

We modify the call to inheritsFeature/2 so that ExtR is a pure variable. Since the theta grid for likely is grid([],[proposition]), ExtR will be instantiated to [], and therefore the code will not look any further down the tree.

externalRole(XP,Role) :-
	\+ inheritsFeature(XP,blockTheta),
	inheritsFeature(XP,grid(ExtR,_)),	  % external Role
	ExtR = [Role].

Block Theta

Have taken out Block Theta from "be" chain.
Not sure what part of the "be" code this is.

Inherent Case

Changed Inherent Case assignment to SPEC of AP, to prevent getting genitive assigned into the SPEC of AP. Before change, wasn't getting:
     I am proud of John.
     I am proud that John is here.
This is because genitive was being assigned to the SPEC.
Added code:
assignsInherentCase(a,NP,gen):- NP has_feature compl.

Control

CONTROL code. Within CONTROL code, changed partOfNonargumentChain(NP) .
% Need to account for
% [IP it [nonarg(+)] [VP t [V2  ..PRO..]]]
% where t m-commands PRO and t is part of a chain headed by 'it':

partofNonargumentChain (NP):-
     NP has_feature nonarg (+)
     -> true
     ; partofChain (NP),
     headOfChain(Head,NP),
     Head has_feature nonarg(+).

Lowering Filter

Lowering Filter added.
Gets rid of derivations in which the wh-element is lowered.

Code for Lowering Filter:

%%% Lowering Filter
%%%
%%% Prohibition against chains with lowering

loweringFilter parser_operation 'Lowering Filter'
      with_components
            io             given_by [ss],
            dep            given_by resolveTraces,
            controlType    given_by filter,
            %optional...
            sCueFilter.

loweringFilter in_all_configurations CF where
        isTrace (CF), downPath (CF,Path)  then Path = [].

Also changed parser, dispalay, defaults (of parser) to include lowering filter. Paser is now j5c parser. This is now included in the defaults file.

Problems

[Spec,VP] now lower in tree than secondary verb complements

[Spec,VP] should be higher than all internal arguments.

Also Binding problem:

Implementation of Larson's analysis of double object constructions will fix this.

Observations


1. The distribution of "each of the NP" and "NP each" is different, with the former being totally free and the latter not. Examples of the freedom of the former:

a) Each of the boys left.
b) I saw each of the boys.
c) I gave a book to each of the boys.

The third one of these allows "each" essentially to have scope over the implicit determiner of "a book": i.e. For each of the boys, I gave him his book.

2. The distribution of the stranding of "each" in the "NP each" construction follows the distribution of "NP each" in the base itself. That is, if "NP each" is possible in a given position in the base form, then the stranding of "each" is possible in that same form, with only slight degradation. But "NP each" is not possible in every position. Some relevant position are:

a) VP internal subject postion: "NP each" generated and "each" strands.
Ex.: The boys may each talk to Bill.

b) AP internal subject position: "NP each" generated and "each" strands.
Ex.: The boys are each intelligent.

c) VP internal position after "to": "each" strands.
Ex.: The boys seem to each like Mary.

d) Subject position of complement under raising verb. Good, maybe very slightly odd.
Ex.: The boys seem each to like Mary.

e) Object postion, no other arguments. Stranding totally out. Also, base form totally out. (As noted earlier, the stranding follows the base form in ungrammaticality: it is really a question of whether the base form is grammatical.)
Ex.: *The boys were hit t each.
*The boys were hit t each quickly.

f) Object position, another PP argument. Stranding marginal (??). Also, base form marginal (one follows the other). Base form slightly better than stranded form.
Ex.: ??I put the books each on the shelf.
?*The books were put t each on the shelf.
??I gave the books each to Mary.
?*The books were given t each to Mary.
? I sent the books each to Mary.
??The books were sent t each to Mary.

g) First Object position in the V NP1 NP2 construction. Perfect, or close to it.
Ex.: I gave the boys each a book.
The boys were given t each a book.
I bought the boys each a book.
The boys were bought t each a book.

3. Note that "each of the boys" and "the boys each" have very differing scopes, with the former able to take scope much more freely:
a) I gave a book to each of the boys.
b) *I gave a book to the boys each.


Back up to the PAPPI homepage