Content uploaded by Francis Cornish
Author content
All content in this area was uploaded by Francis Cornish on Apr 07, 2017
Content may be subject to copyright.
Published in J.L. Mackenzie & M.A. Gómez-González (eds.), Studies in Functional Discourse Grammar, Berne:
Peter Lang, 2005, pp. 21-47.
FRANCIS CORNISH
Null Complements, Event Structure, Predication and Anaphora: A Functional
Discourse Grammar Account
1. Introduction
The theme of this chapter is the possible existence, and if so, interpretation, of zero or null complements
of predicates which may take one or two internal arguments (i.e. either an A2 or an A2 and an A3),
realizable syntactically.1 The chapter aims to show how this phenomenon may receive a satisfactory
treatment within Functional Discourse Grammar (FDG). I am concerned here only with null complements
having nominal values, leaving aside predicational zeros (as in VP ellipses such as […] and Peter was
too). There are three essential issues concerning the possibility of occurrence and the type of interpretation
of null complements: first, what are the conditions under which they may occur with various types of
transitive verbs?; second, what are the semantic and referential values which these null complements may
assume in different contexts?; and third, what are the principles which make these values possible?
Clearly, the occurrence of null complements needs to be licensed – it is not just any transitive verb, in any
type of context, which may allow its direct and/or indirect complement(s) to be unrealized syntactically. It
is these questions which I will be addressing in what follows. A satisfactory account of the possibility of
non-realization of one or both of a predicate’s internal arguments syntactically, and when this is possible,
of the way in which they receive an interpretation, requires recognizing the existence of an interaction
amongst lexical-semantic structure, the construction selected as a whole, and various discourse-contextual
factors.
Zero forms and deletion are not recognized in the FG framework; and indeed, evidence will be
provided that the former must be the result of the non-instantiation of one or more argument positions in a
given predicate frame (or predication frame, in García Velasco / Hengeveld’s 2002 account: see Section 5
below), with its consequences for the event-type denoted. They are not derived via an underlying
representation of the term filling the argument position, a term which is subsequently deleted. As in the
case of indexical expressions generally (cf. Cornish 1999, 2002a), zero forms have properties which are
not predictable from those of the term or predicate which they might be said to replace; and in any case,
the predication created from a predicate all of whose argument positions are filled with lexical terms is not
necessarily identical in value to one in which one or more of these positions is unfilled, as we shall see in
particular in the final part of this chapter.
2. The semantic vs. syntactic valency of predicates
A characterization of the phenomenon of null complements requires distinguishing a predicate’s semantic
valency from the syntactic valency of the verb, adjective or preposition corresponding to that predicate;
hence, a formal treatment of the possibility of occurrence and type of interpretation in context of null
complements presupposes the distinction made within FDG between the Representational and the
Structural levels of analysis, and between abstract meaning definitions of lexemes and their linking to
predication frames (cf. García Velasco / Hengeveld 2002). Furthermore, where a discourse referent
evoked or retrieved via a null complement is involved, this would seem to require representing at the
Interpersonal level, with contributions from the Cognitive and the Communicative Context components
(see Section 5 below).
Briefly, a predicate’s arguments may form part of its semantic valency qua predicate at the level of
lexical-semantic structure, as well as of the syntactic valency of the lexeme which corresponds to it at the
morphosyntactic level.2 That is, a predicate’s array of argument positions in terms of lexical-semantics (cf.
Mairal Usón / Faber’s 2002 notion of “lexical templates”) may well correspond to the syntactic arguments
it takes when realized by a given lexeme. This is illustrated by the examples in (1) below.
(1) a. John saw the “No Entry” sign.
b. The postman placed the packet in the tray.
c . The car hit the railing.
Clearly, the second or second and third arguments of the predicates see, place and hit are required both
semantically and syntactically, as the examples in (2) show:
(2) a. *John saw.
b. ?The postman placed the packet.
*The postman placed in the tray.
*The postman placed.
c. *The car hit.
However, the 3-place predicate place is somewhat different from the 2-place predicates see and hit, in that
it may well occur in context without one of its “internal” arguments being instantiated: here its A3. This
then becomes “tacit”, recoverable from the context of utterance of the clause involved. In this particular
case then, we see that the semantic and syntactic valencies of a predicate may diverge – its syntactic
valency being reduced to 2, while its semantic valency remains at 3. But even verbs like hit may occur in
context with only their A1 instantiated, as in this attested example from the genre of journalism:
(2c´) [Context: article about a 1.3kg meteorite which crashed into a New Zealand couple’s living room
shortly before breakfast]
…Even at that speed [several hundred metres per second] it would have been moving fast enough to kill on the spot and
the Archers’ one-year-old grandson had been playing in the room moments before it hit. (The Guardian, 14.06.04, p. 2)
Note here that the inanimate pronoun it would not be appropriate as a substitute for the null complement
of hit in this example (it would tend to be interpreted as referring back to “the room in which the Archers’
one-year-old grandson had been playing”). With the (basic 2-place) predicates in (3), on the other hand,
their syntactic valency may be reduced by 1, this having particular effects on the semantics of these
predicates. This is the so-called “absolute” use of transitive verbs (see also the use of the 2-place verbal
predicate kill in (2c´)).
(3) a. Ron sawed, and Mildred pruned.
b. Hilda read, while Jim wrote.
Here, the emphasis is on the activities of sawing, pruning, reading and writing, respectively, on the part of
the individuals involved, and clearly not on the thing(s) sawed, pruned, read or written. These are all
“incremental-object” verbs (cf. Van Hout 1999). But it would be a mistake to believe that these
predicates’ A2 in their transitive use has disappeared at the lexical-semantic level when they are not
realized syntactically, since it is quite possible for the interlocutor to question this entity: …I wonder what
Ron sawed/Mildred pruned? for (3a), and …I wonder what Hilda read/Jim wrote? in (3b) (cf. Fillmore’s
1986 test).
