P. 1
The Generative Lexicon_James Pustejovsky

The Generative Lexicon_James Pustejovsky

|Views: 1,554|Likes:
Published by Jong Lung
For Semantics Study
For Semantics Study

More info:

Published by: Jong Lung on Apr 01, 2011
Copyright:Traditional Copyright: All rights reserved

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF or read online from Scribd
See more
See less

07/20/2013

pdf

as

shown

in

( 12

)

below

,

where

D - ARG

is

a

default

argument

,

and

S - ARG

is

a

shadow

argument

.

a

[

ARGl

=

. . .

]

( 12

)

ARGSTR

=

ARG2

=

. . .

D - ARGl

=

. . .

S - ARGI

=

. . .

For

example

,

the

lexical

semantics

for

the

verbs

discussed

above

can

now

be

partially

represented

with

argument

structure

specifications

,

as

illustrated

in

( 13

) - ( 15

) .

68

Chapter 5

[ARGSTR

:=: ARG}, ARG2, ..., ARGn]

[EVENTSTR

:=: EVENT

}, EVENT2

, ..., EVENT
n]

AYAXAe[build(e, x, y) 1\ (}l (e, x) 1\ (}2(e, y)]

This assumes an atomic view on event structure, where internal aspects
of the event referred to by the single variable are inaccessible. Moens
and Steedman (1988) and Pustejovsky (1991b) argue that finer-grained
distinctions are necessary for event descriptions in order to capture some
of the phenomena associated with aspect and Aktionsarten. Assuming
this is the case, we need a means for both representing the subeventual
structure associated with lexical items while expressing the necessary
relation between events and the arguments of the verb. In Pustejovsky
(1995a), a mechanism called Orthogonal Parameter Binding is outlined,
which allows us to bind into an expression from independent parame-
ter lists, i.e., argument structure and event structure. Given a listing
of arguments and an event structure represented as a listing of event
variables,

we can view the semantics of the verb as being centrally defined by the
qualia, but constrained by type information from the two parameter
lists. The predicates in the qualia refer directly to the parameters:

[QUALIA = [ ... [Qi = PRED(EVENT

j ,ARGk)] ...]

I will return to the specifics of qualia structure in the next section. First,
let us turn to the exact nature of the event argument list illustrated
above, and what motivation exists for its structure.

Proceeding from our earlier discussion in chapter 2, I assume that
events can be subclassified into at least three sorts: PROCESSES

, STATES,
and TRANSITIONs.6 F\1rthermore, I assume a subeventual structure to
these event sorts as well. This has the advantage of allowing principles
of predicate-argument binding to refer to subevents in the semantic rep-
resentation, a move which has significant theoretical consequences (cf.
Grimshaw, 1990, and Pustejovsky, 1991b). As shown in Pustejovsky
and Busa (1995), however, evidence from unaccusativity and the varied
nature of causative constructions shows that this notion of event struc-
ture does not fully capture the underlying semantics of unaccusative
constructions, with respect to how the subevents project to syntax.

This definition states that the event e3 is a complex event structure
constituted of two subevents, el and e2, where el and e2 are temporally
ordered such that the first precedes the second, each is a logical part of
e3, and there is no other event that is part of e3. Verbs included in this
description are causatives as well as inchoatives, as argued in chapter 9.
An event tree structure is not restricted to representing strictly se-

quential relations between subevents, but structures other orderings as
well. For example, an event composed of two completely simultaneous
subevents can be lexicalized with a relation we will call "exhaustive over-

lap part of," cx::. It can be defined as follows (cf. Kamp, 1979, and Allen,
1984):

69

The Semantic Type System

Within an event semantics defined not only by sorts but also by the
internal configurational properties of the event, we need to represent
the relation between an event and its proper subevents. Extending the
constructions introduced in van Benthem (1983) and Kamp (1979), we
interpret an "extended event structure" as a tuple, < E, j , <, , ~ ,* >,
where E is the set of events, :S is a partial order of part-of, < is a strict
partial order, 0 is overlap, ~ is inclusion, and * designates the "head"
of an event, to be defined below. An event structure with structured
subevents such as that in Pustejovsky (1988, 1991), shown in (16) below,

(16)

e < cx:

el

e2

(18) a. [e3 el cx: e2] = def cx:({el ? e2} , e3)

can be constructed in such a model as follows. Let us define an event
tree structure in terms of the relation of "exhaustive ordered part of,"
< .7
cx.

