Presented at the LSA by AEG, Jan 7, 1999
Obligatory Adjuncts
Adele E. Goldberg and Farrell Ackerman
University of Illinois and University of California, San Diego1
In this paper, it is argued that the existence of “obligatory adjuncts” in both predication and
modification is best understood as following from general conversational pragmatics; no
grammatical stipulation is necessary. In the case of clausal predication, the requirement
corresponds to the familiar idea that every utterance requires a focus; adjuncts are just one of
several ways in which the focal requirement can be satisfied. It is argued that the focal
requirement itself is best understood in terms of Grice's maxim of Quantity or Horn's RPrinciple, and that these latter principles also account for the appearance of obligatory
adjuncts with nominal modification structures.
1. Introduction
Consider the contrasting clauses with short passives in (1) and nominal modification structures in
(2), first observed by Grimshaw and Vikner (1993). When uttered with “neutral” intonation,
adjuncts are required in order to avoid a sense of anomaly: impressionistically, the (a) structures
leave us yearning for more to be said, while the (b) structures somehow satisfy this yearning:
1
a. #This house was built.
b. This house was built last year.
2
a. # a built house
b. a recently built house
We will be examining these so-called “obligatory adjuncts”: the adjuncts such as last year and
recently which seem to rescue (1b) and (2b). We will critically examine a grammatical account
offered by Grimshaw and Vikner (1993) and suggest instead a pragmatic account of obligatory
adjuncts. We will argue that: predication or modification of an argument is only licensed when
it is informative in the discourse context.
Our account of clausal predication as in (1a,b) is motivated by the widely accepted claim that
utterances require an information focus providing an assertion that makes the contribution worth
1
uttering in a conversation (Bolinger 1965; Rooth 1992; Kiss, 1998; Halliday 1967; Lambrecht
1994; Polinsky 1999).
By information focus, we follow Halliday (1967:204) who defines it as, “one kind of emphasis, that
whereby the speaker marks out a part (which may be the whole) of a message block as that which
he wishes to be interpreted as informative."
Lambrecht (1994:207) likewise defines the focus of a proposition as “the UNPREDICTABLE or
pragmatically NON-RECOVERABLE element in an utterance”.
An explanation for the unacceptabilty of (1b), (#The house was built) then is simply that the
utterance does not contain an information focus, in that it conveys no new or informative
information. More specifically, a clause with a definite subject presupposes the existence of the
subject referent (Strawson 1964). Therefore it is possible to infer that at some point in the past, the
referent was created.2 Nothing informative is being said that can't be calculated by knowing how
presupposition works in definite NP subjects and the meaning of house. The fact that adjuncts are so
often required for passive verbs of creation is expected, since we generally assume that artifacts are
created in default ways.
As a heuristic for isolating the information focus, we can rely on Erteschik-Shir and Lappin's
(1979) “lie” test. The focus of an utterance (the “dominant” part of an utterance in Erteschik-Shir's
terminology) is that part which is denied by the assertion “That’s a lie.” Consider how the lie test
operates on the short passives involving verbs of creation:
A: This house was built in 1992.
B: That’s a lie! It was built in 1972!
Focus = `in 1992'
B’: That’s a lie! # It was not built.
1
We would like to thank Chris Barker, Georgia Green, Masha Polinsky, David Perlmutter, Svetlana Godjevac and the
audience at the Finnish Symposium on the Relationship between Syntax and Semantics, held in Helsinki, September
1999, for very helpful discussion.
2
It is important to note that we are not claiming that English or natural languages generally have no tolerance for any
redundancy. This point will be clarified below.
2
Without a special context, it is more natural for B to deny the date in which the building was
erected than to deny that the house was built at all. This is because, in a non-contrastive context, the
mere fact that the house was built is not what is asserted by A's utterance. It is instead
presupposed, and therefore, like presuppositions generally, cannot easily be negated.
