Open Journal of Modern Linguistics
2013. Vol.3, No.3, 233-251
Published Online September 2013 in SciRes (http://www.scirp.org/journal/ojml) http://dx.doi.org/10.4236/ojml.2013.33032
Semantic Features in Argument Selection
Izchak M. Schlesinger
Hebrew University, Jerusalem, Israel
Received February 27th, 2013; revised April 2nd, 2013; accepted April 9th, 2013
Copyright © 2013 Izchak M. Schlesinger. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
One of the problems that has to be dealt with by theorists of early language acquisition theory is the mis-
match between semantic constructs, like Agent, and syntactic ones, like subject. It is proposed that the
linguistic system is based on semantic features that are more fine-grained than thematic roles, and that se-
lection of subject and direct object can be accounted for by merely four semantic features. These features
are conceived of as properties of participants in the lexical entries of verbs, and in this respect, too, they
are unlike thematic roles, which are ascribed to NPs in sentences. Thematic roles play a part only in the
realization of certain other arguments, notably, the oblique object. It is shown that this different treatment
of direct and oblique objects permits a parsimonious explanation of certain linguistic regularities that have
posed problems for other theories. Early language acquisition can be explained in terms of the acquisition
of these semantic features, and this account thus supersedes the semantic assimilation hypothesis pro-
posed previously to deal with the lack of congruence between thematic roles and syntactic categories.
Keywords: Semantic Features; Thematic Roles; Linking Rules; Language Acquisition
Introduction
Several theories of argument selection have followed Fill-
more’s (1968) seminal paper and dealt with the semantic-syn-
tactic interface in terms of what have been variously called the-
matic relations, thematic roles, or θ-roles. The construct of the-
matic roles and that of a thematic role hierarchy have come un-
der criticism from various quarters (e.g., Davis & Koenig, 2000;
Dowty, 1989, 1991; Gropen et al., 1991a; Hoekstra, 1992; Le-
vin & Rappaport Hovav, 2005; Newmeyer, 2002; Schlesinger,
1995: pp. 28-29; Zubizaretta, 1987, among others). As Andrews
(2007: p. 140) observes, “… no presently known system of se-
mantic roles can be applied in a comprehensive and convincing
manner”1. Instead of thematic roles, Van Valin (2005) therefore
posits more global “macro-roles”. Some other theorists propose
less global notions, such as semantic features (see, e.g., McRae
et al., 1997; Zubizaretta, 1987). In Jackendoff’s (1987: p. 409)
system, thematic roles are derived notions “argument positions
in conceptual structure” that are constituted of more basic se-
mantic concepts which are akin to features. Others view fea-
tures as elements into which thematic roles are decomposed
(Rozwadowska, 1988), as “entailments” characterizing “proto-
agent” and “proto-patient” roles (Dowty, 1991), or as occupy-
ing a level preceding that of thematic roles (Schlesinger, 1995).
In the present paper it is argued that realization of the subject
and the direct object can be accounted for without resorting to
thematic hierarchies or to conceptual structures like those pro-
posed by, e.g., Jackendoff (1990b) and Kiparsky (2001). In-
stead, rules defined over certain semantic features are formu-
lated, which link NPs to the subject and the direct object. The-
matic roles are conceived of as playing a part only in the reali-
zation of certain other arguments. This theory affords a more
parsimonious explanation of linguistic phenomena and solves
certain problems encountered by a theory of early language ac-
quisition2.
The first section of this paper introduces the semantic fea-
tures figuring in the present system. Then a theory is developed
in which features residing in the lexical entries of verbs deter-
mine the selection of subject and direct object. Certain linguis-
tic regularities are accounted for by assuming that the realiza-
tion of oblique objects differs in that it involves thematic roles.
The final section shows how the language learning child can ac-
quire such a feature-based grammar.
Features of Subjects
The purpose of this section is to introduce the features that
function in my system. Examples will be given of features that
characterize subjects, but some of these features appear not
only in subjects. The linking rules that select subjects and direct
objects will be discussed in the section on “Selection of sub-
jects and objects”.
Cause
The subjects of the following sentences pertain each to the
cause of the activity referred to by the verbs:
1And Chomsky (1982: p. 89) once remarked in an interview: “I never know
how people are able to pick out thematic relations with such security, I
can’t”.
2The present theory is an extensive modification of some ideas first ex-
p
ounded in Schlesinger (1995). Among others, the system of features has
been much simplified, the hypothesized level of “cases” mediating between
features and syntactic structures has been abandoned. Some criticisms o
f
that book by Rosta (2002) have also been taken into account.

I. M. SCHLESINGER
234
1) a. Sybil drank her coffee.
b. Ron pushed Jack.
c. My uncle refused the offer.
d. Sheila is swimming.
e. The girl cried.
f. Bill was snoring on the couch.
Each sentence in 1) describes an activity caused by the entity
referred to by the subject, and so the feature Cause is accorded
to each of the subjects (or, more precisely, to the elements in
the lexical entries of the respective verbs that are linked to the
subjects. Cause characterizes internal causation, as in 1d), 1e),
and 1f), as well as agents causing a change in another entity3.
Control
The animate subjects in 1) are not only the cause of the ac-
tivity referred to, but also have Control over it (see Siewierska
1991: 47-49 on this feature). Our feature Control differs some-
what from Dowty’s (1991) entailment “volitional”, as is shown
by the following examples:
2) a. Celia refrained from interfering.
b. Ronald is sleeping in his armchair.
c. Our community tolerates eccentric behavior.
d. My aunt underwent surgery.
Celia is in control in so far as she can desist from refraining
(that is, she can start to interfere), and one probably can, at least
in some cases, voluntarily stop sleeping, and decide to behave
intolerantly. And the lady referred to in 2d) by my aunt may
refuse to be operated on or escape from the hospital.
Cause and Control are closely related features; they differ
mainly in the phases of the activity which they focus on. Stated
roughly, Cause initiates the activity and Control steers it or
even terminates it. In fact, Cause and Control usually go to-
gether. Most of the subjects in 2) have also the feature Cause.
Refraining is “caused” by Celia, just as crying in 1e) is caused
by the girl. An exception is 2d): the lady does not cause any-
thing (surgery is performed by surgeons), but she may be said
to be in control. There are also subjects that have only Cause
and no Control, as for instance the subject of snore; see 1f).
Change of State or Location
When a NP refers to an animate or inanimate entity under-
going a change of state, we will say that it has the feature
Change.
3) a. The ice will melt.
b. Their son has grown up.
c. The hamster died in his cage.
d. The rainbow slowly vanished.
e. The waiter slipped on the wet floor.
The ice, their son, and the hamster are said to undergo a
change of state, the rainbow and the waiter, a change of loca-
tion4.
Some of the verbs in 1) also involve movement in addition to
Cause. Further on, I discuss the question of whether they have
the feature Change.
Attributee of State
Some verbs, like those in 4), refer to states or to properties
that are attributed to the subject5.
4) a. This box contains small coins.
b. This car needs an overhaul.
c. Smith is a pauper.
d. You really deserve a special treat!
e. The organization lacks funds for rebuilding.
f. Ken likes ice cream.
g. She fears him.
h. He has a mole on his right shoulder.
I propose to introduce a feature that has not been discussed in
the literature, to my knowledge: attributee of a state or property,
or Attr-St, for short. There is no cause-and-effect relationship
between the subjects and predicates of these sentences. One does
not conceive of the box as being the cause of containing any-
thing or of the car as causing its needing anything; instead, con-
taining and needing are states or properties attributed to them6.
Examples 4f) and 4g) are of so-called mental verbs (or
“psych-verbs”) in which the experiencers are sentence subjects.
The subjects of these and other such verbs admire, respect, pity,
hate, loathe, and dread, among others have the feature Attr-St.
Some mental verbs can appear in the imperative form, as in 5).
Presumably this is because one may have some degree of con-
trol over one’s emotions, which suggests that these verbs have
also the feature Control7.
5) a. Don’t pity me!
b. Don’t hate your enemy!
Conversely, the subjects in 2a), 2b), and 2c) should be as-
signed the feature Attr-St in addition to Control, because the
verbs attribute properties to their subjects.
The subjects of verbs denoting posture, like those in 6), may
be viewed as having several features.
6) Herb is sitting/lying on the couch.
Usually, sitting and lying result from one’s deliberately as-
suming a posture, and one can also desist from assuming it (or
determine one’s manner of doing so). Subjects of the verbs sit
and lie thus have the features Cause and Control, in addition to
Attr-St, and so have some other stative verbs, like stand, and lean8.
There are also some unclear cases. For instance, does the
subject of find have the feature Cause (i.e., is finding an activity,
like seeking) or Attr-St? One should not assume the existence of
a sharp boundary line between these two features.
On the Nature of Features
Three classes of verbs have been described here: activity
verbs, the subjects of which have Cause and/or Control, proc-
ess verbs, and state verbs, the subjects of which have Change
and Attr-St, respectively. Not included in our list of features are
3These distinctions between kinds of causes, as well as those proposed in
Schlesinger (1995), are not needed for the purpose of accounting for subject
and object selection. The subjects in 1) seem to be close to what Van Valin
and Wilkins (1996) call “effectors” and Levin and Rappaport Hovav (1995:
p. 135) call “immediate causes”.
4The set of verbs having subjects with the feature Change overlaps more or
less with that of unaccusative verbs.
5Phrasal predicates, like that in 4h) will not be further discussed in the pre-
sent paper.
6The direct objects in these sentences (or in any sentence, for that matter) do
not have this feature. Nothing is expressly attributed to small coins in (4a);
one can only infer that they are in the box.
7In a rating study (Schlesinger,1992), it has been shown that adult speakers
j
udge subjects of these verbs to have a greater degree of control than their
direct objects.
8These verbs may be also be predicated of inanimate subjects by a kind o
f
metaphorical extension. See section on NP matching for the mechanism that
operates here.

I. M. SCHLESINGER
some of those found in the literature, like animacy and sen-
tience9. These and some other features can possibly account for
some linguistic phenomena that are not explainable in terms of
the above four features, but they are not needed for stating the
linking rules that select subjects and direct objects. In the fol-
lowing, this will be shown to be true for English. Not only lan-
guages that do not have subjects (like Tagalog, which has a pre-
dicate-topic structure; Schachter & Otanes, 1972), but also
other languages with a subject-predicate structure may require
additional features (see, e.g., Hopper & Thompson, 1980).
The theory proposed in the following sections is more parsi-
monious than theories based on thematic hierarchies, and ac-
cordingly the difficulty of learning the basic linking rules of the
native language is diminished, as shown in the last section. It
might be argued that the foregoing four features are thematic
roles in disguise. Note the considerable overlap of features Cause
and Control with Agent, of Change with Theme or Patient, and
of Attr-St with Experiencer. However, unlike thematic roles,
features are relatively unequivocal and easily identified. To de-
cide whether something undergoes a change is simpler than de-
ciding whether it is a Theme. This is partly due to there being a
larger number of thematic roles than of the features proposed
here; the latter are more basic concepts, each of which may be
common to a number of roles. Further, while definitions of the-
matic roles are typically vague, and as a result cannot be ap-
plied easily “outside classes of the verbs which have been used
to motivate the θ-roles or are extended in unprincipled ways to
new cases” (Rappaport & Levin, 1988: p. 8), this will be seen
not to be true of features.
Another basic difference between thematic roles and our fea-
tures is that while the former are usually conceived of as apper-
taining to NPs, features are characteristics of participants in the
lexical entries of verbs, as will be seen in the next section.
The Linking Problem
In the next section it will be shown how linking rules for
subject and direct object can be formulated in terms of the four
features discussed in the foregoing: Cause, Control, Change,
and Attr-St. In the present section, I show that—unlike the rules
in current theories, which are defined in terms of characteristics
of the NPs—the rules that select the subject and direct object
can be defined on features of lexical entries of verbs. But first,
some of the concepts figuring in the present theory will be ex-
plained.
Lexical Entries
Participants and Features
Consider two simple sentences:
7) a. Doris broke the window pane.
b. Dan spilled the wine.
The subjects and direct objects of these sentences are deter-
mined by the lexical entries of the verbs break and spill, re-
spectively. Every lexical entry of a verb contains a “scene”. The
scene for break, for instance, is a mental representation of
breaking, and it includes mental representations for each of the
participants in the breaking activity, namely, the one who
breaks and the object that is broken10.
It is important to bear in mind that in the present system the
term “participant” does not refer to arguments, like subject
and object (as Hopper and Thompson 1980, for instance, use
this term). Neither does it refer to NPs or to their referents:
instead, NPs are mapped into participants, as will be shown
below.
There are many kinds of breaking: one can break chairs,
pencils, windows, threads, etc. Therefore the mental representa-
tion of the action of breaking must be schematic (just as dictio-
nary definitions ought to be sufficiently abstract to accommo-
date most instances of the defined term). The participants in-
cluded in the mental representation will also be schematic:
there are not only several ways of breaking but also kinds of
breakers and of things that can be broken.
Each participant has certain syntactically relevant features.
The following examples illustrate formulas for lexical entries
(the left-to-right sequence of the participants being a matter of
convenience and having no theoretical significance whatsoever):
break: “breaker” (Cause, Control), the thing broken (Change)
spill: “spiller” (Cause, Control ), the material spilled (Change)
No further decomposition of predicate verbs is needed for
subject and object selection. The linking rules selecting subjects
and direct objects are defined for features of the participants in
the lexical entry and will be called feature linking rules. One
such rule determines that the feature Cause is linked to the
subject and another one that the feature Change may be linked
to the direct object; see the next section for a discussion of fea-
ture linking rules.
The participant in a lexical entry which, after the appropriate
feature linking rules have applied, corresponds to the subject (in
active sentences) e.g., the “breaker” and the “spiller” will be
called here subject participant. Similarly, the participant that
ends up as direct object (the thing broken, the material spilled)
will be called the object participant. These terms are merely
convenient labels; the participants themselves have neither sub-
ject nor object character prior to the application of feature link-
ing rules.
An assumption made by the present theory is that a lexical
entry of a verb can include at most two participants: the subject
participant and the object participant. Oblique objects are not
realizations of participants but rather are arrived at by a differ-
ent kind of linking rules; the rationale for this will be discussed
in the section on Roles of NPs (below). Transitive sentences
contain two participants, each of which may be realized by one
NP, as in 7), or by more than one, as in 8):
8) Jim and John broke the vase, the saucers, and the salt
shaker.
The lexical entries of intransitive verbs contain only one par-
ticipant, which will be realized as subject, e.g.,
cry: the one who cries (Cause, Control)
melt: the material that melts (Change)
Verbs that have a subject participant with the feature Change,
like cry and melt, are usually intransitive, but there are some
exceptions, e.g.,
9) The pensioner sustained an injury.
sustain: affected entity (Change), source of affecting (ø)
Here the object participant does not have any one of the fea-
tures introduced so far—Cause, Control, Change, and Attr-St.
The symbol ø—the “default feature”—stands for the absence of
9McRae et al. (1997) propose “features” which, unlike our features, are close
to what Dowty (1989: pp. 76-77) calls “individual thematic roles”, i.e. verb-
specific (though each may characterize a large proportion of verbs in the
language).
10Note that “scene” is used here for the mental representation of a verb, and
not in the sense used b
y
Fillmore
(
2003:
pp
. 209-213
)
.

