.

L

 

Mk‘

‘Q

N l5

»
é
o
.

,
' o
.
In

1'

"A

 

,‘.,_,.

 

INS-"5‘

"'1,"’r ('l v‘

 

 

 

 

1. H
r
.,|

rurr
' n
I:
ml.

 

 

 

 

 

 

 
 

- .
.lcwwrfiflott. Ir \: I}, L
cl 5!!! In. "I!!!
. Of? ' I
0.! 7) left. I. LQ-“fl VVIH. .lOOA'C'!‘ ll
.0; . .lvrj
U A .15 l. o '00
it; .I I
I:
.......»um3i..l:z:.!. .lsal.lxuwu|mrflfl

22941707

mallllg‘lllglllWill

mefiv

Michigan State
University

.1
R S

 

 

This is a, certify that the
thesis entitled
A SEMANTIC EXPLORATION
0F LEXICAL DOUBLETS

presented by
David Joseph Kathman

has been accepted towards fulfillment
of the requirements for

éajor professor

Date—MEL.—

 

 

MSU

 

 

RETURNING MATERIALS:
Place in book drop to remove this
checkout from your record. FWES

 

UBRAR'ES will be charged if book is returned
.g_—_2 —- after the date stamped below.
if“; :3 q 1‘0"?
J r;

 

 

 

A SEMANTIC EXPLORATION OF LEXICAL DOUBLETS

by

David Joseph Kathman

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF ARTS

Department of Linguistics and Germanic, Slavic,
Asian and African Languages

1 989

CPQ
5-(\

CED/7'53}

ABSTRACT
A SEMANTIC EXPLORATION OF LEXICAL DOUBLETS
By

David Joseph Kathman

This thesis examines the behavior of lexical doublets in English,
defined as pairs of verbs which refer to the same action or situation,
but differ in the syntactic position of two or more nominal arguments.
The focus is on three-place doublets, i.e. those having three nominal
arguments; oft-cited examples are 'sell'l‘buy' and 'give'l‘receive'. After
a survey of previous work on lexical doublets, a stratified model of
language is presented, including a creative semantic level in which .
similarly structured conceptual domains are seen as belonging to
parallel semantic "fields”. There follows a survey of English
three-place verbs (those having both indirect and direct objects), and it
is suggested that all such verbs capable of forming doublets have a
conceptual ”path” present, either of whose endpoints can be realized
syntactically as the subject (subject to restrictions according to the

semantic field involved).

Copyright by
DAVID JOSEPH KATHMAN
1 989

3.1
3.2
3.3
3.4

4.1
4.2

5.1
5.1.1
5.1.2
5.2
5.2.1
5.2.2
5.2.3
5.2.4
5.2.5
5.2.6
5.2.7

TABLE OF CONTENTS

List of Figures
Overview and Goals
Background

Assumptions

The model in general

Nodes

Strata

A few words about processing

The semantic stratum
The nature of semantic structure
Semantic fields

Three-place verbs and lexical doublets
Defining terms

Lexical doublets

Three-place verbs

Types of three-place verbs

”Sell” type

'Throw"-type

"Bring'-type

Abstract transferee: 'Teach"-type
Other abstract transferees
Non-doublet—forming three-place verbs
Other uses of "give"

iv

13
13
15
17
20

21
21
27

3O
30
3O
33
34
35
44
48
50
54
60
64

TABLE OF CONTENTS (continued)

Summary and Conclusions
Appendix: Diagrams of some conceptual traces

Bibliography

69

72

76

UST OF FIGURES

Figure 1. Fillmore's diagrams for 'buy' and 'sell'

Figure 2. A stratificational model of language

Figure 3. Stratificational node types

Figure 4. Part of a conceptual tactic diagram

Figure 5. Conceptual traces of two sentences

Figure 6. Semological traces of two sentences

Figure 7. Conceptual representation for 'x sells y to 2'
Figure A1 . 'John threw Mary a cantaloupe'

Figure A2. 'Pete told me a joke about phoneticians'
Figure A3. 'Being on trial taught Ollie a valuable lesson'

Figure A4. 'Bill gave Leopold a black eye'

vi

14

15

26

27

32

41

72

73

74

75

W

There are a number of interesting questions which have come up
fairly often in linguistic discussions over the years, but have defied
simple analysis. One such problem is the analysis of what Dowty (ms)
calls 'lexical doublets' - pairs of verbs which can be used in
sentences describing the same situation, but with the participants in
different syntactic positions. One type of doublet is illustrated by the

verbs in the following pairs of sentences:

lla-chansnldannstrichnuaam.

In both of these cases, the first sentence entails the second; if
Roman sells an ostrich to Noam, it must also be true that Noam buys an
ostrich from Roman. In addition, the participants involved are the
same in each pair; the difference is in the syntactic positions occupied
by these participants. lntuitively, each pair of sentences describes
the same event, with either of two arguments capable of being
realized as the subject.

Not all similar states of affairs can be realized in more than one
way, however. For instance, one might be tempted to say that ’throw'
forms a doublet with 'catch', but on further reflection this is not quite
so. For one thing, it is awkward to use 'catch' with three arguments

(cf. ?Bnmanmucht.thaarmadillqimm_unam. although some authors.
such as Gruber (1976), apparently find no difficulty with this

construction). N80. Neamfluenthaarmadilmacman does not

entail WW; Roman might have dropped it.

Furthermore, there are some three-place verbs (i.e. verbs which take
an indirect as well as a direct object) which do not even form
pseudo-doublets similar to 'throw'l'catch‘; there do not seem to be any
partners for such syntactically analogous verbs as 'owe', 'deny', and
'promise'.

This thesis will explore these verbs in the context of a general
model of language with an emphasis on their behavior in lexical
doublets. More specifically, it will address the question implicit in
the above paragraph: Why is it that some three-place verbs (e.g. 'sell')
appear in doublets, others (e.g. 'throw') appear in ”pseudo-doublets',
and others (e.g. 'owe') do not appear to be paired with any similar verb
at all? Another way to look at this question is in terms of "subject
selection“ principles, such as those found in Fillmore (1968), which
determine which of a verb's arguments will be lexicalized as the
subject of the sentence. In which situations (e.g. a commercial
transaction) can either of two participants be realized as the subject,
and in which similar situations (e.g. person A promising something to
person 8) can only one participant be so realized? In order to answer
these questions, recourse to semantics appears to be inevitable, since
these verbs' syntactic behavior is identical or nearly so. Thus part of 1
this thesis will be devoted to outlining a plausible semantic theory,
followed by application of this theory to three-place verbs and
doublets involving them in order to better understand such
constructions. The main focus will be on describing the situations

where doublets do and do not occur, since few previous analysis have

considered the full range of three-place doublets in any detail. Some
partial explanations will also be offered, although many questions will
have to be left unanswered for now.

A caveat may be in order here. Obviously the sentences W
an Wu and Bomanhouohtanostrichfiomfloam are not
freely interchangeable; they focus on different participants, and they
would usually be used under different circumstances. Lexical doublets
need not be pragmatically and stylistically the same; in fact, they
generally are not, since otherwise they would be redundant. What is
important is that they refer to the same event or state of affairs; we
are interested only in the fact that two separate, syntactically
different verbs exist in English to describe this event or state. The
pragmatic differences behlveen 'buy' and 'sell' are certainly worthy of

study, but these differences are beyond the scope of this thesis.

Lflackomund
Before we start any analysis of lexical doublets, it will be a good

idea to see what other linguists have had to say about them. Pairs
such as 'buy'/'sell' have been mentioned over the years in discussions
of semantics, and a number of different approaches have emerged. In
some cases, there are merely observations, with little or no attempt

at formal analysis, but numerous authors have tried more structured
approaches. The following account is only a sample of some
representative analyses, since an exhaustive catalog of all the
different attempts to deal with lexical doublets would be very large
indeed.

The early days of transformational grammar saw several attempts
to relate doublets by means of transformations or information in the
lexicon. These attempts generated some controversy at the time,
although in retrospect they seem to be quaint artifacts of the heady,
transformation- happy days of the 1960's. Lyons (1963) refers to
doublets as “converse terms" and states: ”The lexical substitution of
'buy' for 'sell', etc., can be thought of as associated with a set of
automatic syntactic transformational rules, which carry the sentence
containing 'buy' over into the corresponding sentence containing 'sell'”
(1963:72). Steel (1967) goes on to suggest that such relations be
incorporated into the lexicon. He proposes that the lexical entry for
'sell' should contain the following information:

(39”, [+V, +_NP {Mj-toNP {M, buy-from H)
where M indicates the possibility of passivization. This information
would indicate that the sentences WW and Mary
W are paraphrases. Bar-Hillel (1967) accepts

some of Staal's conclusions but claims that the relations would be
better handled by means of “meaning rules“ of the type “For all X, Y,
and Z, X sells Y to Z <-> Z buys Y from X" (based on Camap 1956).
Several problems with the above analyses come to mind. First of
all, treating the relationship between 'buy' and 'sell' as syntactic, as
Lyons and Staal appear to do, stretches the definition of syntax to the
breaking point (although such overzealous use of transformations was
far from uncommon in those days). Katz (1972:296n) makes
essentially the same objection, and Gruber (1976:42-3) points out a

number of problems with an analysis in terms of transformations. All

three of the above analyses, furthermore, seem to invite redundancy; it
would be better if we could come up with a general rule involving
some kind of semantic feature of the verbs rather than having to
provide a separate meaning rule for each related pair. Such
arbitrariness makes it nearly impossible to answer the question posed
by this thesis: what kinds of verbs, in general, can be related in this
way?

In many instances, there has been an attempt to analyze doublets in
terms of "cases” or “thematic roles“, as papularized by Fillmore (1968)
and elaborated in such works as Anderson (1971). Fillmore's
widely-read article addresses the issue of subject-selection
principles, but chiefly in terms of sentences containing the same verb
accompanied by different sets of roles (e.g. Wm
vs. WWW. His discussion of the verb
'give' in terms of these principles focuses on the various surface
configurations involving this verb W

W. We
mm, but unfortunately for our purposes here, he does not go

on to explore the relationship between pairs like 'give' and 'receive' in
terms of case grammar.

Subsequent work in many varieties of case grammar did sometimes
try to analyze doublets; as a result of this and similar developments,
many case grammarians began assigning multiple cases (or “roles') to
single arguments in an effort to capture generalizations. Longacre
(1976), for instance, presents a system whereby 'sell' and 'buy' both
have case-frames involving a Source, Goal, and Patient, the difference

being that 'sell' assigns an additional Agent case to the Source, while
with 'buy', the Goal is the Agent. Anderson (1971) proposes a similar
system, though his terminology is totally different (e.g. 'ergative',
'ablative', and 'locative' for Agent, Source, and Goal).

Recognition of limits on the assignment of multiple roles led some
subsequent researchers to posit disjunct "levels“ or ”systems” of
cases. For example, Van Valin and Foley (1980) have two different
case systems in their model: ”derived“ (corresponding to traditional
roles such as agent, patient, and instrument) and ”basic“ (consisting of
an opposition between just two roles: Actor and Undergoer). Thus in
analogous pairs of sentences containing 'sell' and 'buy' (Jnhmldjhn
Insomniac/WW. the participants would
have identical roles in the derived system, but would differ in the
basic system as to which participant is the Actor. This does help
account for the intuitive semantic similarity of such sentences, but it
does not address the question of why some verbs and not others can be
paired in such a way. Various other authors, such as Dik (1980) and
Culicover and Wilkins (1986) have made similar proposals for
multi-tiered role systems.

Katz (1972) devotes an entire chapter to lexical doublets (which he
calls 'converses"), and he develops a very complicated formalism for
handling the time relations involved. (Katz's chapter is an expansion
of some ideas sketched in Katz 1967.) Essentially, Katz sees a verb
such as ’sell' as referring to a series of states, each connected to an
index representing a point in time; no thematic relations as such are

involved. Since he assumes an interpretive view of semantics, he sees

the problem as: how do we make sure that the semantic
interpretations of, say, 'John sold the book to Mary“ and “Mary bought
the book from John“ are the same, at least in terms of truth
conditions? His solution involves variables in the semantic
interpretation of the verb and conditions for the replacement of those
variables with semantic interpretations of the various noun phrases in
the sentence. In greatly simplified terms, part of the semantic
interpretation of 'sell' would be 'X possesses Y at time i; 2 possesses Y
at time j", where j is later than i, and part of the interpretation of

'buy' would be '2 possesses Y at time i; X possesses Y at time i". The
variables X, Y, and 2 above stand respectively for the subject, direct
object, and indirect object of the sentence's deep structure; the
semantic readings of the appropriate noun phrases are then
substituted in, with the resulting semantic interpretations being
identical for the 'buy' and 'sell' sentences. This analysis does have its
attractive points, such as its attempt to decompose the meanings of
the verbs in order to account for regularities, but on the other hand it
is formidable in its complexity.

