Disentangling two distinct notions of NEG raising

Abstract

In this paper we consider two analyses of NEG raising phenomena: a syntactic approach based on raising NEG, as recently advocated in Collins & Postal 2014, and a semantic/pragmatic approach based on the Excluded Middle Assumption; see Bartsch 1973. We show that neither approach alone is sufficient to account for all the relevant phenomena. Although the syntactic approach is needed to explain the distribution of strict NPIs and Horn clauses, the semantic/pragmatic approach is needed to explain certain inferences where syntactic NEG raising is blocked.

Full text

Semantics & Pragmatics Volume 11, Article 5,2018
https://doi.org/10.3765/sp.11.5
This is an early access version of
Collins, Chris & Paul M. Postal. 2018. Disentangling two distinct notions of
NEG raising. Semantics and Pragmatics 11(5). https://doi.org/10.3765/sp.
11.5.
This version will be replaced with the final typeset version in due course.
Note that page numbers will change, so cite with caution.
©2018 Chris Collins and Paul M. Postal
This is an open-access article distributed under the terms of a Creative Commons Attribution
License (https://creativecommons.org/licenses/by/3.0/).

early access
Disentangling two distinct notions of NEG raising
Chris Collins
New York University
Paul M. Postal
New York University
Abstract
In this paper we consider two analyses of NEG raising phenomena:
a syntactic approach based on raising NEG, as recently advocated in Collins
& Postal 2014, and a semantic/pragmatic approach based on the Excluded
Middle Assumption; see Bartsch 1973. We show that neither approach alone is
sufficient to account for all the relevant phenomena. Although the syntactic
approach is needed to explain the distribution of strict NPIs and Horn clauses,
the semantic/pragmatic approach is needed to explain certain inferences
where syntactic NEG raising is blocked.
Keywords:
interclausal NEG raising, excluded middle assumption, strict NPIs, Horn
clauses
1Introduction
Various linguists and philosophers long ago noticed a distinctive property
of negative constructions in various languages involving a relatively small
subset of main predicates taking complement clauses. An English instance of
this phenomenon is seen in (1):
(1) a. I don’t think this course is interesting.
b. I think this course is not interesting.
While (1a) has, given the presence of main clause negation (the syntactic
element we call NEG), an expected reading which simply denies that I have
a particular thought, the relevant characteristic is that it appears to have
another reading equivalent to that of (1b). On that reading, it is stated that I do
have a definite thought, namely, that represented by the negated complement
clause in (1b). The restriction of this ‘extra’ reading to a limited class of main
clause predicates is illustrated by the fact that no ‘extra’ reading is associated
with the cases in (2a), none of which shares a meaning with the corresponding
examples in (2b):

early access
Chris Collins, Paul M. Postal
(2) a.
I don’t overlook/pretend/reaffirm/swear that that this course is
interesting.
b.
I overlook/pretend/reaffirm/swear that that this course is not inter-
esting.
We will call the facts involving the relations between main clause negation and
the semantics of embedded complement clauses illustrated in (1) and (2)NEG
scope fixing in order to give them a label that does not prejudge the analysis.
These phenomena include the fact that (1a) appears to be ambiguous, that on
one interpretation (1a) is equivalent to (1b), and that none of (2a) is ambiguous
in the same way as (1a).
Background history of the recognition of NEG scope fixing drawn from
several languages and many relevant comments are found in Horn 1978:129-
131,Horn 1989:308-312,Horn 2014a. Hereafter, we refer to readings like that
of (1a), in which no thought about the complement proposition is attributed
to the matrix subject, as the weak reading and to readings like that of (1a)
equivalent to (1b) as the strong reading. We freely extend these terms to
verbs/predicates distinct from think like believe,suppose, etc.
A fundamental question about NEG scope fixing is, evidently, how can
cases like (1a), with only a main clause instance of NEG, have a reading
involving a complement clause understanding of negation, that is, a strong
reading.
Roughly two main descriptive and theoretical approaches to NEG scope
fixing have been advanced. First, there is a syntactic approach, formally
initiated in Fillmore 1963 and extensively defended in Collins & Postal 2014;
hereafter: CP (2014). Under this conception, (1a), on the reading taken as
equivalent to that of (1b), has been analyzed in terms of syntactic raising of a
NEG from the embedded clause. This schematic specification by no means
defines a unique syntactic approach. For instance, Collins & Postal 2017a
presents and justifies a syntactic view which, while also based on syntactic
NEG raising, differs significantly from that in CP (2014). Since the analysis in
CP (2014) is more or less equivalent to traditional syntactic analyses and is
more widely known, for convenience’s sake, we adopt it in this article. But
our conclusions would also apply to any account involving syntactic NEG
raising, such as that of Collins & Postal 2017a. Even more than that in CP
(2014), the analysis in Collins & Postal 2017aappeals heavily to NEG deletion;
see the cited works for details.
2