I. M. SCHLESINGER
236
any one of these features11.
Lexical entries of verbs that have expletive subjects (like
rain) have no featured participants.
Assigning Features
A comment is in order regarding the feature Change. This
feature is broadly conceived in the present system. Both af-
fected and effected objects have Change:
10) a. David damaged this machine.
b. David constructed this machine.
Change may stand for change of state, as in the object par-
ticipant in 10a), or for change of location, i.e. motion, as in the
object participant of spill. There are also mental changes (as in
the direct objects of frighten, and surprise), changes of status
(as in the direct objects of elect, nominate, appoint, promote,
convict, indict). Further, as illustrated in 10b), Change may
apply to entities which come into being as a result of the activ-
ity referred to by the verb, and these may also be abstract enti-
ties (cf. pass a resolution, devise a strategy, deliver a lecture).
Changes can be scalar (as in mature or learn) or non-scalar (as
in break or laugh). All these distinctions do not appear to be
linguistically relevant as far as subject and object selection go,
and will be ignored in this paper.
In the case of many verbs it is not immediately clear whether
their lexical entries contain a given feature. For example, the
subject participants of many verbs involve a limited degree of
movement in addition to Ca use and/or Cont rol. To break some-
thing, one moves one’s arms (or legs) and spilling is normally
accompanied by some motion. The feature Change character-
izes the subject participants of only those verbs that refer to
activities in which movement or a change of state are essen-
tially involved, such as swim, run, walk, enter, jump, and not of
verbs like break, spill, or drink, where the movement or change
of state are incidental. Movement is not part of the meaning of
the actor of break, spill, and drink, whereas in the case of swim,
run, and walk, change of location is part of the meaning. The
statement He ran without moving is an oxymoron (the first part
of the sentence contradicts the second one), but He drank wine
without making the slightest movement describes, at most, a
physical impossibility. I propose this here as a rule of thumb,
but naturally there will be many unclear cases.
Similarly, Change is assigned to an object participant only
when there is an “essential” change in it. For instance, the ob-
ject participant of break has, but that of hit does not have the
feature Change. This is because 11a) is self-contradictory due
to the meaning of break, which involves a change in the object
broken, whereas 11b) asserts what is perhaps a physical impos-
sibility, but the sentence is not self-contradictory.
11) a. *Doris broke the window pane, but it remained intact.
b. Doris hit the TV screen, but it remained intact.
There will of course be borderline cases, namely verbs where
it is hard to decide whether the change (in either the subject or
the object participant) is essential or not, but as should become
clear further on, this will not affect the application of linking
rules12.
Some verbs may refer to events that either do or do not in-
volve a certain feature. For instance, a person who is lying on
some surface is obviously in a state, and so the subject partici-
pant of lie has Attr-St; cf. example 6). But against the claim
made there that lie also has the features Cause and Control, one
might want to argue that an unconscious person, for instance,
may be put on a stretcher or on a bed (and so does not cause
anything) and left to lie there (without having any control over
the situation). A related problem, which will be dealt with fur-
ther on, may arise with verbs that may take either animate or
inanimate subjects. However such objections overlook the fact
that features are assigned to lexical entries and not to verbs in
particular contexts13. The lexical entry of a verb is determined
by its standard use, and the verb lie, as used standardly, has
subjects referring to the person or animal that is the (internal)
cause of assuming this posture and is in control of it.
There will be verbs which may raise doubts as to their stan-
dard use. Fortunately, this will normally not be problematic for
subject selection, as will be seen from the formulation of the
subject linking rule further on.
Features Are Independent of Context
Features of participants reflect the meaning of the verb, and
this means that they are independent of the sentence in which
the verb appears. This is an important point that is easy to over-
look, as the following examples show.
12) a. The police suspected the pickpocket.
b. The police accused the pickpocket.
Suspect is a mental verb, and its subject participant has the
feature Attr-St; cf. 4f) and 4g). Accuse has a subject participant
with Cause and Control. In both 12a) and 12b), the police is
therefore the subject. But what about the pickpocket? Plying his
trade led to his being suspected or accused, and so one might
maintain that the pickpocket caused the event referred to by the
verb (and if so, the pickpocket competes with the police for
subject position). A similar question might arise also with many
other verbs, like blame, praise, arrest, punish, and reward.
But this argument is based on a faulty assumption. Lexical
entries are not determined by the context in which the verb
appears. It is not NPs or their referents that have the feature
Cause, but participants in the lexical entry. Sentences like 13)
show that the meaning of suspect or accuse does not imply that
the person suspected or accused is necessarily the cause of what
the verb refers to.
13) a. The police suspected an innocent bystander.
b. The police accused an innocent bystander.
Each of the lexical entries of suspect and accuse therefore
includes one participant eligible for subject position and an-
other participant with the “default feature”, and so have the
lexical entries of blame, praise, arrest, punish, and reward.
suspect: “suspecter” (Attr-St), “suspectee” (ø)
accuse: “accuser” (Cause, Control), “accusee” (ø)
It is important therefore to keep in mind that features are
context-free. In this respect, thematic roles are usually concei-
ved of differently; they are “components of the mental repre-
sentation of objects and concepts” (Wilkins, 1988: p. 191), and
not components of the meaning of verbs, like features. As a
result, truth-conditionally equivalent sentences have identical
11Cf. Fillmore’s (1968) “Objective”, which is a catch-all case. Some writers
use “Theme” in a similar manner.
12In a lexical entry formula, I will sometimes leave out Change, even where
a good case can be made for including it. In most instances, adding Change
to features of the subject participant will not make any difference to linking,
as will become clear in the section on sub
j
ect selection.
13Although it is the participants in lexical entries, and not NPs that have
features, it will be convenient to use the somewhat imprecise expression that
“the NP… has the feature F”, which should be taken to mean “the partici-
pant which is realized as the NP… has the features F”. Occasionally I will
write “the sub
j
ect
/
direct ob
j
ect has the feature
F
”.

I. M. SCHLESINGER
role assignments. On this view, assignment of a thematic role
may depend on the sentential context. Thus, when the sentence
states that an activity was performed accidentally, the actor may
be assigned a thematic role that differs from that of a sentence
which states that the activity was performed for some purpose.
The assumption that thematic roles are sensitive to the part
played by a NP in the situation described may give rise to
problems. For instance, Grimshaw (1990: pp. 19-21) discusses
two sentences describing the same situation:
14) a. The storm frightened us.
b. We feared the storm.
Assuming Baker’s (1988: p. 37) theta criterion, Grimshaw
argues as follows. The storm is Theme in both sentences, since
the sentences are near-paraphrases of each other. Likewise, we
(us) is Experiencer in both. Since Experiencer precedes Theme
in the thematic hierarchy, an explanation is required for the fact
that in 14a) the storm, the Theme of frighten, is subject. Grim-
shaw thus focuses on the situations described by the NPs in
these sentences, rather than on the verbs. Her reasoning is based
on the assumption that if a NP has the thematic role R in a gi-
ven sentence S, it must have the same role in every sentence that
paraphrases S, or as Dowty (1991: 563) writes: “… any truth-
conditionally equivalent sentence has the same role assignments”.
In a section the heading of which is called, appropriately,
“The inadequacy of thematic reanalysis”, Grimshaw accounts
for the fact that subject selection in 14a) does not conform to
the hierarchy of thematic roles, by stating that the storm has
causal status, which is why it can be subject14. According to her
analysis, then, thematic roles cannot do the job by themselves;
the feature Cause comes in through the back door. Further on, it
will be shown how 14) is dealt with in the framework of the
present theory.
NP Matching
In the present system, then, features are attributes of partici-
pants in lexical entries, not of NPs. NPs may match participants
in the lexical entry of a verb. For the purpose of a schematic
illustration that does not pretend to mirror actual performance
of language users, let us assume that a speaker intends to state
that Doris has broken a window pane. Expressing this idea
involves choosing a verb that describes the activity, and break
fills this bill15. The lexical entry of this verb is break: “breaker”
(Cause, Control), thing broken (Change).
This lexical entry prescribes that two NPs pertinent to the
event be selected, one for each participant, the “breaker” and
the “thing broken”. Doris can be matched with the “breaker”
and the window pane, with the “thing broken”. The Feature
linking rules to be formulated further on, link Doris to the sub-
ject and the window pane to the direct object. The resulting
sentence is
15) Doris broke the window pane.
Matching May Ignore Features
Usually a NP matches a participant not only in respect to its
general sense, but also in respect to the participant’s features. In
15), Doris corresponds closely to the “breaker” participant of
break, because Doris is the cause of the breaking and presuma-
bly has some control over it. The window pane undergoes a
change as a result of the breaking activity and so the window
pane matches with the “thing broken” participant, which has
Change. But such a close correspondence is not a necessary
condition for NP matching. Take, for instance,
16) a. The captain passed the straits.
b. The passengers passed the straits.
Presumably the subject participant in the lexical entry of pass
has the feature Control:
pass: “passer” (Cause, Control), location passed (ø)
But although only the captain, and not the passengers, has
control over the event, the mental representations of the pas-
sengers and the “passer” participant in the verb’s lexical entry
of pass are sufficiently similar to permit matching in 16b) de-
spite the absence of Control.
In general, although inanimates lack control over an activity,
NPs referring to them can match with a participant having the
feature Control. Breaking, for instance, is normally predicated
of persons, and the lexical entry of break has therefore a subject
participant with the feature Control. Although it cannot be said
to have any control, the falling branch can be the subject of
break in 17), because of its similarity to the subject participant
in its lexical entry.
17) The falling branch broke the window pane.
Similarly, the verb extinguish has a participant with Control,
but can have not only an animate subject, as in 18a), but also an
inanimate one, as in 18b).
18) a. The boy extinguished the camp fire.
b. The rain extinguished the camp fire.
Another feature that may be ignored in NP matching is
Change. Consider
19) a. He broke his promise.
b. They dropped the plan.
In 19a), break is used in an extended sense to refer to the
breaking of something immaterial: a promise is conceived of as
something that is broken. Although the object participant of
break, the “thing broken” has the feature Change, his promise
can match with it, although no change occurs in the promise.
Likewise, the object participant of drop has the feature Change
(a dropped object changes its location), but although a plan
cannot be said to change its location by being dropped, the plan
in 19b) matches with the object participant of drop. In general,
metaphorical extensions of verb meaning, as in 18)-19), are due
to such “partial” matchings that ignore a specific feature. This
is in line with the principle that features are not properties of
NPs but of participants in the lexical entry of a verb.
Similarity in Matching
Not just any conceivable similarity suffices for NP matching
(after all, practically everything is similar to everything else in
some respect or other). Take for example the verbs stir and eat:
stir: “stirrer” (Cause, Control), stirred material (Change)
eat: “eater” (Cause, Control), eaten material (Change)
An instrument can be subject of stir, although only the per-
son who wields it has control over the activity. Thus, in certain
contexts, 20) may be acceptable.
14A different solution to this problem has been proposed by Belletti and
Rizzi (1988).
15In some cases, the NP determines the choice of the verb. In sentences like
have fun, make an effort, give evidence, the effected objects incorporate
most of the meaning in the verb phrase and the “light” verbs have, make,
and give make only a subsidiary contribution to the meaning. The nouns that
serve as effected objects of these verbs are usually not deverbal: fun, effort,
and evidence do not have corresponding verbs (*to fun, *to effort, *to evi-
dence). Even when a corresponding verb exists, it will have a slight differ-
ence of meaning. Thus, the baby can have a bath, but hardly bathe, and
when someone gives a talk he (hopefully) does more than just talking.

