University of Pennsylvania Working Papers

in Linguistics
Volume 28
Issue 1 Proceedings of the 45th Annual Penn Article 13
Linguistics Conference

11-18-2022

Against the Argument/Adjunct Distinction

Andrew McInnerney
University of Michigan

Follow this and additional works at: https://repository.upenn.edu/pwpl

Recommended Citation
McInnerney, Andrew (2022) "Against the Argument/Adjunct Distinction," University of Pennsylvania
Working Papers in Linguistics: Vol. 28: Iss. 1, Article 13.
Available at: https://repository.upenn.edu/pwpl/vol28/iss1/13

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/pwpl/vol28/iss1/13

For more information, please contact [email protected].
Against the Argument/Adjunct Distinction

Abstract
This paper suggests that much of the empirical basis for the argument/adjunct distinction (A/AD) is
weaker than traditionally believed. First, I argue that successful argumenthood diagnostics should both (i)
identify a distinction among dependents which resembles the conventional A/AD, and (ii) draw a
distinction which is not already predicted independently of the A/AD. Focusing on PPs within the VP in
English, I argue that purported diagnostics for argumenthood, including omission, 'do so'-substitution, and
iteration, do not meet these criteria, therefore do not provide empirical support for the A/AD. After briefly
discussing the prospects for several further diagnostics, I conclude that it is plausible that the A/AD
might be eliminable. Though this would raise technical questions in various areas, it would be desirable
from a Minimalist perspective.

This working paper is available in University of Pennsylvania Working Papers in Linguistics:

https://repository.upenn.edu/pwpl/vol28/iss1/13
 Against the Argument/Adjunct Distinction

Andrew McInnerney*

1 Introduction
The Argument/Adjunct Distinction (A/AD) has come to be a widely accepted component of syn-
tactic theory. Though its detailed formal implementation varies across frameworks, the basic idea
that the dependents of a given head can be divided into two classes, a class of arguments, which in
some sense form the core of the projection, and a class of more peripheral adjuncts, is adopted
near-universally. Despite this, the null hypothesis is always to reject distinctions like this. As such,
the A/AD, or, more specifically, the syntactic apparatus postulated to implement it, is a hypothesis
which must be supported by evidence.
Traditionally, the A/AD has been supported by diagnostics for argumenthood, syntactic phe-
nomena whose application draws a distinction between dependents, where this distinction has
been identified with the A/AD. For example, some dependents of a given head can be omitted, and
others cannot be omitted. The omissibility diagnostic thus draws a distinction between elements
that are omissible and those that are not. To the extent that this distinction is related to the A/AD,
omissibility may provide evidence in favor of the A/AD. The same applies for other traditional
diagnostics, such as do so-substitution, iteration, and islandhood.
The goal of this paper is to establish a framework for evaluating whether a given set of diag-
nostics successfully motivates the A/AD, and to apply this framework to several traditional diag-
nostics for argumenthood. Section 2 first describes two criteria which any diagnostic must meet to
count as motivation for the A/AD. Sections 3-6 then evaluate traditional diagnostics by these crite-
ria. Assuming that, if there is an A/AD, it will apply in the VP, this paper focuses on the applica-
tion of A/AD diagnostics to dependents of V. As such, sections 3-5 focus on the phenomena of
omission, do so-substitution, and iteration as they apply within VP. For reasons of space, some
potential diagnostics cannot be discussed in detail here; abridged discussion of additional diagnos-
tics is contained in section 6. Section 7 concludes it is plausible that the A/AD could be eliminated
in favor of alternative, more accurate explanations for effects traditionally attributed to it.

2 Diagnostics for Argumenthood as Evidence for the A/AD

The A/AD is traditionally thought to be identified by a collection of diagnostics, including those
listed in (1). See Schütze (1995), Huddleston & Pullum (2002), Needham & Toivonen (2011) and
numerous other sources for similar lists.

(1) a. Omission: The adjuncts to a verb can be omitted, but a verb’s arguments (usually) can-
not be omitted.
b. Do so-substitution: The adjuncts in a VP can occur with do so taking that VP as its ante-
cedent, but the arguments in a VP cannot occur with do so.
c. Iteration: The adjuncts to a verb can be freely iterated, but the arguments cannot be iter-
ated.
d. Islandhood: The adjuncts to a verb are opaque for the purposes of extraction, but the
arguments are (by default) transparent for extraction.
e. Extraction from weak islands: The adjuncts to a verb cannot be extracted out of a weak
island, but a verb’s arguments can (at least marginally).
f. Reconstruction for Condition C: The adjuncts to a verb do not reconstruct for Condition
C under extraction, but a verb’s arguments do reconstruct under extraction.
g. Permutation: The adjuncts to a verb can be freely permuted, but permutation of a verb’s
arguments degrades acceptability.

*
This work has benefitted from helpful discussions with Acrisio Pires, Ezra Keshet, Lisa Levinson,
Daniel Seely, and Richard Lewis. I thank Natasha Abner, Marlyse Baptista, Alan Ke, Yushi Sugimoto, Lucy
Chiang, Peter Culicover, Rafaela Miliorini, Omer Preminger, and audiences at PLC45 for helpful comments.