early access
Disentangling
Second, there is a semantic/pragmatic approach, proposed most influ-
entially in Bartsch 1973, whereby (1a) is taken to logically entail (1b) under a
particular assumption, the Excluded Middle Assumption (hereafter EMA), an-
alyzed by Bartsch as a ‘pragmatic presupposition’. We elaborate her approach
briefly in Section 2. Other related approaches include those of Gajewski 2005
and Romoli 2013.
One might be tempted to assume a priori that there is a unique correct
conception of NEG scope fixing and that at least one of the theoretically
distinct views represented by syntactic NEG raising and appealling to the
EMA must simply be wrong. It could be claimed that theoretical simplicity
concerns support this exclusionist position. The title of Bartsch’s extremely
influential article Negative transportation gibt es nicht, that is, ‘NEG raising
doesn’t exist’ (see Section 2) clearly suggests an exclusionist position. And
no doubt other researchers, of both the syntactic and semantic/pragmatic
schools, have also assumed exclusionist positions. There has essentially been
debate over which unitary view is correct.
The position of CP (2014) on this matter was noncommittal. While that
work contains no statement to the effect that EM inferences play no role
in NEG scope fixing, there was also no explicit statement that they do. And
the overwhelming emphasis in the monograph on the need for a syntactic
treatment of a range of NEG scope fixing cases might have given some the
impression that we regarded appeal to the EMA as unnecessary everywhere.
Although an exclusive, unitary view of NEG scope fixing is, we believe,
widely shared (see Crowley 2016 for one recent non-exclusionist proposal),
we are aware of no attempt from either conceptual point of view to argue
for the correctness of an exclusionary view. Such question-begging is the-
oretically critical since it facilitates incorrect overgeneralization. That is, it
leads mistakenly from a supported conclusion to the effect that case X of
NEG scope fixing requires an analysis of one type to the conclusion that every
case does, which is a non-sequitur. But what is needed at this stage of inquiry
is to debate the appropriateness of one or the other types of mechanism for
particular subsets of NEG scope fixing data. This was done in CP (2014) for
specific subsets, e.g. Horn clauses, strict NPIs, islands, parentheticals, etc.
Only at some future point when every known instance of the phenomenon
had been analyzed could a truly universal exclusionist position be justified.
In the present paper, we argue that even the currently known facts related
to NEG scope fixing preclude any exclusionist view. That is, for each approach
there is good evidence that there are some instances of NEG scope fixing that
3

early access
Chris Collins, Paul M. Postal
it cannot properly characterize. Specifically, even if one accepts the CP (2014)
arguments that a syntactic account of NEG scope fixing is required for certain
phenomena, one must still grant that a unitary exclusionist account of all the
facts is not viable.
The remainder of the present paper is organized as follows.
First, Section 2reviews Bartsch’s (1973)exclusionist semantic/pragmatic
view of NEG scope fixing. Section 3sketches a limited portion of the argu-
ments in CP (2014) that, contra Bartsch 1973 and other exclusionist non-
syntactic views of NEG scope fixing, appeal to syntactic NEG raising is needed
in English grammar. We concentrate on the CP (2014) discussion of what we
call Horn clauses. Section 4cites three cases where EM inferences are needed,
but where evidence shows that syntactic NEG raising is not possible. Section
5returns to the issues of the equivalence of (1a) and (1b). Section 6is the
conclusion.
2Bartsch’s proposal
To explain the distinction which we claim exists between two different types
of phenomena within the overall NEG scope fixing domain, it is useful to
briefly discuss Bartsch’s (1973)proposal. Her position, contrary to that in-
troduced in Fillmore 1963, was that the strong reading of (1) had nothing to
do with syntax, that there was no syntactic NEG raising and that the NEG
appearing in the main clause in (1a) was in no sense a syntactic constituent
of the complement clause.
Bartsch’s novel, alternative and non-syntactic account of the strong read-
ing was that simply by associating predicates like think, but not those like
any in (2), with a particular pragmatic presupposition, the relation between
(1a, b) could simply be reduced to a question of logic. The idea was that for
any individuals denoted by the subject x of the main predicate P and for any
proposition p, x stands in the relation P to p or x stands in the relation P
to not(p). Thus for (1a), the pragmatic presupposition would be that I either
think the course is interesting or that I think the course is not interesting.
This excludes the possibility that the subject denotation has no opinion, has
not thought about the matter or has never heard of the course, etc. That is, it
excludes the weak reading of (1a), that not shared with (1b).
Bartsch characterizes the pragmatic presupposition as follows. We give
both the original German and the English translation (thanks to Christina
Behme for the latter):
4