Clearly in these cases, the non-instantiation of the A2 has had the effect of turning an
accomplishment predication (where all the predicates’ arguments are lexically instantiated) into an activity
one (where their non-instantiated A2 arguments are construed as indeterminate or generic). Such a
relationship would be treated by Dik (1997b: ch. 1) in terms of a Predicate Formation Rule; but see García
Velasco / Hengeveld (2002) for arguments against this treatment once their construct of “predication
frame” (see Section 5 below) is allowed to replace the standard predicate frame. It is the highly specific
selection restriction imposed on the instantiation of this argument position which is responsible for
transferring this semantic property to the non-instantiated argument positions,3 resulting in the
understanding that, for example, “Ron sawed logs” and “Mildred pruned roses” in an utterance of (3a),
and that “Hilda read books/magazines/newspapers” and “Jim wrote letters/his diary/articles” in one of (3b)
(the context of utterance is clearly responsible for delimiting the specific understanding of these null A2s
in such instances – see also (4) below).
3. Three semantic or discourse-referential values realizable by null complements
There are three distinguishable semantic or discourse-referential values realizable by null complements
(“generic” or “indeterminate”, “referential-(in)definite”, and “anaphoric (contextually-definite)”). I
present and illustrate each in turn, and will then focus particularly on the third subtype.
3.1. “Generic” or “indeterminate” null complements
This value has already been illustrated in this chapter, in the shape of examples (3a, b) as well as the use
of the normally transitive, accomplishment verb kill in (2c´), where the non-instantiation of the A2 of the
transitive variants of the verbs concerned had the effect of converting the predication to an activity one,
the internal arguments having an indeterminate reference – but constrained by the context of utterance of
the clause in question, as we have seen. Example (4) provides an attested illustration, where four normally
transitive verbs have precisely this value:
(4) “See, try, admire or buy at London’s Motor Show.” (Advertisement, The Sunday Times, 9.10.83, p. 9)
(5) [Notice on individual dustbins on pavements in a street in Canterbury, UK:] “Recycling is so easy when it’s collected from
your doorstep.”
The types of things which the reader of the advertisement in (4) is enjoined to “see, try, admire or buy”
(note the imperative mood of these predications, which favours non-realization) are clearly new models of
motor vehicles and their accessories, exhibited in the Motor Show to which the utterance is referring. We
thus have a set of (fairly general, in all these instances) selection restrictions transferred by the senses of
each of the predicates involved to their non-instantiated A2s, senses which are delimited to a denotation
type via the context in which the text occurs – an advertisement for a Motor Show. As Ricardo Mairal
Usón (p.c.) points out, such arguments would not receive a macro-role in Van Valin / LaPolla’s (1997)
Role and Reference Grammar model, no discourse referent being introduced in order to be the object of
subsequent predications. And as Mejri / François (to appear) point out, it is necessary to distinguish
between indeterminate (as I shall call this sub-type) and generic values of null complements. The type in
(5) could be argued to be generic, since the (initial) clause as a whole is generic: the tense is the (gnomic)
present, and the predication attributes a property to an event type as opposed to token. Whereas in (4), the
four predications are eventive, the conjuncts each being in the imperative mood, and the actions enjoined
being located within a specific commercial event. The null complements thus all have an indeterminate,
rather than generic value here. In both cases, it is the event (token or type) denoted by the verbal predicate
which is highlighted by the null complement realization, its participants being backgrounded thereby. See
also the use of kill in (2c´), whose null complement would appear to have an indeterminate (human-
denoting) value here (“people, whoever they might be”).
3.2. “Referential (in)definite” null complements
Another value assumable by null complements is that of evoking an identifiable ((in)definite) entity which
may later be retrieved, under certain conditions, via an anaphor. What is crucial here is that the zero
form’s intended referent be identifiable by the addressee, or at least that it be treated as such by the
speaker. Whether or not it is salient at the point of occurrence in the co-text is immaterial. This
corresponds to both of Mejri / François’ (to appear) subtypes “latent-identifiable” and “latent-identifiable
and salient”. Unlike the generic or indeterminate type we looked at in Section 3.1, this subtype may evoke
a discourse referent. Deictic occurrences in the context of the imperative form of the host verbs are an
initial type of example, as seen in (6):
(6) Eat!/Watch!/Mind!/Smell!/Taste!
In each such case, the interlocutor’s attention is being specifically drawn to the thing or event involved,
which is available within the situational context. As such, it is thereby made salient for both participants.
(7) I wrote to you a week ago, you know, but you never answered!
In (7), the context indicates that the verb write is being used in its ‘correspond’ sense; the predicate at
issue therefore has three arguments, the second of which is unrealized syntactically. This non-instantiated
A2 argument (“a letter sent by the speaker to his/her interlocutor a week before the time of utterance”) is
clearly referential, owing to the definite past tense chosen here, and the reference to a specific event which
occurred prior to the utterance of (7). It is equally clearly indefinite, being an introductory reference (even
though the intended addressee may already be aware of the existence of such a letter), the referent
constituting discourse-new information in context. Unlike the “generic/indeterminate” value of null
complements seen in Section 3.1, the “referential-(in)definite” use does introduce a discourse referent (and
would clearly warrant the assignment of a “macro-role” in terms of RRG).
3.3. “Contextually-definite, anaphoric” null complements
A third possibility is where the implicit internal argument is not only referential and identifiable, as in the
case of the null A2 complement of wrote in (7) above, but anaphoric. The second non-instantiated A2
argument in (7), the referent of the null complement of answered in the second conjunct, is contextually
definite, in contrast, as it is construed as referring back anaphorically to the letter introduced by the null
complement in the initial conjunct. Fillmore’s (1986) test for indefinite (non-referential, of the type seen
in Section 3.1 above) null complements would be clearly negative here: …#I wonder what you never
answered. See also the null complement of hit in (2c´).
Given that I include under the heading “anaphora” exophoric uses of potentially anaphoric
expressions,4 I would subsume under this value such uses of zero complements – as in the case of labels
on bottles of pharmaceutical products, instructions for use etc. of the type: Take with precaution (label on
bottle of medicinal pills); Break in an emergency (notice displayed on a glass panel behind which is
placed an alarm). The host verbs in such examples are in the imperative mood, like the deictic examples in
(6). However, it is clear that the intended referent of the zeros in such “label” cases is not only
identifiable, but salient: the addressee’s attention is assumed, in such caption-like instances, to be already
centred on the object on or under which the notice is placed. Thus the implicit argument is contextually-
definite, and the reference is anaphoric (cf. the infelicitous queries: #…I wonder what should be taken with
precaution/#…I wonder what should be broken in an emergency, respectively).