U. Penn Working Papers in Linguistics, Volume 28.1, 2022

2 ANDREW MCINNERNEY

Before evaluating these diagnostics, this section seeks to establish what a successful diagnostic for
the A/AD would look like. I propose two criteria. The first is a heuristic. The A/AD has a long
history, and it is associated with certain canonical categorizations. For example, NP dependents of
the verb, including grammatical subjects and direct objects, are almost always considered argu-
ments. Some PPs are canonically considered arguments, and some are canonically considered ar-
guments, with lines typically drawn thematically; e.g. recipient PPs, addressee PPs, subject-matter
PPs, and goal PPs are typically considered arguments, while PPs denoting manner, temporal, loca-
tive, benefactive, and instrumental information are typically considered adjuncts (Schütze 1995,
Huddleston & Pullum 2002, Koenig et al. 2003, Dowty 2003). Additionally, dependents which are
non-omissible are conventionally considered arguments, virtually as a matter of definition (Hud-
dleston & Pullum 2002:221-222, Boeckx 2003:96, Patejuk & Przepiórkowski 2021:4-5).
This conventional A/AD draws distinctions among dependents independently of the particular
implementation one adopts for this distinction. Therefore, to say that some diagnostic supports the
A/AD means that it provides linguistic evidence for a distinction resembling the conventional no-
tion presupposed by prior literature. A heuristic criterion naturally emerges: a successful defense
of the A/AD would have the results of each diagnostic align with the predictions which come from
the conventional A/AD. That is, a diagnostic should treat most of the conventional arguments one
way, and most of the conventional adjuncts another way.
As for the second criterion of evaluation, a diagnostic must make a distinction among PPs
which is not motivated independently of the A/AD. For example, suppose that a potential diagnos-
tic is shown to draw a distinction between prosodically heavy phrases prosodically light phrases.
Because prosodic heaviness has nothing to do with argumenthood, this diagnostic could not be
taken as evidence for a syntactic A/AD. More concretely, below I argue that the iterability diag-
nostic distinguishes locative/temporal PPs (regardless of their argumenthood) from other PPs.
Since locatives and temporals form a natural class independently of the A/AD, iterability can be
attributed to that distinction, with no need to invoke the A/AD. Thus, diagnostics motivating the
A/AD should be sensitive to properties specific to the A/AD, not attributable to other phenomena.
A corollary of this second criterion is the following: for a diagnostic to be relevant to the syn-
tactic A/AD, it must be sensitive to properties which cross-cut categorial distinctions. If a diagnos-
tic draws a distinction only across categories (e.g. between NPs and PPs), but does not draw dis-
tinctions within categories (e.g. treating all NPs equally and all PPs equally), then there is no need
to appeal to an A/AD to explain these facts; the differences can simply be attributed to the already-
needed category distinction. For this reason, my focus will primarily be on prepositional phrases
here – by sticking to a single category, it will be possible in principle to identify differences in
empirical behavior which would need to be explained by something like the A/AD. Thus, for the
purposes of this paper, a given diagnostic can be relevant for the A/AD only if it detects a differ-
ence of some sort among PPs.
In order to provide motivation for the syntactic A/AD, a diagnostic must meet both criteria.
Imagine, for instance, that a given diagnostic meets criterion (ii), the no-independent-motivation
criterion, but not criterion (i), the heuristic criterion. In this case, we would be looking at a diag-
nostic which motivates some novel theoretical concept, but it would be an error to identify this
concept with the A/AD. The conventional A/AD exists independently of any particular theory of
the distinction. Thus, even if a diagnostic were to identify a syntactic distinction between verbal
dependents, if the distinction does not resemble the conventional A/AD in some way, then it
makes more sense to call the new distinction by a new name; there would be no reason to relate it
to the A/AD if it did not bear some resemblance to the conventional, off-the-shelf concept.
What if a given diagnostic satisfies the heuristic criterion, but fails to do so with respect to the
no-independent-motivation criterion? In this case, the appropriate conclusion depends on the be-
havior of other diagnostics. If a number of other diagnostics pattern the same way, all reducible to
the same, independently motivated property, while at the same time passing the heuristic criterion,
then we would conclude that the A/AD could be reduced to this independent property. On the oth-
er hand, if it is not the case that multiple diagnostics reduce to the same principles – even if they
partition dependents into similar but crucially not identical classes – then we must conclude that
the A/AD is not supported. That is, multiple diagnostics passing the heuristic criterion while fail-
ing the no-independent-motivation criterion for different reasons would not motivate the A/AD. In
this case, several independent properties would happen to pattern somewhat closely, but there
 AGAINST THE ARGUMENT/ADJUNCT DISTINCTION 3

would be no reason to unify those properties under the larger umbrella of the A/AD (as this would
violate principles of parsimony).
The following sections demonstrate that, by these criteria, certain traditional diagnostics do
not provide evidence for a syntactic A/AD. I conclude that, with the traditional set of diagnostics,
it is plausible that the syntactic A/AD might be eliminable.