early access
Disentangling
Aber erst in neuerer Zeit wurden pragmatische Faktoren in die Bestim-
mung der Bedeutungen von Sätzen einbezogen, so dass auf diese Weise formal
dargestellt warden kann, wie Aussagen, die satz-und wort-semantisch nicht
äquiuvalent sind, doch zusammen mit Sprechsituationsbestimmungen dieselbe
modell-theoretische Interpretation haben können und damit diselbe Informa-
tion übermitteln können. In dem hier zu betrachtenden Fall geht es jedoch
nicht um die Einbeziehung der Sprechsituationsbestimmungen hinsichtlich der
indexkalischen Ausdrücke, d.h. der in ihrer Denotaton von der Sprachsituation
abhängigen Ausdrücke sondern um generelle Anwendungsbedingungen für
bestimmte Arten von Ausdrücken die darin liegen dass für die ‘normale’ Ver-
wendung diese Ausdrücke bestimmte pragmatische Präsuppositionen gemacht
werden. Ich gebrauche hier ‘pragmatische Präsupposition’ im Sinne von R.
THOMASON(1973), der eine Präsupposition für die Verwendung eines Aus-
drucks ‘pragmatisch’ im Gegensatz zu ‘semantisch’ nennt, wenn normaler-
weise diese Präsupposition erfüllt sein muss, wenn der Ausdruck [in nicht
zitierter Weise] verwendet wird, ween es aber doch Ausnahmessituationen gibt,
in denen der Ausdruck auch ohne diese Präsupposition [in nicht zitierter Weise]
verwendet werden kann. Dagegen muss eine semantische Präsupposition stets
erfüllt sein. Bartsch 1973
‘But only recently have pragmatic factors been included in the determina-
tion of the meaning of sentences, so that in this way it can be formally shown
how propositions which are not sentence-semantically and word-semantically
equivalent still have the same model theoretic interpretation and transfer
the same information when the context is also taken into consideration. In
the case under consideration here, however, we are not interested in the
inclusion of context regarding indexical expressions, i.e. in the denotation of
expressions that depend on the speech situation but in the general conditions
for the use of certain kinds of expressions which are based on certain prag-
matic presuppositions for the ‘normal’ use of these expressions. I use here
‘pragmatic presuppositions’ in the way suggested by Thomason [1973] who
calls a presupposition for the use of an expression pragmatic as opposed to
semantic when under normal conditions the presupposition must be fulfilled
when the expression is used [not in cited form]; but when still exceptional
conditions exist under which the expression can be also used [not in cited
form] without those presuppositions. But a semantic presupposition must
always be satisfied.’
Two things to note about this characterization are that the pragmatic
presupposition can be cancelled (just as a conversational implicature can
5

