ON PARSING PREFERENCES
Lenhart K. Schubert
Department of Computing Science
University of Alberta, Edmonton
Abstract. It is argued
that
syntactic p r e f e r e n c e
principles
such as Right A s s o c i a t i o n and Minimal
Attachment
are
unsatisfactory
as
usually
formulated.
Among
the
difficulties
are:
(I)
dependence
on
ill-specified
or
implausible
principles
of p a r s e r operation; (2) d e p e n d e n c e on
questionable a s s u m p t i o n s about syntax; (3) lack Of
provision,
even in principle, for i n t e g r a t i o n with
semantic and p r a g m a t i c p r e f e r e n c e
principles;
and
(4) apparent counterexamples, even when d i s c o u n t i n g
(I)-(3).
A possible
approach
to a
solution is
sketched.
while
the
latter
chooses
the
longest reduction
among
those
possible
reductions
whose
initial
constituent
is "strongest" (e.g., reducing V NP PP
to VP is p r e f e r r e d to reducing NP PP to PP).
In
(5), M i n i m a l
Attachment
would
predict
association
of the PP on that rack with wanted,
while the actual p r e f e r e n c e is for a s s o c i a t i o n with
dress. Both Ford et al.
and S h i e b e r
account
for
this fact by appeal
to lexical preferences: for
Ford et al., the strongest form of want takes an NP
complement only, so that Final A r g u m e n t s
prevails;
for Shieber,
the NP the dress is stronger than
wanted,
viewed
as a V requiring
NP
and
PP
complements,
so
that
the
shorter
reduction
prevails.
I. Some p r e f e r e n c e p r i n c i p l e s
The
following
are
some
standard
kinds
of
sentences
illustrating
the
role of
syntactic
preferences.
(I) John bought the book which I had selected for
Mary
(2) John p r o m i s e d to visit frequently
(3) The girl in the chair with the
spindly
legs
looks bored
(4) John carried the g r o c e r i e s for Mary
(5) She wanted the dress on that rack
(6) The horse raced past the Darn fell
(7) The boy got fat melted
sentence (6) leads most people "down the garden
path",
a fact
explainable
in terms
of Minimal
Attachment or its variants.
The e x p l a n a t i o n
also
works
for
(7)
(in the c a s e of Ford et al. with
appeal to the a d d i t i o n a l p r i n c i p l e that r e - a n a l y s i s
of complete p h r a s e s requiring r e - c a t e g o r i z a t i o n
of
lexical c o n s t i t u e n t s is not possible). Purportedly,
this
is an a d v a n t a g e over Marcus' (1980) p a r s i n g
model,
whose
three-phrase
buffer
should
allow
t r o u b l e - f r e e p a r s i n g of (7).
(I)
(3) illustrate Right A s s o c i a t i o n of PP's
and adverbs, i.e.,
the p r e f e r r e d a s s o c i a t i o n
of
these modifiers with the rightmost v e r b (phrase) or
noun
(phrase) they can modify (Kimball 1973). Some
variants of Right A s s o c i a t i o n
(also c h a r a c t e r i z e d
as Late Closure or Low Attachment) w h i c h have Dean
p r o p o s e d are F i n a l A r g u m e n t s (Ford et al. 1982) and
Shifting Preference (Shieber 1983); the
former
is
roughly
Late
Closure
restricted
to the last
o b l i g a t o r y c o n s t i t u e n t and any
following
optional
constituents
of v e r b phrases, while the latter is
Late Closure w i t h i n the context of an LR(1)
shiftreduce parser.
2. Problems w i t h the p r e f e r e n c e p r i n c i p l e s
2.1 D e p e n d e n c e on i l l - s p e c i f i e d or i m p l a u s i b l e
p r i n c i p l e s of parser operation.
Frazier
&
Fodor's
(1979)
model does
not
c o m p l e t e l y s p e c i f y what
structures
are
built
as
each
new w o r d is accommodated. C o n s e q u e n t l y it is
hard to tell e x a c t l y
what
the
effects
Of their
p r e f e r e n c e p r i n c i p l e s are.