In contrast to Katz's extreme attention to detail, Fillmore (1977)
presents an informal but intuitively appealing proposal involving the
“scenes” associated with utterances. He points out that many verbs
presuppose things about the context in which the action occurs; for
example, the verb ‘write' implies the existence of a trace-leaving
implement going over a surface, and the result of the activity is
presumed to be linguistic in nature. He then goes on to present an
Informal notation system to show the relationship between the syntax

 

._ euv —-fii SELL——
o A
FROM TO
0 0
Foo Foe

 

 

 

 

 

 

of an utterance and its scene. As an example he uses a commercial
transaction, which can be described using the old standbys 'buy' and
'sell' as well as a variety of other verbs such as 'spend', 'charge‘, and
'cost'. Figure 1 illustrates some verbs describing such a transaction.
Here A, B, C, and D stand respectively for the buyer, the seller, the
money, and the thing being sold; the numbers 1, 2, and 3 stand
respectively for the subject, object, and indirect object of the
sentence, with appropriate propositions indicating non-central
participants.

Jackendoff (1972) also briefly considers lexical doublets (using the
standard example 'buy'/‘sell'), and his discussion there is a prelude to
his later, more fully developed ideas. He starts out by considering an
analysis in terms of thematic roles similar to Longacre's, but rejects

it in favor of a partial decomposition of the meanings of the verbs into
the functions CAUSE and CHANGE, with a system of indexing similar to
Katz's variables relating the semantic and syntactic structures. In

such a system, the notion of Agent or Source is not primitive, but is
defined as a relation in semantic structure. In later writings,

Jackendoff further refines the idea of semantic decomposition along
lines developed earlier by such authors as Gruber (1976) and Miller and
Johnson-Laird (1976). Jackendoff (1983 and 1987a) presents a system
in which many abstract areas of the conceptual system have an

internal organization parallel to that of more concrete areas; for
instance, notions of relations in time (“It happened at 6:00“; “We

moved the meeting from Tuesday to Wednesday“) have a semantic
structure parallel to that for spatial relations (“It happened at John's
house“; “We moved the table from the living room to the dining room“).
This concept will be explored in more detail later, but as an

illustration, the semantic structure for 'buy' would be roughly the
following in Jackendoffs revised system:

[EMTCAUSE ([Thlngx]! Em GOP“ ([ThInQYIr FROMPogquhlngzD ) )1

Path TOPoee([Thlngx])

Path ToPou‘lThInqzl)
(The symbol '$' means that the element in question must be a sum of
money, and the variables x, y, and 2 stand for entities in the mental
world of the language user.) The idea of semantic fields, in the sense
used by Jackendoff, will play an important part in the analysis
presented below. Furthermore, Jackendoff discusses possible

1O

constraints on how semantic representations may be lexicalized;
although he does not specifically address the issue of lexical doublets
as defined above, his comments will provide fertile ground for
discussion.

There is, however, one further proposal involving lexical doublets
which we must consider: the very interesting one given in Dowty (ms).
Dowty's discussion of doublets is part of a larger argument involving
thematic roles; he complains that current work involving roles is often
idiosyncratic and inexact, and he makes some radical suggestions for
remedying the situation. Essentially, what Dowty proposes is that
instead of traditional roles such as Agent, Patient, Recipient, etc.,
there are just two proto-roles, (Proto-) Agent and (Proto-) Patient.
These are defined in terms of sets of prototypical characteristics
rather than necessary and sufficient conditions, so that an argument
can be Agent-like or Patient-like to various degrees; whichever of a
verb's arguments is the most Agent-like will be realized as the
grammatical subject, and whichever is the most Patient-like will be
realized as the direct object. The Proto-Agent characteristics given by
Dowty (ms) are:

a. volition

b. sentience (and/or perception)

0. causes event

d. movement
and his Proto-Patient characteristics are:

a. change of state (including coming to and going out of being) .

b. incremental theme (i.e. determinant of aspect)

11

c. causally affected by agent

d. stationary (relative to movement of Proto-Agent)

It is possible for more than one argument to have properties from one
of the above lists; but according to Dowty's theory, whichever argument
has the most Proto-Agent properties entailed by the meaning of the
predicate will be lexicalized as the subject, and whichever has the

most Proto-Patient properties entailed will be the direct object.

What happens when two arguments have an equal number of Proto-
Agent characteristics? It is in such a case, Dowty claims, that we get
lexical doublets, since either argument can be realized equally well as
the subject. As an example, consider the following pair of sentences,
which illustrate a three-place doublet:

3) a. W.

b. WWW!!-
In these sentences, John is entailed to be sentient (Proto-Agent
property b) and to act volitionally (property a) and causally (property
0); however, Ethelred appears to have the same properties, since a
commercial transaction such as this presumably requires the mutual
consent of the two parties. Since there seem to be no other relevant
entailments involved, and each argument has an equal number (three) of
Prom-Agent properties, either argument can be realized as the subject.
Two-place doublets such as 'like' and 'please' can be analyzed in a
similar way, according to Dowty. In a pair such as mum
moons/WW, 'John' is presumably sentient, while
the bon-bons are causing a reaction in John; since these are both
Proto-Agent properties, either argument can occur as the subject.

12

On closer inspection, however, the above explanation of three-place
doublets runs into some problems. While this theory seems to work for
'buy' and 'sell', it fails to account for similar doublets. Consider the
following sentences, whose verbs form a doublet:

4) a. W

b. WW
Ron, In these sentences, appears to have the three Proto-Agent
properties mentioned above: volition, sentience, and causation.
Mikhail, on the other hand, does not need to be either volitional or
causative; he can be an entirely passive participant. Thus it appears
that either Ron or Mikhail can be lexicalized as the subject, even though
the former has three of Dowty's Proto-Agent characteristics while the
latter has only one. Dowty recognizes this as a problem and tentatively
suggests some explanations in terms of these verbs being few in
number and acquired later by children than other verbs. Admittedly
there might be some way to revise the theory to better account for this
type of case, and overall Dowty's proposal is intriguing, but the
situation is actually more complicated than it seems from Dowty's
examples. This thesis will thoroughly explore the various types of
three-place verbs and doublets in semantic terms, and will attempt to
at least lay the groundwork for a more adequate explanation of the
existence of lexical doublets.

13

Mammalian:
3J_Ihe.rnodal.in.oanaral

The linguistic model used in this thesis Is a form of stratificational
grammar (SG), although it differs in some respects from many of the
models which have borne that name in the past. The term
“stratificational', in the broad sense, means “having more than one
creative, autonomous level of structure.“ By this definition, both
tagmemics and systemics are broadly stratificational, since they
explicitly recognize independent principles of organization for such
levels as grammar and phonology (Pike & Pike 1976, Halliday 1985). On
the other hand, neither classical transformational grammar nor
government and binding theory are stratificational, since they treat
syntax as the only creative component of a grammar, with semantics
and phonology being interpretive components (Chomsky 1965 and 1981 ).
In recent years, however, some generativists have abandoned
syntactocentrism in favor of a more Stratificational approach. For
example, Jackendoff (1983: 9) presents a bistratal view in which
semantics is creative and independent of syntax, but phonology is still
interpretive; in Jackendoff (1987a) he claims the autonomy of
phonology as well, resulting in a tri-stratal system. McCawley (1988)
includes in his version of generative grammar “gross surface
combinatorics“, which are essentially tactics specifying the surface
order of constituents, independent of the unordered underlying
structures he also includes in his model.

Figure 2 is a broad sketch of the model of language used in this
thesis. There are four strata, each of which contains a tactic pattern

14

gnostotactics

t

conceptual semotactics
structures {

 

 

 

 

 

 

 

 

sememic lexotactics
structures
\ t _
lexemic .
| Structures] phonotactIcs
\ L
phonemic
structures
Eigllm 2 e 51 lot. I.

specifying the well-formed structures of that level; these are

connected by alternation patterns (represented here by double-headed
arrows) which specify the correspondences between adjacent strata
(Arguments for independent patterning at various levels of language
can be found in Lockwood 1972: chapter 2, Jackendoff 1987b: chapter 5,
and McCawley 1988: chapter 10). Both the tactic patterns and the
alternation patterns are networks of relationships, usually represented
graphically by means of lines and nodes; the points where the tactic
patterns and the alternation patterns intersect are known as 'emes',
since they correspond roughly to traditional morphemes, phonemes,
etc. For the sake of clarity, some explanation of the system and its
formalism is given below, though this formalism will generally not be

a crucial part of the analysis which follows.

15

 

 

A A
B C B C B C
a. ordered and b. simultaneous and c. associative and
A A
B’ C B C
d. unordered or e. ordered or
3.2.Nodas

Graphic representations of the major types of nodes are shown in
figure 3. Four types of nodes have traditionally been recognized in SG
- ordered and. simultaneous and, ordered gr, and unordered gr. (See,
for instance, Lamb 1966 or Schreyer 1980). Other nodes have been
proposed at various times by various people, and some of these may be
necessary. (Christie 1977 presents an especially long list of nodes,
some of them quite esoteric.) The present work will introduce just one
node beyond the basic four -- an associative and, represented by a black
triangle (Lockwood ms).

The three triangular nodes (3a-c) are syntagmatic nodes, meaning
that they specify how two or more elements are related to one another
in combination. With the ordered and, this relationship is linear order;

16

Figure 3a says that B precedes C. Figure 3b, a simultaneous and,
specifies that B and C occur simultaneously. This would be used, for
example, to specify the relationship among the members of a bundle of
phonetic features; in this thesis, it will be used mainly for the
relationship between a part of a tactic pattern and a label. The
associative and of Figure 3c says that B and C are not simultaneous in
terms of processing, but there does not seem to be any justification for
ordering them. This node is used mainly in the semotactics and
gnostotactics, where linear order does not come into play but
dominance relations must be specified.

The next two nodes are ”paradigmatic nodes, representing a choice
among two or more alternatives. Figure 3d, an unordered m, specifies
that either 8 or C, butnot both, occurs. The ordered gr, shown in Figure
So, says that B occurs only if its line has been “activated“ from
elsewhere in the linguistic system, and otherwise C occurs. This node
is used to handle things such as gender concord, but it will not be
necessary for our purposes here. _

As noted before, each stratum has a tactic pattern consisting of a
hierarchically arranged network of these nodes. At the top of each
tactic pattern is an initial symbol, usually represented by a small
circle, and immediately below this is a node which dominates all the
other nodes in the network. At the bottom of the tactic pattern are the
terminal nodes or “emes“ of that level - for the lexotactics they are
called lexemes, for the semotactics sememes, and so on. Every node
between the initial symbol and the terminal nodes must have a singular

side, which has one line coming from it, and a plural side, which has

17

two or more lines coming from it. If a node's plural side is oriented
toward the initial symbol, it is an upward node; if the plural side

points toward the terminal nodes, it is a downward node. Since tactic
diagrams are generally drawn with the initial symbol at the top,
downward nodes look like those in Figure 3; upward nodes are the same
but upside down. It should be noted that it is possible in SG for a line

to loop back up to a higher part of the diagram, usually to an upward 91;

this is how recursion is handled.

W
The number and functions of specific strata have varied in

stratificational analyses of language, so some explanation of the
system used here may be helpful. The topmost stratum (i.e., the one
closest to meaning) is the conceptual (or gnostemic), which specifies
how a potential utterance is organized in terms of some basic
conceptual principles. Many of these principles are taken from the
ideas of Ray Jackendoff (1983, 1987a, 1987b), though they have been
adapted to stratificational notation so they will fit in with the rest of

the system. Essentially, this level deals with the semantic relations
holding between the various elements of a sentence or discourse, and
the internal semantic structure of those elements. Such matters have
received scant treatment within the stratificational paradigm

compared to the attention given to more accessible areas of language
such as phonology and morphology, but I believe the analyses presented
here have some merit; I do not share the view of some linguists (e.g.
Sampson 1980:157) that semantics must be anecdotal and vague.