early access
Chris Collins, Paul M. Postal
be cancelled) and that the pragmatic presupposition is associated with a
particular expression (e.g., a verb like think).
We call such assumptions the Excluded Middle Assumption. We need
not take a stand on whether the EMA is a conversational implicature or a
pragmatic presupposition or something else; for various views see Gajewski
2005,Romoli 2013 and Horn & Bayer 1984. Furthermore, we take no stance on
whether the EMA is part of the definition of specific lexical items (e.g., think)
or is just a general (non-lexically specific) assumption that can be invoked
in certain contexts with certain utterances to give rise to certain inferences.
Under the lexical theory, one needs an account of which lexical items are
associated with the EMA. Under the non-lexical theory, one needs an account
of when the EMA is invoked to support EM inferences. Either way, additional
assumptions are needed. Given the presence of EMA, the apparent equivalence
of one reading of (1a) to (1b) follows via logic (disjunctive syllogism) from
the fact that the main clause NEG, interpreted exclusively in the main clause,
eliminates the possibility that the first disjunct of EMA holds. The logic is
symbolized as follows:
(3) a. F(x,p) ∨F(x,¬p) Excluded Middle Assumption
b. ¬F(x,p) Matrix Clause Negation
c. F(x,¬p)
Logical Consequence of (3a,b) by
Disjunctive Syllogism
Given that examples like (1a) permit the weak reading not equivalent to
(1b), the existence of EM inferences must be considered a mere option, as
Bartsch (1973)recognized. Otherwise, (1a) would, wrongly, be claimed to have
only the strong reading equivalent to that of (1b) every time it is used.
However, the logical considerations Bartsch invoked by no means yield a
determinative conclusion about NEG scope fixing. As argued extensively in
CP (2014) many kinds of data require a syntactic account.
3Collins and Postal’s (2014) syntactic evidence
3.1Remark
While the literature claiming to support the syntactic character of NEG scope
fixing has been dominated by appeal to strict NPIs, CP (2014) documents
a broader range of syntactic evidence. Three other types of evidence were
6

early access
Disentangling
presented. These involved what were there called Horn clauses, island phe-
nomena and parenthetical constructions.
Since analysis of each of these bodies of evidence involves considerable
complication and need for detailed background discussion, the reader seeking
a solid grasp of the syntactic support at issue should consult that work.
Rather than provide a sketchy account of all four types of syntactic evidence,
we will consider here only Horn clauses, permitting greater analytic depth
than if we attempted to cover the other types as well. The purpose of this
discussion is to provide a sense of the strength of the evidence for a syntactic
treatment of some instances of NEG scope fixing.
3.2Horn clauses
Horn (1975:283) cited example (4) he had heard during a broadcast:
(4)
I don’t think that ever before have the media played such a major role
in a kidnapping.
Recognizing cases like (4) amounted to the discovery of an English sen-
tence type involving a unique sort of complement clause we have called a
Horn clause. The specific instance in (4) is highlighted. Horn clauses can be
initially characterized as that clause complements containing an extracted
phrase, e.g. ever before in (4), in the initial position of a clause manifest-
ing subject-auxiliary inversion. Moreover, critically, the extracted phrase is
systematically based on a negative polarity item; in (4) this is ever before.
Horn (1975:284) took Horn clauses to support a syntactic view of NEG
scope fixing. But Horn (1989:315) shows that his later work was part of the
general trend that rejected a syntactic view. However, to our knowledge, as
the assumption that NEG scope fixing is uniformly a non-syntactic matter
took hold, neither Horn nor any other advocate of a non-syntactic view gave,
or even sought to give, an account of Horn clauses consistent with their
properties.
Other instances of Horn clauses are seen in (5) and (6), where the negative
polarity item in each is highlighted:
(5)McCawley (1998:598)
a. I don’t suppose that under any circumstances would he help me.
b. We didn’t anticipate that any student would our decision confuse.
7