I. M. SCHLESINGER
238
20) The spoon stirred the soup.
Because a spoon can be viewed as “doing” the stirring, the
spoon in 20) matches with the “stirrer”. But subjectivization of
the instrument is not possible for some other verbs, like eat,
because the similarity between the instrument and the subject
participant is not large enough:
21) *The spoon ate the soup.
More is involved in eating than just bringing food to the
mouth, which is all a spoon can do. The spoon therefore does
not match the subject participant of eat. The same factor also
accounts for the oddness of the asterisked sentences in
22) a. He walked into the hall with a crutch.
a’. *The crutch walked into the hall.
b. They listened to the music with earphones.
b’. *The earphones listened to the music.
c. She jumped over the fence with a pole.
c’. *The pole jumped over the fence.
Further, the similarity between a given NP and a participant
must be not only large enough, but also greater than the simi-
larity of any other NP in the sentence to that participant. This
rules out
23) *The spoon stirred the soup by Fred.
The subject participant of stir has both Cause and Control.
Fred is not only the cause of stirring, but is also in control, while
the spoon can be viewed as causing the stirring but not as con-
trolling it. Thus Fred fits the subject participant better than the
spoon. Omitting by Fred results in the grammatical sentence 20).
As stated, Feature linking rules are defined in terms of fea-
tures of participants in the lexical entry. However, such a rule
does not operate directly on a participant with the required fea-
tures, but on the NP that matches with this participant. For ex-
ample, a Feature linking rule specifies that a NP that matches
the participant with Cause is linked to sentence subject (i.e., is
positioned before the verb, etc.). When Doris breaks the win-
dow pane, Doris matches the participant “breaker”, and since
the latter has the feature Cause, Doris has to be subject, as in 15).
The next section deals with linking in greater detail.
Selection of Subject and Direct Object
Subject Selection
In English and in many other languages, the lexical entries of
almost all verbs include a participant that is linked to the sen-
tence subject. We now take a closer look at the feature compo-
sition of the sentence subject and the direct object. Table 1
shows the combinations of subject and object features that we
have encountered so far.
This table shows that a participant with the “default feature”
ø never becomes subject. A necessary condition for a partici-
pant of a given verb to be linked to the sentence subject in an
active sentence is that it has at least one of the features Cause,
Control, Change or Attr-St. But is this also a sufficient condi-
tion? I now show that it is not.
Feature Conflict?
There are verbs with lexical entries having one of the above
four features in both their participants, and there might appear
to be a “conflict” between the features: Which participant is
linked to the subject? There are four classes of such verbs:
a) Object participants with Change: The object participants
of many verbs have Change, which is a feature that permits
Table 1.
Features of subjects and of direct objects.
Subject participant Object participant Example
Cause, Control, Change (none) run
Cause, Control, Attr-St (none) sit
Cause, Control (none) cry
Cause (none) snore
Cause, Control Change break
Cause Change frighten
Cause, Control ø accuse
Control, Attr-St ø pity
Control ø undergo
Change ø sustain
Attr-St ø contain
linking to the subject. The object participants of spill, pull,
throw, and send involve changes of location; in those of break,
mix, eat, and repair a change of state occurs; and the activity
denoted by a verb like build, paint, sculpt, or write results in
something coming into being, i.e., changing from non-existence
to existence. These verbs might seem to involve a conflict be-
tween features that license linking to the subject: Change, on
the one hand, and Cause and Control (the features characteriz-
ing the “spiller”, “breaker”, “builder”, etc.), on the other.
b) Both subject and object participants with Cause: In the
lexical entries of a few verbs, both the subject and the object
participants have the feature Cause. For instance,
24) The guide is leading the tourists.
Leading implies that both the leader and his followers move,
that their movements are (internally) caused by them and are
under their control; that is, the followers, like the leader, have
the features Cause, Control, and Change. This appears to con-
stitute a feature conflict, since only one of the participants can
be linked to the subject16. Another verb that has an object par-
ticipant with Cause and Control is incite.
c) Intransitive verbs used causatively: For instance,
25) a. The lady is walking her dog.
b. The sergeant marched his men across the square.
When one walks a dog, the latter is walking, too. Likewise,
not only the sergeant, but also the men march. The movements
of the dog and of the men are internally caused, and in each 25a)
and 25b) there are therefore two participants with Cause, only
one of which can be realized as subject. Davis (2001) criticizes
Dowty’s (1991) theory for not accounting for sentence with
such causative verbs.
d) Location subjects: Some verbs may have a location as
subject, e.g.,
26) a. This room sleeps two students.
a’. Two students can sleep in this room.
b. This hall seats 150 listeners.
b’. 150 listeners can sit in this hall.
26a) and 26b) describe properties of the room and the hall,
16One might argue that the activity of the tourists is itself caused by the
leader, and that such cases are best dealt with by introducing the construct
“prior cause”. A somewhat different solution will be proposed further on.
The verb follow does not raise the same problem. The lexical entry of follow
includes only one participant with Cause and Control: the follower (not the
leader). A detective can follow a person without the latter leading him, and
many of us occasionally follow instructions, rules, or examples.

I. M. SCHLESINGER
and so their subjects have the feature Attr-St. As a result of
these properties, the students and the listeners, respectively, are
able to sleep or sit, as stated in 26a) and 26b). Arguably, there-
fore, the object participants of the transitive sleep and seat (re-
alized here by two students and 150 listeners) have the features
Cause and Control17. Here, again, there seems to be a conflict
between features, and a Feature linking rule has to determine
which NP becomes the sentence subject.
Resulting Features
Note that in the above four cases of apparent feature conflict,
the features of the direct objects result from the event described
by the verb. The object participants of spill, pull, break, mix,
and so on, have the feature Change, because of the change that
occurs as a result of the activity referred to by the verb. Simi-
larly, in 24), the event referred to by lead results in the tourists’
following the guide, and so the features Cause, Control, and
Change of the object participant (matched with by the tourists)
are resulting features. The features Cause, Control, and Change
of the object participants of the causative walk and March in 25)
are also resulting features. In 26), the students’ potential sleep-
ing in the room and the listeners’ potential sitting in the hall
result from the characteristics of the room and the hall. Partici-
pants with such Resulting features are not eligible for becoming
the subject of an active sentence.
The term “resulting feature” can be defined as follows:
27) A resulting feature is a feature of a participant in the le-
xical entry of the predicate verb V, if and only if that feature is ac-
corded to that participant as a result of the event referred to by V.
The occurrence of resulting and non-resulting features is
summarized in Table 2. A feature can be a resulting feature in
the lexical entry of one verb and a non-resulting one in that of
another. Cause, Control, and Change can be either resulting
features, as in some causative verbs (and then the participants
to which they pertain become direct objects) or non-resulting
ones (and then they become subjects). The features ø and
Attr-St are always non-resulting; the former is linked to the
direct object and the latter to the subject.
A linking rule for the subject can now be formulated in terms of
four features and the distinction between resulting and non-
resulting features:
28) A participant with at least one non-resulting feature
Cause, Control, Change or Attr-St is linked to the subject of an
active sentence.
This rule applies to subjects of both transitive and intransi-
tive verbs. It bars participants with the feature ø and those with
resulting features from being linked to the subject. Passive sen-
tences will be discussed below.
“Symmetrical” Verbs
Some verbs might seem to present difficulties for this for-
mulation of the subject linking rule. Let us look at the verbs
equal and resemble.
29) a. One yard equals .914 meters.
b. John resembles his father.
The subject participant of equal has the feature Attr-St; in the
first sentence, something is attributed to one yard. One might
want to argue that the object participant, .914 meters, also has
Attr-St, and that there is thus a feature conflict. But note how
awkward .914 meters equals one yard sounds. To provide in-
formation about meters it would be more natural to say One
meter equals… The sentence does not say anything about .914
meters, although one can of course infer from it something
about this measure of length. The object participant of equal
has therefore the “default feature” ø and not Attr-St. The same
reasoning applies to resemble. 29b) attributes a state to John,
not to his father (John’s father resembles John would sound
strange). On the psychological asymmetry of similarity state-
ments see Tversky (1977: p. 38).
The moral is that one must distinguish between what is said
in a sentence and what is implied by it. Feature assignment is
independent of whatever can be inferred from the event or si-
tuation described by the sentence.
There are also quasi-symmetrical activity verbs, as in
30) a. Erwin married my sister.
b. My sister married Erwin.
Both these sentences have the same truth value, but in many
situations they are not interchangeable. When Erwin’s parents
are furious that he married my sister, one does not say that they
are furious that my sister married him. The two sentences 30b)
and 30a) do not say the same; one can at most infer one from
the other. Only the subject participant of marry, and not its
object participant, has the feature Cause18.
The same holds true of the lexical entries of meet (in one of
Table 2.
Examples of resulting and non-resulting Cause, Control, Change, and Attr-State*.
Feature Subject participant Object participant
Non-resulting feature Non-resulting feature Resulting feature
Cause and/or Control cry
undergo - lead, seat, walkcausative
Change melt - break, lead, walkcausative
Attr-St contain - -
ø - accuse -
17The causative form illustrated in 26) is found in only a handful of verbs. Neither the elevator stands four adults nor the tent lies six boy scouts is
grammatical. Children learning English as their native language sometimes over-generalize the causative construction to such verbs (Bowerman &
Croft, 2008).
18In a psycholinguistic experiment, Kasof and Lee (1993) found that when presented with a sentence like Car A collides with car B, people judged
Car A to have been moving faster than Car B. A series of experimental studies of symmetrical and apparently symmetrical verbs has been conducted
b
y Gleitman et al. (1996).

I. M. SCHLESINGER
240
its uses) and fight. The subjects and objects in sentences with
these verbs cannot always be interchanged with impunity: If a
young boy met a pop star, it would be odd to say that the pop
star met him, and if the government fights poverty, it is not the
case that poverty fights the government.
Direct Object Selection
Semantics of the Object Category
We saw in Table 2 that a participant with the “default fea-
ture” ø is always linked to the direct object. It would be desir-
able to find a way to define this feature positively in terms of a
property or properties common to all object participants having
ø; but such a characterization in semantic terms seems to be
unattainable. Already Jespersen (1933, Section 11.3) observed
how difficult it is to define the relation of the object to the verb.
That an entity is referred to by a direct object does not imply
that it is in motion or undergoes a change of state. In measure
the width of the table, the width is neither effected nor affected,
nor is it the Goal of the action in any sense. The object partici-
pant of measure shares the “default” feature ø with the object
participants of many other verbs, for instance, with those un-
derlying lack money, answer the door, relate the incident , avoid
the rush, imitate a novelist, and with the direct object partici-
pants of deny , describe, forgive, revenge, regret, study, admit,
prove, see, like, and many more. There seems to be no common
semantic denominator of all these object participants (see Levin
1993; Levin & Rappaport Hovav, 2005: pp. 27-28; Schlesinger,
1995: pp. 164-166). Hence the “default feature” can be defined
only negatively, as the feature that is neither Cause, Control,
Attr-St, nor Change.
How, then, is the direct object selected? I propose that fea-
tures of participants determine selection of the subject accord-
ing to rule 28), and the remaining participant, if there is one, is
linked to the direct object19. In other words, the direct object is
a default category (cf. Levin & Rappaport Hovav, 1995: p. 154).
This view of object selection is based on the already mentioned
assumption that a lexical entry of a verb has at most two par-
ticipants (but see note 26 for possible exceptions). This as-
sumption is motivated also by other, independent, considera-
tions, as will be shown in the section on Roles of NPs, below.
But first let us clarify some questions concerning object selec-
tion.
Converse Verbs
Above I discussed Grimshaw’s (1990) analysis of the pair of
verbs frighten and fear. Let us see now how these verbs are
analyzed in the present framework. Grimshaw’s example 14) is
repeated here:
14) a. The storm frightened us.
b. We feared the storm.
The relevant lexical entries are:
frighten: the entity that frightens (Cause), the one who is
frightened (Change)
fear: the one who fears (Attr-St), the feared entity (ø)
In 14a), “the entity that frightens” has the feature Cause, and
so is selected as subject, and the remaining participant, “the one
who is frightened”, is linked to the direct object. In 14b), “the
one who fears” is said to be in a state and so has Attr-St, which
qualifies we for subject position, and so “the feared entity”, the
storm, becomes direct object.
Two kinds of objections to this analysis might be made:
a) A state is attributed to “the one who is frightened” (in the
lexical entry of frighten). Hence this participant should have the
feature Attr-St, just like the participant “the one who fears” in
the lexical entry of fear. Accordingly, Attr-St would here be a
resulting feature of “the one who is frightened”.
However, an activity or event cannot result in a state, but
only in a change of state. “The one who is frightened”, is said
to have changed from not fearing (i.e., may have evinced no
fear before the frightening event) to fearing, and so has the re-
sulting feature Change. Attr-St is never a resulting feature (see
Table 2).
b) “The feared entity” in the lexical entry of fear, the storm
in 14b), is what causes the fear and so should be assigned the
feature Cause , rather than ø (just like “the entity that frightens”
in the lexical entry for frighten). “The feared entity” would thus
compete with “the one who fears” for subject position.
But the storm referred to in We feared the storm is not the
cause of the fear; rather, it is its content, just as it can be the
content of mental states like expecting, imagining, enjoying,
and hating, and not their cause:
31) a. We expected the storm.
b. We imagined the storm.
c. We enjoyed the storm.
d. We hated the storm.
The storm cannot be the cause of expecting, since causes do
not operate backward in time, and a storm does not cause any-
one to imagine it. Verbs describing cognitive states other than
ones referring to emotions also have object participants per-
taining to content rather than to causes:
32) a. He remembers his uncle.
b. He forgot his wallet.
The uncle may have long since died, and the wallet can
hardly be held responsible for being forgotten.
The pair of verbs please and like will be analyzed like fri-
ghten and fear, and so will persuade and belie ve (Peter persu-
aded Bill and Bill believed Peter)20. Some other converse verbs
are dealt with in the section on Roles of NPs, below.
Object Elision
There are sentences in which the object participant of the
lexical entry of a verb is not realized. Some transitive verbs,
among them eat, drink, help, le ave, permit elision of the direct
object. Both Percy is eating an apple and Percy is eating are
grammatical.
“Symmetrical” verbs, like marry, fight, etc., also permit eli-
sion of the direct object. If their subject is a conjoint, the verb
may be understood in one of two ways: either as reciprocal, as
in 33a), where each other is elided, or not, as in 33b), where
19Oblique objects will be discussed in the section on Roles of NPs, below.
Cognate objects are not realizations of any participant in the lexical entry.
The verbs laugh, live, and die do not have an object participant (they are
normally used intransitively), but each of them may have a direct object that
repeats the information in the verb: laugh an anguished laugh, live a
g
oo
d
life, die a violent death; consider also sculpt an impressive sculpture. It
appears that such cognate objects result from a realization rule that intro-
duces a modifier of the verb, and they occur mainly when no adverb can do
the job: Living a good life is not the same as living well, sculpt an impres-
sive sculpture is not the same as sculpting impressively, and an anguishe
d
laugh cannot be replaced by *laugh anguishedly.
20Cf. Fillmore (1968). The fact that there is no Actor in either please or like,
led Jackendoff (1987: p. 397) to conclude that the difference between them
is “lexically determined”.