18

The next stratum down from the conceptual is the sememic. A level
of this name is included in most SG analyses, and it is usually
described as part of semantics (e.g. by Lamb 196620 or Sullivan
1980:305). However, the sememic level proposed here is more properly
part of syntax (as it is understood by most linguists); among other
things, it specifies the number and type of arguments a given verb
takes, as well as how different syntactic categories may be associated.
For example, the fact that the verb 'put' requires two nominal
arguments and a propositional phrase argument would be specified in
the semology; so would the fact that prOper names in English do not
occur with determiners except in very limited cases. One thing which
is not specified in semology is linear order; in this way it differs from
most versions of deep structure in classical transformational grammar.

The lexemic stratum is what determines the linear order of words
and constituents; it also indicates the presence of semantically empty
words such as j: in W. The lexotactics simply specifies the
relative order (and, where relevant, hierarchical structure) of all the
possible surface elements of a sentence; which of these elements occur
in a given sentence has to be determined by the semology. In the terms
used by McCawley (1980:174), semology more or less includes
“semantic combinatorics“ (i.e. the valence of verbs, etc.) while lexology
includes “syntactic combinatorics“ (i.e. linear order).

Of the level(s) below the lexology, little needs to be said here, since
they are not directly relevant to the problem of lexical doublets. Most
versions of SG have a morphemic stratum as well as a phonemic, but

the diagram in Figure 1 shows only the latter. It seems that most, if

19

not all, of the phenomena generally considered morphological can be
included as an extension of the lexology: Lamb (1983) explicitly argues
for this position, and Jackendoff (1987a:372-3) makes a similar claim.

I will leave the question open here, since it does not affect the

analysis one way or another.

A few additional remarks about this model might be in order.
Practitioners of 86 have generally stressed that their system consists
solely of relationships; such terms as “lexeme' and “phoneme“ are
merely convenient labels for points in a network, defined by their place
in the network and having no existence as objects in their own right.
This view will be assumed here, but with one stipulation: if the model
is to be truly stratificational as well as capable of encoding and
decoding, then the emes of each stratum must be fundamentally
different from the other nodes in the system. Otherwise there will be
no way to build structure on any single level, and if this is not possible
there is no reason to have separate levels at all. (Indeed, Reich 1969,
1970 has presented a model based on relationships but lacking strata.)
However, since different levels of stmcture appear to be desirable, I
will assume that emes are not quite the same as the nodes described in
section 2b above. One way of looking at them is as special nodes
capable of “remembering“ a signal during processing, so that a
structure on a given level can be “built“. Actually, terminology does not
matter as long as we recognize that emes are fundamentally different
from and and n: nodes.

20

WW

Essentially the same inventory of nodes described above appears in
the alternation patterns which link adjacent strata. (In actual practice
there are probably some differences, but we need not worry about that
in this brief survey.) The emes of each stratum are connected by these
patterns to the emes of the strata above and below, mapping structures
onto one another. In a fully specified stratificational system, each
possible structure on a given stratum will be mapped onto one or more
structures on each adjacent stratum; this system can thus be seen as a
model of ideal language use ('performance') as well as of passive
linguistic knowledge (competence).

A processing model based on a model similar to the one presumed
here Is described and defended in chapter 6 of Jackendoff (1987b).
Jackendoffs model has three strata (conceptual, syntactic, and
phonological), and after discussing several alternatives, he proposes an
interactive model of processing. In such a model, a given level need not
be fully specified before it starts transmitting information to the next
level, and both top-down and bottom-up processing occur
simultaneously. For example, in speech perception, the part of
phonological structure which has already been constructed generally is
used to determine the syntactic structure; but in certain situations,
the syntactic structure may be needed to determine part of the
phonological structure. Such a situation might be a conversation in a
noisy room full of people, in which the conversants “hear“ words which
have not actually been perceived by their ears, due to their brains _
unconsciously reconstructing the missing parts of the speech signal.

21

(Warren 1970 describes this phenomenon.) One part of the 86 model
which might have to be adapted to account for such phenomena is the
bidirectionality of the lines in the alternation patterns; if both

top-down and bottom-up processing are going on simultaneously, two
separate unidirectional alternation patterns may be needed. Actually,
such an idea has been proposed by some stratificationalists; it is
found, for instance, in the “micronode“ notation described by Christie
(1977), and dates back to unpublished work of Lamb from the late
1960's.

ilbasemamicstratm
WWW

According to the sketch given above, semantics in the present model
is presumed to be an autonomous level, with a structure determined by
its own internal principles. What is the nature of this structure?
Various linguists, including Gruber (1976), Miller and Johnson-Laird
(1976), Jackendoff (1983, 1987a, 1987b), Rappaport and Levin (1988),
and generative semanticists such as McCawley (1968), have argued that
semantic structure can be fruitfully analyzed as containing semantic
units organized in some principled way. These authors differ, though,
as to how these units are arranged and how their structures relate to
the rest of language. Following Jackendoff (1983) and Rappaport and
Levin (1988), I will take semantic structure to be arranged
hierarchically, in much the same way syntax is generally represented.
A similar assumption was made by the generative semanticists (cf.
McCawley's (1968) famous decomposition of 'kill' as 'cause to become

22

not alive'). but the present approach differs from early generative
semantics in claiming that syntax and semantics are two separate,
equally creative parts of language structure.

FIrst, consider sentences involving spatial location and motion, such
as the following:

5) a. W

Jackendoff (1983) breaks the conceptual organization of these
sentences into such elements as Places, Paths, Events, and Things,
nested inside one another and related by “functions“. (These are not
necessarily functions in the mathematical sense, but semantic
predicates; to avoid confusion on this point, the term “predicate“ will
henceforth be used where Jackendoff uses “function“.) For example,
sentence (5a) might have a semantic structure roughly as follows,
using Jackendoffs notation:

[Star-BE ([mngELV'SL WKALAMAZOOD]

This is a State as opposed to an Event; it contains the State—predicate
BE, which takes two arguments: a Thing (the object or person being
located) and a Place (the location of this Thing). With an Event,
different predicates occur. One of these is GO, whose arguments are a
Thing and a Path; this Path in turn contains one or more
Path-predicates (such as FROM or T0), whose arguments are Places.
This can be seen in the following representations of the semantic
structure of (5b) and (5c):

23

[Emco ([mmELVIS], [MFROM ([meALAMAzool)
TO ([p.mLANSING])])]

[EvaO ([ThinoMOUSE], [pamTO ([meNDER ([mmTABLEDDD]

Another possible Event-predicate is CAUSE, whose arguments are a

Thing (usually) and an Event, as in the semantic representation of (5d):

[EMCAUSE ([mngMARLYS], [Emco (Immnorxsn,
[PathTO ([Placs'N ([ThlngOVEND])])])]

The CAUSE predicate differentiates this sentence from mm:
mm whose representation would correspond to the innermost
Event above. Somewhat related to this is the predicate Jackendoff
calls LET, to account for such sentences as Who
W as opposed to W-

A few words of caution may be appropriate here. The above
decompositions are not meant to be exhaustive; they only represent
part of the meaning of an utterance, although it is an important part in
many cases. The above formalism, even as it is most fully developed in
Jackendoff (1983), does not distinguish among W
W Western. and monsters:
all reduce to roughly the following:

[EventGo ([ome'l-Ll. [PathTo ([PlaceAT (lowSTORElllllll

It is difficult to know where to go from there in terms of

decomposition. As many researchers (e.g. Fodor, Garrett, Walker, and
Parkes 1980) have noted, semantic decomposition can only go so far
before one runs into residue which does not seem to be decomposable

any further without ad hoc devices. One well-known example is color

24

terms; after abstracting out a feature [+color], how do we distingush
’red' from 'green'? And how should one distinguish 'run' from 'walk' and
'skip' using features? Katz and Fodor's (1963) solution was to call this
residue a “distinguisher' and claim that it does not play any part in
ambiguity or other formal properties of utterances; however, Bolinger
(1965) pointed out that such formal properties sometimes do need
access to material in the distinguisher. Some linguists (apparently
including Fodor himself; see Fodor 1975) have thus decided that the
whole idea of decomposition should be abandoned, since it is not
sufficient to fully describe natural language meanings.

However, this negative conclusion is reached only if one assumes
that a decomposition must be necessary and sufficient to describe a
word's meaning. It may be that the differences between 'red' and 'green'
or 'run' and ’trot' are to be found in other areas of cognition which
complement a decompositional “calculus of thought“. For instance,
much has been written about the prototype model of word meanings
(Rosch et al. 1976); it is easy to think that prototypes and
decomposition are mutually exclusive, and that a choice must be made
between the two. However, Jackendoff (1987b: chapter 10) discusses
how something very similar to prototypes (based on Marr's (1982) 3-D
model of visual structure) could be integrated with the lexical
decompositions given above, and elsewhere (1983:ch. 7,8; 1989) he
describes other systems which contribute to lexical meanings in a
complementary way. I will assume that such systems exist to
supplement the rough decompositions given below, and will thus not try
to make these decompositions necessary and sufficient.

25

Jackendoff uses phrase-structure type rules to specify the possible
combinations of conceptual categories, but these combinations can also
be specified by means of a tactic diagram such as that shown in Figure
4. The associative and nodes in this diagram generally represent
Jackendoffs States, Events, Places, Paths, and so on. Note that these
are not labeled as such; the category a node belongs to is determined by
its connections in the tactics. A node (such as the black one above GO
in FIgure 4) which leads downward to a choice of Things, a choice of
predicates including GO, and a choice of Paths, represents, by
definition, an Event. However, when there is more than one argument
for a function, these arguments have labels (arbitrarily given as 'A' and
'B') so that they can be differentiated for the purpose of realization
between strata. This is really a notational variant of Jackendoff's (and
Miller and Johnson-Laird's) method of listing arguments in an arbitrary
but fixed order; here, 'A' corresponds to his first argument and 'B’ to his
second argument.

This graphic notation can also be used to represent the conceptual
structure of individual sentences, as seen in Figure 5. These
conceptual traces are made from the full tactics by making choices at
all g: nodes so that only and nodes are left. Such traces are equivalent
to the bracket notation introduced earlier; in order to save space,
semantic structure will generally be represented here using bracket
notation, eliminating any labels or brackets which are unnecessary for
the immediate point being made. For instance, the traces in FIgure 5

could be represented as follows:

26

 

 

 

(PLACES)

(THINGS)

E':I:iIEII IIII'I'

27

 

E
B
A
KALAMAZOO
ELVIS
a. EMamKalamazoo b. Madmaunnamasumhamn
MW

[BE ([ELVIS], [KALAMAZOOD]

[CAUSE ([MARLYS], [GO ([ROAST], [INTO OVEN])])]
if it is obvious from the context what conceptual categories are
involved and the condensed structure is irrelevant for the point at hand
(cf. the fully specified representations of these sentences given

earlier).

I 2 S l' f l I
All of the decompositions discussed above deal with the location and
movement of objects in physical space. However, Gruber (1976) goes
on to suggest that this same structure is used for more abstract
domains. Consider the following sets of sentences (based on examples

in Jackendoff forthcoming):

28

6) a. W-
b. W.
e. W
7) a. Ibaodzemnttafillbatt
D. W
o. Wharton
8) 8- W.
b. Warm.
o. W-
9) a. W-
D. W-
c. WWW-

There are some obvious lexical and syntactic similarities here, even
though these sentences are describing quite different things: the
sentences in (6) describe physical location and motion; those in (7)
describe possession; those in (8), properties of Susan; and those in (9),
the scheduling of a party. As a guide to determining the semantics of
such examples, this thesis will adopt an assumption about semantics
which Jackendoff calls the Grammatical Constraint; this holds that
“one should prefer a semantic theory that explains otherwise arbitrary
generalizations about the syntax and the lexicon“ (1983:13). We appear
to have an instance of such generalizations here; a reasonable
assumption would thus be that all these various domains are organized
in parallel fashion, involving the same types of categories and
principles of organization that we saw in the semantics of spatial
expressions. If the conceptual organizations of the different domains

29

are similar, than the similarities seen above simply reflect this
parallel organization.