3 Omission
Perhaps the most prototypical diagnostic for argumenthood is the diagnostic from omission. Omis-
sibility was one of the earliest formal diagnostics for argumenthood (see e.g. Harris’s 1970:275fn1
application of the diagnostic in Structuralism), and it has continued to be a primary test for argu-
menthood. Huddleston & Pullum’s (2002:222(c)) grammar of English, which places great empha-
sis on empirically justifying grammatical postulates, assigns omissibility an important role in iden-
tifying adjuncts. Indeed, Boeckx (2003:96) refers to omissibility as “adjuncts’ core property.”
Despite this, it is also well-known that the omissibility diagnostic has major problems, with
some authors already noting that it does not provide good motivation for the A/AD (e.g. Culicover
2009:69, Needham & Toivonen 2011:9). I argue in support of this idea here. First, I discuss the
independent motivations for omission, concluding that some mechanism like subcategorization is
required to capture patterns of (non-)omissibility. Then, I argue that the omission diagnostic does
not provide evidence for the A/AD because it does not identify a distinction aligning with the con-
ventional understanding of argumenthood.

3.1 Factors Conditioning Omissibility

The omission diagnostic for argumenthood is thought to apply in contrasts such as (2).

(2) Lee devoured *(something) (yesterday).

The direct object of devour is non-omissible, while a temporal like yesterday is omissible. Accord-
ing to the omission diagnostic, this is explained by the A/AD: direct objects are arguments, while
temporals are adjuncts.
However, in many cases, patterns of omission arguably arise for reasons unrelated to the
A/AD. Since such data are explained independently of the A/AD, they do not provide motivation
for the A/AD. I will briefly discuss two A/AD-independent factors which condition omissibility in
this subsection. First is idiomaticity. Certain verb + preposition combinations, like result in, de-
pend on, or deal with, are idiomatic. I take such examples, in which the preposition is invariable
and semantically bleached, to be idiomatic following Pesetsky (1995:135), who writes “… when a
predicate takes an ‘idiosyncratic’ preposition, such as on in depend on …” this represents “… a
special case of idiomaticity.”
If a particular verb + preposition combination were idiomatic, then we would expect the pres-
ence of the preposition (and hence the PP) to be obligatory to complete the idiom. For data like (4),
then, and other data like these, we expect the PP to be non-omissible due to its participation in an
idiom (independently of its alleged argumenthood).
Another factor conditioning omission is pragmatics. Some have suggested that some canoni-
cal adjuncts are syntactically non-omissible for pragmatic reasons. For example, Goldberg &
Ackerman (2001), building on ideas of McConnell-Ginet (1994) and Ernst (1984), suggest that
some PPs are non-omissible because without them, the Gricean maxim of quantity is violated.
Take the PP that occurs with dress, which has traditionally been characterized as an “obligato-
ry adverb” (Jackendoff 1972).

(6) Lee dressed *(like a clown).

Goldberg & Ackerman suggest the reason the PP is non-omissible here is that, without it, the
pragmatic contribution of the sentence is insufficient, at least out of the blue. The assertion that
someone “dressed,” in the sense that they simply wore clothing of some unspecified style, is not
sufficient to satisfy the maxim of quantity. Verbs like dress “give the appearance of needing an
4 ANDREW MCINNERNEY

adverb only because in most contexts they do not provide sufficient information by themselves,”
(Ernst 2001:273, emphasis mine). Thus, pragmatics arguably offers a further independent motiva-
tion for non-omission, independent of the A/AD.
Still, some data are left unexplained by these two independent factors (idiomaticity and prag-
matics). Take verbs like slog, lumber, trudge, tramp, and traipse (all meaning roughly “walk ef-
fortfully”). These all take a non-omissible PP denoting a path or goal, as shown in (8) with trudge.

(8) Lee trudged *(through the field).

Non-omissibility does not follow from idiomaticity with these verbs, since a variety of preposi-
tions can be used. Consider into, around, out of, and so on in place of through in (8). Further, it is
clear that non-omissibility does not come from pragmatics either, both because altering the context
does not facilitate omission, and because paraphrases (with walk) do not require a goal PP.
While relatively rare, it is possible to identify other examples of PPs (like those with the verbs
put, reside, and hand) whose non-omissibility is not attributable to idiomaticity or pragmatic felic-
ity. The existence of these cases implies that something like syntactic subcategorization is at work,
as a lexical property of verbs. That is, it is not possible to deduce, from one verb to the next, which
dependents will be syntactically omissible and which will not. As mentioned above, the verbs slog,
trudge, etc. have a non-omissible goal PP, while related verbs like walk and travel do not.

3.2 Heuristic Problems for the Omission Diagnostic

While omissibility can be partially explained by independently-motivated principles, it cannot be

entirely explained by them. Some notion of subcategorization is needed to directly capture certain
examples of non-omissible dependents. Here, I show that the diagnostic fails to meet the heuristic
criterion described in section 2, and as such does not qualify as supportive of the A/AD.
Before demonstrating the heuristic problems with the diagnostic, it is important to flag the
fact that omissibility comes in two flavors. One version, which I will call the strong omissibility
diagnostic, assumes a two-way relationship between omissibility and argumenthood, as in (9).

(9) Strong omissibility diagnostic:

a. Arguments are always non-omissible.
b. Adjuncts are always omissible.