We will see a little later that adjuncts are just one of several ways in which the focal
requirement can be satisfied. The focal requirement itself can be understood in terms of Grice's
maxim of Quantity or Horn's R-Principle, “make your contribution necessary; say no more than
you must,” since it is not necessary to utter a sentence that contains only redundant or recoverable
information, i.e., uttering a sentence without a focus violates the maxim of quantity. Following
previous work by Ackerman and Goldberg (1996), we will see that this conversational principle
also accounts for the appearance of obligatory adjuncts within nominal modification structures as in
(2a,b).3
2. An Event Structure Account
Grimshaw & Vikner (1993) provide a rare previous attempt to account for the contrasts in (1a-b)
and (2a-b). Their proposal relies on the complex event structure of accomplishment predicates, so
we will refer to this as the event structure account. Accomplishments consist of two subevents, a
process and a state. G&V suggest that each subevent must be “identified" by some element in the
sentence.
They observe that only certain accomplishment predicates require obligatory adjuncts as in (3a-b);
these involve verbs of creation. No adjuncts are needed with other types of accomplishment verbs
such as verbs of destruction, as in (4), nor with verbs in which a performance object is created such
as record in (5).
3
Grimshaw and Vikner cite David Pesetsky (personal communication) as suggesting that the adjuncts are required
because one must “say something.” The following discussion can be viewed as an attempt to take up Grimshaw and
Vikner’s challenge to “characterize the notion of ‘say something’ in a revealing way.” (1993:154).
3
3a. #This house was built/created/made/designed.
b. This house was built/created/made/designed by a 16th century architect.
4. This house was destroyed.
5. Your conversation was recorded.
Given that all of these predicates are accomplishments, and therefore consist of two subevents: a
process and a state, the complex nature of the event structure itself cannot be sufficient to explain
the different behaviors of the various verbs.
G&V hypothesize that in the case of verbs of creation, the theme argument (y in the examples in 6
and in Figure 1) can only identify the state, since the theme does not exist until the process is
complete. Therefore the process component of the complex event must be identified, they reason,
either by the causer argument in an active sentence (x in 6a), or by an adjunct in a passive sentence
(6c).
Figure 1:
6a. x builds y
b. *y was built.
th
c. y was built in the 20 century.
event
process
state
In the case of other accomplishment verbs, e.g., destroy, record, they suggest that the theme
argument (corresponding to the y variable in example 7a,b and in Figure 2) simultaneously
identifies both subevents, since, for example, the sound exists even as it is being recorded.
Because both subevents are identified, no adjunct is required (7b).
4
Figure 2:
7a. x records y
b. y was recorded.
event
process state
x, y
y
3. Problems for an Event Structure Account
Although this basic idea has been adopted by a number of theorists (e.g., van Hout 1996; Rappaport
Hovav & Levin 1998, Levin 19994), there remain outstanding empirical problems with the Event
Structure account. There are two basic questions that we will focus on here:
1. Is the class of participating predicates really only accomplishment predicates?
2. What is the nature of the split behavior within the class of accomplishments?
2.1.Statives and activities as well as accomplishments can require adjuncts
The event structure account depends crucially on the complex nature of the event structure of verbs
of creation. And yet we find predicates with simple event structures, both statives (8) and activities
(9-10), also often require some type of adjunct when they appear in short passives.
8a. # The claim was believed.
b. The claim was believed by many/in the seventh century/in the South.
9a. # The book was read.
b. The book was read by many/yesterday/over the airwaves/in the shower.
Contrary to Grimshaw and Vikner, Rappaport Hovav and Levin assume that only arguments can “identify” a
subevent. This stronger claim is also counterexemplified by the data presented here since many examples are provided
in which no adjunct or argument is required to identify a second subevent.
4
5
10a. #The television program was watched.
b. The television program was watched all over the country/by millions/with anticipation.
3.3. Constructive accomplishments do not uniformly require adjuncts
Certain changes in tense as in (11), which was noted by G&V, can obviate the need for an
adjunct. A further problem for the event structure account stems from the fact that in addition to
changes in tense, variations in modality (12), polarity (13), and emphatic uses of auxiliaries as in
(14) can eliminate the need for an adjunct as well.