Shieber's
(1983)
shift-reduce
p a r s e r is welldefined. However, it p o s t u l a t e s
complete
phrases
only,
whereas
human p a r s i n g
appears
to involve
integration of c o m p l e t e l y
analyzed
phrases
into
larger,
incomplete
phrases.
C o n s i d e r for e x a m p l e
the following sentence Deginnings:
(8) So I says to the ...
(9) The man r e c o n c i l e d h e r s e l f to the ...
(10) The news a n n o u n c e d on the ...
(11) The r e p o r t e r a n n o u n c e d on the ...
(12) John beat a rather hasty and u n d i g n i f i e d ...
People p r e s e n t e d with
complete,
spoken
sentences
beginning
like
(8) and
(9) are able to signal
detection
of
the errors
about
two
or
three
syllables
after
their
occurrence. Thus a g r e e m e n t
Regarding
(4), it would seem that a c c o r d i n g to
Right A s s o c i a t i o n
the PP for Mar~
should
be
preferred
as p o s t m o d i f i e r of g r o c e r i e s rather than
carried; yet the opposite is the case.
Frazier
&
Fodor's
(1979) e x p l a n a t i o n is based on the assumed
phrase structure rules VP -> V NP PP,
and
NP ->
NP PP:
attachment
of the PP into the VP m i n i m i z e s
the resultant number of nodes.
This p r i n c i p l e
of
Minimal
Attachment
is assumed to take p r e c e d e n c e
over Right Association. Ford et al's (1982) variant
is Invoked Attachment, and Shieber's (1983) variant
is Maximal Reduction; roughly speaking, the
former
amounts to early closure of no___nn-final constituents,
247
features appear to p r o p a g a t e upward from i n c o m p l e t e
constituents.
(10)
and
(11)
suggest
that
even
semantic
features
(logical
translations?)
are
propagated
before
phrase
completion.
The
"premature"
recognition
of
the
idiom
in
(12)
provides further e v i d e n c e for early i n t e g r a t i o n
of
partial structures.
the
theory;
(however, see Crain & S t e e d m a n 1981).
Two views that seem to u n d e r l i e some d i s c u s s i o n s of
this issue are (a) that syntactic
preferences
are
"defaults"
that
come
into
effect
only
in
the
a b s e n c e Of s e m a n t i c / p r a g m a t i c p r e f e r e n c e s ;
or
(b)
that
alternatives
are tried in o r d e r of s y n t a c t i c
preference, w i t h semantic tests s e r v i n g
to
reject
incoherent
combinations.
Evidence
against
both
p o s i t i o n s is found in s e n t e n c e s in w h i c h
syntactic
preferences
prevail
over
much
more
coherent
alternatives:
(21) Mary saw the man w h o had lived w i t h her
w h i l e on m a t e r n i t y leave.
(22) John met the tall, slim, a u b u r n - h a i r e d g i r l
from M o n t r e a l that he m a r r i e d at a d a n c e
(23) John was named a f t e r his twin sister
W h a t we
apparently
need
is
not
hard
and
fast
decision
rules,
but
some
way
of
trading
off
s y n t a c t i c and n o n - s y n t a c t i c p r e f e r e n c e s of
various
s t r e n g t h s against each other.
These
considerations
appear to favour a "fullpaths" parser which integrates each s u c c e s s i v e w o r d
(in p o s s i b l y
more
ways
than
one)
into
a
comprehensive
parse
tree
(with
overlaid
alternatives) spanning all of the text processed.
Ford
et
al.'s
(1982)
parser
does
develop
complete
top-down
paths,
but
the nodes on these
paths d o m i n a t e no text. Nodes p o s t u l a t e d
bottom-up
extend only one level above c o m p l e t e nodes.
2.2 D e p e n d e n c e on q u e s t i o n a b l e
ab____out syntax
assumptions
2.4 A p p a r e n t
The
successful
prediction
of
observed
p r e f e r e n c e s in (4) d e p e n d e d on an
assumption
that
PP p o s t m o d i f i e r s are added to c a r r i e d v i a the rule
VP -> V NP PP and to g r o c e r i e s via the rule
NP ->
NP PP.