4. The anaphoric potential of null complements, event structure and predication
Let us now concentrate essentially on the first and the last of the three sub-types set out in Section 3, in an
attempt to make precise the latter’s anaphoric potential, as compared with that of unaccented third person
personal pronouns. It will be shown that this is a function of an interaction amongst the event-type
designated by the clause as a whole, the host predicate’s selection restrictions, the choice of zero vs.
pronoun as complement where either is possible, and wider contextual factors.
There would seem to be two main conditions which must hold in English for a null complement to
occur under an anaphoric interpretation: 1) there must exist a specific selection restriction upon the
internal argument(s) subject to non-realization in terms of syntax; and 2) the null complement’s referent
must be contextually salient at the point where it occurs. This condition is a necessary, though not
sufficient one, as (8) shows:
(8) Martin liked the look of the pair of walking shoes displayed in the store window: he went and bought *ø/them without trying
*ø/them on.
Here, the verb buy is used in the definite past tense, and the reference is clearly to a specific occasion of
buying something – a pair of walking shoes – and of not trying that something on. But even though this
referent is contextually salient, this is not sufficient to permit the non-instantiation of the internal argument
of the two verbs concerned under an anaphoric (coreferential) reading. It would seem to be the non-
specific nature of the selection restriction associated with the lexical-semantic structure of these verbs
which prevents this type of functioning. The verbs buy and try on seem to have only very general selection
restrictions (respectively, <commodity> and <clothing>). In spontaneous spoken French, however, the
equivalent verbs may well occur with a null complement, under an anaphoric interpretation: …Il est allé
acheter ø/les acheter sans essayer ø/les essayer. In English, only an overt pronoun may occur as
complement of the verbs in such a context.
Other verbs, having more specific selection restrictions as well as different Aktionsart properties,
permit both types of form. Let’s look first at a pair of examples presented, but not further analysed, by
Groefsema (1995: 156):
(9) a. John picked up the glass of beer and drank ø.
b. John picked up the glass of beer and drank it.
Here, the choice of a zero complement of drank in (9a) induces a partitive interpretation. The zero is
anaphoric, but the null instantation of this predicate’s A2 has had the effect of changing the
accomplishment Aktionsart evident in (9b) with the pronoun into an activity predication. What John is said
to have drunk in (9a) is some, not necessarily all, of the beer in the glass evoked in the initial conjunct. In
(9b) in contrast, John is stated as having drunk all of the beer in the glass (the overt pronoun, enabling the
accomplishment event structure to be specified, induces a holistic interpretation). The anaphoric, and not
“generic”, value of the zero in (9a) is determined by the fact that the two conjuncts of this example
designate a sequence of two specific events (note the definite past tense borne by the verbs in each
conjunct) which each form an integral part of a more global event. This is not the case in (10a) below,
where each predication denotes an atemporal property (via the simple present tense in each conjunct and
the lack of an overt article in the NP gin), the second property holding independently of the first. Thus the
predication in the second conjunct of (10a) does not continue the situation established in the first, and so
the null complement is not anaphoric in value. We thus have the generic value, delimited via the reference
to ‘gin’ in the initial conjunct to ‘alcoholic beverages’, which we saw in Section 3.1 (see also example (5)
above).
(10) a. John drinks only gin, but I won’t drink ø.
b. John drinks only gin, but I won’t drink it.
(Lehrer 1970: 245, examples (67) and (68),)
I would argue that this systematic difference in interpretation arises because of the fact that overt pronouns
are nominals which, because of their inherent definiteness and (potential, at least) referentiality, change
the Aktionsart of the verb whose internal argument they instantiate, from an atelic activity into a telic
accomplishment. Thus with definite complement pronouns, the emphasis is no longer on the activity of
drinking (or eating), but on the nature of the thing drunk (or eaten). But this may also be the case even
with “incremental-object” verbs like eat and drink, whose Patient argument is understood to be
progressively affected by the process involved, even when their internal argument is unexpressed
syntactically. Compare (11a) and (11b) in this respect:
(11) a. Mary ate at noon.
b. The waiter served the main course. Mary ate hungrily.
In both (11a) and (11b), the tense is the definite past, and the reference is to a specific event which
occurred prior to the act of utterance. In (11a), the presence of the contextualizing PP at noon, a ó2
localizing satellite, induces the culture-specific stereotype ‘midday meal’ as the frame in which the event
denoted is to be set, so that the predication is telic, and not atelic via non-instantiation of the A2 (though
the more basic “activity” reading is still co-present here). But in (11b), the presence of the ó1 manner
satellite hungrily highlights the basic activity sense corresponding to the predicate’s inherent value (this
satellite being a predicate satellite). There is no implicature available such that Mary actually finished
eating the main course which she has been served. Thus the zero has the indeterminate, non-referential
value which we saw in the case of the verbs in (3) and (4). If we apply Fillmore’s “I wonder what X V-ed”
test, it is positive in (11b) (‘I wonder what Mary ate’), but negative in (11a) (with the predicted response
#She ate a midday meal; however, with the type of response expected and normal for the same query on
(11b) – e.g. […] chicken and noodles – it is positive). The bounded/non-bounded adverbial tests (see (12)
and (13) below) also discriminate the two occurrences: for (11a), Mary ate ø ?#for ages/in an hour at
noon; and for (11b) [...] Mary ate ø hungrily for ages/#in an hour .
Rappaport Hovav / Levin (1998: 104ff.) and Brisson (1994) also claim that it is the aspectual
structure of the verbal predicates concerned which determines whether or not their internal argument may
be left unrealized: however, it is not the static, inherent lexical-semantics of each individual verb which is
relevant here, but the compositional semantics of the predicative unit as a whole (verb + possible extra
adverbial element) – see also Ritter / Thomas Rosen (1998) and Van Hout (1999), as well as the contrast
between (11a and b) above. In the (a) examples below, the verbs are simple activity predicates, while in
the (b) ones, the presence of a resultative PP or aspectual particle determines an accomplishment event
structure ((12) and (13) are my own examples):
(12) a. John ran (for half an hour/*in half an hour).
b. John ran to the river bank (?for half an hour/in half an hour).