In other words, the inability to be omitted is a necessary and sufficient condition to identify a giv-
en dependent as an argument. (And the converse, the ability to be omitted is a necessary and suffi-
cient condition to identify a dependent as an adjunct.) There is also a weak version of omissibility.

(10) Weak omissibility diagnostic:

Adjuncts are always omissible.

The weak omissibility diagnostic states only that omissibility is a necessary condition for adjunc-
thood. Thus, weak omissibility is only a one-way diagnostic; it is capable of identifying non-
omissible dependents as arguments, but it cannot identify omissible dependents either way.
As I mentioned at the beginning of this section, previous authors have already recognized the
heuristic problems faced by the omission diagnostic (see e.g. Culicover 2009:69, Needham & Toi-
vonen 2011:9, Sailor & Schütze 2013). These criticisms primarily apply to the strong version. Alt-
hough strong omissibility states that arguments are non-omissible, the more general case seems to
be that PPs are omissible, even PPs that are canonically considered arguments. This applies to
even the most canonical of argument PPs, as with look at, talk to, run to, and think about. I con-
clude, along with prior literature, that the strong version of the omission diagnostic is untenable.
As such, the diagnostic in its strong form cannot be taken as evidence motivating the A/AD.
The standard reaction to the strong omission diagnostic’s failure is to adopt the weak version
instead. However, while the weak version of the omission diagnostic might be consistent with the
existence of the A/AD, it does not provide support for it. To see why, consider how weak omissi-
bility could be implemented. Let’s begin by accepting that omissibility directly identifies the phe-
 AGAINST THE ARGUMENT/ADJUNCT DISTINCTION 5

nomenon of subcategorization; i.e., if an element is non-omissible (for reasons other than idio-
maticity/pragmatics), then it is subcategorized. From here, there are two basic approaches to im-
plement weak omissibility. On one hand, we could identify subcategorization with argumenthood
(i.e., subcategorized ↔ argument), but stipulate the existence of some other feature (or set of fea-
tures) F on top of subcategorization which makes a subcategorized element omissible. In other
words, this first approach states that omissible arguments both (i) are subcategorized and (ii) bear
some additional feature F, while non-omissible arguments are subcategorized elements lacking F.
On the other hand, a second approach could reject the identification of subcategorization with
argumenthood, and instead assume merely that argumenthood is necessary for subcategorization.
There must then be some additional feature (or set of features) F which is both identifiable with
argumenthood (i.e., +F ↔ argument) and necessary for subcategorization, but which is not suffi-
cient for non-omissibility. In other words, this second approach states that omissible arguments are
F-bearing elements that are not subcategorized, while non-omissible arguments are F-bearing ele-
ments that are subcategorized.
In both cases, we must stipulate an additional relationship between subcategorization and ar-
gumenthood which omissibility by itself does not justify. In approach 1, we must stipulate that, in
conjunction with some additional feature F, subcategorized elements become omissible. Note that
if we didn’t stipulate some F in approach 1, we would predict all subcategorized elements (hence
all arguments) to be non-omissible. In order to make omissibility only a one-way reflection of ar-
gumenthood, we need F. Similarly, in approach 2, we must stipulate that subcategorization re-
quires the presence of some additional feature F. Again, some F is required above and beyond the
direct implication of subcategorization.
Therefore, although omissibility may directly identify subcategorization (i.e. non-omissible
→ subcategorized in both approaches), by implementing omissibility as a one-way diagnostic for
argumenthood we are forced to say that subcategorization is related to argumenthood in a way that
omissibility alone is not capable of diagnosing. In other words, we must assume both (i) that there
is a three-way distinction between non-omissible arguments, omissible arguments, and adjuncts,
and (ii) that omissibility itself identifies only non-omissible arguments, leaving the additional nec-
essary distinction between omissible arguments and adjuncts unmotivated.
Therefore, omissibility must be paired with some independent diagnostic identifying a distinc-
tion between omissible arguments and adjuncts. In the absence of that additional evidence, omis-
sibility does not motivate the A/AD. Either way, omissibility does not motivate the A/AD by itself.
It may motivate subcategorization, but the connection from subcategorization to argumenthood is
not motivated by omissibility alone; that connection requires additional evidence. The remainder
of this paper makes the case that existing argumentation in this direction is not compelling.

4 Do so-Substitution
Another common diagnostic for argumenthood involves do so-substitution. This diagnostic holds
that do so can have dependents which correspond to adjuncts in its antecedent, but cannot have
dependents corresponding to arguments in its antecedent. For example, in (11a), since tomorrow
can occur with do so, the do so-substitution diagnostic concludes that the corresponding dependent
in the antecedent clause is an adjunct (i.e. last Tuesday is an adjunct in the VP take a trip last
Tuesday). In (11b), onto the wagon cannot co-occur with do so; therefore, the corresponding de-
pendent in the antecedent clause (i.e. onto the truck) is considered an argument.

(11) a. John [VP took a trip last Tuesday], and I’m going to do so tomorrow.
b. *John [VP loaded a sack onto the truck], and I did so onto the wagon.