However,
these rules fail to do justice to
certain
systematic
similarities
between
verb
phrases and noun phrases, evident in such pairs as
(13) John
loudly
quarreled
with
Mary
in
the
kitchen
(14) John's loud q u a r r e l w i t h Mary in the k i t c h e n
When the a n a l y s e s are a l i g n e d
by
postulating
two
levels
of
postmodification
for
both
verbs
and
nouns,
the
accounts
of
many
examples
that
supposedly
involve
Minimal A t t a c h m e n t (or M a x i m a l
Reduction) are spoiled. These include (4)
as
well
as
standard
examples
involving
non-preferred
r e l a t i v e clauses, such as
(15) John told the girl that he loved the story
(16) Is the block sitting in the box?
counterexamples.
There
appear
to
be
straightforward
counterexamples
to
the
syntactic
preference
principles
which
have
been
proposed, even if we
discount evidence
for
integration
of
incomplete
structures,
accept the s y n t a c t i c a s s u m p t i o n s made,
and r e s t r i c t o u r s e l v e s to cases w h e r e none
of
the
a l t e r n a t i v e s show any s e m a n t i c anomaly.
The
following
are
a p p a r e n t c o u n t e r e x a m p l e s to
Right A s s o c i a t i o n (and S h i f t i n g Preference. etc.):
(24) John stopped s p e a k i n g f r e q u e n t l y
(25) John d i s c u s s e d the girl that he met with his
mother
(26) John was a l a r m e d by the d i s a p p e a r a n c e of the
a d m i n i s t r a t o r from head o f f i c e
(27) The d e r a n g e d i n v e n t o r a n n o u n c e d that he had
p e r f e c t e d his d e s i g n of a clip car shoe
(shoe car clip, clip
shoe
car,
shoe
clip
car, etc.)
(28) Lee and Kim or Sandy d e p a r t e d
(29) a. John removed all of the fat and some of
the bones from the roast
b. John removed all of the fat and sinewy
pieces of meat
The
point
Of
(24)-(26) should De clear. (27) and
(28) show the lack of r i g h t - a s s o c i a t i v e
tendencies
in
compound
nouns
and c o o r d i n a t e d phrases. (29a)
i l l u s t r a t e s the n o n - o c c u r r e n c e
of
a garden
path
predicted
by
Right
Association
(at
least
Dy
S h i e b e r ' s version); note
the
possible
adjectival
r e a d i n g of fat and ..., as i l l u s t r a t e d in (29b).
2.3 Lack of p r o v i s i o n for i n t e g r a t i o n w i t h
semantic/pragmatic preference principles
Right A s s o c i a t i o n and
Minimal
Attachment
(and
their
variants)
are
typically
presented
as
principles
which
prescribe
particular
parser
choices.
As such, they are simply wrong, since the
choices often do not coincide
with
human
choices
for
text
which
is
s e m a n t i c a l l y or p r a g m a t i c a l l y
biased.
For example, there are c o n c e i v a b l e
contexts
in
which the PP in (4) a s s o c i a t e s with the verb, or in
which (7) is trouble-free.
(For the latter, imagine
a
story
in w h i c h
a young worker in a shortening
factory toils long hours melting down
hog
fat
in
clarifying
vats.)
Indeed, even isolated sentences
d e m o n s t r a t e the effect of semantics:
(~7) John met the girl that he married at a dance
(]8) John saw the bird with t~e y e l l o w wings
(!9) She w a n t e d the gun on her night table
(20) This lens gets light focused
These sentences should be contrasted w i t h (I), (4),
(5). and (7) respectively.