I. M. SCHLESINGER
direct object NPs are elided.
33) a. John and Jill married.
b. John and his sister married.
In such “symmetrical” verbs, each NP in a conjoint must
match in the same way with the subject participant of its predi-
cate verb. The reason why The drunk and the lamppost embra-
ced, is ungrammatical, is that the subject participant of embrace
has the features Cause and Control, the drunk matches with this
participant in respect to both these features, whereas the lamp-
post would match it merely “partially”.
Object elision in several other verbs, e.g., dress, shave, bathe,
wash, hide, and feed, leads to the verb’s being construed as
reflexive, as in 34b):
34) a. Paula dresses her doll.
b. Paula dresses.
A Note on Dowty’s Theory
On the present theory, four features and the distinction be-
tween resulting and non-resulting features are sufficient for sub-
ject and direct object selection. The definitions of the four fea-
tures differ in some respects from those of Dowty’s (1991)
features of the Proto-Agent, and some of Dowty’s features (e.g.,
“sentience”) are not among the features included here.
Kako (2006) has found that the grammatical function of a NP
in a sentence could serve as a clue as to Dowty’s features, thus
providing evidence for the psychological validity of the latter.
However, that speakers are aware of these syntactic-semantic
correspondences does not mean that subject selection is based
on them. The rule in 28) is simpler than Dowty’s rules, since it
states that any one of the specified features is a sufficient con-
dition for linking to the subject, whereas in Dowty’s system the
number of features characterizing his Proto-Agent and Proto-
Patient has to be taken into account (and this has also implica-
tions for first-language acquisition; see the last section).
Roles of NPs
As stated, a lexical entry has either one or two participants,
and these are realized as subject and (in entries with two par-
ticipants) direct object. Other NPs in a sentence, notably oblique
objects, do not match participants. In the following we will see
what motivates this claim.
The Insufficiency of Features
Suppose for a moment that not only two, but all arguments of
a verb are realizations of participants. Then there would be se-
veral serious difficulties for subject and object selection, as will
be shown now.
Causal Phrases
A prepositional object may refer to the cause of the event or
situation described by the verb. Here are some examples:
35) a. The king died by the sword.
b. The old man died through neglect.
c. The explorer died from exhaustion.
d. The child thrives with love.
Suppose that the sword in 35a) were to be matched with a
participant of die; then it would have a non-resulting feature
Cause (the sentence mentions no other entity that causes the
sword’s activity). Accordingly, the sword would have to be the
subject, because the king matches with a participant with a
resulting feature (Change). The same goes, mutatis mutandis,
for 35b), 35c), and 35d).
Sentences like these are also an obvious embarrassment for a
thematic-roles theory. The NPs in the prepositional objects of
35) refer to causes, and these should be higher in the hierarchy
than the thematic roles of what are here realized as subjects.
Compare 35a), 35b), and 35c) to their near-paraphrases, The
sword killed the king, Neglect killed the old man, and Exhaus-
tion killed the explorer. Why, then, should the sword, neglect,
exhaustion, and love in 35) not be subjects?
Dowty’s (1991) theory apparently cannot provide an expla-
nation either. True, the subjects of 35) are all animate and sen-
tient, and might be good candidates for proto-agency. But the
subjects in 36) are inanimate, hence not sentient; they do not
cause a change, nor do they necessarily move, and so don’t
even come near to being proto-agents. In fact, the oblique ob-
jects in 36a)-36d) do have some of these entailments and thus
should be proto-agents:
36) a. The plants withered for lack of water.
b. The law passed thanks to the prime minister.
c. Prices rose due to foreign investors.
d. The alarm went off because of our cat.
e. The shack collapsed of old age.
Converse Verbs, Again
Converse verbs present what Dowty (1991: p. 379) calls a
“puzzle” for a thematic role approach. The two sentences in 37)
are near-paraphrases; Ben seems to have the same thematic role
in both, and so has Amy. What accounts for the fact that the two
sentences have different subjects? The same question arises in
the framework of feature theory. When Amy sells a car, she
causes the selling and has control over it, regardless of whether
the transaction is described as in 37a) or as in 37b), and so Amy
should be subject in both. A similar argument can be made
concerning the buyer, Ben.
37) a. Amy sold a car to Ben.
b. Ben bought a car from Amy.
Alternating Verbs
Alternations present difficulties for thematic hierarchies. Some
kinds of alternations appear to pose problems also for our fea-
ture-based approach, for instance,
38) a. The fountain spouts water.
b. Water spouts from the fountain.
39) a. He hit the wall with a hammer.
b. He hit the hammer against the wall.
If spout in 38b) had two featured participants, the fountain
would match with the subject participant just as it does in 38a).
So why is the fountain in subject position in 38a) and not in
38b)? Similarly, in 39), a hammer may be thought to have the
same features in both sentences, and so should have the same
syntactic function in both.
Instruments
40) a. This key will open the door.
b. The door will open with this key.
If the key in 40b) were a realization of a participant in the
lexical entry of open, it might be expected to have the same
features in both a. and b. Why, then, is the key subject in a. but
prepositional object in b.? A theory based on thematic hierar-
chies encounters the same problem, because in both sentences,

I. M. SCHLESINGER
242
the key has the Instrument role, and in both the door has the
same role (Theme?).
Any theory based on a thematic hierarchy will presumably
provide a separate solution for each of the difficulties posed by
35)-40). Grimshaw’s (1990) treatment of converse verbs like
fear and frighten can perhaps be adapted to cover cases like
sell and buy in 37), but will not be applicable to causal preposi-
tional phrases like 35)-36). Likewise, treatments of the locative
alternation (e.g., Beavers, 2006; Dowty, 1991: pp. 587-592;
Hopper and Thompson 1980: 262) and Fillmore’s (2003: pp.
222-224) account of verbs like hit and break are not applicable
to other cases discussed in the foregoing. In the next sub-sec-
tion, I propose a single solution to all the difficulties described
here. The linking system, it will be argued, relies neither on fea-
tures alone nor on thematic roles alone; both features and the-
matic roles are needed.
Roles without Hierarchies
The lexical entry of a verb, I have argued, contains at most
two participants with features, and these are linked to the sub-
ject and (when there are two) the direct object. A sentence may
of course include NPs besides subject and object, notably obli-
que objects, but these do not match any one of the participants
in the verb’s lexical entry (even in those verbs that have an ob-
ligatory oblique object). Instead, they express thematic roles, or
roles, as they will be called here for short. In other words, only
arguments that are realized as subjects and direct objects (and
not adjuncts) appear in the lexical entry as participants with fea-
tures. Thus, give has three arguments, but only two featured
participants (the “giver” and the thing given); the Recipient is a
role that is not based on a participant.
Subjects and direct objects are selected on the basis of Fea-
ture linking rules and thus do not depend on precedence rela-
tions between thematic roles. All other arguments have roles
and are realized in accordance with Role Linking Rules, to be
discussed further on. To illustrate:
41) a. After lunch, Anne threw the ball over the fence.
b. After lunch, Anne threw the ball to Clara.
The lexical entry of throw is
throw: “thrower” (Cause, Control), object thrown (Change)
Feature linking rules assign the NP Anne, which matches the
participant with Cause and Control, to subject position, and the
ball, to direct object position. Of the remaining NPs, lunch has
a temporal role, the fence in 41a) has a locative role, and Clara
in 41b), a Recipient role. Role linking rules determine that after
precedes lunch, over precedes the fence in 41a), and in 41b) to
precedes Clara.
Problems Solved
Positing roles alongside features provides a straightforward
solution for the difficulties discussed in the foregoing:
a) Causal phrases: In 35)-36), the lexical entries of die, thrive,
wither, pass, rise, go off, and collapse have each a single parti-
cipant, which is linked to the subject. The NPs realized as pre-
positional objects are adjuncts; they have causal roles and do
not match with any featured participants. Hence they are not
candidates for subject-hood.
b) Converse verbs: The lexical entries of sell and buy are:
sell: seller (Cause, Control), thing sold (Change)
buy: buyer (Cause, Control), thing bought (Change)
In addition to subject and direct object, the sentences in 37)
contain NPs with roles: the Recipient or Goal (the buyer, Ben),
and the Source (the seller, Amy). These are adjuncts realized as
prepositional phrases. Similar analyses can be made of the uses
of other pairs of converse verbs: teach and learn, lend and bor-
row, and give and receive.
Gropen et al. (1991a) present an analysis of fill and pour in
terms of event structures. On the present approach, these do not
have to be resorted to. Consider:
42) a. John fills the glass with water.
b. John pours water into the glass.
The lexical entries of these verbs are:
fill: the one who fills (Cause, Control), the receptacle filled
(Change)
pour: the one who pours (Cause, Control), the material pour-
ed (Change)
The “receptacle” in fill undergoes a change of state (from
empty to full), and the “material” in pour, a change of location.
42a) shows that a sentence with fill may have, in addition to the
subject and direct object, a NP with a role (Theme? Material?),
and 42b) shows that a sentence with pour may include a NP
with the role Goal.
c) Alternating verbs: Most alternations apply to only a small
class of verbs and are based on different sub-entries in the lexi-
cal entry of the verb. Spout, for instance, has a sub-entry con-
taining two participants: “source of spouting’ (the internal Cause)
and “the material spouted” (with Change). This sub-entry un-
derlies The fountain spouts water 38a). Another sub-entry of
spout underlies Water spouts from the fountain 38b) and has
only one participant with the feature Change. The prepositional
object from the fountain is not based on a participant of spout,
but has a role, and is realized as an oblique object. The verb hit
also has two sub-entries. In the sub-entry underlying 39a), the
object participant is “the location hit” and in the one underlying
39b), “the instrument of hitting”. Both sentences have NPs with
roles (a hammer and the wall, respectively), and roles do not
compete for subject and direct object positions21.
d) Instruments: The verb open has two lexical sub-entries, a
transitive one, as in 40a), and an intransitive one, as in 40b):
open1: the opener (Cause, Control), the thing opened (Change)
open2: the thing that becomes open (Change).
In 40a), this key matches “the opener” of open1 (by partial
matching, because the key has no control). 40b), however, is
based on open2, and only the door realizes a participant; this
key has a role (Instrument), and so its syntactic function is not
determined by the linking rules for subject and direct object.
Passives with by-Phrases
By-phrases are prepositional objects. They can be assigned
various roles, as shown in Table 3.
By contrast, the subjects of the corresponding active sen-
tences do not have roles, but features. Table 3 shows that when
the direct object of an active sentence has the feature Change,
the subject of the corresponding passive sentence also has
21As pointed out by Fillmore (1970), there is no similar alternation for the
verb break (*He broke the hammer against the fence). The reason seems to
be that broke the hammer can be construed as something done to the ham-
mer, and speakers avoid such a construal. For hit the hammer this is not a
likely construal, since people rarely hit hammers (they are more likely to
b
reak them). There is some psycholinguistic evidence showing that the
plausibility of an erroneous construal in the course of processing does affect
the judged acceptability of a sentence (Schlesinger, 1977: pp. 211-215).