In the terminology of Gruber (1976), the sentences in (6)-(8) are
Positional, Possessional, and ldentificational respectively; Jackendoff
(1983) calls these “semantic fields“ and refers to them as Positional,
Possessive, and Identificational, adding Temporal for the sentences in
(9). This terminology will be adopted here, except that Positional will
be called Spatial and Possessive will be broken into several fields as
discussed below. The categories involved are the same in each field,
but they refer to a wide variety of things. For instance, the first
sentence in each of the above trios seems to involve “movement“ from
one conceptual location to another. Again using Jackendoffs notation,
conceptual representations of these sentences might be roughly as

follows:

[EventGOPosit ([ThingEl-VISL [PathFROMPoeit ([PlaceKAl-AMAZOOD
Tom. Immusmonm

[EventGoPoee ([ThingPR'ZEL [PathTOPoes ([pmG'LBERTlllll

[E,,,,,,c.o.,..,m (Immsusm), [meROMM ([FMHAPPW)

To“... (IPMSAonm
[EmpAUSE ([TthE],
[5,,“ch ([EmeARTY], [MFROMTW ([nmruesom)
701...... (InmSAmRDAYDDDl

Notice that the arguments of the Path-predicates (FROM and T0) are of
widely varying types, yet the overall structures of the representations
are very similar. Here semantic fields are indicated by subscripts on

30

the predicate names, but nothing hinges on this notation; in fact, in this
thesis fields will generally be indicated in the bracket notation (where
relevant) by means of a single subscript outside the outermost bracket,
in order to avoid undue clutter.

Why should conceptual information be organized in semantic fields?
One answer is economy. In dealing with an abstract domain such as the
ascription of properties to things, the mind does not have to create an
entirely new structure; it merely takes over, with certain minor
modifications, the structure used for a more concrete domain such as
physical location. Actually, it is not necessary to think of the spatial
field as more 'basic' than the others, although it is the most intimately

tied to sensory information.

5 II _ l I l I . I I I l |
5 I D I' . |

With the above outline of the conceptual system in mind, we can now
turn to the various three-place verbs in English and see what light a
semantic analysis can shed on their behavior in forming doublets.
First, though, it will be a good idea to define more precisely what we

mean by a “lexical doublet“ and a “three-place verb“.

Walsh
A preliminary definition of a lexical doublet for our present
purposes might be as follows:
Two three-place verbs form a lexical doublet if: 1) they can be used

in sentences containing exactly the same nominal arguments; 2)

31

these arguments do not all occupy the same syntactic positions in

the two sentences; and 3) the sentences refer to the same event,

which generally means that they entail each other.
Admittedly the third condition is somewhat subjective, but it should
not cause too many problems. It is intended to distinguish between
pairs such as 'sell'/'buy', which can generally be considered paraphrases
except for the difference in focus, and pairs such as 'throw'l'catch',
which refer to separate though related events and cannot always be
used for the same situation. These differences will be discussed
further below.

Note that there is no specification that the verbs have to be

different; a single surface form can function as both halves of a

lexical doublet, as in IbaJandlothantethssmmnmaoartmem

students and Win
landlord. There is a problem, however, in that this definition, strictly

interpreted, seems to say that the verbs in active-passive pairs and
other similar variations are doublets. Although the noun phrases in
such pairs differ in their surface position in the sentence, a
transformational grammarian of the old tradition would say that they
occupy the same positions in deep structure. In the most traditional
stratificational terminology, the noun phrases in question are said to
have identical semological roles, with the differences in surface
positionshowing up in the lexology. This position will be adopted here,
although with a change in terminology. Instead of the usual SG labels
'Agent', 'Patient', and 'Recipient’, 'S', 'O', and 'I' will be used to refer to

the arguments which in the most usual cases are realized as the

32

   

 

 

SELL BUY
8 o
S o I Hm
VLADIMIR
3'” VLADIMIR BILL A
CANTALOUPE CANTALOUPE
afiiILsoldacantalouoa Ill-WW
I III I' . I 8'"
WW

subject, direct object, and indirect object respectively. This approach
is empirically similar to one in which the arguments of a predicate are
given in an arbitrary but fixed order (i.e. 'buy (x,y,z)'), in much the same
way that the use of 'A', 'B', 'C', etc. to label arguments in conceptual
structure parallels Jackendoff's arbitrary ordering of the same
arguments.

As an illustration, Figure 6 shows semotactic traces corresponding

to the sentences Wimmamalouna and MladimlLbouohta

Will respectively. These show the semological
structures of individual sentences, which in this case illustrate the

doublet 'sell'/'buy'; they are specified from a semotactic diagram in
the same way that the conceptual traces shown earlier in FIgure 5 are
specified from a conceptual tactic diagram. From this illustration it

33

should be evident how we can revise the definition given earlier: at
least some of the nominal arguments associated with the verbs of a
lexical doublet must have different semological labels. This condition
is satisfied here, since both 'Bill' and 'Vladimir' have different labels in
these two diagrams: Bill is 'S' in 6a and 'A' (for 'adjunct') in 6b, while
Vladimir is 'I' in 6a and 'S' in 6b. In an active/passive pair such as lel
..,. ..".. . ,m . .z, ..".. ,. . .. ,-. - . .5 . -' ,the
arguments would have identical semological labels in the two
sentences (Bill and Vladimir would be 'S' and 'I' respectively in both
sentences), with other factors such as focus accounting for the surface

discrepancies.

W

The term “three-place verb“ should be taken to mean a verb which
takes both a direct and an indirect object; Green (1974) calls these
“dative-movement verbs“, but this term will be avoided here since it
implies the existence of a dativo-movoment transformation. It is a
well-known fact that in English, many indirect objects can occur either
between the verb and the direct object (internal indirect object) or
after the direct object in a propositional phrase (external indirect
object). However, there are some cases in which one of these generally
seems less natural, as in the following examples:

10) a. Wm

b. ?Ldonated.me.SalxatioaAtmv_soma.clomas.

11) a. 7W.

b. WW

34

We need not be overly concerned with such cases, since the
acceptability of such “odd“ sentences varies widely in different
contexts and with different speakers; for the most part, internal and
oxtomal indirect objects will be used interchangeably below. However,
the following discussion will not deal with “for“ datives, as found in
such sentences as WWW
cake. There do not appear to be any doublets which involve this
construction, so for simplicity's sake we can ignore these verbs here.

In order to make the following discussion easier to follow, it may be
a good idea to introduce some terminology here. In a pair of sentences
such as WWW
Agnes, 'Agnes' will be considered the Source and 'Bill' will be
considered the Goal. These labels will not always be accurate in a
literal sense, but they are a convenient way of referring to the
appropriate noun phrases in a doublet. 'Warthog' will be referred to as
the Transferee; again, this label will not always be literaly accurate,
but it is convenient Gruber (1976) and Jackendoff (1983) use the term
“theme“ for the transferred item; unfortunately, the same term is used
in systemic grammar (Halliday 1985) in a much different sense. In
order to avoid undue confusion, the more neutral label “Transferee' will
be used here instead of “Theme“.

MW
With the above definitions in mind, let us now turn to a survey of
three-place verbs with a view to noting and explaining their behavior in

terms of forming doublets. As it happens, Green (1974) contains a

35

fairly lengthy survey of throe- place verbs, dividing them into classes
based on‘rough decompositions of their meanings. Green proposes five
general classes of three-place verbs (if we exclude the for-dativo

verbs as noted above); these resemble the informal classes noted by
Josperson (1933(1964):114). Green's classes can be used as a general
guideline for our survey, although they will not be considered

definitive.

One of Green's classes includes, among other verbs, 'sell' and 'rent';
each of these can be paired with another verb ('buy' and 'rent'
respectively) to form a doublet:

12) a. W

D. W

13) a. Wanna

o. W.
As noted earlier, such pairs meet the conditions for a doublet: they
contain the same nominal arguments, two of these arguments have
differing semological labels (see Figure 6), and the sentences refer to
the same event; it is difficult to imagine a situation where Bill sells a
cantaloupe to Vladimir but Vladimir does not buy a cantaloupe from
Bill. In fact, this last observation suggests a quick test for
doublothood: if it is possible to conjoin one of the sentences involved
with the negation of the other, and have the result be semantically

well-formed, then the two verbs do not form a doublet. Thus the

sentence ?Bi|| sold a ganlalgugg lg Vladimir, but Vladimir didn't buy it

36

is very odd, giving us evidence that 'sell' and 'buy' form a doublet; but
the sentence WWW
gatghjt seems all right, suggesting that 'throw' and 'catch' should not
be considered a true doublet. Admittedly this is not a perfect test,
given the notoriously subjective nature of acceptability judgements,
but it can nevertheless be helpful.

Among the other verbs which Green includes In the same class are
'give' and “lend. Both of those form doublets, as the following examples

show:

13)a. W

The one thing which these verbs all seem to have in common is that
they involve a transfer of possession. Notice that this does not
necessarily mean that the thing being transferred actually moves in

space- In the sentence DarleneJusLsoldahousmlnomar. it is hard to

imagine the house actually going anywhere, yet this sentence is
structurally identical to Wm, in
which it is difficult (If not impossible) to imagine the watermelon not
moving. With the notion of semantic fields, though, those facts are
easily explainable: 'sell' describes an event In the possessional field,
while 'throw' describes an event in the spatial field. Movement In the
former does not necessarily imply movement In the latter.

Actually, the situation Is a little more complicated than that, since

there are several different kinds of possession; Miller and

37

Johnson-Laird (1976:562-5) present a useful outline of these. First,
there is the distinction between “alienable“ and “inalienable“
possession. Inalienable possession Is inherent in the nature of the
thing being possessed, and it cannot normally be transferred; the most
common examples of this are family members (“my mother“) and body
parts (“my arm“), and it can also be used with inanimate subjects (Ina
WW. Alienable possession is transferable; all
the doublets noted above ('sell'/'buy', 'rent'l‘ront', etc.) involve some
sort of alienable possession. It thus turns out that doublets involving
possession appear to always refer to alienable possession; because of
this, inalienable possession will not be a major factor in our
discussion.

However, there are also several different types of alienable
possession. Miller and Johnson-Laird identify three such types:
“inherent“ (referring to ownership of something), “accidental“

(referring to having temporary control of something without
necessarily owning It), and “physical“ (referring to the physical
location of something in someone’s hands or on their person). The pairs
'sell'/'buy' and 'ront'l'rent' differ in that the former refers to a transfer
of inherent possession in exchange for money, while the latter refers
to a similar transfer of accidental possession; in neither case Is it
necessary for there to be an actual transfer of physical possession.
'Givo'l'receivo' and 'Iend'l'borrow' are parallel to 'sell'/'buy' and
'ront'l‘ront', except that here there is no money involved. Katz (1972)
notes that all four of these doublets involve a transfer of possession,
but are distinguished by two parameters: whether or not a transfer of

38

money is involved, and what type of possession (inherent or accidental)
Is being transferred.

This is not quite accurate, though; 'give' can apparently refer to
transfer of any of the three types of possession, as the following
sentences demonstrate:

15) 8- W

D. WW

0. W?
(In addition, 'give' can be used in more abstract ways, as In ,lglmgayg
W or W, but discussion of these special
cases will be deferred until later.) The three sentences in (15) refer to
the transfer of inherent, accidental, and physical possession
respectively. However, In the last two sentences above, there are
synonyms which can be substituted for 'give' with no apparent change in
meaning ('Iend' in (15b) and 'hand' in (15c)), while no such synonym
comes readily to mind for (15a). Also, WEE
seems odd If it refers to a situation where Bill simply handed me the
book. If we substitute 'get' for 'receive', the situation does not change.
'Get' Is ambiguous between an agentivo and a non-agentivo sense
(roughly synonymous with ‘obtain' and 'receive' respectively); if we
interpret it In the non-agentivo sense, it appears to only refer to a
transfer of inherent possession. In other words, the sentence ngLa
W11 refers either to a situation where I actively obtained the
book from Bill, or to one where the book was a gift from Bill to me; It
does not refer to Bill simply handing me a book which I may already

own (there is another potential reading of 'receive'/get In die sense

39

of, for example, receiving something in the mail; this will be discussed
when we get to the Spatial field). Thus it does not seem unreasonable
to consider the inherent-possession meaning of 'give' to be paired with
'receive' In a doublet, allowing It to fit into Katz's paradigm.