The
following
are
apparent c o u n t e r e x a m p l e s to
M i n i m a l A t t a c h m e n t (or Maximal Reduction):
(30) John a b a n d o n e d the attempt to p l e a s e M a r y
(31) Kim o v e r h e a r d John and Mary's
quarrel
with
Sue
(32) John
carried
the
umDre!la, the t r a n s i s t e r
radio, the bundle of old magazines, and
the
g r o c e r i e s for Mary
(33) The boy got fat spattered on his arm
While
the
account
of
(30)
and
(31)
can be
rescued by d i s t i n g u i s h i n g s u b c a t e g o r i z e d
and
nonsubcategorized
noun
postmodifiers,
such
a
move
would lead to the
failures
already
mentioned
in
section
2.2.
Ford
et
al.
(1982)
w o u l d have no
While the reversal of choices
Dy
semantic
and
pragmatic
factors is regularly acknowledged, these
factors are rarely a s s i g n e d any
explicit
role
in
248
e x p e c t a t i o n p o t e n t i a l to the total p o t e n t i a l of the
node.
The e x p e c t a t i o n
p o t e n t i a l c o n t r i b u t e d by a
d a u g h t e r is maximal
if the d a u g h t e r
immediately
follows
the m o t h e r ' s head lexeme, and d e c r e a s e s as
the distance (in words) of the d a u g h t e r
from the
head lexeme
increases.
The d e c a y of e x p e c t a t i o n
p o t e n t i a l s with d i s t a n c e
evidently
results
in a
r i g h t - a s s o c i a t i v e tendency. The maximal e x p e c t a t i o n
potentials
of the d a u g h t e r s
of a node are fixed
p a r a m e t e r s of the rule i n s t a n t i a t e d
by the
node.
They can be thought Of as e n c o d i n g the "affinity"
of
the
head
daughter
for
the
remaining
constituents, with "strongly expected" c o n s t i t u e n t s
having r e l a t i v e l y large e x p e c t a t i o n potentials. For
example, I would assume that verbs have a generally
stronger
affinity
for
(certain kinds
Of)
PP
adjuncts than do nouns. This a s s u m p t i o n can explain
P P - a s s o c i a t i o n with the v e r b in examples like
(4),
even
if
the
rules
governing
v e r b and noun
postmodification
are
taken
to be
structurally
analogous.
Similarly
the
scheme
allows
for
c o u n t e r e x a m p l e s to R i g h t
Association
like
(24),
where the a f f i n i t y of the first v e r b (stop) for the
frequency
adverbial
may
be
assumed
to De
sufficiently great c o m p a r e d to that of the
second
(speak)
to o v e r p o w e r
a weak
right-associatlve
effect resulting
from
the d e c a y
of e x p e c t a t i o n
p o t e n t i a l s with distance.
t r o u b l e with (30) or (31), but
they,
too,
pay a
price:
they would e r r o n e o u s l y predict a s s o c i a t i o n
of the PP with the object NP in
(34) Sue had d i f f i c u l t i e s with the teachers
(35) Sue wanted the d r e s s for Mary
(36) Sue returned the dress for Mary
(32) is the
sort
of example w h i c h
motivated
Frazier
&
Fodor's
(1979)
Local
Attachment
principle, but their
parsing
model
remains
too
sketchy for the i m p l i c a t i o n s of the p r i n c i p l e to be
clear.
Concerning
(33),
a small-scale experiment
indicates that this is
not a garden
path.
This
result
appears
to i n v a l i d a t e the a c c o u n t s of (7)
based on irreversible closure at fat. Moreover, the
d i f f e r e n c e between (7) and (33) cannot De explained
in terms of one-word
lookahead,
since a further
experiment has indicated that
(37) The boy got fat spattered.
is q u i t e as difficult to understand as (7).
3. Towards an account of p r e f e r e n c e t r a d e - o f f s
My
main
objective
has been to point
out
deficiencies
in current
theories
of
parsing
preferences,
and
hence
to spur their revision. ]
conclude with my own rather speculative
proposals,
w h i c h represent work in progress.
I suggest
that
the effect
Of
semantics and
pragmatics can in p r i n c i p l e be c a p t u r e d
through
a
semantic
potential
contributed
to each
node
potential by s e m a n t i c / p r a g m a t i c p r o c e s s i n g
of the
node.