(17) a. [e3 el b. Vel,e2,e3[ Ve [e ~ e 3 -t e == e 1 V e = e2]]

70

Chapter 5

b. Vel, e2, e3[ocx:({el, e2}, e3) ~ el ~ e3 .I\e2 ~ e31\el ~ e21\e2 ~
ell \ 3e[e ~ ell \e ~ e21\e = e3] /\\ie[e :S e3 -1- e = el Ve = e2]]

This event structure is denoted by verbs such as accompany, which in-
volve two subevents occurring simultaneously. This can be illustrated
in terms of an event structure tree as follows:

(19)

eooc

el

e2

The interesting

thing

about

such a verb is that , because

it makes

refer -

ence to an implicit

event , it is aspectually

underspecified

, and assumes

both

telic and atelic

interpretations

, depending

on the context .

( 20 ) a . John will accompany

you to the store . ( telic )

b . Mary

accompanied

me while

I was walking . ( atelic )

In this sense , it is aspectually

similar

to the verb go , which

also admits

of multiple

event

interpretations

.

Finally , the lexicalization

of two basically

simultaneous

subevents ,

where

one starts

before

the other , called

" exhaustive

ordered

overlap ,"

< cx, can be defined

as follows , where

init

is a function

over events ,

returning

the initial

part of that

event , and end is a function

returning

the final part of the event :

( 21 ) a . [e3 el < cx: e2 ] = def < cx: ( { el , e2 } , e3 )

b . Vel , e2 , e3 [< cx: ( { el , e2 } , e3 ) f--+ el =5 e3 1\ e2 =5 e3 /\ el 0 e2 1\

init ( el ) < init ( e2 ) 1\ end (el ) == end ( e2 ) /\ Ve [e ::s' e3 --+ e ==

el V e = e2 ]]

The relation

< o ()( above

defines

an event

containing

two subevents , el

and e2 , where

el starts

before

e2 . The event

structure

tree is shown

below :

[El
=

Ti ]

E2
=

Ti
RESTR
=

cx
:

The Semantic Type System

71

(22)

e < ooc

el

e2

(25)

We shall see that because of this partial ordered relation , a type of

causative relation

exists between the subevents , but not in the same way

as with the relation

this subeventual

structure , where two motion processes are structured

in

an overlapping relation ; that is, the efficient motion of the legs bringing
about the final motion of the body . I will argue that this relation is

also present with control interpretations for aspectual predicates such
as begin, with the associated entailments regarding causation .

Given what I have said so far , there are two facets of an event tree

structure that need to be represented for a lexical structure : the specific
events and their types ; and the ordering restriction over these events.

This is illustrated schematically in (23).

a

(23

) EVENTSTR
= [ ~~

:: ::: ]

RESTR = . . .

For example , the verb build is typically analyzed as involving a develop-

ment process and a resulting state (cf. Dowty, 1979, Moens and Steed-
man, 1988, Pustejovsky, 1991b), ordered by the relation "exhaustive

ordered part of ," < cx:.

build

[ El

=

process
]

(24) EVENTSTR

= E2 = state

RESTR = < cx:

. . .

Unlike build , however , which constrains the types of its two subevents

to PROCESS and STATE, the verb accompany permits either telic events ,

TRANSITIONS , or PROCESSES. The typing constraint , however , is similar

to that for a coordinate structure in that they must be of like type :

accompany

EVENTSTR

=

Of course, there are many more possible relations between subevents

than are actually realized in lexicalized forms in natural languages;8
one of the principal goals of a semantic theory must be to constrain the
model to reflect these restrictions .

The structural information discussed thus far for event structure , al-
though necessary, is not sufficient to capture lexical distinctions that

languages systematically make with respect to the relative prominence

or importance of the subevents of a larger event. Talmy (1975,1976) and
others have long noted that the event information conveyed by a verb
can be much richer than the "sequence of events" structure encoded in

the representations above. These grammatical observations , however,
can be accounted for in terms of something I will call event headedness

(cf. Pustejovsky , 1988). Event headedness provides a way of indicating a
type of foregrounding and backgrounding of event arguments . An event

structure provides a configuration where events are not only ordered by
temporal precedence, but also by relative prominence . One instance of

prominence for an event, e, is provided by the HEAD marker , annotated
as e*. The conventional role of a head in a syntactic representation is to
indicate prominence and distinction . Rules of agreement, government ,
etc. militate in favor of marking structures in terms of heads of phrases.
Within the interpretive domain of events, when viewed in a structural

or configurational manner , the possibility of referring to heads becomes
available . Informally , the head is defined as the most prominent subevent
in the event structure of a predicate , which contributes to the "focus" of

the interpretation . We can view * as a relation between events, *(ei, ej )
("ei is a head of ej" ), where ei .-:::S ej :

72

Chapter 5

(26) *(ei, ej) ==def [e. ... e; ...]

.1

Headedness is a property of all event sorts, but acts to distinguish the
set of transitions, specifying what part of the matrix event is being fo-
cused by the lexical item in question. Adding the property of headedness
to the event structure gives the following representation:

a

EVENTSTR

=

.

.