In order to meet the no-independent-motivation criterion described in section 2, the do so-

substitution must identify a distinction among PPs which lacks an A/AD-independent explanation.
Since Lakoff & Ross (1976), it has been widely held that the A/AD is indeed needed to explain
which PPs can occur with do so and which cannot. The idea is that the A/AD entails distinct syn-
tactic position-types for arguments vs. adjuncts in the VP; arguments attach to a lower projection
in the VP than adjuncts do. Following Harley (2007, 2014), I will use the label √P for the lower
projection to which arguments are thought to attach, and vP for the higher projection to which ad-
6 ANDREW MCINNERNEY

juncts are thought to attach. If do so is analyzed as a pro-form taking the lower projection as its
antecedent (Hankamer & Sag 1976, Jackendoff 1977), it follows that adjuncts but not arguments
can occur with do so.
This analysis of do so provides support for the A/AD, because it depends crucially on a syntactic
property of the A/AD, namely the difference in attachment height between arguments and adjuncts.
Unfortunately, this analysis must be rejected; alternatives which do not depend upon the A/AD
appear to be superior. First, it is known that the antecedent for do so cannot be a syntactic constit-
uent. For example, Culicover & Jackendoff (2005:125) find that do so can refer to discontinuous
material for which a constituent-source is difficult to identify (without recourse to unmotivated
vacuous movements). This is illustrated in (12). Additionally, Kehler & Ward (1999:13) find that
do so can refer to material split across clauses, such that it is not possible to identify the antecedent
with a single constituent, as shown in (13).

(12) a. Robin slept in the bunkbed for twelve hours, and Leslie did so on the futon.
b. Robin slept for twelve hours in the bunkbed, and Leslie did so for eight hours.
(13) a. What I am suggesting is that when we delay, or when we fail to act, we do so inten-
tionally...
b. Fortunately, the first person to die in 1990 and the first couple to file for divorce in 1990
were allowed to do so anonymously.

These data require a semantic account of do so-anaphora resolution. For more detailed discussion,
see the cited sources of these data (see also Miller 1992, Przepiórkowski 1999, Houser 2010).
One might suppose that, even if do so seeks a semantically-defined antecedent, its potential
antecedents might still be constrained by the A/AD. For example, perhaps a given sub-event e is a
potential antecedent for do so only if e specifies all its arguments. This alternative approach is
contradicted by attested data (from the Corpus of Contemporary American English, Davies 2008)
such as (14).

(14) a. In any case he had decided to, because an eruption of hardcore, coke-and-speed-headed
paranoia could destroy them all. It had done so to many others.
b. It is so much more tragic when we control others through physical harm than by
more passive means, and it is worst when we do so to our children.
c. It is no more unkind or inaccurate to typify teachers as losers than it is to do so to intro-
verts.
d. No country has officially been named a currency manipulator by the US since Bill Clin-
ton’s administration did so to China in 1994.
e. It’s ok to make you both look like idiots, not so much to do so to your spouse alone.

Take (14a). The antecedent for do so must be simply destroy, not destroy them all. Otherwise, the
resulting interpretation of the do so-clause would be “destroy them all to many others,” which is
not what we observe. This shows that the antecedent to do so can exclude semantic arguments,
since, if there is an A/AD, the direct object them all certainly qualifies as an argument of the verb
destroy. Therefore, it is not the case that the semantic predicate serving as do so’s antecedent must
include all its arguments.
What then determines which PPs are acceptable with do so and which are not? By the reason-
ing just described, we should expect that, in (11b), a potential antecedent for do so should be load
a sack, omitting the “argument” PP onto the truck. Why then is (11b) unacceptable? I follow Mil-
ler (1990, 1992) in taking the answer to involve the selectional properties of the verb do. Just like
any other verb, do is compatible with some types of PPs but not others. It can occur with a to-
phrase, as in do something to someone, but not with a goal PP, as in do something into something.
Miller points out that a PP’s compatibility with do so mirrors exactly the PP’s compatibility with
the lexical verb do. This is what explains which PPs can occur with do so; do so is headed by lexi-
cal do, and thus has the selectional properties of lexical do. Thus, any PP which is compatible with
lexical do can occur with do so, and any PP which cannot occur with lexical do cannot occur with
do so.
I conclude that the A/AD-free analysis of do so-substitution is superior to the traditional
 AGAINST THE ARGUMENT/ADJUNCT DISTINCTION 7

A/AD-based analysis. Therefore, the do so-substitution diagnostic does not provide evidence for
the A/AD, as it fails the no-independent-motivation criterion described in section 2; its effects are
expected entirely independently of the A/AD. The do so-substitution diagnostic, as well as the
related diagnostics from do what-pseudoclefting and VP-fronting, are discussed in more detail in
McInnerney (2021a), including data demonstrating critical heuristic problems for these diagnostics.

5 Iteration
This section discusses the diagnostic from iteration, which states that adjuncts of the same type
can be repeatedly iterated, while arguments of the same type cannot (Hornstein & Nunes 2008:59-
60, Schütze 2012:4-5, Zyman 2020:1). Example (15) gives a typical contrast which might be used
to illustrate the diagnostic.