The
semantic
potential
of a terminal node
(i.e., a lexical node w i t h a p a r t i c u l a r
choice
of
word
sense
for the w o r d it dominates) is high to
the extent that the a s s o c i a t e d word sense refers to
a f a m i l i a r (highly consolidated)
and
contextually
salient
concept
(entity, predicate, or function).
For example, a noun node d o m i n a t i n g
star,
with a
translation
expressing
the a s t r o n o m i c a l sense Of
the word,
presumably
has
a
higher
semantic
p o t e n t i a l than a similar node for the s h o w - b u s ~ n e s s
sense
Of
the word,
when an a s t r o n o m i c a l context
(but
no
show-business
context)
has
been
established;
and vice versa. P o s s i b l y a spreading
a c t i v a t i o n m e c h a n i s m could account for the c o n t e x t dependent part
of the
semantic
potential
(of.,
Quillian
1968,
Collins
& Loftus
1975, C h a r n i a k
1983).
In summary,
the
p r o p o s e d model involves (I) a
full-paths
parser
that
schedules
tree
pruning
decisions
so as to limit the number of a m b i g u o u s
constituents
to three;
and
(2) a
system
of
numerical
"potentials"
as a way of i m p l e m e n t i n g
p r e f e r e n c e trade-offs. These p o t e n t i a l s (or "levels
of activation") are a s s i g n e d to nodes as a f u n c t i o n
of their
syntactic/semantic/pragmatic
structure,
and the p r e f e r r e d structures are those w h i c h lead
to a g l o b a l l y high potential. The total
potential
of
a
node
consists
of
(a) a negative
rule
potential~ (b) a p o s i t i v e semantic potential,
(c)
positive
e x p e c t a t i o n p o t e n t i a l s c o n t r i b u t e d by all
daughters following the
head
(where these
decay
with d i s t a n c e
from
the head
lexeme),
and
(d)
transmitted p o t e n t i a l s passed on from the d a u g h t e r s
to the mother.
I have a l r e a d y argued for a full-paths
approach
in which
not
only
complete p h r a s e s but a l s o all
incomplete
phrases
are
fully
integrated
into
(overlaid)
parse
trees d o m i n a t i n g all of the text
seen so far.
Thus
features
and partial
logical
translations
can
be p r o p a g a t e d
and checked for
c o n s i s t e n c y as early as possible, and a l t e r n a t i v e s
chosen or d i s c a r d e d
on the basis of all of the
available information.
The s e m a n t i c p o t e n t i a l of a n o n t e r m i n a l node
is
high
to the
extent
that its logical t r a n s l a t i o n
(obtained
by
suitably
combining
the
logical
translations
of
the
daughters)
is
easily
t r a n s f o r m e d and e l a b o r a t e d into a d e s c r i p t i o n of a
familiar
and
c o n t e x t u a l l y relevant kind of object
or situation. (My a s s u m p t i o n is that an u n a m b i g u o u s
m e a n i n g r e p r e s e n t a t i o n of a phrase is c o m p u t e d
on
the b a s i s
of its initial logical form by c o n t e x t dependent
pragmatic
processes;
see S c h u b e r t
&
Pelletier
1982.)
For
example, the sentences Time
flies, The y e a r s pass swiftly,
The m i n u t e s
creep
by,
etc., are i n s t a n c e s of the f a m i l i a r p a t t e r n of
predication
< p r e d i c a t e of l o c o m o t i o n > (<time term>),
and as such are easily t r a n s f o r m a b l e
into c e r t a i n
commonplace
(and unambiguous)
assertions
about
one's p e r s o n a l sense of p r o g r e s s i o n
through
time.
Thus
they are likely to be a s s i g n e d high s e m a n t i c
The rule p o t e n t i a l
is a n e g a t i v e
increment
contributed
by a phrase structure rule to any node
which i n s t a n t i a t e s that rule. Rule p o t e n t i a l s
lead
to a m i n i m a l - a t t a c h m e n t
tendency: they "inhibit"
the use of rules, so that a parse
tree
using
few
rules will
generally
De p r e f e r r e d
to one using
many. Lexical p r e f e r e n c e s can be captured by m a k i n g
the rule p o t e n t i a l
more
negative
for the more
unusual
rules
(e.g.,
for N --> fat,
and
for
V -~ time).