(13) a. Sandy drank (for ten minutes/*in ten minutes).
b. Sandy drank up (?for ten minutes/in ten minutes).
The durative time adverbials for half an hour and for ten minutes are possible modifiers of the activity
predicates run in (12a) and drink in (13a), while the bounded temporal PPs in half an hour and in ten
minutes are clearly unacceptable. In the case of the accomplishment event type denoted due to the
presence of the goal PP to the river bank in (12b) and the telic aspectual particle up in (13b), the bounded
temporal PP is a possible modifier, though the durative one only results in an iterative reading of the
events denoted (that is, there were repeated events of John’s running to the river bank and of Sandy’s
drinking up during the time spans indicated, rather than one single unbounded event). This is typical of
accomplishments, as Brisson (1994: 91) points out in the case of write-type verbs.
Now, as for the non-realization of internal arguments, both Brisson and Rappaport Hovav / Levin
claim that this is only possible when the argument at issue is a “content”, and not a “structural” one. The
distinction involves that between the aspectual or “event structure” of the sentence as a whole, where the
participants involved are “structural participants”, and the lexically-specific content of the predicate which
“heads” that structure. This predicate brings with it a certain array of arguments in terms of its meaning:
these are the “content” arguments (cf. García Velasco / Hengeveld’s 2002 abstract meaning definitions, in
the context of FDG). Thus there may be a discrepancy between the two types of structure. Brisson argues
that (“activity”) verbs of the type illustrated by sweep (plough, pack, dust, vacuum, clean, mow, rake…)
may occur with either a durative or a bounded time adverbial, whether with or without a syntactically-
realized object NP, and whether this NP is definite or indefinite. This would give the predicate sweep (as
well as the others in its class) the “content” structure sweep (x,y) but the event structure ‘activity (x)’.
From Brisson’s account, one can infer that the “y” argument in the content structure is more or less totally
determined by the meaning of this predicate (stereotypically, a “floor” of some kind) – an “inherent”
argument, then; whereas, given that this verb is basically an activity predicate (as indicated by the various
tests applied to it), there would be no second, internal argument at this event-structural level at all. Now,
given that this is the case, the sole internal “content” argument, not being a “structural” one, need not be
realized syntactically – so long as its essential content is contextually recoverable. This is the case with
verbs of the type represented by sweep, where general knowledge tells us that it is typically floors that are
swept (see also plough à fields, pack à suitcases, dust à furniture, vacuum à carpets, clean à
artefacts, mow à lawns, rake à leaves, etc.). This general class of predicates is characterized by
Rappaport Hovav / Levin (1998: 99) as “verbs of surface contact through motion”. The situation described
here would seem to characterize the first type of non-realization of internal arguments seen above in
Section 3.1 (generic or indeterminate argument types, as illustrated in (3a, b), (4), (5), (10a) and (11b)).
On the other hand, where a given predicate has an achievement or accomplishment interpretation,
there is necessarily a binary event structure involved (cf. García Velasco / Hengeveld’s 2002 predication
frames), consisting of a causing event (an activity) and a resulting state. Thus there are inevitably two
structural arguments, both of which must be realized syntactically, according to these authors. This would
explain then why such predicates (for example, English break, as we have seen) cannot leave their internal
argument unrealized. Break is an “externally-caused change of state verb”, according to Rappaport Hovav
/ Levin (1998: 99). The content structure of break would then be break (x,y), and its event structure
‘activity (x) CAUSE [BECOME broken y]’, where broken represents both the essential content of the
predicate break and its status as “resulting state” of the macro-event involved here. (I have inserted the
abstract operators CAUSE and BECOME here, as well as the square brackets, which Brisson does not do;
indeed, she does not represent the structure of break in her article.) It can be argued that the intransitive,
inchoative use of break (as in The vase broke) is more basic, and that the transitive-causative use is
derived from it by rule. I have attempted to formalize these two types of structure under (14a and b)
below, drawing inspiration from the notation system used in Van Valin / LaPolla (1997) (the segment in
parenthesis following ‘¬ intact’ (y)’ is intended to capture the selection restriction imposed on its single
argument by the meaning of this predicate. For the inchoative “activity” or “process” value, the segment
preceding the operator BECOME in (14b) would not be instantiated).
(14) a. “Content structure” of core sense of break, after Brisson (1994) and Rappaport Hovav / Levin (1998):
BECOME ¬ intact’ (y) ((partially_)rigid_object, y)
b. “Event structure” of break, after Brisson (1994) and Rappaport Hovav / Levin (1998):
[[activity (x)] CAUSE [BECOME ¬ intact’ (y) ((partially_)rigid_object, y)]]
Brisson’s (1994: 97) two licensing conditions on the possibility of unexpressed objects are as follows:
(15) a. Grammatical licensing condition: structure arguments must be expressed.
b. Contextual licensing condition: the unexpressed object must be understood.
Rappaport Hovav / Levin (1998) formalize and further develop Brisson’s essential insights. They
propose two well-formedness conditions on the syntactic realization of event structures, as follows:
(16) Subevent Identification Condition
Each subevent in the event structure must be identified by a lexical head (e.g. a V, A or P) in the syntax. (Rappaport Hovav /
Levin 1998: 112)
(17) Argument Realization Condition
a. There must be an argument XP in the syntax for each structure participant in the event structure.
b. Each argument XP in the syntax must be associated with an identified subevent in the event structure. (Rappaport
Hovav / Levin, 1998: 113)
Condition (17a) makes more precise Brisson’s (1994) “Grammatical licensing condition” given under
(15a) above; while Condition (17b) ensures that all argument expressions in the syntactic realization be
relatable to a predicate corresponding to an identifiable subevent in the event structure associated with the
sentence. Condition (16) completes the picture, in that it ensures that each predicate marking a subevent in
the event structure be relatable to a relevant lexical head.
However, it would seem that these conditions are much too rigid and absolute: they do not take
enough account of the surrounding co-text or context of occurrence of the verbal predicates at issue here –
i.e. of the way in which these predicates are actually used. If we take the (causative-accomplishment)
verbal predicate break as a typical verb having a binary event structure, with two content and two
structural arguments, the prediction is that non-realization of the internal argument is excluded – as we
have seen. In the case of the non-referential use of the verb, of course, the authors could claim that what
we have is an activity, such that there is no longer an internal structural argument since the binary event
structure is no longer available. This could occur in the following kind of situation: imagine a warehouse
full of trestle tables on which are piled substandard mass-produced cups, plates and saucers etc., all
containing imperfections of one kind or another. Members of the public are invited to break these items
into pieces, so that they can more easily be recycled. On the entrance to the warehouse is pinned a large
notice, with the words:
(18) Break ø to your heart’s content!