(15) a. Lee sang in the city (at the big theater) (on the main stage).
b. Lee sang to a big crowd (*to all their friends) (*to Terry and Casey).

Iterability says that the locative PPs in (15a) can be iterated because they are adjuncts, while the
PPs in (15b) cannot be iterated because they are arguments.
I argue that this diagnostic does not offer support for the A/AD because the distinction it iden-
tifies does not resemble the conventional A/AD, and because this distinction appears to follow
from properties unrelated to argumenthood. To demonstrate this, observe first that most conven-
tional adjuncts cannot be freely iterated. (16) illustrates non-iterability of several typical adjuncts.

(16) a. *I built it [PP with my tools] [PP with a hammer].

d. *I went to the Halloween party [PP as a pirate] [PP as a zombie].
e. *The house was built [PP by amateurs] [PP by my cousins].

There are some conventional adjuncts, specifically benefactives and comitatives, which can occur
twice in the same verb phrase (Ernst 2001:134). However, benefactives and comitatives are lim-
ited such that no more than two can occur in the same sentence. (17) illustrates with benefactives.

(17) a. ?I baked it [for you] [for your parents].

b. ?*I baked it [for you] [for your parents] [for a friend].

In this respect, benefactives and comitatives differ from true iteration, occurring with locatives and
temporals, where iteration is unbounded.

(18) I left it [in Ohio] [near Columbus] [at the library] [in the back room] [under the desk] …

The acceptability of (17a) and unacceptability of (17b) shows that the term “benefactive” is too
coarse. Instead, there are at least two distinct types of thematic relations. It supports this conclu-
sion that numerous authors have independently argued for distinguishing “benefactives” into two
types. See e.g. Van Valin & La Polla (1997:383-384), Kittilä (2005), and Bosse (2015:121-122)
for a distinction between “plain” benefactives and “recipient” benefactives. Thus, (17a) is ac-
ceptable because it contains two distinct types of benefactives. In contrast, true iteration as in (18)
involves repetition of thematically uniform phrases. A similar argument can be made for sentences
with up to two comitatives.
While most conventional adjuncts, as demonstrated above, cannot be iterated, locatives and
temporals can be iterated, as example (19) demonstrates.

(19) a. Terry laughed [last week] [on Monday] [at noon] …

b. Terry cried [at home] [in their bedroom] [in the closet] …
c. Terry boiled the eggs [on the stove] [in a big pot] [in salty water] …

I therefore conclude, in agreement with Ernst (2001:134ff), that, in comparison to other conven-
tional adjuncts, locatives and temporals display distinct behavior with respect to iterability.
8 ANDREW MCINNERNEY

Among the conventional adjuncts, only locatives and temporals clearly allow unbounded iteration.
Critically, the same generalization applies for PPs which are conventionally considered argu-
ments. Among conventional arguments, locatives and temporals can be iterated, and other types of
PPs cannot be iterated. That most conventional arguments cannot be iterated, as the iteration diag-
nostic predicts, is illustrated in (20). In contrast, that locatives (including both spatial and direc-
tional locatives) can be iterated, even when they would conventionally be considered arguments,
is illustrated in (21).

(20) a. Terry laughed at Lee (*at Lee’s hair).

b. Terry talked to the students (*to Lee’s friends).
c. Terry wrote about the war (*about the treaty).
(21) a. Terry kept the book [in the library] [on the shelf] [next to Aspects] …
b. Terry lives [in France] [near a large city] [in a small cottage] …
c. The ball rolled [under the tree] [next to the trunk] [in front of the hollow] …

The possibility of iteration therefore depends not upon argumenthood, but on the semantics of the
iterated expression. Locatives and temporals permit iteration, and other thematic elements do not.
Ernst’s (2001:135) explanation for iteration of locative and temporal adjuncts can be extended to
iteration of locative and temporal arguments. He suggests that iterability of locatives and tempo-
rals follows from their being conceived of as “spaces” which can be geometrically nested within
each other. Non-iterable PPs, like recipients and benefactors, are not readily conceived of in geo-
metrical terms, and therefore lack the property of nestability.
I conclude that iterability meets neither the heuristic criterion nor the no-independent-
motivation criterion as discussed in section 2, and as such this diagnostic does not provide motiva-
tion for the syntactic A/AD.

6 Further Diagnostics
For reasons of space, I cannot discuss the remaining diagnostics of islandhood, extraction from
weak islands, reconstruction for Condition C, and permutation in detail in this paper. However,
this section briefly discusses each of these remaining diagnostics. The goal is to demonstrate plau-
sibility for alternative analyses which do not depend upon the A/AD.
First, consider islandhood. It is widely believed that islandhood of the bracketed constituents
in (22) derives from their status as adjuncts. However, notice that the island properties of these
constituents remains constant when they are recast as conventional arguments. The via-phrase in
(22a) would be considered an argument in (23a), and the after-phrase in (22b) would be consid-
ered an argument in (23b).

(22) a. *This is the path that we proceeded [via __ ].

b. *Who did Terry cry [until Lee hit __ ]?
(23) a. *This is the path that we walked [via __ ].
b. *Who did the party last [until John hit __ ]?