Each "expected" d a u g h t e r of a node w h i c h follows
the node's head lexeme c o n t r i b q t e s
a non-negative
249
potentials,
and
so will
not
easily admit
any
alternative
analysis.
Similarly
the phrases met
[someone] at a dance (versus married [someone] at a
dance) in sentence (17), and bird with
the y e l l o w
wings (versus saw [something] with the yellow w i n g s
) in (18) are easily i n t e r p r e t e d as d e s c r i p t i o n s of
familiar
kinds
of o b j e c t s and situations, and as
such c o n t r i b u t e semantic p o t e n t i a l s
that help to
edge Out competing analyses.
semantic
potential),
preliminary
investigation
suggests that they can do justice to examples
like
(I)-(37).
Schubert
&
Pelletier
1982 b r i e f l y
d e s c r i b e d a f u l l - p a t h s parser w h i c h
chains
upward
from the c u r r e n t word to current " e x p e c t a t i o n s " by
"left-corner stack-ups"
Of rules.
However,
this
parser
searched a l t e r n a t i v e s by b a c k t r a c k i n g only
and did not handle
gaps
or coordination.
A new
version
designed
to
handle
most
aspects
of
G e n e r a l i z e d Phrase Structure G r a m m a r (see Gazdar et
al., to appear) is currently being implemented.
Crain
& Steedman's
(1981)
very
interesting
suggestion
that
readings
with
few
new
presuppositions
are p r e f e r r e d has a p o s s i b l e p l a c e
in the p r o p o s e d scheme: the m a p p i n g
from
logical
form to u n a m b i g u o u s
meaning
representation
may
often be r e l a t i v e l y simple when few p r e s u p p o s i t i o n s
need to De a d d e d to the context.
However,
their
more g e n e r a l p l a u s i b i l i t y p r i n c i p l e appears to fail
for e x a m p l e s like (21)-(23).
Acknowledgements
I thank my unpaid
informants
who patiently
a n s w e r e d strange q u e s t i o n s about s t r a n g e sentences.
I have also b e n e f i t e d from d i s c u s s i o n s with m e m b e r s
Of
the
Logical
Grammar
Study
G r o u p at the
U n i v e r s i t y of Alberta,
especially
Matthew
Dryer,
who
suggested
some
relevant
references.
The
r e s e a r c h was supported by the N a t u r a l S c i e n c e s
and
Engineering
Research
Council
of C a n a d a
under
O p e r a t i n g Grant A8818.
Note
that
the
above
pattern
of t e m p o r a l
p r e d i c a t i o n may well be c o n s i d e r e d
to v i o l a t e
a
selectional
restriction,
in that p r e d i c a t e s
of
locomotion cannot l i t e r a l l y a p p l y
to times.
Thus
the nodes with the highest semantic p o t e n t i a l are
not n e c e s s a r i l y those c o n f o r m i n g
most
fully w i t h
selectional
restrictions.
This
leads to
some
departures
from
Wilks'
theory
of
semantic
preferences
(e.g.,
1976), a l t h o u g h I suppose that
normally the most e a s i l y i n t e r p r e t a b l e
nodes,
and
hence
those with
the highest semantic potential,
are indeed the ones that c o n f o r m w i t h
selectional
restrictions.
References
Charniak, E. (1983). Passing markers: a theory
of
contextual
i n f l u e n c e in l a n g u a g e comprehension.
C o g n i t i v e Science 7, pp. 171-190.
Collins, A. M. & Loftus, E. F. (1975). A spreading
activation
theory
of
semantic
processing.
P s y c h o l o g i c a l R e v i e w 82, pp. 407-428.