11a The house will be built.
b. The house is being built
c. The house has been built.
12a.
b.
The house might be built.
The house should be built.
13. The house wasn't built.
14, The house WAS built.
It is not obvious that these factors alter either the event structure of accomplishments or the nature
of identification associated with them.
One might be led from this data to claim that all
expressions with future or perfect tense, negative polarity or modals must have a simple event
structure5; that is, the event structure account could be defended by claiming that none of the
examples in 11-14 actually involve a complex event. We will suggest, however, that there is no
need to make this drastic move.
5
It is hard to imagine anyone claiming that the emphatic auxiliary somehow results in a simple event structure, so
example (14) would still remain a problem for the event structure account.
6
To summarize, the event structure account is both too specific and too general, in that it
does not explain why examples like (8a), (9a) and (10a) require an adjunct nor why examples such
as (11-14) do not.
3. Pragmatic Proposal
Recall the pragmatic proposal described at the outset: the basic idea in the case of clausal
predications is that utterances require a focus; adjuncts are just one way of satisfying this
requirement. The idea that asserting something that is already presupposed is infelicitous conforms
to Stalnaker's essential conditions on the rational employment of assertion in communication. He
writes (1978:325):
"To assert something incompatible with what is presupposed
is self-defeating; one wants to reduce the context set, but not to eliminate
it altogether. And to assert something which is already presupposed is to
attempt to do something that is already done." [emphasis ours]
Notice that if a contrastive context is invoked, and we assume that what is asserted is that the
house was in fact actually built (with stress on was), no adjunct is required:
15. This house WAS built.
In this case, there is an implicit contrast with a negative proposition, and the positive polarity of the
copula verb provides a contrastive focus for the clause. Constrastive focal stress on the subject
argument or on the verb can also, as expected, render bare passives felicitous since a focus is
provided:
16. The HOUSE was built, (not the garage).
17. The house was BUILT, (not just designed).
There are various other ways of providing a focus in simple sentences as well. In fact, and contrary
to the blanket prohibition against constructive accomplishments in the event structure account, if
7
the method of creation is somehow unusual, a verb of creation can itself play the role of focus,
without emphatic stress.
18. This cake was microwaved.
19. These diamonds were synthesized.
With respect to tenses, those other than the simple past serve to inform the listener that the creation
took place before, after or during a particular reference time:
20a. The house will be built.
b. The house has been built.
c. The house had been built.
d.The house is being built.
The pragmatic account does not depend on the event structure of the predicate. The examples in
(8a), (9a) and (10a) repeated below each lack an information focus. We assume claims are
believed, books are read and television programs are watched by someone or other; in a neutral
context, these examples do not convey anything informative.
8a.# The claim was believed.
9a. #The book was read.
10a. #The television program was watched.
Since only definite subjects presuppose the existence of their referent, and therefore presuppose
that the referent was created, an indefinite subject can render the short passive acceptable:
21a. A MANSION was built.
b. MANSIONS were built.
These sentences can be used to introduce one or more mansions into the discourse, as in the
following context:
8
22b. The city expanded. MANSIONS were built. Roads were paved.
In this case, the subject argument must receive the sentence accent indicating that it is, or is part of,
the focus domain (see Lambrecht 1994 for discussion). It cannot receive the default sentence final
accent, except, as just mentioned, in a contrastive context (see 16):
23. # A mansion was BUILT.
To summarize, there are many ways of making a clause informative (i.e., providing a focus). An
adjunct phrase is just one way. Without an adjunct or some other focal information, the clausal
predication is infelicitous.
Middles
The same pragmatic explanation extends naturally to the English middle construction as well. It has
frequently been observed that English middles often require some type of adjunct:
24a. #The car drives.
b. The car drives like a boat/easily /365 days a year/only in the summertime.