Given these and further data, McInnerney (2021b) concludes that the A/AD is not a plausible fac-
tor leading to islandhood in (22). Instead, the same, internal-structural properties are more likely to
be responsible for islandhood in both (22) and (23).
Extraction from weak islands has sometimes been characterized in terms of the A/AD, such
that arguments but not adjuncts can escape a weak island. However, much subsequent work has
reframed the data, such that the A/AD is no longer considered an accurate way of characterizing
patterns of extractability. Instead, syntactic category (NP vs. PP, Cinque 1990) or referentiality
(Kroch 1998) may be relevant. See Szabolcsi (2006) and Miliorini (2019) for overviews of ap-
proaches to weak islands which reject a role for the A/AD.
Condition C reconstruction is also sometimes analyzed in terms of the A/AD (Lebeaux 1988),
but the judgments on which the generalization is based have never been considered robust (cf.
Reinhart 1986, Speas 1990), and experimental work (Adger et al. 2017, Bruening & Al Khalaf
2019) has discovered that the data are better characterized in terms of a categorial distinction: de-
 AGAINST THE ARGUMENT/ADJUNCT DISTINCTION 9

pendents of N do not reconstruct for Condition C, while dependents of P and A do.

Finally, previous authors have questioned the role of the A/AD in controlling permutation of
PPs in the VP. Hawkins (2000) and Wasow (2002) report a great deal of flexibility among post-
verbal constituents, with prosodic weight playing the most pivotal role. While slight weight-
independent ordering preferences among PPs can be detected (Wiechmann & Lohmann 2013,
Neeleman & Payne 2020), these apply to argument and adjunct PPs alike (Schweikert 2005,
Cinque 2006), meaning permutability arguably does not reflect the A/AD.

7 Conclusion
The A/AD is a hypothesis which must be justified by evidence. This paper considered several pur-
ported diagnostics for argumenthood, specifically omission, do so-substitution, and iteration, and
asked whether those diagnostics can provide support for the A/AD. I concluded that omission pro-
vides evidence for a mechanism of subcategorization, but that this can only be connected to the
A/AD in conjunction with independent evidence for a distinction between omissible arguments
and adjuncts. Do so-substitution and iteration do not provide such evidence; both these diagnostics
are better analyzed on grounds entirely independent of the A/AD.
For reasons of space, I did not discuss further potential diagnostics in detail. However, section
6 suggested that each additional diagnostic may be better analyzed along independent lines, not
relying upon the A/AD. If that’s the case, it would mean that much of the traditional evidence
thought to motivate the A/AD does not in fact do so. Instead, other sources of evidence would be
needed to support the distinction. I conclude it is plausible, depending on whether relevant evi-
dence can be provided, that the A/AD might be eliminable. Though elimination of the A/AD
would raise many analytical questions (since the A/AD plays a role in the analysis of a variety of
syntactic phenomena), it would be desirable from a Minimalist point of view, as it would reduce
the number of principles and operations postulated to comprise the computational system of hu-
man language.

References
Adger, David, Alex Drummond, David Hall, and Coppe van Urk. (2017). Is there Condition C reconstruction?
In Proceedings of the 47th Annual Meeting of NELS, eds. A. Lamont & K. Tetzloff, Amherst: GLSA,
21-30.
Boeckx, Cedric. 2003. Review to ‘The Syntax of Adjuncts.’ Studies in the Linguistic Sciences 32: 95-101.
Bosse, Solveig. 2015. Applicative Arguments: A Syntactic and Semantic Investigation of German and English.
New York: Peter Lang Publishing.
Bruening, Benjamin, and Eman Al Khalaf. 2019. No argument-adjunct asymmetry in reconstruction for
Binding Condition C. Journal of Linguistics 55: 247-276.
Cinque, Guglielmo. 1990. Types of Ā-dependencies. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 2006. Complement and adjunct PPs: Implications for clause structure. In Restructuring
and Functional Heads, ed. G. Cinque, 145–166. Oxford: Oxford University Press.
Culicover, Peter W. 2009. Natural Language Syntax. Oxford: Oxford University Press.
Culicover, Peter W., and Ray Jackendoff. (2005). Simpler Syntax. Oxford: Oxford University Press.
Davies, Mark. 2008-. The Corpus of Contemporary American English (COCA). Available online at
https://www.english-corpora.org/coca/.
Dowty, David. 2000. The Dual Analysis of Adjuncts/Complements in Categorial Grammar. ZAS Papers in
Linguistics 17:53 - 78
Ernst, Thomas. 1984. Towards an Integrated Theory of Adverb Position in English. Bloomington, Indiana:
Indiana University Linguistics Club.
Ernst, Thomas. 2001. The Syntax of Adjuncts. Cambridge: Cambridge University Press.
Goldberg, Adele E., and Farrell Ackerman. 2001. The pragmatics of obligatory adjuncts. Language 77:798-
814.
Hankamer, Jorge, and Ivan Sag. 1976. Linguistic Inquiry 7:391-428.
Harley, Heidi. 2007. One-replacement, unaccusativity, acategorial roots and Bare Phrase Structure. Ms., Uni-
versity of Arizona.
Harley, Heidi. 2014. On the identity of roots. Theoretical Linguistics 40:225-276.
10 ANDREW MCINNERNEY

