(17) a. The wind opened the door
b. The door was opened by the wind
There is no motivation for introducing a further semantic relation, ‘Force’
(Huddleston 1970), in this latter instance either, or for its assimilation to
Instrumentals (Fillmore 1971:§5(b)). The ‘displaced’ Force in (17b) is marked
with by, as a (non-propositional) ‘agent’.
‘Instrumental’ is only circumstantial. The diVerentiation between the pre-
verbal arguments in the Kewa sentences in (18) (from Palmer (1994: 48), citing
Franklin (1971: 62)) thus doesn’t reXect a distinction in participant role:
(18)a.a
´
a
´
-me
´
re
´
pena po
´
a
´
-a
man-
AGT tree cut-did
(‘The man cut the tree’)
b. raı
´
-mi ta
´
-a
axe-

this inconclusive review of ‘case criteria’.
On the basis of such arguments as we have looked at in this chapter, and
others, Cook (1978; 1979) envisages Wve propositional ‘cases’, which he pre-
sents as in (20):
(20) (Experiencer)
(Agent) (Benefactive) Object
(Locative)
The brackets indicate optional presence in a proposition; Cook assumes that
the Object(ive) is obligatory (cf. Gruber 1965; Anderson 1971b: 37 ; Taylor 1972;
Starosta 1978; 1988:§4.2.1.4), and we shall return to this. But the three ‘cases’
presented vertically in (20) are regarded as mutually exclusive. This oVends
against syntagmatic contrastiveness, however: ‘cases’ cannot be complemen-
tary. And it does not seem to be correct as an observation. Consider, for
example, (21):
(21)JeV derived considerable pleasure from the expedition
Here we seem to have, from right to left, a propositional Locative, an
Objective, and a Benefactive or Experiencer.
The situation is a little more complex, then, though there is something to
Cook’s suggestion: these ‘cases’ are related in some way. Attempting to
describe this relationship will bring us on to the approach to the identiWcation
of ‘cases’ given as (c) in (1), which we take up in §5.4:
(1) The identiWcation of case(s)
c) the content of case
The relationship underlying Cook’s suggestion emerges in one particular
attempt to address (1c) that we’ll look at there.
Let us note also that the kind of perception that underlies Cook’s sugges-
tion has also led to the recognition of the some variety of ‘componentiality’
for ‘case’. Nilsen (1972), for instance, argues for a characterization of some-
thing like the traditional ‘cases’ as bundles of semantic feature values. And
suggestions that some ‘cases’ at least could be combined (as initially in this

interpretation at least, is a straightforward Agentive, and even in its case
Agentive is combined with Objective, since the denotatum of the subject
combines the source of the action with the entity subjected to it. The subject
of (22b) is a straightforward Objective, though the verb also, by virtue of its
meaning, imposes an animacy requirement on the Objective argument. The
subject of (22c) is what Fillmore (1968a) called a Dative (roughly what has
subsequently been known as an ‘Experiencer’). We can again associate Agen-
tive with the subject of (22d), but in this instance it is not combined with
Objective (but rather, in terms of the analysis developed in the following
chapter, a Goal). There is no ‘problem’ of accounting for why all of these ‘sub-
roles’ can be combined into the role of ‘Agent’: only (22a) and (22d) involve
Agentive, and each of them involves in addition a diVerent relation. There is
no such ‘combining’; there is no ‘problem’.
90 Modern Grammars of Case
Dowty (1989) also reintroduces the familiar examples of ‘symmetric(al)
predicates’, ‘psych predicates’, and pairs like buy and sell (discussed in, for
example, Fillmore 1972). Concerning these last, Dowty observes, concerning
the pair in (23), as expressed by Ackerman and Moore (2001: 24), that ‘on an
intuitive level one would assume that, e.g., Max is an AGENT and Mary is a
RECIPIENT in both the (a) and (b) sentences’:
(23) a. Max sold the piano to Mary for $1,000
b. Mary bought the piano from Max for $1,000
Appeals to ‘an intuitive level’ are always suspect; linguists’ intuitions (which
are of diverse origins) are not ev idence. And in the present case (as discussed
in §4.2.2), it doesn’t follow from the assumption that the ‘vendor’ in (23a) is
an Agentive, as source of the immediate action described by the verb, that the
‘vendor’ in (23b) is presented as an Agentive. In the latter instance it is rather
the ‘customer’ that is presented as the source of the immediate action, even
though the same ‘real-world’ event may be being referred to by (23a) and
(23b). To maintain otherwise is to succumb to ‘the objectivist’s misconcep-