Given these four basic three-place doublets, similar pairs come to
mind. 'Loaso'l'loase' is obviously parallel to 'rent'l'rent', and for
several of the other verbs it is possible to make a trivially new
doublet by substituting a synonym: 'purchase' for 'buy', 'loan' for 'Iend',
etc. A number of pairs can be found which are similar to
'give'/'receive'. Such pairs include 'boqueath' and 'inherit' and possibly
'pay' and 'receive':

16) a. WWW

17) a. W

D. W
In the above situations there are selectional restrictions on the
transferee, as well as various other conditions or presuppositions, but
nevertheless these pairs appear to be doublets. In some cases (e.g.
'award', 'code', as well as 'pay' above) verbs similar to 'give' do not
seem to be pairablo with anything other than (possibly) 'receive'; there
are In fact considerably more verbs of the 'give' type than there are of
the 'receive' type. Such a fact does not seem too surprising, since the
source starts out possessing the transferee, and thus is better able to
control any action involving its transference. We will see later that
similar generalizations can be made about other types of three-place

verbs.

40

How should this group of verbs be represented conceptually? First
of all, we appear to be dealing with two abstract semantic fields,
which we can call “inherent posssessional“ and “accidental
possessional“. The verbs discussed above seem to involve movement in
one of those fields, from one possessor to another. Note that the only
“locations“ possible In these fields are not places in the usual sense of
the word, but people (or, possibly, other animate beings: animals,
personified organizations, etc). Also, as Jackendoff (1983:192) notes,
these locations are discontinuous. An object cannot be halfway
between being possessed by John and possessed by Mary; it is
possessed by one or the other, and so a transfer from one to the other
can be thought of as Instantaneous In conceptual terms (though in real
life, of course, there Is often physical movement of the object over a
period of time).

With all this in mind, we might represent 'sell' as shown in Figure 7
and below in bracket notation (from now on, bracket notation will
generally be used in order to save space, and diagrams of selected

semantic representations can be found In the appendix):

[CAUSE ([Parsonxlrlrpos. [GO ([Y]. [PamFROM “me1) T0 (WDN
[GO ([$]. [pamFROM ([Personzl) T0 (IPMXDDI )1

This states that x (the subject of 'sell') causes two events in the
Inherent Possessional field (abbreviated here as Inposs): y going from
x to z, and some unspecified amount of money going from 2 to x. (The
same representation could be used for 'rent' by changing the Inherent
Possessional field to the Accidental possessional field.) The symbol '$'

indicates a selectional restriction on the secondary transferee; if it is

41

 

lexically expressed, it must be an amount of money, and If it is not, it
is assumed to be present implicitly. (Jackendoff 1987a:384-7
discusses such restrictions in terms of “argument fusion“) Note that
the arguments of the Path-functions FROM and T0 are Persons, rather
than Places. A Person, in this conceptual sense, does not have to
literally be a person; it can be any volitional being, such as a person, an
organization, an animal, or something similar. Persons are distinct
from Things, which are physical objects. Compare the following pairs
of sentences:
18) a. W.
b. Mkamnsmresfiadlllacs
19) a. W
b. ??Mllss:s.bodiLoims_thtea_Qadillacs.

42

The word 'Mike' in (18) seems at first glance to be referring to same
thing in both sentences, but as (19) shows, this Is not quite true. 'Mike'
can refer either to a physical entity located in space, in which case it
Is conceptually a Thing, or to a volitional entity capable of possession
(among other things), in which case it is conceptually a Person. The
'Iocations' In the Inherent Possessional field can thus be considered
Persons, since Things are incapable of inherent possession.

'Buy’ can be represented conceptually as follows:

[CAUSE (famous... [GO (tyi. [pamFROM (trams) To Imam]
[GO (I$]. Ip....FRou «9......21) To remain] )1

(The second sense of 'rent' could be represented in the same way, but In
the Accidental Possessional field.) This is exactly like the

representation for 'sell' except that 2, rather than x, is seen as

Initiating the action and Is thus the first argument of CAUSE. Note that
in both cases, the participant who appears in this position is

lexicalized as the unmarked subject, or, in the terminology being used
here, Is realized in the semology as '3'. Thus we might tentatively say
that a doublet occurs when two conceptual representations are the
same except for the Initiator of the action; whichever participant is

the first argument of CAUSE will be realized as '8'.

However, consider the following conceptual representation:

[CAUSE ([me]. InpoaeIGO ([yI. [p.mFROM ([Porsonxl) T0 (IPMZIllllll
Such a situation might conceivably be described by either W
1 or W; in both cases the same person initiates the
action, although this fact is more central in the case of 'give'. In fact,

it is probably more appropriate for 'receive' not to involve the CAUSE

43

predicate at all, so that W would be something like
this: ‘

.mico (Iyi. [PamFROM (emu) To (cmzmim
This is essentially the representation Jackendoff (1983:192) uses for
'receive', and it seems intuitively plausible. (Note that it is possible to
not mention he Initiator at all, as in W; In
such a case, the FROM predicate would simply not be a part of the
conceptual representation.) Yet the above representation does not
allow our tentative subject selection principle to work, since there Is
no CAUSE predicate; there is thus no way to tell which argument is
realized as the unmarked subject.

Furthermore, consider the following representation of 'take', as In x
takesxfromz:

[CAUSE (lama. .nmico (M. [PamFROM (Ipenle) To remain»:
This is identical to the above representation of 'give', except that 2,
rather than x, causes the action; in other words, the goal rather than
the source is the initiator. Also, just as 'receive' can be paired with
'give', 'lose' can be compared with 'take', as In the following sentences:

20) a. W-

b. GormamrlosLAlsacaLorraInotoEranoo
With both 'lose' and 'receive', there does not seem to be any overt
causation involved, since the focus is on the non-controlling
participant and the presence of the causer is optional Wu

armadillo. GormanflostAlsaotLorraIno) A conceptual representation
for W; thus might be as follows:

44

.mtco (Iyl. [pamFROM (ipmxn To remain]
Note that this Is the same representation proposed above for M
urea”. If this is the case, then there is a double problem for the idea
that only the initiator of an action is realized as the subject. On the
one hand, as we noted before, there is no CAUSE predicate above to
determine the subject; on the other hand, In the absence of such a
predicate, either of two participants (x and z in the above example) can
be realized as the subject.

There Is another way of looking at the problem which seems more
promising. Notice that in all the above examples, each of the possible
subjects is one of the endpoints of the Path taken by the transferee.
The argument of FROM loses possession of the transferee, while the
argument of T0 gains possession; there is a kind of reciprocal relation,
especially if we think of the transfer as a single, instantaneous event.
As a general rule, we could say that either endpoint of a Path in a
Possessional field can be realized somologically as the 'S' argument
(I.e. the one which will appear as the grammatical subject in the most
usual cases) without necessarily being the Initiator. Before we try to
elaborate any further, though, it might be a good idea to look at some
other types of three-place verbs and see how they relate to the above

discussion.

W
One of Green's categories contains many verbs which form ‘pseudo-

doublets'; that is, they meet the first two criteria for doublethood, but

do not refer to exactly the same event. One example which has been

45

mentioned several times already is 'throw'/'catch'. Throwing a ball to
someone does not necessarily mean that they will catch it; we have
already seen that it is reasonable to say MW
Whit. Similar statements could be
made about such pairs as 'send'/'receive' or 'mail'l'receive' (of. 1

Some of the verbs put in this class by Green ('roll', 'fling', 'fonrvard',
etc.) do not seem to form pseudo-doublets except possibly with 'catch'
or 'receive'. This is much like the situation with Possessional verbs, in
that there are more distinct verbs with a source subject (e.g. 'send')
than there are verbs with a goal subject (e.g. 'receive'). A related
observation is that the source appears to be the unmarked initiator;
there are more verbs where the source causes the action ('give', 'send')
than there are where the goal causes the action (e.g. 'take'). It is
difficult, for example, to find a goal-initiator verb corresponding to
'send' or 'throw' in the same way that 'take' corresponds to 'give';
presumably this is because of the difficulty of exerting control over
something from which one is separated in space. ('Send for' is a '
possibility, but It is two words and Introduces additional
complications, such as an implied third party.)

Despite our revised subject selection principle, it is still the case
that the Initiator, it it is specified in conceptual structure, is realized
as semological '8'. Thus the above generalizations are related and can
be collapsed; if the source is more likely than the goal to be the
initiator, and the Initiator is always realized as unmarked subject, it

follows that the source is more likely to be the unmarked subject. It

46

is still possible for the goal to be the initiator, but such cases are
relatively unusual.
There are numerous pieces of evidence for placing the 'throw' group
of verbs in the Spatial (Jackendoffs Positional) semantic field. One Is
that the transferee must actually move in physical space; it is
impossible to make sense of X throwing a ball to Y if the ball remains
stationary. (On the other hand, we have seen that it is perfectly
possible to sell something without that thing moving at all.) Because
of this, the events described by the members of a pseudo-doublet such
as 'throw' and 'catch' are necessarily separated In time; this explains
why one of these events can be negated without making the sentence
anomalous (BflLthreIrLaoantaloupoJoMadImLoutMladImILdidnzt
millil- Also, note that verbs such as 'send' and 'throw', unlike 'sell'
and 'give', can normally be used with any directional propositional
phrase, and are parallel in this regard to spatial uses of 'go':
21) a. BIIIJbrentbaoaIlJmomoJaIrino
b. ?EIlLsold.ttIe_oalLInto.ttIe.ravino
o. W

22) a. Usasentthaoaclsaoooutonhmumm
b. ?Llsa.oaire.ths.oaclsads.oumf.tho.oountor
c. Lisamnoutonhoooumnr-

There Is, however, a restriction: the subject of a verb like 'receive'
must normally be a Person. Note the following:

23) 8- Johannsontaoackadattmate.

b. Katarscoixedaoackaoatromdobann.

47

24) a. JonamLsanLaoaclsaomuenxdrls.
D. 7W.

A sentence such as (24b) seems to only be interpretable in a sense
where ‘New York' is understood to mean some person or organization In
New York, as when a businessman uses it to mean 'our New York office'
or 80019009 says WNW
government. This tendency is not iron-clad; it is sometimes possible
to say something like W in which
the status of the subject as a Person Is marginal, but in general it Is
Person endpoints which are realized as subjects.

With all this in mind, we can represent 'send' conceptually in the
following way:

[CAUSE (rams. salaico (IyI. [p.anou (Ipmxn To (Ipmsmnn
And, following our earlier practice, 'receive' in the Spatial sense could
be given the following representation, lacking only the causative

predicate:
Spatial'flO ([YL [PathFROM ([Personxl) TO ([Personzmn

In fact, essentially the same representations could also be used for -
'throw'/‘catch', as well as for any similar verbs in this group, with the
differences in meaning showing up elsewhere in cognition. The above

representation Is appropriate only when both endpoints are Persons; the
representation of WW would be roughly:

[CAUSE ([BILL], wgieo ([BALL], FROM ([anBILLI)
PathTO ([PIaca'N ([rhingRAV'NEDD )D]

The argument of T0 here is a Place rather than a Person, and it cannot

48

be realized as 'S' in the semolooyz we cannot say mum];
mammal or 7W. These ‘
observations suggest a slight revision of our subject selection

principle: only a Path endpoint which Is a Person may be realized
somologically as '8'. When both endpoints of the same Path can be
realized as 'S', and the rest of the representations are the same except
possibly for the presence of an Initiator, a doublet (or a pseudo-
doublet, according to the semantic field involved) occurs.

Another of Green's classes includes such verbs as 'take', 'bring',
'drag', and 'push', in which the subject accompanies the transferee to
the goal. These are similar to verbs of the 'throw'-class in that they
Involve the Spatial field: the transferee necessarily moves in physical

space, and it is possible to have a directional propositional phrase

instead of an indirect object (Ltooldtnahiopoootamusintomaolsxator.

3 o.oo:,ou: no,” 0 s“:n| : o o Izolnsgge :
However, this group of verbs is somewhat troublesome in terms of
semantic analysis. At least some of these verbs appear to form
pseudo-doublets with the ubiquitous 'receive':

25) a. JohoomuohtadozamlonmstammodmssstManr-

b. We

This can only be a pseudo-doublet rather than a true doublet in that
(25a) does not necessarily imply (25b); Mary could refuse to accept the
roses, for example. Note that, as with the 'throw' type, such a

pseudo-doublet is only possible with a Person as indirect object:

49

26) 8- JonnorouohtanostrIoMLManr'saoartmom
b. 7W.