I. M. SCHLESINGER
Table 3.
Passive sentences with by-phrases.
Active Passive
Subject particip. Object particip.Example Subject particip.Role of by-phraseExample
1 CA CO CH Doris broke the window. CH Agent The window was broken by Doris.
2 CA (CO) CH The tsunami broke the window.CH Natural Force The window was broken by the Tsunami.
3 CA (CO) CH A stone broke the window. CH Instrument The window was broken by a stone.
4 CA (CO) CH The storm frightened us. CH Stimulus We were frightened by the storm.
5 ATT ø His friends admire him. ATT Experiencer He is admired by his friends.
6 ATT ø This room sleeps two students.- - -
7 ATT ø One yard equals .914 meters.- - -
8 CA CO CH ø They arrived at the airport. - - -
9 - - - ATT Manner The solution was found by chance.
Abbreviations: CA = Cause; CO = Control; CH = Chang e; ATT = Attr-St. (CO), in parenthesis, indicates that the subject participant has Control which is lacking in the
specific NP (and supplied by NP matching).
Change22. When the direct object of an active sentence has the
feature ø, the subject participant of the corresponding passive
sentence has the feature Attr-St. When a by-phrase has the
Manner role, as in line 9 of the table, there is no corresponding
active sentence.
As shown in lines 6 - 8, there are also active sentences with-
out corresponding passives (either with or without by-phrases).
The reason is that unlike all subjects of passive sentences,
which have either Change or Attr-St, the suppositional subjects
of such passive sentences would not match with any one of
these features: Nothing of interest would be asserted of .914
meters by .914 meters are equaled by one yard, or of the airport
by the airport was arrived at by them, and so it would be infe-
licitous to ascribe to these NPs the feature Attr-St. This is also
the reason why the verbs possess, have, lack, suit, become, and
fit normally do not passivize. See Schlesinger (1995: pp. 131-
138) on this and similar constraints on passivization.
Distinguishing Direct from Oblique Objects
The theory proposed here is more parsimonious than one
based on thematic hierarchies. This simplification comes at the
price of a distinction between two sets of semantic notions,
features and roles, and two corresponding classes of rules: Fea-
ture linking rules and Role linking rules (to be dealt with fur-
ther on. Let us see now whether there is any semantic criterion
determining which arguments are represented by features in a
lexical entry and which are not.
In many sentences, direct objects seem to be, as it were, more
intimately related to the verb than oblique objects. One can
throw a ball without throwing it to any specific place or person,
and so the ball in 41) is direct object, whereas over the fence
and to Clara are not. That direct objects are more central to the
verb’s meaning is the view of, inter alia, Jackendoff (1990a:
449), who states that the first object of a ditransitive verb (e.g.,
John in give John a book) is not a direct object, since it “ex-
presses a semantic role that is not an essential part of the verb’s
meaning” in the way direct objects are.
But essentiality is anything but a reliable criterion. Thus, for
filling to take place, there has to be, in addition to the agent, a
specific vessel that is filled and a material it is filled with. All
three are essential for filling to occur, but only the vessel ap-
pears as direct object; cf. 42). Buying requires, besides the buy-
er and an item bought, a seller and money or a commodity
given in exchange. In the absence of any one of these, buying
cannot be said to have taken place, but the seller and the money
given in exchange are expressed as oblique objects23.
Another distinguishing property that has been proposed is af-
fectedness. A case in point is the locative alternation:
43) a. She sprayed the wall with paint.
b. She sprayed paint onto the wall.
The first sentence, but not the second, suggests that the wall
is wholly covered with paint. It has been argued that the direct
object carries with it a sense of completeness (Anderson 1971;
Fillmore 2003: 226), as evidenced also by the pair search in the
room, se arch the room, and denotes the “completely affected”
(Hopper & Thompson, 1980: p. 262; see also Beavers, 2006),
“wholly involved” (Levin & Rappaport Hovav, 2005: p. 209)
entity. As Gropen et al. (1991a: p. 162) remark: “… the most
natural interpretation of a state change is that it is the entire
object that undergoes the change, rather than one part…”. Ac-
cordingly, when the wall is completely painted, 43a) is more
appropriate than 43b). This does not take us very far, however.
It does not explain other alternations. For instance, why can the
hammer and the wall each be either a direct or an oblique object
in 39)?
A factor related to affectedness is success of the activity.
Examples of the conative alternation, like push the table, push
at the table and shoot the bird, shoot at the bird, indicate that
often the direct object, but not the oblique one, signals success
of the activity. This factor can account only for alternations where
one sentence has “verb + at” (like, e.g., push at) and the other
only “verb” (push).
A semantic distinction between direct and oblique objects
22Or, in the case of an adjectival passive, Attr-St (e.g., The vase is broken, in
the sense of “broken vase”). One might conceive of active and the corre-
sponding passive sentences as stemming from two sub-entries in the lexical
entries of the verb V, V and be V en, and the adjectival passive as based on
yet another sub-entry. Alternatively, the passive may be viewed as a result
of a generalized lexical rule. This question will be left open here.
23True, in these particular instances the oblique objects can be elided when
their identity is known or is not of sufficient interest to be reported. But this
does not indicate that they are less essential than direct objects; in fact,
direct objects are also often elided (as in He is eating), whereas the oblique
objects of many verbs cannot be (cf. put the book on the shelf). Consider
also that some verbs, for instance, rely and belong, require a prepositional
object, but do not have a direct one.

I. M. SCHLESINGER
244
has been proposed by Dowty (1991: p. 576) in his Corollary 2:
The non-subject argument with the greater number of Proto-
Patient properties becomes direct object and the one with a
smaller number becomes oblique or prepositional object. This
proposal, again, does not account for alternations where the
direct and prepositional objects can be interchanged. For in-
stance, in both 39a) and 39b), the hammer has the Proto-Patient
properties “undergoes change of state” (it is in motion) and
“causatively affected”, whereas the wall has only one: “station-
ary”. On Dowty’s proposal, this would rule out 39a), where the
hammer is not the direct object but a prepositional object. Nei-
ther can Dowty’s hypothesis account for crosslinguistic differ-
ences: There are English verbs taking a direct object and their
French translation equivalents, an oblique one (discuss, direct
object, discuter, oblique, or enter vs. entrer), or vice versa
(long, oblique, vs. desirer, direct object)24. Further, the alter-
nates of the following pairs do not differ in respect to Proto-
Patient entailments: blame X for Y, blame Y on X, present X
with Y, present Y to X, and strip X of Y, strip Y from X. Finally,
some verbs can have either a direct or a prepositional object:
John meets X, John meets with X, Jack consults X, Jack con-
sults with X, J erry confesses his theft, J erry confesses to his
theft, Julie ponders the alternatives, Julie ponders over (or on)
the alternatives (Schlesinger, 1995: pp. 175-179). Apparently,
the direct and prepositional objects in each of these pairs do not
differ in respect to their “entailments”.
It appears therefore that the lexical entries of many verbs
stipulate what becomes a direct rather than an oblique object25.
In the final section, we will see how the language learner mas-
ters the distinction between direct and prepositional objects.
Role Linking Rules
Feature Linking Rules and Role Linking Rules
A Feature linking rule is defined in terms of features. It states
that a participant with the non-resulting feature F will be real-
ized as the sentence subject. Each Feature linking rule operates
on the NP that matches with a participant with the feature F.
Role linking rules, by contrast, do not take such an indirect
route and do not require NP matching. A Role linking rule may
state, for instance, that a NP having role R is preceded by the
preposition P and can appear in such and such a syntactic com-
bination. While features pertain to participants and do not take
into account the sentential content, roles are independent of the
lexical entry of the verb and its participants, and Role linking
rules depend entirely on the role expressed by the prepositional
phrase: a location (in the garden) a purpose (for the garden), a
cause (because of the garden), etc.
Alternative Role Linking Rules
Sentences with ditransitive verbs, like 44), are apparent
counter-examples to the claim that lexical entries have one or,
at most, two participants:
44) Anne threw Clara the ball.
In this sentence there are three arguments: besides the “thrower”
and the “object thrown” there is the recipient, Clara. Both
Clara and the ball, are often regarded as direct objects, but
Jackendoff (1990a: p. 449) argues against this. Hudson (1992)
presents further evidence against regarding the indirect object
as a sort of direct object, and also argues against regarding it as
allied to oblique objects. Instead, he conceives of the indirect
object as a separate kind of object.
It is proposed that 44), like any other sentence with a transi-
tive verb, has only two participants, a subject participant (here
Anne), and an object participant (here the ball), and Clara has a
role, the Recipient, just like Clara in
45) Anne threw the ball to Clara.
The Recipient role, then, has two alternative realizations: one
with a preposition and positioned after the direct object, as in
45), and one without preposition and positioned before the di-
rect object, as in 44)26.
Some Constraints on Role Linking Rules
Only some verbs permit a choice between the double object
construction, as in 44), and these belong to certain semantically
defined classes, such as change of possession verbs (like give
send, buy, sell) and verbs of communication, like tell, show, ask,
teach, write, read, e-mail. Within the latter class there are se-
mantically homogeneous sub-classes, for instance, “manner of
speaking” verbs (e.g., shout, murmur, scream), which rule out
the double object construction (Gropen et al., 1989; see also Jack-
endoff, 1990b: pp. 194-200; Pinker, 1989; and Shibatani, 1996.
Evidence for a processing factor has been found by Schlesinger,
1977: pp. 204-215). The Recipients of some other verbs, like
ask, call, and elect, require the double object construction and
are usually not expressed as prepositional objects27.
A Role linking rule, or one of two alternative Role linking rules,
then, may be blocked for certain semantically defined classes of
verbs (and in some cases, for certain individual verbs). This
shows that Role linking rules may be sensitive to the meaning
of the main verb, where “meaning” refers not just to the parti-
cipants, note that tell and shout have similar participants with
the same features, but to the entire mental representation of the
verb’s meaning in the lexical entry.
24Witman (2000) found such discrepancies in close to 10 percent of a sam
p
le
of pairs of English and French verbs that were translation equivalents. In
Hebrew, the tendency of verbs to take an oblique object is stronger than in
either one of these languages, and the discrepancies were larger: 11.5 per-
cent for Hebrew and English and 14.6 percent for Hebrew and French.
25Many of the facts discussed in this sub-section are also difficult to explain
in a non-ad hoc manner by any theory based on a thematic hierarchy, and
p
resumably such a theory will also have to resort to the assumption that they
are stipulated by the lexical entry. In recent years, several writers have
shifted the burden of accounting for linguistic regularities from general rules
to lexically specific ones (see, e.g., Chomsky, 1995; Ninio, 2006).
26See the extended discussion by Rappaport Hovav and Levin (2008), and
Beavers (2006) for a different treatment of indirect and oblique objects. In
German, indirect objects, like most oblique objects, are in the Dative case,
unlike direct objects, which are in the accusative. However, some verbs, like
lehren (teach) fragen (ask), and kosten (cost) are exceptions: In the German
translation equivalents of Jessica taught my son math, she asked my son a
silly question, The yacht cost my son a fortune, the NP my son is in the
accusative case. Similarly, the verbs teach and ask in the Hebrew translation
equivalents of these sentences require my son to be preceded by et,which
introduces the direct object, rather than by the preposition l-, which intro-
duces an oblique object. Further, English indirect objects can be paraphrased
by prepositional objects with to and for (give this to…, cook this for…),
whereas asked my son is paraphrased by of (ask… of my son), and cost my
son cannot be paraphrased with a prepositional object. In fact, unlike verbs
having indirect objects, ask and cost do not involve any Recipient or Bene-
ficiary. Perhaps, therefore, the lexical entries of ask, cost, and teach should
be regarded as exceptions in that they have two object participants.
27But see Rappaport Hovav and Levin (2008). A further complication: For
certain kinds of direct objects (mainly those denoting immaterial entities),
the double object construction will be unacceptable with some ditransitive
verbs. For instance, one cannot say *Jenny gave a headache/ trouble/
g
oo
d
advice to Richard. This may be due to the fact that the verb is used here in a
s
p
ecial sense.