(25) a. Your brother resembles a horse
b. There is a horse that resembles your brother
But this has to do with the avoidance of indeWnites as subjects. And (25b)
illustrates that, on a non-generic interpretation, it is possible for the indeWnite
to function ‘indirectly’ as the subject of such a verb. Likewise, though the
equative (26a) is not obviously reversible, (26b) allows ‘indirect subjecthood’:
(26) a. Her lover is a plumber
b. It’s a plumber who is her lover
This also illustrates incidentally that we cannot equate the predicative versus
equative distinction with deWnite versus indeWnite: (26a) has both a predica-
tive and an equative reading.
Such arguments as Dowty oVers are based on a fundamental misunder-
standing: linguistic representations do not represent the ‘real world’; they
don’t even represent ‘our perceptions of the real world’, but only one per-
spective on our perceptions. ‘Real world’ situations do not determine linguis-
tic representation. Dowty’s programme for denying case relations an
independent role, and speciWcally a combined semantic and a syntactic role,
has nevertheless been pursued in various ways, as evidenced already by Acker-
man and Moore’s (2001) approach and by developments in ‘role and reference
grammar’ (Foley and van Valin 1984: esp. ch. 2).
At a somewhat later time van Valin (1993a: 43) states his position thus:
In Fillmore’s original proposal (1968[a]), the ‘case frame’ of a verb, e.g[A (I) O], was
intended to be a partial representation of the meaning of the verb, and it also fed into
the operation of grammatical rules, e.g. the subjectivization, objectivization and
raising rules. In R<ole and> R<eference> G<rammar>, thematic relations have
only the second function; the L<logical> S<tructure> of the verb is its semantic
representation, and the role labels like ‘eVector’ and ‘theme’ are mnemonics for the
argument positions in LS.
But the motivation for this weakening of Fillmore’s proposal is unclear.
Indeed, the counting of ‘case’ valency as part of the meaning of an item

Before considering some attempts to eliminate case relations from a central
role in expressing and linking semantics and syntax, we have looked in this
chapter at the kind of criteria that can be invoked in support of the positing of
individual ‘cases’, and at various principles of contrastivity and complemen-
tarity. However, there has not been any general agreement on the implemen-
tation of either apparatus discussed here (in §5.1 and §5.2 respectively), nor
have they been consistently and persistently applied. I shall suggest that this is
not surprising, in principle.
The Identity of Semantic Relations 93
5.4.1 The insuYciency of ‘criteria’
The combination of principles of (1b) is distributionally based, though they
also rely on semantic substance, speciWcally semantic similarit y (the necessity
for which we shall return to). If they are appropriate (or could be made
appropriate), the combination should, when applied consistently, lead to the
establishment of a set of semantic relations language by language, so that these
also correlate with syntactic criteria associated with particular ‘cases’ in a
particular language. But in itself this provides no account of why the ‘cases’
constitute (if they do) a universal set, nor why the set is the size it is, why it
comprises the semantic relations it does. And it therefore still leaves some
scope for the ex tempore proliferation of ‘cases’ in relation to particular
languages. It doesn’t tell us what in principle it takes to be a ‘case’.
Moreover, as revealed by the short discussion of ‘unaccusativity’ in §4.2.1,
‘criteria’ are in practice diYcult to apply, and may even be contradictory.
Particular morphosyntactic properties tend to reXect several (say, categorial)
distinctions at once; they are not ‘pure’. This is also exempliWed by con-
straints on coordination (§5.2), or by the construction illustrated by (2)in
§5.1: the Wnal . . . at it has been invoked as evidence for the analysis of
progressives as locative (Anderson 1973a), but its occurrence is limited to
agentive expressions. ‘Criteria’ are explicanda rather than deWnitive of
some aspect of structure; and they are insuYcient as ‘criteria’ unless we can