In general, this group is similar to the 'throw' class, the main
difference being that here the source acompanies the transferee to the
goal. In terms of conceptual representation, we might have something
like the following for 'bring', where [x&y] means that x and y traverse
the Path together:

[CAUSE “me]! SpgfladGO ([Penonx & Thlmylr [PathTO ([Pgmonzlnnn
The corresponding sense of 'receive' would lack the CAUSE predicate, as

with the other 'receive'-type verbs we have seen:
SpatiallGO ([perscn" 3* rhingYL [pamTO ([PcnonzDD]

However, it is not at all clear that there is a separate sense of 'receive'
with such a conceptual representation; it is difficult to interpret Mam
reoolired_adozonJono;stammad_mses.tromJoho so that it uniquely
refers to the same situation descibed in Wm
WM. If so, then it is doubtful whether 'bring'
and 'receive' should be considered a pseudo-doublet.

Another way of looking at the situation is to interpret the sentence
with 'bring' as referring to two separate but related events: John and
the roses physically moving to where Mary Is located (an event in the
Spatial field), and John giving the roses to Mary (an event in the
Inherent Possessional field). Thus 'bring' in the above sentence would

have a representation like the following:

[CAUSE ([JOHN], Wico ([JOHN & ROSES], [TO ([MARY])])]
InhpouIGO ([80353]. [FROM ([JOHND T0 ([MARYDlllll

50

The 'receive' sentence, however, would just refer to the second part of
the action, as follows:
mp0,,[GO ([ROSES], [FROM ([JOHN]) TO ([MARY])])]

This analysis, with an implied 'give' in the meaning of ‘bring', is
intuitively more satisfying than our first one, and it will be taken here
to be the correct one. 'Bring' and 'receive' thus do not form a doublet as
we have defined it, since the main action is not the same in both cases.
This is consistent with the observation that Wazoo

WW cannot refer uniquely to the same
situation described in W-

All of our examples so far have dealt with concrete transferees;
that Is, the thing being sold or sent or brought has been a conceptual
Thing. There are, however, many Instances of three-place verbs
involving abstract transferees. In some cases these do not behave
exactly like verbs with concrete transferees; later on we will see some
instances where something other than a Person Is the initiator of a '
Path and can thus appear as a subject in a doublet, and there are some
fuzzy cases which are difficult to analyze.

One of the groups of verbs recognized by Green involves some type of
Information as transferee. Consider the verbs in the following
sentences:

27) a. JothauohtfioanIsDJdMilllo

b. MIIflsJaamsdfioanianromJohn

 

If we apply the negation test, the results are not very clear and vary
with different verbs:
30)|l| IIS 'llll'll'lll I'l'll 'I.
31)E|||| 'l I II I" IIII'I'II 'I
32)“ III I'll I II. I II” I'I'II'I
mm.
Green (1974) argues at length that (In the terminology being used here)
'teach' and 'Ieam' form a true doublet only when 'teach' has an internal
Indirect object; In other words, she claims that of the sentences below,
a entails 9 but :1 does not:
33) a. Jammmmnefioanisn
b. Jonmtauomfioanismwllio.
c. MIIIIaleamstoanisMromJohn

My own Intuition Is that 'teach' is unclear as to whether it Implies a
sentence such as g, but not along the strict lines suggested by Green;
for me, both a and b above can mean either that Millie learned
something or that she didn't. This is a somewhat subjective judgement,
but the point can be made clearer by adding an adverbial phrase:
34)a-l|| ||ll°ll°S 'II III l'l'll
b-lll IIS 'IIII'II'I III l'l'll

52

Other verbs of this type are also unclear; both Green (1974) and Miller
and Johnson-Laird (1976) confidently make many acceptability
judgements which seem questionable at best. Actually, the question of
whether or not these are true doublets or pseudo-doublets is not a very
pressing issue; it seems safe to say that the above pairs can at least be
considered pseudo-doublets.

What kind of conceptual representation might these verbs have?
first of all, they are syntactically analogous to the other verbs we have
been considering, and intuitively, some kind of information seems to be
“going“ from one person to another. Based on these facts and the
Grammatical Constraint, we could postulate a basic structure similar
to that for 'give'/'receive':

[CAUSE (IXI. Ico (Iyl. [pathROM (IXI) To ([2])lllll

This analysis is further supported by the fact that many of the verbs of
physical location and possession which we have already discussed can
also be used with an information transferee, as in the following
examples from Green (1974:121):

35) IhistsooLoamofromasixtoomnsomuouext

36) Ltoolslbosofldurostromihohdordsmanual.

37)I III. 'II II ..

The problem Is in determining what semantic field we are dealing
with, as well as what conceptual categories the variables belong to.
First of all, notice that when Bill, for instance, tells me a joke, he does
not cease to “have“ the joke; after he tells it, we both know it. This
suggests that the transferee in such a sentence should belong to a
different conceptual category from the transferee of 'sell'; we can call

53

this Information, as opposed to Person or Thing. Also, note that when
we have a doublet such as those in sentences (35)-(37), both endpoints
of the Path tend to be Persons. This is hardly surprising in the case of
the goal endpoint, since only an animate, sentient being is capable of
having thoughts and being aware of information. As for the source
endpoint, there are (as we will see later) some apparent Instances of
non-Persons in this role, but only in cases where the action is not a
volitional one.

It is difficult to say what semantic field these verbs should belong
to; since the transferee is an abstraction, there is little basis for
comparison with the other three-place verbs we have seen. It does not
seem very plausible that they should belong in the Spatial field, since
an abstraction is incapable of physical movement. Some kind of
possession IS perhaps more plausible, but both inherent and accidental
possession seem strange when applied to an Information transferee; if
John tells me a joke, I don't own that joke, nor do I have any obligation
to ”give it back“ to John. Perhaps the best solution is to postulate a
different semantic field and call it Cognitive Possession, although no
particular significance should be read into this choice of terms; Green
(1974:121) apparently takes such a course of action when she
decomposes these verbs into something like “Subject CAUSE i.o. to
HAVEcog d.o. BY CAUSE-Ing do. to GO TO i.o. BY VERB-ing d.o.“.

Cognitive Possession involves Information as transferee, and an
indefinite number of people can cognitively possess the same
Information simultaneously.

Based on all this, a conceptual representation of 23m; and z

54

mm respectively would be as follows:
[CAUSE ([Pgrsonxlt Cogposs[GO (llnfoylr [FROM ([Personxl) To ([Pemonzl)])])]

These are exactly parallel to the representations for 'give' and 'receive'
except that the internal event is in the Cognitive Possessional rather
than the Inherent Possessional field, and the transferee y is
Information rather than a Thing. This Information may be expressed as
a sentential complement rather than a noun, as in Jaaammglaat
Wags. Our subject selection principle still holds here, Since
these verbs are essentially like 'give' except that the transferee Is
abstract rather than concrete. However, as we will see below, an
abstract transferee allows various situations which cannot occur when

the transferee is a physical object.

52W

'Give' can be used with Information as transferee, in an apparent
Instance of the 'tell' type of verbs:
38) a. Wow.

b- Lootrrecoiiredjlmanswouomxoroolomiromdohn.
However, it is also possible to have a Thing or an Event as the subject
of 'give' in such a case, and the same is true for a few other 'tell' verbs
such as 'teach':

39) a. loofalhodramoaxamsdhaansmommuroolom
b- Looflrsooixedihaansnemmuroolomiromjhsialflno
Lala.

55

40) a. W
b. QlflaIeamad.axaluablonsson.tmm.balno.on.trIal

There is a difference between (38) and (39)-(40), however. Sentence
(38a) could mean that John unintentionally caused me to think of an
answer (for instance, if my watching or thinking about John suddenly
inspired me to come up with a solution), but it is more likely to refer
to a situaton where John acted volitionally, knowing the answer and
then willingly imparting that information to me. In other words, John
causes the action in both readings, but only In the second one Is this
causation volitional. Sentence (39a), though, can only have the first
reading; rain obviously cannot do anything volitional, since it Is
inanimate. Similarly, (40) refers to a situation where the experience
of being on trial caused Ollie to learn a valuable lesson, but this
causation cannot be intentional, since the subject Is an Event.

These differences suggest some revisions in the conceptual
representations we have been using; in particular, some indication of
volition Is needed. Many authors (e.g. Miller and Johnson-Laird 1976,
Talmy 1976, and Gruber 1976, to name just a few) have tackled the
question of intentional vs. unintentional actions in a variety of ways.
Jackendoff (1987a) postulates a separate “action tier“ (in addition to a
“thematic tier“ such as we have already seen), baSed on similar
suggestions in Culicover and Wilkins (1986) and Talmy (1985). This
specifies (where appropriate) the Actor and Patient of a situation, as
well as whether or not the action is volitional. In Jackendoff's
terminology, an Actor need not be animate, as long as it is the focus of
some action; for instance, Mall would be the Actor in W

56

W, and the action tier of this sentence would look like this:

[ACT ([BALL])]
If an action is volitional (obviously possible only with an animate
Actor), this is indicated in the action tier; thus the sentence BllLatoaa
aaplo would have an action tier something like the following:

[ACT ([BILL], [APPLE])]

VOL
Furthermore, some sentences will lack an Actor and thus have an empty
action tier, as in W.

The exact notation for indicating volition will not be a major
concern here; Jackendoff's thematic tier is as reasonable a suggestion
as any. The Important thing is that all the causative three-place verbs
we saw earlier (e.g. 'give', 'sell', 'bring', etc.) involved volitional
actions. On the other hand, sentences (39a) and (40a) are clearly not
volitional, since the subjects are inanimate, yet they apparently are
causative. These facts can be captured by having the conceptual
representations of both types of sentences Involve a CAUSE predicate,

with the difference that the action involved is specified as a volitional

one in the case of, for example, WW,
and a nonvolitional one in the case of W
ansnemmiLoroblom.

There is another question to be confronted, though. In a sentence
such as W, it does not seem right at
first glance to consider the falling rain a “source“ in the sense we have
been using, i.e. something at which the answer originates before being

transferred to Stan; rather, it is more like a stimulus which causes

57

Stan to “have“ the answer indirectly. (There is, however, some
independent evidence for considering Stan a goal in such cases, e.g. Iaa
3W.) On the other hand, treating such a
noun phrase (or gerundive verb phrase, such as baiagoajrial) as
anything other than a source leads to potential problems.

For example, the sentences MW and
Wain would seem to illustrate a

doublet, at least when the second is interpreted in a non-agentivo
sense. One possible conceptual representation for the first sentence
would involve the falling rain only as the initiator and not as source, as
in the following:

[CAUSE ([FALLING RAIN], [GO ([ANSWER], [TO ([STAN])])]”
This creates a problem for the 'get' sentence, since we must either
introduce a CAUSE predicate into its conceptual representation and
distort our principle that initiatbrs are always the unmarked subject,
or find somewhere else in the representation to put the rain. However,
if we follow the pattern used with concrete uses of 'give' and 'receive',
the conceptual representations of these two sentences would be the
same except for a CAUSE predicate in the first one, as shown below:

[CAUSE ([FALLING RAIN], [GO ([ANSWER], [FROM ([FALLING RAIN])

T0 ([STAN])])])I

[GO ([ANSWER], [FROM ([FALLING RAIN]) TO ([STAN])])]

Here the falling rain is treated as a source, in order to account for its
presence in an oblique phrase in W
rain. It might be possible to consider the rain something else

conceptually (perhaps an instrument?), but there is no clear reason to

58

do so. In fact, the Grammatical Constraint says that arbitrary
generalizations involving syntax (such as the parallel structures in the
concrete and abstract uses of 'give' and 'receive') should be reflected in
semantic structure unless there Is a good reason to do otherwise, and
here there appears to be no such good reason. The idea of the rain as a
source Is not so implausible if we expand the notion of conceptual
“source“ beyond its most literal interpretation. A source can be not
only a location from which “motion“ proceeds, but anything serving as
the origin of the transferee in a broad sense; this would include
non-agentivo initiators such as the falling rain above.

There is another group of expressions involving abstract 'give' which

seem to parallel those in (39)-(40). Consider the folowing sentences:

41) a. JQDD'S criticism gave Marx an informal! complex.
b.” I? 'I'IHI II II,

Here, as in (39)-(40), it does not make much sense to speak of the
subject of 'give' as the literal source of the transferee, but it can be
considered the source in a more general, abstract way. Note also that
it is not necessary for any of the participants to be Persons, although

they can be. What seems to be common to all the above sentences is

59

that the subject of 'give' causes the indirect object to have some
property, either physical or psychological: an Inferiority complex, a
fresh coat of paint, a black eye, a dent. This looks like a candidate for
another semantic field, which we can call the Property Possessional
field. The “locations“ In this field are Things or Persons, and possibly
also Events (cf. I. :
gigjgaity), and the transferee is a Property. This Property is often a
physical one which is defined by its effect on its “host“; a dent, for
instance, must be a dent in something.