As is evident from (24b), a wide variety of adjuncts can be used to rescue middles from infelicity.
Several researchers have observed that negated middles or middles that are overtly emphasized
often attenuate the need for an adjunct (Keyser and Roeper 1984: 385; Fellbaum 1985: 9; Dixon
1991: 326):
25. That car doesn’t drive.
26. These red sports cars DO drive, don’t they? (Dixon 1991:326).
Fellbaum (1985) notes that the negation serves to supply “non-given” information, while the
emphasized verb serves to indicate non-expectedness. That is, the change in polarity or emphasis
9
provides a focus for the clause, making the expression informative and therefore acceptable. We
assume that cars can be driven so asserting that they cannot be (25) is informative; in (26), the
emphasized auxiliary is used to convey the idea that the cars drive really well or fast or easily.
In accord with this explanation, Rosta (1995: 132) notes that “the more `newsworthy' adjunctless
mediopassives are, the less odd they are." He cites the following examples in this context:
27a.The car will steer, after all.
b. These beaurocrates bribe.
a. Boy did that mountain climb!
b. She sure did interview. (1995:132)
Recognizing the requirement for an adjunct to be a pragmatic requirement allows us to explain
why many middles do not require an adjunct. For example,
27. <How do you close this purse?> It snaps/ It zips/ It buttons.
28. <Where do we enter the secret passageway?> The bookshelf opens.
In a context in which it is informative to assert that people should be able to perform a given action
on the subject argument, no adjunct is required.6
Summary
Sometimes what would otherwise be a fully acceptable sentence, is unacceptable because it doesn’t
count as saying enough to be licensed conversationally: it doesn't contribute more than could be
calculated by knowing the word meanings and presupposition structure of the sentence.
The
conversational context and background assumptions are critical, so that the predications that appear
odd in a “neutral” context can often be rescued by simply finding the right context.
6
As David Perlmutter (personal communication) observes, the present account also extends to instances of cognate
objects, which general require some type of modification since otherwise the verbal predication is wholly redundant:
a. #I dreamed a dream.
b. I dreamed a scary dream.
10
We have relied on the idea that every utterance must contain a focus. As we noted at the outset,
this constraint ultimately follows from Grice's maxim of Quantity, or Horn's R-Principle, since it is
not normally necessary to utter sentences without an information focus, that is, sentences in which
all information is redundant or recoverable. Viewing the focal requirement in these terms allows us
to extend the present account to instance of obligatory adjuncts in certain modification structures.
The use of modification must also be informative within the discourse context.
18. Adjectival Past Participles
In Ackerman & Goldberg (1996), we demonstrated how conversational pragmatic principles
account for obligatory adjuncts in nominal modificational structures. For example, it was argued
that simply modifying house with built as in (29a) is not felicitous because it is not informative: we
assume houses are built:
29a. #a built house
b. a recently built house
As expected, the felicity of a modification structure is dependent on the combination of a particular
property and a particular head noun, not on the event structure of the deverbal adjective. Notice
that the same modifier, paid, is odd in (30a), but fine in (30b). This is because we readily assume
that physicians but not escorts, are paid:
30a. #paid physician
c. paid escort
Notice also that, as we saw in the case of clausal predication, the modifiers can be rescued by a
contrastive context: in this case the modification serves the function of differentiating the state of
affairs from other states of affairs relevant to the discourse. Therefore the modification is
informative, and as expected, acceptability results:
31. That particular physician was unpaid, but these are all paid physicians.
11
The idea that modification must be informative extends beyond instances of adjectival past
participles to modification more generally. Consider cases of denominal adjectives with -ed
suffixes:
32. # headed boy; but red-headed boy
Hirtle (1969) observed, “The very notion underlying boy brings in the notion of head with no
outside help." Finally, underived adjectives also also have to be informative (33a-c):
33a. # dead corpse
d. # liquid water
e. #cold ice
Interestingly, what counts as informative, is somewhat context dependent. While wholly
redundant information is not typically informative, it can be if the listener can infer something
informative from the fact that the speaker bothered to utter it. For example, while out of context,
as expected, cold ice seems odd since ice is always cold, the example in (34) is nonetheless fully
acceptable:
34. The cold ice felt good on her throat.
In this context, mentioning the fact that the ice was cold is informative since it tells the addressee
which aspect of the ice felt good. That is, the modification informs the addressee that it was the
coldness of the ice that was relevant.