El
=
...
E2
=
...
RESTR
=
...
HEAD
=
Ei

(27)

73

The Semantic Type System

Assuming that events have at most a binary event structure , and that

there are three temporal ordering relations realized in language 0(,
CO(, and < CO() there are six possible head configurations with two

events, given a single head; there are twelve possibilities , if unheaded

and double -headed constructions

are included . These are listed below

along with an example of each type , where HEAD is indicated by an
asterisk , and the event tree structure is given in a linear representation :

(28) a. [eO' el* b. [eO' el c. [eO' el* d. [eO' el e. [eO' el* 0(:;(: e2] - buy
f. [eO' el cx: e2*] - sell
g. [eO' el* 0(:;(: e2*] - marry

h. [eO' el 0(:;(: e2] - UNDERSPECIFIED

i. [eO' el* < j. [eO' el < k. [eCT

el* < (x e2*] - ??

1. [eCT

el < (x e2] - UNDERSPECIFIED

Intuitively , structure (28a) represents accomplishment verbs, where the
initial event is headed, focusing the action bringing about a state; (28b)

represents achievement verbs, for which the persistence of the final state

is the focus of interpretation; (28c) illustrates events involving a rela-

tional predicate on each subevent , and characterizes unilateral transi -

tions with three arguments , i.e., a subclass of ditransitive transfer verbs

such as give and take. Predicate pairs such as buy and sell are character -

ized by (28e) and (28f) respectively, where there are two simultaneous

events involved in the transaction , but only one is focused by the lexical

item .

Finally , ordered overlap gives rise to the structure in (28i), where
one event begins, and subsequently gives rise to another process which
continues only while the first event continues to hold .

Notice that the structures in (28d), (28h), and (281) are unheaded

and hence ill -formed without further specification at surface structure

(cf. chapter 9). The role of semantic underspecification will figure promi -
nently in the analysis of verbal polysemy . In terms of event structure ,
polysemy occurs when a lexical expression is unspecified with respect

to headedness, i.e., headless. Headless event structures admit of two

possi ble interpretations. More generally, a predicate should be as many
ways ambiguous as there are potential heads. This representation pro-
vides us with a mechanism for relating the logical senses of polymorphic
verbs such as: causative/ unaccusative predicates such as break and sink,
associated with (28d); argument inversion predicates such as rent, asso-
ciated with (28h); and raising/ control predicates such as begin and stop,
associated with (281).9
There are several motivations for positing a head as part of an event
structure, a matter that is discussed in Pustejovsky (1,988) as well as
in Pustejovsky and Busa (1995). When adjoined to predicates denoting
transitions, prepositional and adverbial phrases not only can modify the
entire event, but can also take scope over the individual subevents. In
particular , heads seem to license certain types of modification. Observe
that the durative adverbials in (29) modify the designated head of the
event rather than the entire event structure.

John ran home for an hour .

My terminal died for two days.
Mary left town for two week .

74

Chapter 5

(29) a.
b.

c .

The event tree structure is illustrated in (30) below, where, until the for-
mal structure of qualia is presented in the next section, I refer informally
to the expressions associated with each event in a tree structure.

(30)

It is generally assumed that only processes and states license durative
adverbials, yet modification by an adverbial in (29) is grammatical, even
though the sentences denote telic events. Obviously, the available inter-
pretation has the adverbial modifying the duration of the final state; in
(29a) John spent an hour at home, in (2gb) the terminal was dead for
two days, and in (29c) Mary was out of town for a period of two weeks.

The Semantic Type System

75

A similar

phenomenon

occurs with left -headed events (e.g ., TRANSI -

TIONS ) , when modified

by certain manner adverbs such as carelessly and

quietly ; namely , modification

is over the initial

(headed ) subevent .

(31 ) a. John built the house carelessly .

b . Mary quietly drew a picture .

In (31a ) carelessly modifies the act of building

which brought the house

into existence ; similar remarks hold for the sentence in (31 b ) .10 In the

event tree structure

below , I assume a predicative

approach to the rela -

tion between 'the object being drawn in the process e1 and the resulting

picture

in e2 , similar to Burge 's (1972 ) treatment

of mass terms . The

constitutive

relation , caNST , defined below as an integral part of qualia

structure , gives the relationship

between these two variables .

(32 )

e< cx

' ~'/ / / / " ' ~" -" " " " - -
ei

e2

I

I

3y [draw (m , y )]

3x [picture (x ) A Const (y , x )]

Evidence

from event modification

is only one of several arguments

in fa -

vor of making reference to a focusing mechanism

within an event struc -

ture (but cf . Pustejovsky

( 1988 ) for more discussion ) .

This complete

our initial

discussion

of how lexical items make ref -

erence to events . In later chapters we will see how this view of event

structure

is expressive

enough to capture the polymorphic

behavior

of

verbal semantics

while also being sufficiently

constrained

to not over -

generate semantic

expressions .

76

Chapter 5

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->