Harris, Zellig S. 1970. Papers in Structural and Transformational Linguistics. Dordrecht: Springer Sci-
ence+Business Media
Hornstein, Norbert, and Jairo Nunes. (2008). Adjunction and labeling in Bare Phrase Structure. Biolinguistics
2:57-86.
Houser, Michael. 2010. The syntax and semantics of do so anaphora. Doctoral dissertation, UC Berkeley.
Huddleston, Rodney. & Geoffrey K. Pullum. (2002). The Cambridge Grammar of English. Cambridge:
Cambridge University Press.
Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.
Jackendoff, Ray S. 1977. X-bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press.
Kehler, Andrew and Gregory Ward. 1999. On the semantics and pragmatics of ‘Identifier So’. In The Seman-
tics/Pragmatics Interface from Different Points of View, ed. K, Turner, 233–256. Amsterdam: Elsevier.
Kittilä, Seppo. 2005. Recipient-prominence vs. beneficiary-prominence. Linguistic Typology 9:269–297.
Kroch, Anthony. 1998. Amount quantification, referentiality, and long wh-movement. University of Pennsyl-
vania Working Papers in Linguistics 5:21-36.
Koenig, Jean-Pierre, Gail Mauner, and Breton Bienvenue. 2003. Arguments for adjuncts. Cognition, 89:67-
103.
Lakoff, George & John R. Ross. 1976. Why you can’t do so into the sink. In Syntax and Semantics, Volume 7:
Notes from the Linguistic Underground, ed. J. D. McCawley, 101-111. New York: Academic Press.
Lebeaux, David. 1988. Language acquisition and the form of the grammar. Doctoral dissertation, University
of Massachusetts, Amherst.
McConnell-Ginet, Sally. 1994. On the non-optionality of certain modifiers. In Proceedings of the 4th Seman-
tics and Linguistic Theory Conference, ed. M. Harvey & L. Santelmann, Linguistic Society of America.
230-250.
McInnerney, Andrew. 2021a. Separability of dependents from VP in English: beyond the argument/adjunct
distinction. Ms., University of Michigan.
McInnerney, Andrew. 2021b. The argument/adjunct distinction does not condition islandhood of PPs in Eng-
lish. Ms., University of Michigan.
Miliorini, Rafaela. 2019. Extraction from weak islands: alternatives to the argument/adjunct distinction. Re-
vista Virtual de Estudos da Linguagem, edição especial 17:37-58.
Miller, Philip. 1990. Pseudogapping and do so substitution. In Papers from the 26th Regional Meeting of the
Chicago Linguistic Society, 293-305.
Miller, Philip. 1992. Clitics and Constituents in Phrase Structure Grammar. New York: Garland.
Needham, Stephanie and Ida Toivonen. (2011). Derived arguments. In Proceedings of LFG11, ed. M. Butt &
T. H. King, CSLI Publications, 401-421.
Neeleman, Ad & Amanda Payne. 2020. PP-Extraposition and the Order of Adverbials in English. Linguistic
Inquiry 51:471–520
Patejuk, Agnieszka, and Adam Przepiórkowski. (2021). Category mismatches in coordination vindicated. Ms.,
University of Oxford.
Pesetsky, David. 1995. Zero Syntax: Experiences and cascades. Cambridge, MA: MIT Press.
Przepiórkowski, Adam. 1999. Case Assignment and the Complement/Adjunct Dichotomy: A Non-
Configurational Constraint-Based Approach. Doctoral dissertation, Universität Tübingen.
Reinhart, Tanya. 1986. The Syntactic Domain of Anaphora. Doctoral dissertation, MIT.
Sailor, Craig and Carson T. Schütze. 2013. What is an adjunct? Ms., UCLA.
Schütze, Carson T. 1995. PP attachment and argumenthood. MIT Working Papers in Linguistics 26:95-151.
Schütze, Carson T. 2012. Implementing argumenthood diagnostics for prepositional phrases: comments on
Merlo and Esteve Ferrer. Ms., UCLA.
Schweikert, Walter. 2005. The Order of Prepositional Phrases in the Structure of the Clause. Amsterdam:
John Benjamins.
Speas, Margaret. 1990. Phrase Structure in Natural Language. Dordrecht: Kluwer Academic Publishers.
Szabolcsi, Ana. 2006. Strong vs. Weak Islands. In The Blackwell Companion to Syntax, Vol. 4, ed. M. Evera-
ert and H. van Riemsdijk, 479-531. Oxford: Blackwell.
Van Valin, Robert D. and Randy J. LaPolla. 1997. Syntax: Structure, meaning, and function. Cambridge:
Cambridge University Press.
Wiechmann, Daniel and Arne Lohmann. 2013. Domain minimization and beyond: Modeling prepositional
phrase ordering. Language Variation and Change 25:65–88.
Zyman, Erik. 2020. In situ mixed wh-coordination and the argument/adjunct distinction. Glossa: A Journal of
General Linguistics 5:30.

Department of Linguistics
University of Michigan
Ann Arbor, MI 48109-1220
[email protected]