I. M. SCHLESINGER
What Roles Are There?
Thematic roles are usually conceived of as determining ar-
gument selection. On the present theory, selection of subject
and direct object is taken care of by features and Feature link-
ing rules, whereas roles are required to account for other lin-
guistic regularities, among them assignment of prepositions. To
ask for a list of roles is therefore tantamount to asking for a list
of semantic distinctions that make some linguistic difference in
the realization of roles (cf. Dowty, 1991: p. 562).
Once the question is formulated in this way, it becomes clear
that there must be a far larger number of roles than those in-
cluded in any thematic hierarchy. A NP describing the purpose
of an activity, for instance, appears with certain prepositions
(for, in order to, etc.), and so Purpose has to be a role by itself,
and so has Manner, (as in with care). These roles are not in-
cluded in any current proposal for thematic hierarchies, to my
knowledge.
Some established thematic roles must be split into several
roles or several sub-roles. This becomes evident on considering
the prepositions that are appropriate in each instance. For in-
stance, the role that goes under the name “Location” is in fact a
set of roles or sub-roles, each requiring a different preposition
(and not only those for the Source and Goal roles, i.e., “Loca-
tion to” and “Location from”). In 46a) and 46b), for instance,
three prepositions appear, and these cannot be interchanged:
46) a. They met in Melbourne on Main Street at the corner.
b. The fire started in Australia on a farm at (or near) the
barn.
c. *They met at Melbourne in Main Street on the corner.
d. *The fire started at Australia at a farm on the barn.
The triplet of prepositional phrases with in, on, at is ordered
according to the size of the specified location. This is also the
case when only a single location is mentioned: One does not
say *at Melbourne or *in Main street. There are thus three dif-
ferent Location roles, corresponding to the size of the specified
location.
The same holds true for temporal roles:
47) a. In February on a Monday at eight in the evening.
b. In 1976 on May 16 at noon.
Again, the prepositions in, on, and at have to be applied ac-
cording to (among others) the extent of time referred to; in 47)
no other permutation of prepositions is permitted. We thus have
to distinguish between three temporal roles (in addition to the
temporal roles expressed by oblique objects with prepositions
like during, since, until, after etc.).
As another example, take the closely related roles, Instru-
ment and Natural Force. In the absence of an animate agent,
NPs denoting the instrument and a natural force can be realized
either as prepositional phrases, as in 48a) and 48b), or (because
they are causes) as sentence subjects, as in 48c) and 48d):
48) a. He scratched the stone with a hammer.
b. The window was shattered by the storm.
c. The hammer scratched the stone.
d. The storm shattered the window.
However, when they function as subjects, the instrument or
the natural force cannot deploy another instrument:
49) a. *The hammer scratched the stone with a chisel.
b. *The storm shattered the window with a tree branch.
Exceptions are instruments and natural forces that effect a
change by dint of a proper part of themselves, as in
50) a. The car scraped the tree with its fender.
b. The Tsunami destroyed the area with its high waves.
“Proper part of Instrument” and “Proper part of Natural Force”
thus function as sub-roles of Instrument and Natural Force.
These examples suggest that the number of thematic roles in
the present system greatly exceeds that of the roles commonly
envisaged by theories involving thematic hierarchies.
To summarize, our system differs from many current ones in
that it does not resort to any thematic hierarchy to account for
the realization of arguments. There are two types of semantic
notions, featured participants in the lexical entry of a verb, and
roles, and correspondingly, two types of linking rules: Feature
linking rules, which select subjects and direct objects, and Role
linking rules, which determine the expression of other syntactic
functions, notably oblique objects. We have seen that some
support for this view comes from the fact that it affords simple
explanations for some linguistic phenomena which pose diffi-
culties for current theories. In the next section I show that it
also affords a simpler theory of first language acquisition.
Implications for Language Acquisition
Grammar and language acquisition theory are interdependent:
A grammar must be learnable, and an acquisition theory must
show how the native language learner attains the adult linguis-
tic system. In the following, I outline the way children might
learn the semantic-syntactic mapping of the English language.
Then the semantic feature theory and a linguistic theory based
on thematic roles will be compared in respect to ease of acqui-
sition.
Acquiring Features and Feature Linking Rules
The semantic features proposed in this paper are rooted in the
child’s cognitive equipment. Notions like change, cause, and
control are deployed by the young child in making sense of
events in her environment (see Leslie & Keeble, 1987 on cau-
sation; Premack, 1990 on animacy and causality; Gelman &
Opfer, 2002 on cause and motion). But that a feature is within
the cognitive compass of the small child does not suffice. The
question that remains for an acquisition theory to deal with is
how the child finds out just which notions out of several possi-
ble ones (like, e.g., “big” and “red”) play a part in the formula-
tion of syntactic regularities. Further, that a child has the notion
of, say, causation and understands that a given event causes
another does not mean that she has a semantic feature Cause
(see Schlesinger, 1995: pp. 4-27 for a discussion of the distinc-
tion between cognitive notions and semantic categories). Lastly,
notions pertain to the events or situations referred to, whereas
features inhere in lexical entries; a participant in a verb’s lexical
entry may have a feature F, regardless of the situation referred
to by the verb. The problem for an acquisition theory, then, is:
How do notions like “cause” and “control” develop into seman-
tic concepts, namely, features?
I propose that the development of a feature concept goes
hand in hand with the formation of Feature linking rules that
apply to it. Observing how a participant is expressed in the
sentence leads to the formation of a linking rule and at the same
time draws attention to the feature of this participant. The fol-
lowing sketchy remarks describe, without hypothesizing any
innate semantic concepts, the steps by which this might come
about.

I. M. SCHLESINGER
246
Mini-Rules
The child who comprehends an utterance like Jane pushes
the baby carriage perceives a scene in which there are a pusher
and a “pushee” between which an activity (pushing) takes place.
The child’s encounter with additional push-sentences gradually
leads to two results: a) the meaning of push, i.e., the “scene” of
this activity and of the participants involved, is stored in her
lexical entry for push, and b) she records a lexically specific
rule, to be called here a mini-rule, to the effect that the word
referring to the pusher appears before the word push28. When
the child then intends to comment on another incident of push-
ing, she will access the lexical entry of push and apply this
mini-rule (the issue of direct-object realization will be discuss-
ed further on. Observe that at this stage no general notions like
causation or control need be involved but merely the mental
presentations of a pusher and a “pushee”, and these are integral
parts of the verb’s meaning.
Some evidence for mini-rules comes from findings showing
that lexically-specific rules are predominant at the earliest stages
of language acquisition; see Braine (1976) and Maratsos (1979:
309 et passim) for early statements, Ninio (1999) and Tomasello
(2000, 2003: pp. 117-121) for comprehensive reviews, and Mc-
Clure et al. (2006) for a more recent review.
The transition from mini-rules to features and linking rules
requires two additional steps, as shown in the following.
Generalization
Mini-rules are verb-specific, but they may be generalized to
other verbs. The child’s early verbs are predominantly activity
verbs, and the lexical entries of most of these, include Cause
and Control. Here I will describe schematically how the notions
“cause’” and “control” develop and become associated with lin-
king rules.
Suppose that a child has experienced the utterance Jane
pushes the baby carriage and later on encounters the utterance
Jimmy eats the cookie. In what way are these two sentences
similar? The similarity the child fastens onto will not be that
between Jimmy and Jane (perhaps Jimmy will be perceived to
be more similar to some boy than to a girl like Jane). Nor are
the activities referred to by the verbs push and eat very similar.
Rather, the similarity is between two relations, namely, the
relation between Jimmy and the activity of eating and that be-
tween Jane and the activity of pushing. Perceiving a similarity
between the two events highlights the notions they have in
common: Jane and Jimmy both initiate the activity i.e., partake
of the notion “cause”, and are in control of it. This similarity
between the described events will be reinforced by similarity
between the utterances expressing them: the words for Jane and
Jimmy appear both in pre-verbal position. Since most activity
verbs involve both causation and control, it is reasonable to
assume that they constitute initially a single complex notion,
and only when the child has to deal with verbs having only one
of them, does differentiation set in.
Between any two situations there may be several relatively
salient similarities. In the case of Jane’s pushing and Jimmy’s
eating, for example, the child might note that both Jane and
Jimmy hold something in their hands. Such similarities, call
them “notions” if you will, will be short-lived, however. They
will not be frequently confirmed by additional utterances hav-
ing similar structures, whereas the notions “cause” and “con-
trol” will be. Thus, when she encounters an utterance like Don
is pushing Bill, Don may be holding nothing in his hands, but
will be conceived of as causing the action and being in control.
Incipient notions that have no function in the rule system of the
language will soon be weeded out.
As the notions “cause” and “control” appear in association
with additional verbs they become gradually more salient, and
the tendency will be strengthened to generalize a mini-rule for
one verb to other verbs associated with the same notions.
On the hypothesis that the child forms mini-rules and then
generalizes them, one may expect her early rules to be re-
stricted to narrow classes of semantically similar verbs. This is
because the greater the similarity in meaning between any two
verbs, the more likely is the mini-rule of one to generalize to
the other. In fact, in the earliest stages of their linguistic devel-
opment, children have often been found to deploy a rule for
narrow categories of verbs, such as experiencer verbs or verbs
of communication; see Braine (1976).
Formation of Features
As mentioned, notions are cognitive constructs and not se-
mantic ones. In the mature linguistic system, linking rules are
defined in terms of features, not of notions. Let us see now how
a notion like “cause” crystallizes into the feature Cause.
A feature is:
a) a relatively well-delimited concept, and
b) accessed every time a given verb is used, irrespective of
the situation it refers to.
Delimitation of features, a), is essentially a process of con-
cept formation, like vocabulary acquisition. To learn the exten-
sion of dog, the child will note similarities (of shape and be-
havior) between, say, the neighbor’s dog and the family’s poo-
dle. But this is not sufficient: a sheep is also similar to the poo-
dle, in some ways perhaps more so than the neighbor’s dog. A
linguistic factor clinches the matter: the child observes that the
neighbor’s dog is referred to by the word dog, whereas a sheep
is not. The linguistic factor that is common to verbs with a
“causer”, for instance, is the way they are realized in sentences
by Feature linking rules. The child notes that sentences with
internal causers are constructed in the same way as those with
external ones, and thus forms the feature Cause. Similarly, the
same Feature linking rules apply not only to verbs referring to
changes of location (like fall) but also to changes of state (as in
melt), and thus the feature Change is delimited29.
Independence of specific situations, b), will be attained by
experience with the notion in question. For instance, the child
will eventually find out that in referring to actions like cough-
ing, one deploys the same linguistic structures even on occa-
sions where the perpetrator apparently has no control over them.
In sum, features develop in the course of acquiring linking
rules, and linking rules are formed on the basis of features; both
are acquired concomitantly. The language learner “scrambles
up an intellectual chimney, supporting himself against each side
by pressures against the others”, in Quine’s (1960: p. 93) apt
phrase.
The child at first uses mainly activity verbs, but she will hear
28The grammatical concept of subject is not a linguistic universal (Schachter
& Otanes, 1972; Evans & Levinson, 2009) and so cannot be assumed to be
available to the child. Linking has to be to surface properties like position in
the sentence, or (in many languages) to a NP with an affix.
29Naturally, the child may for some time underextend a feature category or
conflate it with another one (Cause with Attr-St, for instance), but eventually
the Feature linking rules operating in adult language lead to her zeroing in
on the adult categories.

I. M. SCHLESINGER
and produce some utterances with stative verbs, the subject
participants of which have neither Cause nor Control, but rather
Change or Attr-St. How can the language learner tell the vari-
ous features apart? If all features were acquired simultaneously,
this would be a serious problem, indeed. It appears, however,
that features are acquired successively. In early stages of acqui-
sition, the number of stative verbs (like want, sit and lie) is
rather small compared to that of activity verbs. Presumably, the
initial mini-rules for these verbs will be retained until the fea-
tures Control and Cause have been fairly well established. Only
when stative verbs have been acquired in greater numbers, will
the feature Attr-St begin to be formed. As for Control and
Cause, considering that most activity verbs involve both of
them, these two features are presumably acquired simultane-
ously and constitute a single complex feature, which is bifur-
cated only much later, when the child has to deal with verbs
having only Control or only Cause.
Acquisition of the feature Change proceeds essentially along
the lines outlined for Attr-St. At the time Control and Cause are
being formed, the child merely has mini-rules for verbs with the
feature Change30. But this feature poses an additional problem,
because it may be realized either as subject or as direct object.
This is why the distinction between resulting and non-resulting
features has been introduced. Acquiring this distinction should
not pose any special difficulty, but possibly the child does not
need any rules based on this distinction: she may circumvent it.
She may learn a) the rather simple rule that a participant with
Change does not qualify for subject position when another
participant has Cause or Control, and b) the rules for a handful
of verbs (such as lead, march and sleep) in an item-by-item
manner. The only feature which apparently cannot be acquired
by the abstraction process described in the foregoing is the
negatively defined “default feature”’ ø, but this does not im-
pede acquisition, as will be shown in the next subsection.
Once features have been abstracted and the linking rules per-
taining to them have been formed, they will readily be identi-
fied in subsequently learned verbs. Eventually there will be no
longer any need to form mini-rules for new verbs. To learn how
to construct a sentence with, say, touch, it is enough to learn the
meaning of this verb, i.e., to know that this verb has two par-
ticipants: the “toucher” and the “touchee” and that the “toucher”
has the features Control and Cause. To say that X touches Y,
the child merely has to apply the previously acquired Feature
linking rule so as to position the word or phrase for X before
touch. In comprehension, the Feature linking rule will be ap-
plied in reverse: If a NP is preverbal, then it may stand for the
participant with Control (in this example: the “toucher”, not the
“touchee”).
Acquisition of Objects
Some research on child language suggests that the child’s
first direct objects have the feature Change. Slobin (1979, quoted
in Hopper and Thompson 1980) reports the case of a Russian
speaking child who at first used the accusative suffix only for
verbs denoting “direct physical action on things”, like give, carry,
put, and throw, that is, for verbs with a Change object partici-
pant), and not for verbs like read or see31.
A limited linking rule to the effect that the participant with
Change becomes a direct object in postverbal position may suf-
fice for some time (because verbs like die and grow up, whose
subjects have Change, presumably are not acquired in the ear-
liest stages, and if they do, they will function only in mini-
rules). At the two- and three-word stages, however, appear verbs
like read, buy, ask, forget, love, and like with their direct ob-
jects (see, e.g., Braine, 1976: pp. 42-51; Tomasello, 1992: pp.
119-140, for English speaking; and Fortus, 1996: pp. 88-89, for
Hebrew speaking children). The direct objects of these verbs do
not refer to affected entities. They have the default feature ø,
and ø is the only feature that cannot be abstracted in the way
described above.
There are two possible strategies for mastering the linking
rules for direct objects:
a) The child forms a separate rule for each verb. For instance,
she observes that when the verb want is used, the thing wanted
is treated linguistically just like push and throw, verbs that in-
volve affected entities. Such item-specific learning will have to
be supplemented by the second strategy:
b) The child learns a negative rule (see sub-section on Se-
mantics of the object category): The participant that lacks fea-
tures linking it to the subject is realized as a direct object; the
participant with the criterial features become subjects, and the
other participant becomes the direct object (see Braine & Hardy,
1982: p. 224, for a similar proposal).
It might be objected that well before there is evidence that
they have mastered the subject-verb pattern many children use
the verb-direct object pattern correctly (e.g., Tomasello, 1992).
But no definite conclusion can be drawn from the child’s verbal
production, because she may be aware that an entity having
Cause is involved in the situation she talks about and still fail to
refer to that entity in her linguistic output. She may say, for
instance, push ball without mentioning the agent of pushing
(perhaps to save effort or because the identity of the agent is
self-evident), but she will be aware that there is a “pusher”32. In
this connection it is of interest that in some corpora, SVO pre-
cede VO constructions (Fortus, 1996; Keren-Portnoy in prep.).
An apparent difficulty with the above-mentioned strategy of
learning a negative rule is that it assumes that the child can
distinguish between direct and oblique objects. But this distinc-
tion may be acquired in stages. At an early stage of learning,
oblique objects are presumably treated in the same way as di-
rect objects. In fact, in many corpora of child language oblique
objects appear without prepositions33. The question then be-
comes: How does the child move from this broad non-subject
category to the mature direct object category?
As pointed out above, there is no sure-fire semantic criterion
for the distinction between direct and oblique objects. It is plau-
sible, though, that prepositions serve as a lever for narrowing
down the non-subject category. Suppose the child repeatedly
hears certain NPs preceded by the preposition from (e.g., take
the milk from the refrigerator, pull it from the shelf) and notices
30In a series of experimental tasks, Guberman (1992) found children’s, and
in some cases adults’, performance with stative verbs to be poorer than with
activity verbs. See also Gropen et al. (1991b) for an experimental study
p
ertaining to the children’s acquisition of the concept of change (their term
is “object affectedness”).
31But Ninio (1999: p. 647) found that the first verbs have only “low impact
on their patients, such as change of possession”.
32Goldberg (2006) reports on a study of a child learning Russian who failed
to pronounce a verb in her utterance, while having an implicit knowledge o
f
it.
33Here are some examples culled from the CHILDES corpora: 1. Instrument
without preposition: sweep broom, write pencil, eat hand. 2. Location with-
out preposition: walking street, sitting chair, lie down stool.