 

Conceptual representations for the pairs in (41 )-(44) might be
something like this:
[CAUSE (IXL PropPossleo (Ipmpid. [PamFROM (IXl) T0 ([Zl)])])l

proppmlGO ([pmle. [p.mFROM (IXI) T0 ([le1)]
where x and 2 can be Things, Persons, or Events. This representation is
parallel to what we just saw for certain cases involving Cognitive
Possession, except that z--the goal of the Path-~does not have to be a
Person. lntuitively, this is due to the fact that the “possession“ of
Properties is not limited to Persons, as we saw above (of. Iaoaamasa

dontinjtsjondor as well as Willows). while
Cognitive Possession obviously Is (cf. ?My_gar_kagns_taat_slgar vs.
Mamknowsmatstoorl-

It appears that we will have to revise our principle so that a Path
endpoint which can be realized as 'S' does not always have to be a
Person, but can belong to other categories according to the nature of
the transferee and the semantic field involved. For example, in the‘
fields where the transferee must be a Thing (Inherent and Accidental

6O

Possessional and Spatial), an endpoint must be a Person to be
realizable as unmarked subject; this is presumably because only a
volitional being (i.e. a Person) is capable of possessing and controlling
a physical object. The Spatial field, unlike the other two, can involve
“locations“ other than Persons, but only with Persons as locations is
there a sense of possession; in fact, what Miller and Johnson-Laird
(1976) call the Physical Possession field is really just the part of the
Spatial field which involves Persons as locations.

As for the other fields, the abstract nature of the transferee allows
more latitude in the nature of the endpoints. In the Cognitive
Possessional field, the goal must be a sentient being (a Person), since
the transferee is Information which must by Its nature exist in
someone's mind. However, as we have seen, the source can be anything
(a Person, Place, or Event) if we allow a broad but plausible
Interpretation of “source“. Similarly, in the Property Possessional
field either endpoint can be a Person, Thing, or Event; the transferee Is
a Property, which is not limited in Its possible locations the way

Information is.

WW5
Now let us look at some three-place verbs which appear to form
neither tme doublets nor pseudo-doublets. Consider the following
sentences:
45) a. Wad
b. WINNIE.

61

46) 8- Aorilonesflitaumaordollars
b. Emu—Mandolhrsiroonril

There do not seem to be any English words which will fit In the blanks,
even though words with the appropriate meanings are certainly
conceivable and the (a) sentences are structurally parallel to the other
verbs we have been considering. Why should this be so?

Notice that in both the above cases, there appears to be an implicit
gammlal Path which is not necessarily traversed by the transferee.
Speaking In rough terms, in (45a) Neville obligates himself to cause the
theme to go (in terms of inherent possession) along the Path from
himself to Adolf; in (46a), April Is obligated for some unspecified
reason to cause the transferee to go from herself to Bithi, but there is
no guarantee that she will actually do so. An approximation of the

conceptual structure of (46a) might be something like this:

“.mreE ([APRIL]. [AT ([pmMOBLIcATED

(IPotential EvemCAUSE ([APRIL], [Emco ([320].
[p.mFFiOM (IAPRll-I) T0 (IBITHllllllllllllll

This is admitedly a rather ad hoc representation, but it gets the general
idea across. The transfer of $20 from April to Bithi Is not an actual
Event which has occurred, but a Potential Event which April is
obligated to cause some time In the future; it may be the case that this
Event will in fact never occur.

Now consider the following examples:

47) a. W

b. WWW-

62

48) a. W
b- WWW
Here there apears to be another Potential Event which does not occur
because it is specifically negated. Something like the following might
be a conceptual representation for sentence (47a):

[CAUSE ([BOSS], [EMNOT «pomall Emoo ([PROMOTION],
[p.mFROM ([3035]) T0 (ISAMDDDDI

The Idea here is that a Potential Event in these examples is not
something which Is being asserted, but something about which an
assertion is being made. For example, In (45)-(46) it is being asserted
that the subject is obligated to bring about the Potential Event, and in
(47)-(48) it is being asserted that the subject prevented the Potential
Event from occurring. It appears that as far as our subject selection
principle Is concerned, a Path occurring as part of a Potential Event
does not count, and subjects of doublets can only come from a Path
which Is part of an actually occurring event. Better formalisms could
undoubtedly be developed for stating this generalization, but the above
statements are adequate for our purposes here.

There are a few other three-place verbs which seem not to form part
of doublets, in at least some of their senses, because no conceptual
Path is involved. One example is 'show'. When the direct object of
'show' is abstract lnforrnation, it has a meaning similar to 'teach' and
might be reasonably paired with 'Ieam', or possibly 'find out' in a
non-agentivo sense:

49) a. W

b. EmbamodlhualuuflrioadshiaflomMagoio

63

50) a Jermshowumaaomodrhraastiolssain
o. W
In such cases it seems possible to postulate a conceptual
representation like that for 'teach‘, in which an Information transferee
“moves“ between two Persons. In some instances, though, the direct
object is a Thing, and in such cases no doublet appears possible:
51) a. Madmahomuaaasaarliaomolot
b. QaaL_l.a.saarlIao.ooslot.trom.Marilm
'Loarn' and 'find out' certainly do not work here, and no other verbs fit
the blank very well. This Is because this sense of 'show' can be roughly
paraphrased as “cause to see“ or possibly “allow to see“; a conceptual
representation of (51a) in bracket notation might be as follows:
[CAUSE ([MARILYN], [EvaEE ([DAN], [SNARLING OCELOTJ)])]

This representation contains no Path, because nothing Is “moving“. The
ocelot is certainly not physically moving, and it would be difficult to
argue that information about the ocelot is going from Marilyn to Dan;
Marilyn is simply creating a situation where Dan can perceive the

ocelot of his own accord. As Green (1974) also points out, an external
indirect object generally seems odd with an Information transferee,

but is much more natural with a concrete transferee:

52) a. ?Madoio.saoinred_tho_valuoof_friorEIsaio_to_Boa-
b. Maahrasaonedasaarlradooalonmflaa
Thus there is some evidence for two conceptually different verbs
'show'; one of these contains a conceptual Path and forms plausible
doublets, while the other appears not to Involve a Path and does not
enter into any reasonable doublets.

64

Finally, there are some uses of 'give' which do not fit comfortably
into any of the above categories and vary in their ability to form
doublets with 'receive’ and/or 'get'. One such usage Involves
communication of some sort (not always verbal) between two Persons.
In many such cases doublets are possible, although some such
sentences are better than others:

53) a. W

U W
54) a. Bamaraoaxefisordaadlrtuook.
b. Wm
55) a. W
b. ?MarIo_oo.Ltho.flaooLfrorn_tae.daxer.Ia_tbaaoxt.laao
55) 8- W-
b- ”We:-
In general, these uses of 'give' seem similar to the 'tell' type of verb;
there is Information being communicated here, although it is of a
rather special kind. A conceptual representation for (53a) would thus
be like the following:
[CAUSE ([MARIO], [GO ([A LOT OF FLAK],
[FROM ([MARIO]) TO ([LUIGI])])])]

On the other hand, some cases, such as those in (55)-(56), have
characteristics of idioms - they are difficult to use in constructions
other than their normal one. For example, not only is it strange to

express sentence (55a) with 'get', it is also strange to express it with

passive 0me. cleft (?Imas_tao.fiaoortaat

65

W and pseudo-cleft constructions (7W

MarionesJasjaoor). except perhaps for humorous effect. This
suggests that “give (someone) the finger“ is an Indivisible unit. It is

not easy to draw a line between Idioms and non-Idioms, though.
Expressions can be idiomatic to varying degrees; some expressions
generally thought of as idioms can occur in a variety of guises (Mlleg
jaaojagaulled), and judgements of the acceptability of distorted
Idioms are very slippery and subject to context. This is not the place
for an extended discussion of idioms (Makkai 1972 is one such study),
so we can leave the discussion of sentences (53)-(56) with the
tentative conclusion that (53)-(54) are conceptually like 'tell', while
(55)-(56) are basically idiomatic.
Another type of odd 'give' expression involves what might be called a
“reified action“. These fall into two types, the first of which is
illustrated below:
57) a. IAIadooaILaMarooaaata.
b. WWW.

58) a- W.
b. Ihooabldoflmcshradahissiromfixlvaster

59) a. Wad
b-Il 'II! 'I [I [It

66

Some of the 'get'/'receive' sentences are more natural than others, but
most of them seem to form reasonably good doublets. The second type
of reified action Is illustrated In the following sentences:

62) a. W

b. SoLootLeroeiiredarldsJaaMasoratiiromfiaar
63) a. MaaLoaxaEralanooLataraaLsaarllaoooolot
b. War
The sentences with 'receive' are perhaps less natural than the ones
with 'get', but overall the (b) sentences are certainly conceivable.

Green (1974) points out that the above sentences with 'give' can
generally be paraphrased using other verbs. For (57)-(61) we have the
following:

64) a. madamedMaroo-

b- lexssterkissediaooabx-

Actually, there is a difference: the sentences in (64) can have either a
singular or an iterative moaning, while the sentences with 'give' can

only have the singular, one-time reading. This can be seen by adding an
appropriate adverbial phrase to each of a pair of sentences, i.e. Halon
ouaoaed.Ed_Ia.tae.aoso.tor.mro_aours vs. ?Baloaoaire.Ed_aouaoa.Ia.tho
Winona; (It Is possible, though, to make the direct object of

a 'give' sentence plural and make it iterative, e.g. W
W). Other than this aspectual difference, each

sentence with 'give' means essentially the same as the corresponding

67

sentence without 'give', yet taken as a group, the verbs in (64) seem to
have little in common except the fact that they are transitive.

For the sentences in (62)-(63), there is a rough paraphrase with lot
and a subordinate verb phrase:

65) a. W

b. WW
Again, the correspondence Is not exact, since the sentences with 'give'
imply more of a one-time event than do the sentences with 'lot', but the
idea Is essentially the same.

One way of analyzing the above facts would be to consider the
sentences In (57)-(63) to be merely alternate realizations of the same
ideas expressed in (64)-(65), with an added provision that the action is
a single event when the direct object ('kiss', 'punch in the nose') is
singular. If so, then the sentences with 'give' would have essentially
the same conceptual representation as the corresponding sentences
with other verbs, perhaps with some kind of aspectual specification
added (there is no room here to speculate in detail on what such
specification might look like, but Kathman ms and Jackendoff
1987a:398-402 explore some possibilities). For W
W and Why, this basic representation
might be something like this:

[EMIKISS ([SYLVESTER], [BABY])]

Similarly. both fianroaanolaadolaaMaseratI and GaaLIotSoLrido
mam would be something like the following:

[EMILET ([GARY], [EMISOL RIDE IN A MASERATI])]

The alternate realization with 'give' would then be possible only with

68

the appropriate aspectual specification.

What about the 'get'/'receive' sentences, though? The above analysis
does not seem able to account for them. One possible explanation is
analogy; since 'give' is so often paired with 'receive', speakers might
tend to make a parallel idiom with the latter. However, if this is so,
why do idioms such as “give the finger“ or verbs such as 'refuse' and
'show' not allow a parallel expression? Our subject selection principle
Is In danger of becoming vacuous If we allow such apparently arbitrary
exceptions, although some way might be found around the trouble.

Another alternative, though, is to consider the conceptual structure
of the sentences In question to be like the others we have seen, with a

source, goal, and transferee. Thus, W155

would be as follows:

[CAUSE ([SYLVESTER], [co ([KISS], [FROM ([SYLVESTER])
T0 ([BABY])])DI

The transferee In this case (KISS) could be thought of as an Event which
the source causes the goal to experience, and which, since it must be
singular and discrete as outlined above, can be treated like a Thing for
conceptual purposes. As we have seen already, it is not necessary for
transferees to be concrete, and some evidence for this analysis is
provided by the possibility of using these expressions in some marked
constructions (9-0- MaLBaloaoaxaEdmsaouaoamaaaoso. though
other constructions are not so good). Such examples would probably
have to involve a different semantic field from what we have already
seen, perhaps called the Discrete Action field. '
The disadvantage of the above analysis is that it disregards the

69

advantage of the other one, namely the postulation of (at least
partially) identical conceptual representations for sentences which
apparently describe the same thing. Even so, it might be possible to
indicate the similarity by specifying the Event-transferee in some way,

perhaps by means of indices:

[CAUSE ([SYLVESTER], [GO ([1 KISS j], [FROM ([SYLVESTERJ)
To ([BABlellllll

This approach opens up a Pandora's box of questions, though, so the
matter will not be pursued further here.