´
es sur les signiWcations de chacune des
formes entrant dans la cate
´
gorie. Il faut pouvoir indiquer la signiWcation de la
cate
´
gorie prise dans son ensemble.
The mainstream of modern linguistics inherited no uniWed account of case.
As we have seen, the dominant view was that there were two kinds of case, the
grammatical and the local or notional, as displayed in (2.4), which presents
Holzweissig’s interpretation of the early Indo-European languages:
(1.4)a.grammatical cases: accusative, dative, genitive
b. local cases: ablative, locative, instrumental
And observe again that nominative and vocative stand outside both of these
divisions.
Hjelmslev himself reintroduced the localist tradition (1935/7), which had
been sidelined by the end of the nineteenth century, after a contentious
history of some centuries. The localist theory of the content of case is
articulated in terms of spatial dimensions: all the cases are ‘local’. Anderson
(1971b) argued too that this oVered the most promising theory of case and
‘case’, though his articulation of localism diVers from Hjelmslev’s. Some of the
diVerences are contingent (as implied by Starosta 1981; 1988: 194), depending
on the syntactic-derivationalist orientation of Anderson (1971b) and (1977).
Others are more fundamental, as we shall see.
However, I think that Anderson (1992: 71) provides a reasonably uncon-
tentious summary of the core of the ‘localist’ enterprise as conceived in both
recent and ancient times:
The strong version of this view limits the set of C(ase) R(elations) to those which are
deWned by the semantic components required to express concrete location and direc-

vations in the works of Apollonius and Priscian, and receiving more prominence
among the Byzantine grammarians, notably Heliodorus, but that it received its Wrst
recorded explicit presentation at the hands of Planudes.
It is in our own interest, too, as in general, to have an awareness and
acknowledgement of this history; but I want to retain my focus here on
‘modern grammars of case’.
Hjelmslev provides a fairly detailed overall survey of localist and non-
localist theories of case (1935: pt. I). There he identiWes the ‘proble
`
me’ of the
nature of the category of case, with particular attention being given to the
proposals of the nineteenth century. And it is his work that is most relevant to
more recent developments.
5.4.2 Hjelmslev and localism
Hjelmslev gave the localist theory its most radical interpretation: not only the
‘local cases’ of the standard theory of case at the time but also the so-called
‘grammatical cases’, like dative, accusative, and genitive, and even nominative,
had a ‘local’ content. They were structured by a dimension of directionality,
with respect to which they could be positively or negatively oriented or
neutral between these two poles.
We can, rather crudely (and indeed, ultimately, possibly misleading ly),
illustrate something of the Hjelmslevian system for traditionally ‘local’ cases
with the set from Finnish in Table 5.1, which can be interpreted as showing
respectively negative orientation, neutral and positive.
This is the basic semantic dimension for case systems, one of ‘direction’.
96 Modern Grammars of Case
This presentation oversimpliWes Hjelmslev’s proposals considerably. He
also allows for a distinction between an ‘intensive case’ which is semantically
marked in the particular language, and an ‘extensive’, which is diVuse in
meaning. He says (1935: 114) of an ‘intensive’ case (the genitive in English):

´
s par le rapport casuel sont lie
´
s ensemble’ (1935: 36).
Presence of the second dimension allows for the potential presence of a
third, which involves what Hjelmslev labels ‘subjectivite
´
’ versus ‘objectivite
´
’.
Table 5.1 Hjelmslevian directionality illustrated from Finnish
4 0 þ
talolta talolla talolle
ablative adessive allative
Table 5.2 Hjelmslev’s dimension 2 in Finnish
ad-/in- ab-/ex-
0 þ 4
incohe
´
rent talolla talolle talolta ad/ab-
cohe
´
rent talossa talon talosta in-/ex
-essive -lative
The Identity of Semantic Relations 97
He illustrates this with the French prepositional expressions in (27) and (28)
respectively:
(27) subjective: devant/derrie
`
re