I. M. SCHLESINGER
248
that from marks the expression standing for source of the mo-
tion. From the earlier non-subject category she thus separates
out a new category, Source of motion, which has its own rules
of realization. Later on, the same preposition, from, may appear
with a somewhat different role, as in she got the book from her
brother, and she forms an additional category. In this manner,
non-subject NPs begin gradually to fall into classes according
to the roles they play in the sentence (instrument, location, time,
and so on), and Role linking rules are acquired.
Acquisition of Thematic Roles and of Features
Compared
In the following, the acquisition theory outlined in the pre-
ceding is compared with accounts based on thematic roles.
Features Simpler than Roles
Intuitively, it should be easier to attend to semantic features
than to the more global thematic roles. Discussing the advan-
tage of his “entailments” over thematic roles, Dowty observes.
It is certainly not obvious that in ordinary reasoning and
conversation people directly pay attention to or worry about
whether something really was or was not a Theme or a Source
or an Agent (in some sense of Theme, etc.); but we do concern
ourselves all the time, both in everyday life and in courts of law,
and sometimes to a painstaking degree, with whether an act was
really volitional or not, whether something really caused some-
thing or not, … whether something was moving or stationary,
whether something changed in a certain way or not… (Dowty,
1991: p. 575).
Acquiring Thematic Hierarchies
Suppose that, contrary to the present theory, the selection of
subject and direct object is based on thematic roles. These
would have to be formed, like features, in the course of the
child’s cognitive development. Concepts such as Agent and
Patient do not come ready-made. They start out as cognitive
notions that must be crystallized into well-delineated concepts,
if they are to be deployed in the child’s linguistic system. In
this respect, a system based on features seems to have an ad-
vantage, since the number of features is much smaller than that
of thematic roles.
Furthermore, according to current theories, subject and ob-
ject selection are based on a thematic hierarchy. Acquiring such
a hierarchy would be a much more arduous task than learning
the Feature linking rule in 28). The child is never exposed to
such a hierarchy in its entirety, and so would have to piece it
together by means of pairwise comparisons. Authors are not
agreed on the number of roles in a hierarchy, which may be up
to 10 or 15 (Baker, 1997: pp. 107-108), and even on the unreal-
istic assumption that the child makes maximum use of transi-
tivity (if X comes before Y and Y before Z, then X comes be-
fore Z), mastering the hierarchy would require a large number
of comparisons to be made. For instance, Jackendoff’s (1990b:
pp. 259-260) list of roles for transitive verbs would require 18
such comparisons (e.g., Theme precedes Location, etc.)34. By
contrast, mastering the simple linking rules is a much less ar-
duous task than learning a thematic role hierarchy would be. It
is also simpler than counting (and perhaps assigning weights to)
Dowty’s (1991) “entailments”.
Another advantage of accounting for subject and object se-
lection in terms of features is that it provides a simpler solution
to the problem of the many-to-many relation between semantic
and syntactic concepts. Here I will go into this at some length.
Semantic-Syntactic Mapping—A Previous Account
In the early 1970s, it was proposed that what the child learns
is how language expresses semantic relations like Agent and
Patient, i.e., thematic roles (e.g., Bowerman, 1973; Schlesinger,
1971; see also Slobin, 1970). An objection that was advanced
against this so-called “semantic approach” from very early on
was that such thematic roles are not congruent with formal
syntactic ones. An Agent can be not only the sentence subject,
but also a NP in a by-phrase (in a passive sentence) or in a with-
phrase (I swept the room with Denis), and conversely, subjects
can be not only Agents, but also Patients (the subjects of many
passive sentences), Themes (London Bridge is falling down),
Experiencers (Celia admires Doris), Natural Forces (The storm
devastated the village), and Instruments (The new blower dried
her hair fast). Likewise, there are direct objects that are not
Patients, and conversely, not all Patients are direct objects.
At the time this objection began to be voiced, the concept of
a thematic hierarchy was already taking shape (Fillmore, 1968),
albeit not under this name. But the language acquisition litera-
ture did not resort to it in dealing with this objection, perhaps
because it was felt that mastering a system of precedence rela-
tions would be too arduous a task for the young child to tackle.
Instead, a different solution began to be developed: Subjects
and direct objects that are not Agents or Patients, it was argued,
are absorbed into these semantic categories. Referents of cer-
tain verbs may be conceived of by the child as somehow related
to those denoting straightforward activities carried out by agents.
Schlesinger (e.g., 1974, 1988) proposed that the extension of
semantic categories proceeds on the basis of semantic as well as
of formal similarity. This process, which was dubbed semantic
assimilation, would lead to a gradual expansion of the original
agent category, so as to include experiencers (Schlesinger, 1992)
and instruments (Schlesinger, 1989), until it coincides with the
mature subject category. Similarly, Maratsos (1981) and Toma-
sello (2000) have argued that semantic categories may be ex-
tended so as to comprise atypical instances.
Here are some factors that might account for the semantic
assimilation of subjects that are not, strictly speaking, Agents
(Schlesinger, 1988):
a) Similarity:
The child who hears an utterance like 51a) may conceive of
the bridge as moving across the river just as a person walking
across it would be. 51b) is another example of metaphorical
extension, or of “fictive action”, as MacWhinney (1999: p. 231)
calls it: the shelf is anthropomorphized and viewed as “hold-
ing” objects, just as a human would.
51) a. The bridge crosses the river.
b. This shelf holds twenty books.
b) Contrast:
52) a. Herb is lying on the couch.
b. Herb is standing in the doorway.
Lying is not an activity, and Herb cannot properly be re-
garded as an Agent in the full sense of the term. The same may
be said of standing, which does not necessarily involve a typi-
cal Agent. However, lying and standing are treated as Agents
because they contrast with verbs of bodily movement, such as
34If a hierarchy were to contain a much smaller number of thematic roles,
each role would tend to be more comprehensive, and the child would face
the
p
roblem of formin
g
a more
g
eneral, i.e., abstract, conce
p
t.

I. M. SCHLESINGER
walking and running. Similarly, falling may be regarded in con-
trast to jumping or walking, and therefore when a person who is
asserted to have fallen is referred to by a word in subject posi-
tion, he is conceived of as an Agent. Eventually, not only fal-
ling people, but also falling objects will be regarded as Agents
due to the similarity factor, a).
c) Embeddedness in activity:
53) a. Ronald is sleeping in his armchair.
b. Ronald is waiting for his sister.
The verb sleep will be heard by the child in situations in
which various activities are going on: lying down, closing one’s
eyes, turning round in bed, etc. Formal similarity will also be a
factor here (as in semantic assimilation generally): The child
notes that the word referring to the sleeper is in pre-verbal posi-
tion, and thus comes to regard the sleeper as an Agent. Simi-
larly, waiting is normally accompanied by activities like walk-
ing up and down, making inquiries, and the like, so that 53b) is
conceived of as part of a scene where Ronald is active as an
Agent.
In this manner, so the semantic assimilation hypothesis went,
subjects that are not literally speaking Agents, come to be re-
garded as Agents and expressed linguistically like them. It has
to be admitted that in some instances the proposed explanations,
while plausible, are rather ad hoc. See also Bowerman (1974)
for critical comments on the semantic assimilation hypothesis.
A later finding by Bowerman (1990) is also difficult to rec-
oncile with the claim that semantic categories are gradually
extended to accommodate non-typical instances. Bowerman
argues as follows. If thematic roles like Agent and Patient are
extended in the course of acquisition, one would expect “real”,
unextended Agents and Patients to be acquired before extended
ones. But she found that her two daughters made no fewer er-
rors of word order in their sentences with prototypical Agents
and Patients than in those with non-prototypical Agents and
Patients35.
Let us see now how the mismatch between semantic and syn-
tactic categories can be accounted for by a theory based on se-
mantic features.
Semantic-Syntactic Mapping—A Feature Based Account
The problem the semantic assimilation hypothesis was de-
signed to account for was that semantic categories and syntactic
ones stand in a many-to-many relation. According to the pre-
sent theory, subject selection is based on the presence of one or
more of four features. But these features may appear also in
subjects that are not typical agents (as in examples 2) - 5)). Let
us see now in some detail how this works out.
In 52), Herb matches perfectly the subject participants of lie
and stand, which have Cause, Control, and Attr-St; see 6). Ro-
nald in 53) matches the subject participants of sleep and wait in
respect to Control (and arguably also in respect to Attr-St). In
the case of 51), the present account converges on the one given
by the semantic assimilation hypothesis: As pointed out above,
matching does not require the referent of the NP to have prop-
erties corresponding to all the features in the participant in the
lexical entry, but only a general similarity in meaning between
the NP and the participant. In 51a), the mental representation of
the bridge corresponds with that of the subject participant in the
lexical entry of cross, because a bridge can be imagined as
crossing a river just as a boat or a swimmer may do. In the
same way, this shelf in 51b) matches with the subject partici-
pant of hold.
Bowerman’s (1990) findings, alluded to above, do not con-
flict with feature theory, because our features have a wider ap-
plicability than Agent and Patient roles. Among the verbs clas-
sified by Bowerman as taking non-prototypical roles are draw,
write, look, talk, cook, ride, and play, and all these verbs have
subject participants with the features Cause and Control.
Summary
Linking rules for subject and direct object selection have
been formulated in terms of a set of four semantic features, and
the distinction between resulting and non-resulting features.
Unlike thematic roles, semantic features are not ascribed to NPs
in sentences but to participants in lexical entries. The assump-
tion made here is that the lexical entry of a verb contains either
one or two participants with features. Thematic roles are invol-
ved only in linking rules for other syntactic functions, notably
oblique objects. The theory can account for phenomena that are
difficult to explain in terms of thematic roles and of proto-roles.
In this paper, it has been worked out for subjects and objects in
simple sentences in English, and it is shown how it affords a
viable theory of native language acquisition. It remains to be
seen whether additional semantic features are required for the
explanation of other linguistic regularities and for other lan-
guages.
Acknowledgements
I am very grateful to Moshe Anisfeld, Ainat Guberman, Ta-
mar Keren-Portnoy, Yonata Levy, Michael Maratsos, Anita
Mittwoch, Anat Ninio, Dan Slobin, and Virve Vihman for im-
portant comments on a draft of this paper. Thanks also to Shlo-
mo Danziger for the care taken in preparing the manuscript for
publication.
REFERENCES
Anderson, S. R. (1971). On the role of deep structure in semantic inter-
pretation. Foundations of Language, 7, 387-396.
Andrews, A. D. (2007). The major functions of the noun phrase. In T.
Shopen (Ed.), Language typology and syntactic description (pp. 64-
154). Cambridge: Cambridge University Press.
Baker, M. C. (1988). Incorporation: A theory of grammatical function
changing. Chicago, IL: University of Chicago Press.
Baker, M. C. (1997). Thematic roles and syntactic structure. In L. Hae-
geman (Ed.), Elements of grammar: Handbook in generative syntax
(pp. 72-137). Dordrecht: Kluwer. doi:10.1007/978-94-011-5420-8_2
Beavers, J. T. (2006). Argument/oblique alternations and the structure
of lexical meaning. Stanford, CA: Stanford University dissertation.
Belletti, A., & Rizzi, L. (1988). Psych-verbs and θ theory. Natural Lan-
guage and Linguistic Theory, 6, 291-352. doi:10.1007/BF00133902
Bowerman, M. (1973). Early syntactic development. Cambridge: Cam-
bridge University Press.
Bowerman, M. F. (1974). Discussion summary—Development of con-
cepts underlying language. In R. L. Schiefelbusch, & L. L. Lloyd
(Eds.), Language perspectives—Acquisition, retardation, and inter-
vention (pp. 191-209). Baltimore, MD: University Park Press.
Bowerman, M. (1990). Mapping thematic roles onto syntactic functions:
Are children helped by innate linking rules? Linguistics, 28, 1253-
1289. doi:10.1515/ling.1990.28.6.1253
Bowerman, M., & Croft, W. (2008). The acquisition of the English
causative alternation. In M. Bowerman, & P. Brown (Eds.), Cross-
35Bowerman’s study was intended to test Pinker’s (1984, 1989) hypothesis
that these roles are innate.