Clearly, this would correspond to the indeterminate, non-referential use of transitive verbs with unrealized
complements that we saw in Section 3.1. As already noted, this use would not constitute a counterexample
to Brisson’s and Rappaport Hovav / Levin’s constraints (since in this usage break and similar verbs would
be activity and no longer achievement predicates; as such, they would be only unary event predicates).
But there are two other possibilities with break. The first is the possible deictic use which we briefly
saw in Section 3.2, where the null complement of break has a referent available via the utterance situation.
As an example, consider again the “reject” crockery situation evoked a moment ago. Imagine a situation
where a member of the public has entered the warehouse and has been merrily smashing plates, cups and
saucers for the last 20 minutes. Suddenly, he comes upon a large bowl with an attractive design, which
doesn’t seem to him to be in too poor a condition. As he holds it up to examine it, the attendant walks
towards him and says:
(19) Come on now, break ø! They’ve all got to go, you know!
Here, the referent is clearly present, both at the semantic and the discourse-representational levels – and
yet the internal “event-structural” argument is syntactically null, contrary to conditions (15a) and (17a). I
believe (19) corresponds to a deictic and not purely anaphoric use of the null complement, since the
addressee is tacitly querying the status of the intended referent (thus it consists in introducing the referent
qua “non-breakworthy” item of crockery in this context). If an overt pronoun were to be used in place of
the zero complement here, I believe it would be the demonstrative pronoun that rather than the (purely
anaphoric) third person pronoun it. Granted, there are “extenuating” circumstances here, independently
motivating the null complement of break, since this type of occurrence is restricted to the type of highly
modalized context represented by the imperative mood in (19).
But there is a second type of counterexample to these stipulations, falling within the third of the
three cases outlined in Section 3.3. This is the “exophoric” use of predicates like break, which I claimed
come under the anaphoric, contextually-definite use (as in the caption Break ø in an emergency displayed
above a glass panel covering an alarm handle). Here too there is an argument at the lexical-semantic level
as well as a discourse-referent at the level of discourse. Again, the possibility of such occurrences is a
counterexample to stipulations (15a) and (17a).
In all three types of example involving break with a null complement, the clause is in the imperative
mood. This is no accident, in fact. What this mood induces (contrary to the declarative mood, in
particular) is a focusing of attention on the object of the command – the speech-act type typically
correlating with the imperative form. This serves, then, to enhance psychologically, i.e. to “profile”, the
entity at issue, and thus to allow it to be unrealized syntactically.
5. Towards an FDG account: abstract meaning definitions, interpersonal-level
representations and predication frames
Let us now attempt to describe and account for the properties and behaviour of null complements in
English, as identified in Sections 2-4 above, in terms of the new Functional Discourse Grammar model. I
will assume familiarity with the internal organization of this model (see Hengeveld 2004a, 2004b, 2005,
and Hengeveld / Mackenzie forthc. for details). In García Velasco / Hengeveld’s (2002) account, abstract
meaning representations are operated on by “linking rules” mediating between the lexicon and an initial
syntactically-relevant structure (“predication frames”), in that the lexemes in question are eventually
inserted into the latter as a function of the parallel configuration of predicates and arguments in their
meaning representations.
For arguments against the standard FG construct “predicate frame” and the existence of Predicate
Formation rules which map predicate frames into other predicate frames for derived senses or forms of
given predicates, see GVH (2002) as well as Cornish (2002b). GVH further present and illustrate their
concept of predication frames: see in particular their examples (29) (2002: 110-112), whereby given
lexemes are matched with syntactically-relevant structures (predication frames) in which they may occur.
The authors present examples of their formulation of abstract meaning definitions under their (32) and
(35) for the causative and inchoative senses of the English verbal lexeme open (2002: 114 and 115,
respectively). These formalizations are inspired by Van Valin / LaPolla’s (1997) representation system,
which is also the model on which Mairal Usón / Faber (2002) base their construct “lexical template”.
Abstract meaning representations are subject to “linking rules” mediating between the lexicon
(representations of individual lexemes) and the semantico-syntax (predication frames), in that the lexemes
in question are eventually inserted into the latter. The authors operate here in terms of a very simple
procedure of matching between argument structures in meaning definitions, and transitive or intransitive
predication frames, as a function of the parallel number and positions of the arguments concerned in each
construct. However, as we will see below, such “syn tactic” verb -argument structures do not always match
one-to-one the relevant parallel predicate-argument configurations within the abstract meaning definitions.
(20a, b) give GVH’s (2002: 114, 115, items (32) and (35), respectively) representations of the meaning
definitions of the causative and inchoative forms of the verb open, and (21a, b) present its insertion into
the two predication frames selected by these two representations, respectively.
(20) a. open [V]
[f1: [CAUSE (x1) [BECOME open’ (x2)]]]
b. open [V]
[f1: [BECOME open’ (x1)]]
(21) a. (T1: (f1: open [V] (f1)) (T1)) (R1: (x1)Ag R1)) (R2: (x2)Pat (R2))
b. (T1: (f1: open [V] (f1)) (T1)) (R1: (á1)Pat (R1))
“T” symbolizes an ascriptive act at the interpersonal level in a Functional Discourse Grammar, “R” a
referential act, and “á” denotes any entity type (x, e, p or E) at the representational level. Representation
(20a) selects the transitive predication frame presented in (21a) since both contain two arguments, while
representation (20b) selects the intransitive one given in (21b) since there is only one argument variable in
each representation. However, as suggested in Cornish (2002b), a simplification of the two representations
in (20a, b) is possible and indeed desirable, since (as GVH 2002: 116 themselves point out), (20b) is a
proper part of (20a). By placing in parentheses the causative structure in (20a) under which the inchoative
sub-structure is embedded, a single representation is achieved, which is all that is required.