Crain, S. & Steedman, M. (1981). The use of c o n t e x t
by the P s y c h o l o g i c a l Parser. P a p e r presented
at
the
Symposium
on
Modelling
Human
Parsing
Strategies, Center for C o g n i t i v e Science,
Univ.
of Texas, Austin.
Ford,
M.,
Bresnan,
J.
& Kaplan,
R.
(1981). A
c o m p e t e n c e - b a s e d theory of syntactic closure. In
Sresnan, J. (ed.), The Mental R e p r e s e n t a t i o n
of
G r a m m a t i c a l Relations
MIT Press, Cambridge, MA.
Frazier,
L.
& Fodor,
J.
(1979).
The S a u s a g e
Machine:
a
new
two-stage
parsing
model.
C o g n i t i o n 6, pp. 191-325.
Gazdar,
G.,
Klein, E., Pullum, G. K. & Sag, I. A.
(to appear).
Generalized
Phrase
Structure
Grammar: A Study in English Syntax.
Kimball,
J.
(1973).
Seven
p r i n c i p l e s of s u r f a c e
structure p a r s i n g in natural language. C o g n i t i o n
2, pp. 15-47.
Marcus,
M.
(1980).
A
Theory
of
Syntactic
Recognition
for Natural
Language,
MIT Press,
Cambridge, MA.
Quillian, M. R. (1968). S e m a n t i c memory. In Minsky,
M. (ed.), Semantic Information
Processing,
MIT
Press, Cambridge, MA, pp. 227-270.
Schubert,
L.K.
& Pelletier,
F. J. (1982). F r o m
English to logic:
context-free
computation
of
'conventional'
logical
translations. Am. J. of
C o m p u t a t i o n a l Linguistics 8, pp. 26-44.
Shieber, S. M. (1983). Sentence d i s a m b i g u a t i o n by a
shift-reduce parsing technique. Proc.
Sth
Int.
Conf.
on Artificial
Intelligence,
Aug. 8-12,
Karlsruhe,
W.
Germany,
pp. 699-703.
Also
in
Proc.
of
the 21st Ann. Meet. of the Assoc. for
Computational
Linguistics,
June
15-17,
MIT,
Cambridge, MA., pp. 113-118.
Wilks,
Y. (1976). Parsing E n g l i s h II. In Charniak,
E. & Wilks, Y. (eds.), C o m p u t a t i o n a l
Semantics,
North-Holland, Amsterdam, pp. 155-184.
The d i f f e r e n c e
between such pairs of sentences
as (17) and (22) can now be explained in terms of
semantic/syntactic
potential
trade-offs.
In both
sentences the semantic
potential
of the reading
which a s s o c i a t e s
the
PP with
the first v e r b is
relatively high. However, only in (17)
is the PP
close enough
to the first verb for this effect to
overpower the r i g h t - a s s o c i a t i v e
tendency
inherent
in the d e c a y of e x p e c t a t i o n potentials.
The
final
contribution
to the potential of a
node is the t r a n s m i t t e d potential, i.e., the sum of
potentials
of the daughters.
Thus
the
total
potential
at
a
node
reflects
the
syntactic/semantic/pragmatic
properties
of
the
entire tree it dominates.
A crucial
question
that
remains concerns the
scheduling Of d e c i s i o n s to discard
g l o b a l l y weak
hypotheses.
Examples
like
(33) have convinced me
that Marcus
(1980)
was
essentially
correct
in
positing
a
three-phrase
limit
on
successive
ambiguous constituents. (In the context of a fullpaths parser, a m b i g u o u s constituents can be d e f i n e d
in terms
of "upward or-forks" in phrase structure
trees.) Thus I propose
to discard
the globally
weakest
alternative
at the latest when it is not
possible to proceed rightward without
creating
a
fourth
ambiguous
constituent.
Very
weak
alternatives
(relative to the others)
may
be
discarded
earlier, and this a s s u m p t i o n can account
for early d i s a m b i g u a t i o n in cases
like
(10)
and
(11).
out
Although
these
proposals
are not fully worked
(especially with regard to the d e f i n i t i o n
of
250