that have, whatever else, no obvious concrete interpretation are characterized
and distinguished with respect to the dimension of direction.
It is, however, important to observe concerning Hjelmslev’s proposals that
he rather goes out on a limb in including the nominative as directly reXecting
directionality. This is not a position adopted by his localist predecessors,
Byzantine or later (notably Hartung (1831) and Wu
¨
llner (1827; 1831)). Hjelm-
slev comments (1935: 43):
Dans le syste
`
me localiste e
´
tabli par Wu
¨
llner, il y a une chose qui surprend: c’est
l’absence du nominatif. Wu
¨
llner et Hartung ont ici adopte
´
la me
ˆ
me manie
`
re de voir
que les localistes de l’e
´
cole grecque: il ont mis le nominatif a
`
part. Wu

recent times by Fillmore (1968a: §1.1) and elsewhere) the subject seems to
display a variety of semantic ‘orientations’ with respect to its predicator. This
is illustrated by the set of examples in (31), of a familiar character, all w ith the
same name of a human as subject:
(31) a. Bill read the book
b. Bill fell to the ground
c. Bill Xew to China
d. Bill lay on the Xoor
e. Bill lived in China
f. Bill slipped
g. Bill was clever/a peasant
h. Bill knew the answer
i. Bill acquired a new shirt/outlook
j. Bill suVered from asthma/delusions
In (31a) we have an Agentive, in (b) an Objective, whereas (c), as usually
interpreted, seems to combine the two—as I’ll come back to. (31d) is an
Objective again, presumably, but here introducing the argument that refers
to the located entity rather than the moved entity, as in (31b, c). (31e) seems to
combine located entity, Objective, and Agentive, as typically interpreted. In
(31f), Bill is presumably again an Objective, but without attribution of goal or
location; and in (31g) a quality or class is attributed. In (31h) the subject is
apparently neither Agentive or Objective, on both semantic and syntactic
grounds that are familiar; nor is that in (31i) or (31j).
Hjelmslev, however, takes a more ‘abstract’ view of directionality in relation
to the nominative and other traditionally ‘grammatical cases’. Consider his
The Identity of Semantic Relations 99
remarks (1935: 53) on the nominatives in the Russian clause in (32), presented
in his transcription and with his segmentation:
(32)ro
´

may well be appropriate, in some sense, to that domain; but simply collapsing
‘rection’ with these other manifestations of directionality and taking it to
deWne the nominative obscures the neutralization of semantic relations that
we Wnd in (31). It is unsurprising that most localists have not attempted a
localist analysis of subjects as such (as opposed to, in some instances, the
semantic relations that it neutralizes)—but we pursue the characterization of
the nominative in such terms in §6.3.
5.4.3 A localist interpretation of ‘datives/experiencers’
Even if, say, following Fillmore (1968a) and Anderson (1971b; 1977), for
example, we recognize that subjecthood involves something diVerent from
the semantic relations themselves, a neutralization, it is still not clear how we
are to apply the localist hypothesis to the full range of subjects in (31).
Objective, which I’ve associated with the subjects in (31b–g), seems to be
unproblematic, in a negative kind of way: it introduces an argument which
does not denote a location or a goal or a source, which is at most located or
undergoes movement. We can characterize it as, though locatable, lacking the
locational property itself, as it seems to lack everything else: its relation to the
predicator is a kind of default determined by that predicator.
From the point of view of localism, even less problematical in a di
Verent
way, obviously, are the locations and goals which occur as complements of the
100 Modern Grammars of Case
verbs in (31b–e). And the subjects in (31i), at least, (31j) perhaps, might also be
argued to involve a goal, possibly ‘abstract’—though here something else
seems to be involved over and above location of the goal. And I’ll return to
this. Let’s look Wrst, however, at what might look to be the most intractable,
the subjects in (31 a, c, e) and particularly (31h). In what follows I give an
interpretation of the much fuller discussions in Anderson (1971b; 1977).
I associated, fairly uncontroversially, the subject of (31a) with Agentive. In
(31b, c) it is combined with Objective: the action is exerted on the agent itself.