I would agree with García Velasco / Portero Muñoz (2002: 19-20) that 2-place predicates whose A2
is syntactically null, and which denote activities rather than accomplishments when their A2 is unrealized,
remain 2-place predicates semantically. As in the analysis put forward here, the authors provide a meaning
definition for incremental-object activity verbs like eat whereby the second argument-variable position for
such predicates is filled only by the selection restriction transferred via the meaning of such verbs (<food>
in the case of eat). But it is not necessarily the case (and this criticism also applies to Brisson’s 1994 and
Rappaport Hovav / Levin’s 1998 similar approach here) that “If the speaker decides to build up a telic
predication both participant variables in the abstract meaning definition will be projected onto the syntax.
This leads the entry to select a transitive predication frame with two argument positions”. For this would
be to ignore null-complement examples such as (2c´) with hit, where the predication is contextually telic,
or (11a), where the context in which eat occurs induces, precisely, a “telic” predication; or possible
“deictic” examples such as the use of the achievement (hence telic) predicate break as in (19) above.
When such verbs occur in this type of use in (syntactically) intransitive clauses, their A2
must still be
present as a fully referential argument. To illustrate, let us attempt first to represent the meaning definition
of buy, whose meaning definition would be as represented in (22).
(22) buy [V]
[f1: [CAUSE (x1) [BECOME NOT have’ (x2) (x3: <commodity>)] & [BECOME have’ (x1) (x3)] & [BECOME
have’ (x2) (x4: <payment>)]]]
(22) reads (somewhat stiltedly) as follows: ‘(x1) causes (x2) to come to not have (x3), a commodity, and
(x1) to come to have (x3), and (x2) to come to have payment’. Now, only (x1) and (x3) are treated as
nuclear arguments of the verb buy by the syntax, the remaining arguments ((x2) and (x4)) being optionally
realized syntactically as ó2 satellites. This is important, since only the non-realization of the (x3) argument
will give rise to generic or indeterminate, activity predications of the kind seen in (3), (4) and (5). It is as if
the essential meaning of buy were something like ‘(x1) come to have (x3)’, and this is reflected in the
transitive syntax of this verb. Let us see what a representation of the fourth conjunct of (4) would look
like at the initial, interpersonal level in an FDG derivation.
(23) INTERPERSONAL LEVEL
[M: (A1 [DECL-P (P1)S (P2)A (C1: [(T1: (f1: ‘buy’)Foc) (R1: ((P2)A)) (x1: <commodity>))])] (R2: ‘London’s Motor Show’)
The locative ó2 satellite at London’s Motor Show in (4) serves to delimit the domain of “commodities”
denoted by the null complement to new models of motor vehicles and accessories, as we have seen. Here,
there is no “R2” (referential act) at the second argument position of ‘buy’, since no discourse referent is
evoked here, as we have seen, the predication being an “activity” one (cf. also García Velasco / Portero
Muñoz 2002). Contrary to what is claimed by GVH (2002), it is not the particular parallel configuration of
argument variables within the meaning definition in (22) which determines whether the predication frame
selected will be a transitive or an intransitive one, but rather the speaker’s intention to denote an activity in
contrast to an accomplishment SoA. In this type of case, it is not an intransitive predication frame which
will be selected, but rather a transitive one – in which the A2 is not lexically filled. The predication frame
in which the lexical ingredients whose meaning is represented in (22) are inserted might look like (24):
(24) REPRESENTATIONAL LEVEL
([(f1:buy[V](f1))(x1:(P2)A)Ag(gx2:<commodity>)Pat]Activity)
Now, although in (24), the syntactically-relevant argument position is (x2), while its counterpart in the
meaning definition of buy in (22) is (x3), this does not in fact pose a problem, since the mapping between
the two positions (semantic and syntactic) can be achieved via the semantic role of the relevant argument.
In (22), (x3) is the second argument of the first ‘BECOME have’’, and so receives the Patient role (cf.
GVH 2002: 114); and the (x2) argument in (24) is explicitly annotated for the same semantic function. But
under García Velasco / Portero Muñoz’s (2002) account, this argument mapping would not be possible,
since activity-denoting meaning definitions may only select intransitive predication frames. Thus, the
second argument position, as specified in (24), would not be available to be involved in the mapping
procedure. As pointed out earlier, it is the (general) selection restriction placed on this argument variable
in the meaning definition (here <commodity>) which is transferred to the (x2) position in (24), marked
with the term operator g for ‘generic’. As we have also seen, there are also restrictions on the 1
(aspectual) operator selected for the predication as a whole, since this must be IMPF (imperfective). That
the Tense (2) operator need not be NON-PA ST f or this val ue to ob tai n is sh own by e xam ple s li ke (3a, b),
where the Tense is Past but the Aspect is Imperfective.
Now let us examine the representation of the two other types of value associated with null
complements which we distinguished in Section 3. Let us take the two verbs write and answer, as used in
example (7).
(25) a. write [V]
[f1: [CAUSE (x1) [BECOME exist’ (x2: <letter>)] & [INTEND (x1) [go-to’ (x2) (x3:: <human>)]]]]
b. answer [V]
[f1: [CAUSE (x1) [BECOME exist’ (x2: <answer>)] & [INTEND (x1) [go-to’ (x2) (x3:: <original_sender >)]]]]
Both the predicates write and answer as represented in (25a) and (25b), respectively, take three arguments,
in their ‘correspond’ sense. In (7), the second argument of write is unrealized, while both the second and
the third arguments of answer are unexpressed. The second argument of write in (7) evokes a discourse
referent, as we have seen. The initial Interpersonal level representation of each conjunct of (7) would be as
in (26a) and (26b), respectively:
(26) a. INTERPERSONAL LEVEL
[M: (A1 [DECL-P (P1)S (P2)A (C1: [(T1: (f1: ‘write’)Foc) (R1: ((P1)S)) R1) (R2: (i1xi: <letter>) R2) (R3: ((P1)S) R3)])])]
b. INTERPERSONAL LEVEL
[M: (A1 [DECL-T (P1)S (P2)A (C1: [(T1: NEG (f1: ‘answer’)Foc) (R1: ((P2)A) R1) (R2: (Axi: <letter>) R2) (R3((P1)S)R3)])])]
As is evident from (26a), the unexpressed A2
of write is the object of a referential act, whereby a discourse
referent is evoked. The predication as a whole thus designates an accomplishment and no longer an
activity SoA, as in (23). This is yet another counterexample to the claims of García Velasco / Portero
Muñoz (2002: 19), Brisson (1994) and Rappaport Hovav / Levin (1998). And in (26b), the equivalent
argument of answer is equally the object of a referential act – this time anaphoric in character. The (filled)
predication frames selected will be those given as (27) and (28), respectively.