Similarly, the following context rescues #solid rock:
35. After climbing out of the quicksand, she was grateful to sit on the solid rock.
19. Conclusion
12
To conclude, we have argued that the distribution of “obligatory adjuncts” follows from discourse
pragmatics. The general requirement is that:
Predication or modification of an argument is only licensed when it is informative in the
discourse context.
This generalization follows from general conversational principles and thus requires no
grammatical stipulation. Moreover it allows us to account for a much wider range of data than the
event structure account; it explains why changes in tense and polarity as well as changes in context
and background assumptions can all alleviate the need for an adjunct since they can make the
predication or modification informative.
References
Ackerman, Farrell and Adele E. Goldberg. 1996. Constraints on Adjectival Past Participles. In Conceptual Structure,
Discourse and Language, A. E. Goldberg, (ed.) CSLI Publications.
Bolinger, Dwight. (1965) Forms of English. Cambridge: Harvard.
Fellbaum, Christiane. 1985. On the Middle Construction in English. Indiana University Linguistics Club.
Grice, Paul H. 1975. Logic and Conversation. In Cole and Morgan 9eds). Syntax and Semantics volume III: Speech
Acts. NY: Academic Press. 41-58.
Grimshaw, Jane and Sten Vikner. 1993. Obligatory Adjuncts and the Structure of Events. In Knowledge and language
eric Reuland and Werner Abraham (eds). Kluwer Academic Publishers.
Halliday A. K. 1967. Notes on Transitivity and theme in English. Part II. Journal of Linguistics 3. 199-244.
Hirtle, W. H. 1969. -Ed Adjectives like Verandahed and Blue-Eyed. Journal of Language 6:19-36.
Horn, Lawrence. 1984. Toward a New Taxonomy for Pragmatic Inference: Q-Based and R-Based Implicature. In D.
Schiffrin, ed., GURT '84: Meaning, Form and Use in Context. Washington: Georgetown University Press. 11-42.
Keyser, S.J. and T.Roeper. 1984. On the middle and ergative constructions in English. Linguistic Inquiry 15:3. 381416.
Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: Cambridge University
Press.
Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1: 75-116.
13
Comments at LSA meeting 1999:
Larry Horn: it’s true that you can only get “one-headed man” in certain pragmatically determined contexts,
but why is it that you can’t ever just get “headed man”?
Afterwards both Emmon Bach and Chuck noted that the “Ed” ending is not productive. In the right context,
you can get “/leggd/ men” but never “/leggED/ men”
It may be that what is productive is the _____+headED man, this being a lexical item and not simply headED,
because headED has only been produced with the modifier there….So the pragmatic constraint is actually
grammaticalized for that lexical item.
Of course I don’t know why you can’t get the “/eyd/ man” (the man with an eye or eyes) very easily….
Hana Filip: how come you can say:
(1)The claim was contested.
Or
(2)the claim was argued for.
(1) we don’t assume claims will be contested; also “contest” introduced a negative element, and negatives in
general provide the right context (I suppose we don’t generally presuppose negative statements). I don’t think
(2) is actually any better than The claim was believed. Hana later agreed.
Ivan asked about the one focus per utterance constraint: “is that per sentence or per clause?” Ans: it should
really be per clause since we are really saying that EVERY predication must be informative. So what about
cases of parallelism whereby the second clause seems to contain no focus but be present only to preserve the
parallel: “They didn’t know the house had been built, but I knew it was built.”
Georgia Green: So I guess your account predicts that you CAN actually say “the house was built” if it’s
embedded in the right discourse context, and in fact it can…
14