(35) a. Bill read the book erg þ abs
b. Bill fell to the ground abs þ loc(goal)
c. Bill Xew to China abs, erg þ loc(goal)
d. Bill lay on the Xoor abs þ loc
e. Bill lived in China abs, erg þ loc
f. Bill slipped abs
g. Bill was clever/a peasant abs
h. Bill knew the answer ? þ (?)abs
i. Bill acquired a new shirt/outlook ? þ (?)abs
j. Bill suVered from asthma/delusions ? þ (?)abl
The last three obviously remain problematic, despite a possible goal inter-
pretation of the subjects of (35i, j).
These subjects all seem to Wt Fillmore’s (1968a) deWnition of the Dative,
whose formulation was given in (3.10):
(3.10) Dative (D), the case of the animate being aVected by the state
or action identiWed by the verb
This is not obviously localist. Fillmore later (1969; 1971) dispersed what he had
regarded as instances of Dative into Objective and Goal and a new ‘case’
‘Experiencer’:
(36) Experiencer (E), the entity which receives or accepts or experiences
or undergoes the eVect of an action (earlier called by me ‘Dative’)
This reformulation removes, for instance, something of the vagueness and
over-reliance on animacy of (3.10
), but the replacement ‘case’ doesn’t appear
any more amenable to a localist interpretation, despite residual vagueness.
The modiWcation as a whole does at least recognize the locative basis of some
of the former Datives (those that are reanalysed as Goal and Objective). What
it fails to recognize is that the residue of Datives that are reinterpreted as
Experiencers are also locative (Anderson 1971b: chs. 7, 9). Let us look now at
some of the evidence for the locative character of Experiencer.

(39) Bill bought Bella the book
There is, however, some brief discussion of Benefactives in Chapter 13.
Consider here again sentence (38b). Here we seem to have, from right to
left, a Source Locative, an Objective, and an Experiencer, all part of the
valency of the verb. This again illustrates that the situation is a little more
complex than Cook suggests, then. Locative and Experiencer can co-occur if
one is a Source, the other a Goal. This is what characterizes Sources and Goals
in general, as in (40a):
(40) a. Bill Xew from Singapore to China
b. Bill X
ew from Singapore
c. Bill Xew to China
The Identity of Semantic Relations 103
With directional verbs, Source and Goal imply each other, even if one of them
is not overtly expressed, as in (40b, c). (I shall suggest later that such ‘missing’
Goals and Sources are ‘incorporated’ into the verb.) It thus appears to be
more accurate to say that Experiencer shares the joint distribution of Locative
and Goal, in particular.
But we cannot simply identify Experiencer w ith Locative and Goal. The
Experiencers of (12) are diVerentiated from other Locatives and Goals both
syntactically and lexically—and sometimes inXectionally, in the shape of a
distinct ‘dative’ inXection. Other Locatives and Goals are not usually pre-
ferred in subject selection over Objectives/Absolutives, as shown in (35b–e)
and (35j), unless they are holistic, i.e. are also Objective.
The acquisition of apparent Locative-subject verbs with the sense of ‘con-
tain’ and ‘include’ seems to be a late development or a loan even in those
languages which have them, and to be parasitic upon an earlier agentive
meaning. In English, for example, the verbs contain and include are both
late-ish loans. And active sentences with such Locative subjects, unlike actives
with Experiencer subjects, do not have a canonical passive. Compare (41)