(27) REPRESENTATIONAL LEVEL
([(f1: write [V] (f1)) (x1: (P1)S)Ag (i1x2:<letter>)Pat (x3: (P2)A)Rec]Accomplishment)
(28) REPRESENTATIONAL LEVEL
([(NEG (f1: answer [V] (f1))) (x1: (P2)A) (Ax2)Pat (Ax1)Rec]Accomplishment)
6. Conclusion
The constraints on the occurrence of null complements of transitive verbs, adjectives and prepositions in
English would appear to be determined by the need to recover (i.e. to “license”) that or those internal
argument(s). The fact that one or more non-first arguments of a transitive predicate are unrealized in the
syntax does not mean that it is intransitive (i.e. monovalent) semantically.
In two of the three subtypes (the non-referential one and the anaphoric one), the zero complement of
otherwise transitive (or ditransitive) verbs, adjectives or prepositions is licensed by the highly presupposed
nature of its content: in the first case, an “inherent” argument, part of the host predicate’s meaning,
potentially narrowed to a more specific denotation type by features of the co(n)text; and in the second, a
topical (and hence also highly presupposed) discourse referent licensed via the cotext and/or context of
utterance of the host predicate, which is retrieved by the null complement. The former subtype is lexically
presupposed, while the latter is discourse-pragmatically presupposed. As for the third of the three subtypes
of null complement, the referential “discourse-new” one, its existence is licensed via a combination of the
lexical-semantic and Aktionsart structure of the host predicate (making available an appropriate inherent
argument, e.g. ‘letter’ in the case of write in its ‘correspond’ sense) and certain referentially-relevant
features of the host predication as a whole (tense, aspect, temporal or locative “framing” adverbial
modifiers etc.). Only the inherent-argument component of the discourse-new referent evoked is (lexically)
presupposed in such a case. Hence, this subtype may be viewed as the marked member of the set of null
complement interpretation types.
Our brief consideration of García Velasco / Hengeveld’s (2002) account of the structure and
function of their construct “predication frames” within the framework of a Functional Discourse Grammar
(Hengeveld 2004a, 2004b) has shown that the mapping between lexical semantics (the abstract meaning
definitions of head lexemes) and syntactic realization – the first stage of which corresponds to the
selection and completion of a relevant predication frame – is not a simple one-to-one parallel matching
between argument positions in the former and those in the latter. The relation between semantic and
syntactic structure is clearly not isomorphic, as our examination of the lexical-semantic structure of buy in
(22) and its selection of the (syntactically-relevant) predication frame in (24) showed. For the purpose of
indicating non-realization of syntactic arguments, the A2 (and potentially also A3) position(s) within the
latter need(s) to be linked to the relevant argument variable within the head lexeme’s meaning definition. I
have suggested this be done in the case of the A2 argument of buy (see also write and answer) by means of
the identity of semantic function (here the Patient role) between the two arguments, at the lexical and the
representational levels (the completed predication frames). Moreover, a consideration of these three verbs
shows not only that their semantic and syntactic structures may not be isomorphic, but also that the
selection from the meaning definition of a subset of “core” arguments by the syntax (the relevant
predication frames of these verbs) has the effect of profiling the referents of these arguments (cf. Fillmore
1977, Langacker 1991: 304-324, Van Hout 1999: 265-266), showing that the syntactic realization is not
merely “expressive” in value, but has semantic import.
The discussion in Section 5 also shows that the particular interpersonal-level representation
specified at the initial stage in an FDG clause derivation is relevant for the selection (a) of a particular
sense of a given head predicative lexeme, and (b) of a particular predication frame to express its abstract
meaning definition syntactically. A further issue is how exactly the stages of expansion of an underlying
clause structure apply as from the (filled) predication frame. Is this equivalent, for example, to the
“nuclear predication” of the standard FG model? Clearly, more work on these aspects needs to be done
within the FDG framework.
References
Brisson, C. - 1994 - "The Licensing of Unexpressed Objects in English Verbs. In Beals, K. Denton, J.,
Knippen, R., Melnar, L., Suzuki, H. & Zeinfeld, E. (eds) Papers from the 30th Regional Meeting of
the Chicago Linguistic Society. Vol. 1: The Main Session. Chicago, CLS, pp. 90-102.
Cornish, F. - 1999 - Anaphora, Discourse, and Understanding. Evidence from English and French.
Oxford, Clarendon Press.
Cornish, F. - 2002a - Anaphora: Lexico-textual structure, or means for utterance integration within a
discourse? A critique of the Functional Grammar cccount." Linguistics 40(3), pp. 469-493.
Cornish, F. - 2002b - "‘Downstream’ effects on the predicate in Functional Grammar clause derivations."
Journal of Linguistics 38(2), pp. 247-278.
Cornish, F. To appear. "Implicit internal arguments, event structure, predication and anaphoric reference."
In Hedberg, N. A. & Zacharski, R. (eds) Topi cs o n the G ra mm ar-Pragmatics Interface. Papers in
honor of Jeanette Gundel. Amsterdam/Philadelphia, John Benjamins.
Dik, S. C. - 1997a - The Theory of Functional Grammar. Part I: The structure of the slause (Functional
Grammar Series 20). Second, revised edition. Hengeveld, K. (ed.). Berlin/New York, Mouton de
Gruyter.
Dik, S. C. - 1997b - The Theory of Functional Grammar. Part II: Complex and derived constructions
(Functional Grammar Series 21). Hengeveld, K. (ed.). Berlin/New York, Mouton de Gruyter.
Fillmore, C. J. - 1977 - "The case for case reopened." In Cole, P. & Sadock, J. (eds) Syntax and Semantics
Vo l . 8 : Grammatical Relations. New York, Academic Press, pp. 59-81.