I. M. SCHLESINGER
250
linguistic perspectives on argument structure (pp. 279-307). New
York, NY: Lawrence Erlbaum Associates.
Braine, M. D. S. (1976). Children’s first word combinations. Monogra-
phs of the Society for Research in Child Development 41 (1, Serial
No. 164). doi:10.2307/1165959
Braine, M. D. S., & Hardy, J. A. (1982). On what cases there are, why
they are, and how they develop: An amalgam of a priori considera-
tions, speculation, and evidence from children. In E. Wanner, & L.
Gleitman (Eds.), Language acquisition: The state of the art (pp. 219-
239). Cambridge: Cambridge University Press.
Chomsky, N. (1982). The generative enterprise: A discussion with Riny
Huybregts and Henk van Riemsdijk. Dordrecht: Foris.
Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT
Press.
Davis, A. R. (2001). Linking by types in the hierarchical lexicon. Stan-
ford, CA: CSLI Publications.
Davis, A. R., & Koenig, J.-P. (2000). Linking as constraints on word
classes in a hierarchical lexicon. Language, 76, 56-91.
Dowty, D. (1989). On the semantic content of the notion of “thematic
role”. In G. Chiercha, B. H. Partee, & R. Turner (Eds.), Properties,
types and meaning (pp. 69-129). Dordrecht: Kluwer.
doi:10.1007/978-94-009-2723-0_3
Dowty, D. (1991). Thematic proto-roles and argument selection. Lan-
guage, 67, 547-619.
Evans, N., & Levinson, S. C. (2009). The myth of language universals:
Language diversity and its importance for cognitive science. Behav-
ioral and Brain Sciences, 32, 429-492.
doi:10.1017/S0140525X0999094X
Fillmore, C. J. (1968). The case for case. In E. Bach, & R. T. Harms
(Eds.), Universals in linguistic theory (pp. 1-88). New York, NY:
Holt, Rinehart, and Winston.
Fillmore, C. (1970). The grammar of hitting and breaking. In R. A.
Jacobs, & P. A. Rosenbaum (Eds.), Readings in English transforma-
tional grammar (pp. 120-133). Waltham, MA: Ginn.
Fillmore, C. J. (2003). Form and meaning in language, vol. 1. Stanford,
CA: Center for the Study of Language and Information.
Fortus, R. (1996). Semantic relations in the acquisition of language: An
analysis of one child’s first word combinations. MA Thesis, Jerusa-
lem: Hebrew University of Jerusalem.
Gelman, S. A., & Opfer, J. E. (2002). Development of the animate-
inanimate distinction. In U. Goswami (Ed.), Handbook of childhood
cognitive development (pp. 151-166). Malden, MA: Blackwell.
doi:10.1002/9780470996652.ch7
Gleitman, L. R., Gleitman, H., Miller, C., & Ostrin, R. (1996). Similar,
and similar concepts. Cognition, 58, 321-376.
doi:10.1016/0010-0277(95)00686-9
Goldberg, A. E. (2006). Constructions at work: The nature of generali-
zation in language. New York, NY: Oxford University Press.
Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press.
Gropen, J., Pinker, S., Hollander, M., Goldberg, R., & Wilson, R.
(1989). The learnability and acquisition of the dative alternation in
English. Language, 65, 203-257. doi:10.2307/415332
Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991a). Affect-
edness and direct objects: The role of lexical semantics in the acqui-
sition of verb argument structure. Cognition, 41, 153-195.
doi:10.1016/0010-0277(91)90035-3
Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991b). Syntax
and semantics in the acquisition of locative verbs. Journal of Child
Language, 18, 115-151. doi:10.1017/S0305000900013325
Guberman, A. (1992). The development of the verb category in the He-
brew child language. Ph.D. Thesis, Jerusalem: Hebrew University.
Hoekstra, T. (1992). Aspect and Theta Theory. In I. M. Roca (Ed.),
Thematic structure: Its role in grammar (pp. 145-174). Berlin: Foris.
doi:10.1515/9783110872613.145
Hopper, P. J., & Thompson. S. A. (1980). Transitivity in grammar and
discourse. Language, 56, 251-299.
Hudson, R. (1992). So-called “double objects” and grammatical rela-
tions. Language, 68, 251-276. doi:10.2307/416941
Jackendoff, R. (1987). The status of thematic relations in linguistic
theory. Linguistic Inquiry, 18, 369-411.
Jackendoff, R. (1990a). On Larson’s treatment of the double object
construction. Linguistic Inquiry, 21, 427-456.
Jackendoff, R. (1990b). Semantic structures. Cambridge, MA: MIT
Press.
Jespersen, O. (1933). Essentials of English grammar. London: Allen &
Unwin.
Kako, E. (2006). Thematic role properties of subjects and objects. Cog-
nition, 101, 1-42. doi:10.1016/j.cognition.2005.08.002
Kasof, J., & Lee, J. Y. (1993). Implicit causality as implicit salience.
Journal of Personality and Social Psychology, 65, 877-892.
doi:10.1037/0022-3514.65.5.877
Keren-Portnoy, T. (In preparation). Long is easy, short is hard.
Kiparsky, P. (2001). Structural case in Finnish. Lingua, 111, 315-376.
doi:10.1016/S0024-3841(00)00035-8
Leslie, A. M., & Keeble, S. (1987). Do six-month old infants perceive
causality? Cognition, 25, 265-288.
doi:10.1016/S0010-0277(87)80006-9
Levin, B. (1993). English verb classes and alternations: A preliminary
investigation. Chicago, IL: University of Chicago Press.
Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the
syntax-lexical semantics interface. Cambridge, MA: MIT Press.
Levin, B., & Rappaport Hovav, M. (2005). Argument realization.
Cambridge: Cambridge University Press.
doi:10.1017/CBO9780511610479
MacWhinney, B. (1999). The emergence of language from embodiment.
In B. MacWhinney (Ed.), The emergence of language (pp. 213-256).
Mahwah, NJ: Lawrence Erlbaum.
Maratsos, M. P. (1979). How to get from words to sentences. In D.
Aronson, & R. W. Rieber (Eds.), Psycholinguistic research: Implica-
tions and applications (pp. 285-356). Hillsdale, NJ: Lawrence Erl-
baum.
Maratsos, M. (1981). Problems in categorial evolution: Can formal
categories arise from semantic ones? In W. Deutsch (Ed.), The
child’s construction of language (pp. 245-261). London: Academic
Press.
McClure, K., Pine, J. M., & Lieven, E. V. M. (2006). Investigating the
abstractness of children’s early knowledge of argument structure.
Journal of Child Language, 23, 693-720.
doi:10.1017/S0305000906007525
McRae, K., Todd, R., Ferreti, T. R., & Amyote, L. (1997). Thematic
roles as verb-specific concepts. Language and Cognitive Processes,
12, 137-176. doi:10.1080/016909697386835
Newmeyer, F. J. (2002). Optimality and functionality: A critique of
functionally-based optimality-theoretical syntax. Natural Language
and Linguistic Theory, 20, 43-80. doi:10.1023/A:1014290005775
Ninio, A. (1999). Pathbreaking verbs in syntactic development and the
question of prototypical transitivity. Journal of Child Language, 26,
619-653. doi:10.1017/S0305000999003931
Ninio, A. (2006). Language learning and the learning curve. Oxford:
Oxford University Press.
doi:10.1093/acprof:oso/9780199299829.001.0001
Pinker, S. (1984). Language learnability and language development.
Cambridge, MA: Harvard University Press.
Pinker, S. (1989). Learnability and cognition: The acquisition of argu-
ment structure. Cambridge, MA: MIT Press.
Premack, D. (1990). The infant’s theory of self-propelled objects. Cog-
nition, 36, 1-16. doi:10.1016/0010-0277(90)90051-K
Quine, W. Van O. (1960). Word and object. Cambridge, MA: MIT
Press.
Rappaport, M., & Levin, B. (1988). What to do with θ-roles. In W.
Wilkins (Ed.), Syntax and semantics, vol. 21: Thematic relations (pp.
7-36). San Diego: Academic Press.
Rappaport Hovav, M., & Beth Levin, B. (2008). The English dative
alternation: The case for verb sensitivity. Journal of Linguistics, 44,
129-167. doi:10.1017/S0022226707004975
Rozwadowska, B. (1988). Thematic restrictions on derived nominals.
In W. Wilkins (Ed.), Syntax and semantics, vol. 21: Thematic rela-
tions (pp. 147-165). San Diego: Academic Press.
Rosta, A. (2002). Review of the book cognitive space and linguistic
case, by Izchak M. Schlesinger. Journal of Pragmatics, 34, 69-78.
doi:10.1016/S0378-2166(01)00013-3
Schachter, P., & Otanes, F. T. (1972). Tagalog reference grammar.

I. M. SCHLESINGER
Berkeley: University of California Press.
Schlesinger, I. M. (1971). Production of utterances and language acqui-
sition. In D. I. Slobin (Ed.), The ontogenesis of grammar (pp.
63-101). New York: Academic Press.
Schlesinger, I. M. (1974). Relational concepts underlying language. In
R. L. Schiefelbusch, & L. L. Lloyd (Eds.), Language perspectives-
acquisition, retardation and intervention (pp. 129-151). Baltimore,
MD: University Park Press.
Schlesinger, I. M. (1977). Production and comprehension of utterances.
Hillsdale, NJ: Lawrence Erlbaum Associates.
Schlesinger, I. M. (1988). The origin of relational categories. In Y.
Levy, I. M. Schlesinger, & M. D. S. Braine (Eds.), Categories and
processes in language acquisition (pp. 121-178). Hillsdale, NJ:
Lawrence Erlbaum.
Schlesinger, I. M. (1989). Instruments as agents: On the nature of se-
mantic relations. Journal of Linguistics, 25, 189-210.
doi:10.1017/S0022226700012147
Schlesinger, I. M. (1992). The experiencer as an agent. Journal of
Memory and Language, 31, 315-332.
doi:10.1016/0749-596X(92)90016-Q
Schlesinger, I. M. (1995). Cognitive space and linguistic case: Semantic
and syntactic categories in English. Cambridge: Cambridge Univer-
sity Press. doi:10.1017/CBO9780511551321
Shibatani, M. (1996). Applicatives and benefactives: A cognitive ac-
count. In M. Shibatani, & S. A. Thompson (Eds.), Grammatical con-
structions: Their form and meaning (pp. 157-194). Oxford: Claren-
don Press.
Siewierska, A. (1991). Functional grammar. London: Routledge.
Slobin, D. (1970). Universals of grammatical development in children.
In G. B. Flores D’Arcais, & W. J. M. Levelt (Eds.), Advances in
psycholinguistics (pp. 174-186). Amsterdam: North Holland.
Slobin, D. (1979). The role of language in language acquisition. Invited
address to the 50th Annual Meeting of the Eastern Psychological As-
sociation, Philadelphia. Unpublished ms., Berkeley, CA: University
of California.
Tomasello, M. (1992). First verbs: A case study of early grammatical
development. New York: Cambridge University Press.
doi:10.1017/CBO9780511527678
Tomasello, M. (2000). Do young children have adult syntactic compe-
tence? Cognition, 74, 209-253. doi:10.1016/S0010-0277(99)00069-4
Tomasello, M. (2003). Constructing a language. Cambridge, MA: Har-
vard University Press.
Tversky, A. (1977). Features of similarity. Psychological Review, 84,
327-352. doi:10.1037/0033-295X.84.4.327
Van Valin Jr., R. D. (2005). Exploring the syntax-semantic interface.
Cambridge: Cambridge University Press.
doi:10.1017/CBO9780511610578
Van Valin Jr., R. D., & Wilkins, D. P. (1996). The case for ‘Effector’:
Case roles, agents, and agency revisited. In M. Shibatani, & S. A.
Thompson (Eds.), Grammatical constructions: Their form and
meaning (pp. 289-322). Oxford: Clarendon Press.
Wilkins, W. (1988). Thematic structure and reflexivization. In W. Wil-
kins (Ed.), Syntax and semantics, Vol. 21: Thematic relations (pp.
191-213). San Diego: Academic Press.
Witman, S. (2000). Mussa’im yeshirim va’akifim betsarfatit, ivrit
ve’anglit: Mehkar hashva’ati [Direct and oblique objects in French,
Hebrew, and English: A comparative study]. Unpublished paper, Je-
rusalem: Hebrew University of Jerusalem.
Zubizaretta, M. L. (1987). Levels of representation in the lexicon and in
the syntax. Dordrecht: Foris.