Bill secretly slipped abs
g.
*
Bill was secretly clever/a peasant abs
h. Bill secretly knew the answer E þ abs
i. Bill secretly acquired a new shirt/outlook E þ abs
j. Bill secretly suVered from asthma/delusions E þ abl
In order for (44b), (44d), (44f) and (44g) to be viable, the subjects must
be given an Agentive interpretation. I have temporarily Wlled in the
missing subject relations in (44h–j) as E(xperiencer). The capacity to be
modiWed by secretly is shared by sentences with Agentive/ergative and
sentences with Experiencer, whether simple Locative Experiencer, as in
(18h), or a Goal Experiencer (44i, j). It is clearly not enough for the subject
to be animate or even human. Even verbs that necessarily (unless used
Wguratively) take an animate absolutive (or at least one that is a life form)
don’t accept secretly, unless they have an agentive or experiential interpret-
ation:
(45)
*
Bill secretly died
Die is a change-of-state verb not, unlike the suVer of (44j), an Experiencer
verb.
This suggests that, as well as being Locatives, Experiencers share some
property with Agentives. Anderson (1977:§2.6.3) proposes, indeed, that
Fillmore’s Experiencer is a complex role, involving two semantic relations,
locative combined with ergative. Such a distribution for ergative is one reason
for the change of label from Agentive embodied in Table 5.3: Ergative is not
always agentive. But how then is it to be characterized?
We can think of the Agentive as the source of the existence of the action
denoted by the verb: without an Agentive there is no action. Similarly, the

The latter two involve a Goal locative, even though in (48b) the ablative is in
this instance not overtly expressed. The perhaps least promising candidate for
a localist interpretation reveals in its syntax and semantics that such an
interpretation is after all appropriate. Anderson (1977) thus argues that the
set of putative ‘cases’ reduces to a localist group such as that enumerated in
Table 5.3.
5.5 Conclusion and prospect
The Wrst two sections of this chapter looked at attempts to develop and
implement ‘criteria’, semantic and syntactic, for the ‘cases’. These sections
also and (particularly) §5.4.1 look at the insuYciency of these, in the absence
of a theory of the category and its content. Despite this, the basic idea of the
fundamental role of ‘case’ in the grammar is diYcult to dislodge (§5.3). The
immediately preceding section (after §5.4.1) has been largely concerned with
looking at those twentieth-century developments in ‘case grammar’ which, on
106 Modern Grammars of Case
the basis of the assumed semanticity of case, involved the evaluation and
integration of the most comprehensive attempt to describe the content of
‘case’ and its implementation in linguistic structure, the localist theory,
particularly as investigated by Hjelmslev (1935/7) and Anderson (1971b; 1977).
The initial ‘case grammar’ work on this was carried out in the 1970s, and it
is largely some of the fruits of this that are presented in §5.4.3, on the
interpretation of Experiencers. Much else of relevance has scarcely been
touched on here, such as the (re-)extension of the localist theory (and
‘case’) to less obvious areas. And such concerns continue to occupy work
based on the early investigations in the ‘case grammar’ tradition, as well as on
the later developments we shall now look at.
Starosta (1988:§5.2) also espouses ‘localism’ (and see too for such a ‘lexi-
case’ approach to ‘localism’ Starosta (1985 a; 1985b)). Indeed, there it is claimed
that ‘earlier linguistic analyses by Jakobson (1936) and Hjelmslev (1935/
37) . . . are in fact somewhat closer in form and spirit to localism as practised

grammar’ programme but did not even arise as issues in the late 1960s. But
with hindsight there can be seen to emerge from these early discussions
various indirect ‘consequences’: aspects of syntax that might prove problem-
atical for the programme as originally conceived, or emerge from parts of the
programme that were not seen as essential to its initial pursuit, or, more
positively still, aspects of syntax whose analysis might gain insight from rather
obvious extensions of the programme.
I now expand as three members of a set of ‘consequences’ things grouped
rather arbitrarily together as (iv) in (3.11). In early work much of what might
be thought to be implicated in the vague label of (3.11iv) was not overtly
discussed; as we have seen, even the constituency/dependency debate was not
generally resolved. I list these ‘consequences’ along with the (only partially
answered) question of content, which is part of the issue raised in ( 3 .11iii), as
an iconic link between what has preceded and what is to follow. For there have
also been further consequences of the localist enterprise, consequences be-
yond localist interpretation of various domains. These ‘consequences’ as a
whole are what will be pursued in the chapters that follow.
They take the form of a set of questions concerning those consequences of
the ‘case grammar’ programme whose pursuit, it seems to me, is at least
encouraged or even demanded by the basic concepts that we’ve looked at in
previous chapters, as follows:
(49) Consequences of case grammar
a) the question of content
b) the question of category
g) the question of consistency
d) the question of derivationality
These are interrelated; I shall look at them in an order that exploits this, i.e.
the order in which I have just given them. Let me brieXy gloss each question,
and indicate some of the developments it provoked, before (as anticipated)
pursuing the Wrst of them in more explicit detail than in this chapter. Concern