Fillmore, C. J. - 1986 - "Pragmatically-controlled zero anaphora". In Nikiforidou, K. et al. (eds)
Proceedings of the 12th Annual Meeting of the Berkeley Linguistics Society, Berkeley CA, BLS, pp.
95-107.
García Velasco, D. & Portero Muñoz, C. - 2002 - "Understood objects in Functional Grammar". Wo rk in g
Papers in Functional Grammar 76.
García Velasco, D. & Hengeveld, K. - 2002 - "Do we need predicate frames?" In Mairal Usón & Pérez
Quintero (eds), pp. 95-123.
Goldberg, A. E. - 2001 - "Patient arguments of causative verbs can be omitted: The role of information
structure in argument distribution". Language Sciences 23, pp. 503-524.
Groefsema, M. - 1995 - "Understood arguments: A semantic/ pragmatic approach". Lingua 96, pp. 139-
161.
Hengeveld, K. - 2004a - "The architecture of a Functional Discourse grammar". In Mackenzie & Gómez-
González (eds), pp. 1-21.
Hengeveld, K. - 2004b - "Epilogue". In Mackenzie & Gómez- González (eds), pp. 365-378.
Hengeveld, K. - 2005- "Dynamic expression in Functional Discourse grammar". In De Groot, C. &
Hengeveld, K. (eds) Morphosyntactic Expression in Functional Grammar. Berlin/ New York,
Mouton de Gruyter, pp. 53-86.
Hengeveld, K. & Mackenzie, J. L. In preparation. Functional Discourse Grammar. Oxford, Oxford
University Press.
Langacker, R. W. -1991- Foundations of Cognitive Grammar. Vol. II: Descriptive Application. Stanford
CA, Stanford University Press.
Lehrer, A. - 1970 - "Verbs and deletable objects". Lingua 25, pp. 227-253.
Mackenzie, J. L. & Gómez-González, M.A. (eds) - 2004 - A New Architecture for Functional Grammar.
Berlin/New York, Mouton de Gruyter.
Mairal Usón, R. & Faber, P. - 2002 - "Functional Grammar and Lexical Templates." In Mairal Usón &
Pérez Quintero (eds), pp. 39-94.
Mairal Usón, R. & Pérez Quintero, M-J. (eds) - 2002 - New Perspectives on Argument Structure in
Functional Grammar. Berlin/New York, Mouton de Gruyter.
Mejri, S. & François, J. To appear. "Restrictions sémantiques sur l’objet sous-entendu de verbes transitifs
(le cas de boire)." In François, J. et Mejri, S. (eds) Composition syntaxique et figement lexical
(Bibliothèque de Syntaxe et Sémantique, 3). Caen, Presses Universitaires de Caen.
Rappaport Hovav, M. & Levin, B. - 1998 - "Building verb meanings". In Butt, M. & Geuder, W. (eds) The
Projection of Arguments: Lexical and Compositional Factors. Stanford, CSLI Publications, pp. 97-
134.
Ritter, E. & Thomas Rosen, S. - 1998 - "Delimiting events in syntax". In Butt, M. & Geuder, W. (eds) The
Projection of Arguments: Lexical and Compositional Factors, Stanford, CSLI Publications, pp. 135-
164.
Va n H o u t, A. - 1999 - "Event semantics in the lexicon-syntax interface: Verb frame alternations in Dutch
and their acquisition". In Ten ny, C. & P ustej ovsky, J. (eds) Events as Grammatical Objects. The
Converging Perspectives of Lexical Semantics and Syntax. Leland Stanford Junior University, CSLI
Publications, pp. 230-282.
Va n Val i n, R . D . & La P o ll a , R. J . - 1997 - Syntax: Structure, Meaning and Function. Cambridge,
Cambridge University Press.
Notes
1 I would like to thank Christopher Butler for detailed comments on an earlier draft of this paper, as well as for helpful
discussion of some of the issues raised in it at the ESSE7 Seminar on FDG held in Zaragoza (9 September 2004). I also
thank Ricardo Mairal Usón, Anne Grobet, Daniel García-Ve l a sc o , De n is A po t h él o z, M i ch è l e Noailly, Nancy Hedberg and
an anonymous external referee for commenting on an even earlier (and longer) version of this paper (Cornish to appear)
which was not framed within a specifically F(D)G context. All responsibility for any errors remaining is mine alone.
2 This difference is not explicitly or consistently drawn in Dik (1997a): cf. Cornish (2002b: 256-257), where I argue that the
predicate frames of the standard model of FG are hybrids, conflating and thus confusing the two dimensions. Van Valin /
LaPolla (1997: 27-28) do, however, explicitly draw this distinction.
3 Cf. Dik’s (1997a: Section 4.2.6) account of the operation of selection restrictions within standard FG.
4 See Cornish (1999: ch. 4) for arguments in favour of this move. García Velasco / Portero Muñoz (2002: 21) suggest
formalizing the exophoric/ endophoric distinction by means of a distinct operator (“EX” for “exophoric”). However, this
does not take into account my arguments for the conflation of this traditional “geographical” distinction (“antecedent”
within co-text vs. within situational context). Given that both sources of indexical reference presuppose the high saliency
and topicality of the entities concerned, together with the fact that the same expression-types serve to retrieve referents
made available via either of these sources, then it is theoretically more satisfactory to subsume them under the heading
“anaphoric”, formalized by means of a single operator, “A”.
5 See also the ill-formed examples of 2- or 3-place achievement or accomplishment verbs with null complements in (2) above –
where the tense is also the definite past, and the intended referent of the null complement may also be contextually salient.
6 However, my feeling is that examples like Brisson’s (unstarred) (9b) (p. 91) Jack swept in an hour are not fully acceptable.
Interestingly in this regard, Ritter / Thomas Rosen (1998: ex. (48a), p. 160) query the full acceptability of a similar example,
where sweep has a definite object NP: John swept the floor (?in 5 minutes/for 5 minutes).
7 Christopher Butler has suggested there that it is more likely that the null reference would be anaphoric rather than deictic in
(19). I accept this possibility, but my feeling is that either type of interpretation is possible here – i.e. the utterance may be
contextualized in two ways. However, whether it is in the last analysis deictic or anaphoric, in either case there is an A2
“structure” argument present, a fact which contradicts Brisson’s and Rappaport-Hovav / Levin’s stricture.
8 Henceforth GVH.