Each of these proposed analyses has its good and bad points and a
certain amount of evidence In its favor; even so, the second one is'
perhaps slightly more attractive. Accordingly, we can tentatively take
' it to be correct, while recognizing that the analysis with identical
representations has its advantages. Perhaps some other analysis is
possible which combines the best of both worlds, but It will have to
wait for further research. This group of doublets is the most
troublesome we have seen, and it illustrates the difficulties inherent.

in any detailed exploration of conceptual structure.

flammaaraadfioaolusioas

It is evident that the analysis of three-place verbs and lexical
doublets involving them Is more complex than it might seem at first
glance. Those who have bothered to analyze such doublets have tended

to focus on situations involving transfer of possession of a concrete

transferee (i.e. Whoa/WW.
lI llSl II II 'II II ISI I II l'l II

7O

goaciLaolgaLlramNoam), but, as we have seen. these are just the tip

of the Iceberg. The transferee can be abstract as well as concrete, and,
perhaps more Interestingly, several reasonably well-defined groups of
three-place verbs do not enter into doublets at all.

The principles allowing lexical doublets may be summarized as
follows: (1) When there is motion along a conceptual Path, and both
endpoints of the same Path are capable of being realized somologically
as the unmarked subject as specified below, then a lexical doublet
results. (2) The initiator of the motion along the Path is automatically
realized as 'S' if It is specified in conceptual structure, and the
endpoints of the Path can also be realized as '8', subject to different
restrictions according to the semantic field involved. (3) Different
semantic fields place different restrictions on the types of conceptual
entities which can occur as initiator, transferee, and endpoints. (4)

Furthermore, the conceptual motion Involved must be real and not just

potential, as we saw with examples like Nom'llaammisodmnao
Sudstealaad.

The above principles account for the great majority of cases, but
there are still some areas which are ripe for further research. For
example, there is still the question of may the various restrictions
occur. This thesis has been concerned with determining when
three—place doublets do and do not occur and offering some partial
explanations along the way, but it remains to be seen whether still
more general principles could be found which would explain, for

example, why Path endpoints can be realized as unmarked subjects.

The Discrete Action expressions WWW

71

have already been mentioned as presenting problems; space limitations
have prevented a more detailed study of them here, but It would
certainly be Interesting to explore their relation to the more

conventional three-place doublets. Another potential area of inquiry is
two-place doublets (such as 'like'l'please') and how they relate to the
three-place doublets to which this study has been limited. Dowty (ms)
contains some interesting proposals in this direction, but a complete
exploration of two—place doublets Is beyond the scope of the present
thesis.

One of the goals of this thesis has been to systematically explore
the various types of three-place lexical doublets which occur and to
Show that an independent system of conceptual structure can lead to
generalizations about such doublets, and to the extent that it has done
so, it has succeeded. Most of the details of realization between the
semantic and semological levels have been omitted so that more
attention could be paid to the general issues involved, although many of
these details could doubtless be filled in with further research. Any
research Into semantics has a tendency to invite controversy;
nevertheless, we must continue to forge ahead in the hope that
something good will come out of such research. The present thesis has
been one attempt to confront a semantic Issue in a systematic way and

contribute to our knowledge of it.

APPENDIX

72

 

  
 

A CAUSE A

A B
JO-N
SPATIAL
m A

A B

A
A
CANTALOUPE m
JOHN

 

PETE

 

JOKE ABOUT
PHONETICIANS

73

 

CAUSE

 

 

BEING ON TRIAL

A

VALUABLE LESSON

74

 

CAUSE

A

FFIN

BEING ON TRIAL

 

OLUE

75

BILL

BLACKEYE

 

BILL LEG’QD

BIBLIOGRAPHY

76

BIBLIOGRAPHY

Anderson. John M. 1971. We
Ihm. Cambridge: Cambridge University Press.

Bar-Hillel, Yehoshuah. 1967. 'Dictionaries and Meaning Rules.’
EduadatloasoLLaaoIlado 3: 409-14

Bannise, J., and J. Perry. 1985. 'Shifting Situations and Shaken
Attitudes.‘ LIaouIstIcsaadEhIlosooay 82105-61- '

Bolinger, Dwight. 1965. 'The Atomization of Moaning.‘ Languaga 41:
555-73.

Camap, Rudolf. 1956. Moanlngannflaaessily. Chicago: University of
Chicago Press.

Chomsky. Noam. 1965. W. Cambridge.
Mass: MIT Press.

Chomsky. Noam. 1981. LooturssoafioxemmoaLaadfladIao -
Dordrecht: Foris.

Christie, William. 1977. ASIratItioatioaalMIelrLoLLIaouistic
Chanda. Edward Sapir Monograph Series in Language, Culture, and

Cognition, 4. Lake Bluff, Ill.: Jupiter Press.

Culicover, Peter, and Wendy Wilkins. 1986. 'Control, PRO, and the
Projection Principle.‘ Language 62:120-153.

Dik, Simon C. 1980. 'Basic Principles and Application of Functional
Grammar.‘ in Moravcsik and Wirth, 45-75.

Dowty, David. 1988. 'On the Semantic Content of the Notion 'Thematic
Role'.'InG. Chierchia, B. Partee, and D. Tumor, eds. ,Erooony

 

77

Dowty, David. ms. 'Thematic Proto-Roles, Subject Selection, and
Lexical Semantic Defaults.‘ 1987 LSA Colloquium Paper.

Fillmore, Charles. 1968. 'The Case for Case.’ in Emmon Bach and

80de Harms eds. UahrsrsalsJaLiaouistIolaeonr New York:

Holt, Rinohart, and Winston, 1-88.

Fillmore, Charles. 1977. 'Topics In Lexical Semantics.‘ in RogerW.

Cole. ed .CuaeaLlssussJaLIaouistIolhoom Bloomindton

Indiana University Press, 76-138.

Fodor, Jerry. 1975. W. Cambridge: Howard

University Press.

Fodor, Jerry, Merrill Garrett, E. Walker, and C. Parkes. 1980. 'Against
Definitions.’ Qagnltjaa 8:263-367.

Green, Georgia. 1974.
Bloomington: Indiana University Press.

Gruber, Jeffrey. 1 976.
Amsterdam: North-Holland.

Halliday. M. A. K. 1985. AaJatroduotIdatoEuaotIooaLGrammar

London: Edward Arnold.

Jackendoff, Ray. 1972.
Grammar. Cambridge: MIT Press.

Jackendoff, Ray. 1983. W. Cambridge: MIT

Press.

Jackendoff, Ray. 1987a. 'The Status of Thematic Relations in
Linguistic Theory.‘ W 18.3: 369-411.

Jackendoff Ray 1987b. WWW.

Cambridge: MIT Press.

78

Jackendoff, Ray. forthcoming. 'What is a Concept, that a Person May
Grasp lt?’ to appear in Fonlm on 'What is a Concept?', madam
mum -

Josperson. Otto. 1933(1964).Essoatials.of_anlisa_Grammar
University, Alabama: University of Alabama Press.

Kathman, David. ms. 'Aspectual Meaning in Russian.‘ Michigan State
University.

Katz, Jerrold. 1967. 'Recent Issues in Semantic Theory.‘ Eoundations
al.].anguaga 3: 124-194.

Katz, Jerrold. 1972. Samaaljalnaanr. New York: Harper 8. Row.

Katz, Jerrold, and Jeny Fodor. 1963. 'The Structure of a Semantic
Theory.‘ Languaga 39:170-210.

Lakoff. George. 1987. WWW Chicaoo=

University of Chicago Press.

Lakoff, George, and Mark Johnson. 1980. W.
Chicago: The University of Chicago Press.

Lamb. Sydney M. 1966.9.utliaoot51ratiflcatioaaLfirammar
Washington, D.C.: Georgetown University Press.

Lamb, Sydney M. 1983. 'On Determining the Number of Strata' In

Linguistic Structure.‘ in John Morreal,ed.,]]1e_allnln_LAQu_S_Egnim.
Columbia, 8. C.: Hombeam Press, 189-203.

Langacker, Ronald. 1988. 'An Overview of Cognltlve Grammar! In

Brygida Rudzka-Ostyn, ed. ,Ioolcsaanooaitiyafirammar
Amsterdam: John Benjamins, 348.

Lewis, David. 1972. 'General Semantics} in D. Davidson and G. Harman,
eds., Wm. Dordrecht: Reidel, 169-218.

79

Lockwood David 1972 IntmduotioatofitratifcatioaaLLlaouIsfios
New York. Harcourt Brace Jovanovich.

Lockwood.David.ms.latroduotIoa.to.StratifioatIoaaLLlaouistlos
IE . IEI'I' l-

Longacre, Robert. 1976. MW. Lisse: The
Peter de Ridder Press.

Lyons John 1963 StmoturaLSemaatiosLAaAaabrsisoLEaaottao
Wan. London: Oxford University Press.

Lyons, John. 1977. Samanligs. Cambridge: Cambridge University
Press.

Makkai, Adam. 1972. W. The Hague: Mouton.

Marr, David. 1982. Vision. San Francisco: Freeman.

McCawley, James. 1968. 'Lexical Insertion' In a Transformational
Grammar without Deep Structure.‘ in B. Dardon, C. -J. N. Bailey, and

A Davison. eds. WW9

Lingulsljgfigoioar. Chicago: Department of Linguistics,
University of Chlcago 71-80.

McCawley, James. 1980. 'An Un-Syntax.‘ in Moravcsik and Wirth,
167-93.

McCawley. James. 1988.10Wmm
Chicago: University of Chicago Press.

Miller, George A., and Philip Johnson-Laird. 1976. Languagaang
Ramayana. Cambridge: Cambridge University Press.

Moravcsik, Edith A., and Jessica R. Wirth, eds. 1980.91.1er
W New York:

Academic Press.

8O

Pike, Kenneth, and Evelyn Pike. 1976. firammaliaalAnalysja. Dallas:
The Summer Institute of Linguistics.

Redford, Andrew. 1 988. IranslgrmatjanaLGLammar. Cambridge:
Cambridge University Press.

Rappaport, Malka, and Beth Levin. 1988. 'What to Do with
Theta-Roles! In Wendy Wilkins, ed., W
W. San Diego: Academic Press, 7-36.

Reich, Peter. 1969. 'The Finiteness of Natural Language.‘ Language 45:
831-43.

Reich, Peter. 1970. 'The English Auxiliaries: A Relational Network
Description.‘ CanadiaadoumaLoLLiaouistics 16: 18-50.

Rosch, Eleanor, Carolyn Mervis, W. Gray, D. Johnson, and P.
Boyes-Braem.1976.'Basic Objects In Natural Catogories.‘

CooaahraEsxcaoloox 8: 382-439.

Sampson, Geoffrey. 1980. W- London: Hutchinson
8. Co.

Schreyer, Rudiger. 1980. 'Nodes and Nets: Explorations In Relational
Network Notation.‘ In James Copeland and Philip Davis, eds,

W Houston: Rice
University, 163-83.

Staal, J. S. 1967. 'Some Semantic Relations Between Sentoids.‘
EouadatioasoLLaaouaoo 3: 66-88.

Sullivan, William J. 1980. 'Syntax and Linguistic Semantics in
Stratificational Theory.’ in Moravcsik and Wirth, 301-27.

Talmy, Leonard. 1976. 'Semantic Causative Types.‘ in Masayoshi

Shibatini, ed.. IaofirammaLoLQausatlxaCoastmofioasJSxatax
W. New York: Academic Press, 43-116.

81

Van Valin, Robert D., and William A. Foley. 1980. 'Role and Reference
Grammar! in Moravcsik and Wirth, 329-52.

Warren, R. M. 1970. 'Perceptual Restoration of Missing Speech
Sounds.‘ Science 167: 392-3.

"Illllllllllllll'lllllllllll“