recent work, to articulate more precisely an implementation of the localist
theory of ‘case’ whose earlier manifestations have been the concern of the
present chapter. Chapter 7 then considers something of the typology
of linguistic systems involving di Verent neutralizations of the ‘cases’ suggested
in Chapters 5 and 6. After that, discussions focus, in Chapters 8–10,on
questions to do with the categoriality of ‘case’. To begin with, in Chapters
8 and 9, the discussion focuses speciWcally on question (b). Chapters 6–9 are
grouped together as Part II of this work, ‘The Implementation of the Categor y
of Case’.
As I’ve described, a number of researchers adopted the idea that semantic
relations are represented by labelled nodes in a dependency tree. But that
leaves unspeciWed their categorial status: if Agentive, Objective etc. are ‘cases’
or semantic relations, what kind of category is ‘case’ itself? How is it related to
other categories, and how are the representations of individual ‘cases’ related?
To put it another way: say, as a result of the work I describe here in Chapter 5,
we have found a basis for a delimitation of the content of ‘case’ and the extent
of the membership of the set of ‘cases’. We have established, then, if we have
The Identity of Semantic Relations 109
been successful, the set of possible distinctions that can be carried by the
category of ‘case’. We have, in other words, described the secondary categories
of the primary category of ‘case’. These secondary categories are apparently
related to ‘case’ roughly as, say, ‘gender/class’ is to nouns. But what kind of
category is ‘case’? How is it like or unlike ‘noun’ or ‘verb’, say? And how is this
to be represented?
The early ‘case grammar’ programme is not very clear about this. And the
importance of these questions was obscured notationally by the adoption of a
simplex label like Objective or Agentive for individual ‘cases’: the ‘cases’ lack a
categorial feature in common, disguising their status as secondary features
rather than primary, or categorial, features. This is the import of an attempt
to address question (b) that we shall explore in Chapters 8 and 9.

described underlies its role both in deWning the valency of the predicator, thus
specifying its complements and in itself taking (typically) a nominal as a
complement. The morphological and distributional properties are not se-
mantically arbitrary; the semantic relationality of ‘case’ is reXected in mor-
phosyntactic relationality.
Another concept of ‘case grammar’, one which I noted earlier only in
passing, raises the issue of derivationality, or ‘mutation’, question (d). Initial
structures in ‘case grammars’ are unordered, the trees are ‘wild trees’; they are
linearized in the course of derivation. The strongest assumption here would
be the adoption of the assumption that linear order is invariant (Sanders 1970;
1972): it is immutable once assigned (recall Koutsoudas and Sanders (1974:
20), as quoted in §3.1.2). As also noted there, something approaching this
assumption can be seen as underlying Anderson’s (1977) suggestion that
linearization is ‘post-cyclic’, occurring after application of the ‘cyclic trans-
formations’. Thus, in this account, the subject-formation rule corresponding
to Fillmore’s derivation of (7) and (8) is cyclic, so it neither changes nor
assigns linear position; unlike Fillmore’s rule, it simply reattaches the selected
‘case phrase’ rather than also positioning it.
The question that arises here is this: can the assumption of linear invariance
be extended to attachment? Do syntactic structures also show invariance of
attachment? A positive answer to this depends on exploitation of some other
more local ‘concepts’ generally adopted in ‘case grammar’, particularly the
special status of Objective, or Absolutive. BrieXy, Objective is assumed by
many to be obligatory in any predication, even if (though this is not the
general view) not part of the ‘case frame’ of the predicator. Some recent work
has argued that this unsubcategorized-for Objective is the target for multiple
attachments which allow ‘argument-sharing’ between diVerent predicators
(see for example Anderson 1991;Bo
¨
hm 1993). These ‘multiple attachments’