PS-rules for English NP/QP

[George Aaron Broadwell]

 English rules for NP/QP


Dear list participants,

A week or so ago, I asked for some pointers to English PS-rules that are
detailed enough to correctly predict the order and cooccurrence restrictions
for pre-head material in the English NP/QP.  This is for use in an
undergraduate syntax class.  (I made the "mistake" of giving my class real
sentences from the web and soon discovered that our toy grammar was
hopelessly inadequate for phrases like All the United Nations climate
panel's final recommendations on climate change!)

Thanks for all the responses and pointers to literature.   It is clear that
the structure of the pre-head material in English NP/QP is very complex.

1.  References

The most comprehensive discussions that I found were in Jackendoff (1977)
and in Huddleston and Pullum (2002) (where the chapter on NPs is by John
Payne and Rodney Huddleston). I was also referred C.L. Baker's syntax text
and to the Pargram project English grammar.  (For various boring reasons, I
couldn't access these.)  Also helpful was a paper by Robert Berwick and Sam
Pilato (1987) "Learning syntax by automata induction" Machine learning
2:9-38.  I also found useful a chart in the following:

@book{Dekeyser79,
   Address = {Antwerpen/Amsterdam},
   Author = {Dekeyser, Xavier and Devriendt, Betty and Tops, Guy A. J.
and Geukens, Steven},
   Booktitle = {Foundations of English grammar},
   Publisher = {De Nederlandsche Boekhandel},
   Title = {Foundations of English grammar},
   Year = {1979}}

which was kindly forwarded to me by Koenraad de Smedt.


2.  PS-rules

It is not terribly difficult to write PS-rules that will get all the
possible elements in the right order.  The following set of rules will do a
reasonably accurate job.  (I've omitted parentheses, on the assumption that
all the elements are optional).


QP --->  Q NP
NP --->  {Det, NP[CASE GEN]} Num N'
N' ---> AdjP* N' PP-adj CP-rel
N' ---> N-mod NP-descrip  N  PP-obl CP-comp

Q would be a POS label for words like all, half, or both, which precede the
genitive or determiner.  (Payne and Huddleston call these predeterminers;
Jackendoff calls them N''' specifiers)
Num would be a POS label for words like few, many, two, which follow the
genitive or determiner.
AdjP includes ordinary adjectives, and possibly also participles which have
been converted to Adj by lexical rule (e.g. a boring lecture)
PP-adj is for adjunct PPs; CP-rel is for relative clauses
N-mod is for the pre-head N in phrases like box cutter or college
president.  (On some analyses, these might be done in the lexicon.)
NP-descrip is for the 'descriptive genitive'  in phrases like women's
college or men's room.  (Ditto a possible lexical analysis.)
PP-obl is for PPs that seem to act like arguments of the head N, e.g.
description
of John.
CP-comp is for CPs that are arguments of the head N, e.g. claim that the
earth is flat

Flatter analyses would certainly be possible.  There is also disagreement in
the literature about the proper place on the right periphery to attach the
adjunct elements PP-adj and CP-rel.

The challenging thing is to account for which of the prehead modifiers can
cooccur; to say when the head of NP can be omitted; and to account for when
the partitive of appears.  It is not easy to write lexical entries that make
the correct predictions.

The most promising line of analysis to me seems to be a 'fused head
analysis', where Q, Det, NP[CASE GEN] and Num elements have an alternate
analysis where they also express the head N of the phrase.  (This is the
view promoted by Payne and Huddleston.) So in all of the books, all is a
fused Q+N.  This seems to be a better analysis than one with an empty N head
or with movement, since it accounts for the absence of any of of elements
that might normally intervene between Q and N.  (*all John's/those/small of
the books; *I saw John's red).

The best way within LFG to accomplish this sort of analysis would seem to be
the lexical sharing approach of Wescoat (2002).  In Wescoat's dissertation,
he explores this treatment of 'pronominal determiners'  (that, these, etc).
Although he doesn't give an analysis of the other fused head constructions
in NP, it seems like his approach would extend naturally to them.

Pursuing the lexical sharing analysis, we would have lexical entries for
some of the prehead material like the following:

those <-- Det | Det N
all <-- Q | Q N
few <-- Num | Num N

There also appear to be some quantifiers that exclude the presence of a
Det/NP[Case Gen], such as each, every, some  (So *each John's book/*John's
each book).  Jackendoff (1977) treats this group as in the POS category Art
(along with determiners like the, that, those).  These Det quantifiers can
also appear with no following N and with a partitive (each of the books).  A
lexical sharing analysis also seems possible for these:

each <-- Det | Det N


We would also need a general lexical rule that allows a genitive NP to
instantiate both the specifier and head positions of the NP.  However, a
genitive without a head seems to exclude partitive or other subcategorized
complements (I saw many/all/two of the books  vs. *I saw Mary's of the books;
I read Joan's claim that passive is a lexical rule/*I read Joan's that
passive is a lexical rule.)

So perhaps the lexical rule for sharing with genitives is

NP [Case Gen] <-- NP | NP N'

By fusing the genitive with N' rather than N, we would predict no possible
complements for the N.

Finally, it appears that all of the elements that nearly all of the pre-head
elements that are lexically shared with N have optional partitive
complements.  (Those of the books/all of the books/each of the books/many of
the books.)  Many partitive complements seem to be subject to a definiteness
restriction (*all of books) that forces a definite article or a possessor to
occur.

I have not treated here the construction that Jackendoff calls
pseudo-partitive (e.g., a bunch/group of men).  The partitive in these
constructions is not subject to the definiteness restriction, so it either
has a different syntactic structure or (more likely) these nouns impose
different selectional restrictions on their complements.


3.  Conclusion and thanks

There are clearly many remaining issues here!  It was interesting to me in
reviewing the literature to see how difficult it is to manage an analysis of
the pre-head material that does not involve lexical sharing.  Though my goal
here was primarily to find PS-rules good enough to work for English NPs, I
found myself convinced that a lexical sharing/fused head is the best
proposal out there for this.

Thanks to those who wrote with comments and suggestions: Mary Dalrymple,
Joan Bresnan, Stephen Wechsler, Koenraad de Smedt, Alice Gaby, Andrew
Spencer, and Robert Berwick.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lfg/attachments/20070222/bded1f20/attachment.htm>