early access
Chris Collins, Paul M. Postal
(6) a. I don’t believe that at any time did traffic come to standstill.
(theragblog.blogspot.com/. . . /police-state-amerikkka-right-in-my-own. . . )
b. I didn’t expect that for any reason would she agree to that.
Horn clauses can only be complement clauses; main clause versions of Horn
clauses minus their initial complementizer are totally impossible:
(7) a. *Either of them would she be anxious to marry.
b. *Ever again would I agree to such a course of action.
c. *Carvings of any respected deity had he destroyed.
Horn (1975:283,1978:169) claimed that Horn clauses not only must be
complements but, furthermore, they must be complements of just those main
clause predicates manifesting the existence of the strong reading for main
clause negation, like those in (1a, b). This view was endorsed in McCawley
(1998:598). These elements were called Classical NEG Raising Predicates
(CNRPs) in CP (2014), a term we retain here. However, like NEG scope fixing,
the term CNRP is intended to be shorn of any question-begging implication
as to the nature of the NEG scope fixing phenomenon.
Other than in Horn clauses, the combination of an extracted constituent
with subject auxiliary inversion in English embedded clauses is also found in
uncontroversial embedded instances of the Negative Inversion construction
illustrated in (8).
(8) a. I believed that at no time had he lived in Portugal.
b. Sandra felt that none of the proposals could she really support.
c.
Otto guessed that none of the candidates would they actually inter-
view.
d.
Victoria thought that in no sense was Pavil an ideal choice for mayor.
The superficial difference between clauses like those in (8) and Horn clauses
is that the extracted phrases in (8) are overtly negative phrases.
However, the above remarks ignore what CP(2014: chapter 14) called
quasi-Horn clauses. These are illustrated by the highlighted clause in cases
like (9a):
8

early access
Disentangling
(9) a.
The States couldn’t provide us with evidence that at any time had
there been a request for road closure.
b. *
The States couldn’t provide us with evidence that she left until Friday.
CP(2014: Chapter 14.5) took the key contrast between true Horn clauses
and cases like (9a) to lie in the fact that the scope of the fronted quantifier
phrase is systematically in the complement clause in true Horn clauses but
in the main clause in examples like (9b). Moreover, although not noted in CP
(2014), as (9b) illustrates, quasi-Horn clause environments contrast with Horn
clause environments in the distribution of certain strict NPIs, here until.
The remark above (8) also ignores the discovery of a class of apparent
Horn clauses occurring as complements of a class of expressions distinct
from CNRPs. Such examples, illustrated in (10), are discussed in Horn (2014b):
(10) I can’t say that at any time did he show signs of great agitation.
The challenge Horn (2014b) claims cases like (10) provide to the argument
from Horn clauses in CP (2014) is responded to in Collins & Postal (in press).
An obvious question is whether the extraction type found in Horn clauses
is an instance of Negative Inversion or represents a distinct English con-
struction. A priori, simplicity considerations favor the first option if it can
be maintained. And CP(2014: Chapters 13 and 14) argued at length that the
extracted phrase defining Horn clauses is an instance of Negative Inver-
sion. The argument stands on a number of instances of the following form.
There are various constraints Q which block Negative Inversion in certain
environments and Q manifest in Horn clauses; see CP(2014:142-144). Such
arguments combined with the simplicity advantage of reducing Horn clauses
to an otherwise existing construction show unmistakably that Horn clauses
are a subtype of Negative Inversion clause. That conclusion is fundamental
to using Horn clauses to argue for the syntactic nature of one subdomain of
NEG scope fixing.
To proceed with the argument, it is necessary to ask how Horn clauses
can be instances of Negative Inversion. That is initially puzzling because,
under standard views, the fronted NPI phrases in Horn clauses are indefinites
or existentials, and such phrases in general preclude Negative Inversion, even
in complement clauses with main verbs allowing Horn clauses:
9

early access
Chris Collins, Paul M. Postal
(11) a. * I don’t believe that some gorilla did they train.
b. *We didn’t expect that a certain letter would he copy.
c. *Jane didn’t imagine that an internship could she find.
Given the general impossibility of Negative Inversion applying to existen-
tial/indefinite phrases, how can Horn clauses both be instances of Negative
Inversion and yet involve extracted phrases like ever before in (4), any cir-
cumstances,any student,any time and any reason in (5) and (6), forms almost
universally taken to be existential/indefinites?
CP (2014)’s answer is revealed in the sketched analysis of (4) in (12). In the
following, as in CP (2014) and other later works of ours, the notation <NEG>
denotes an unpronounced occurrence of NEG.
(12) a. Underlying Representation of (4):
I do think that the media have [NEG
1
ever before] played such a
major role.
b. The result of Negative Inversion applied to (12a):
I do think that [NEG
1
ever before] have the media played such a
major role.
c. The result of NEG raising applied to (12b):
I do NEG1think that [<NEG
1
> ever before] have the media played
such a major role.
d.
Surface Structure Resulting from the Morphophonemic Interpreta-
tion of (12c):
I don’t think that ever before have the media played such a major
role.
The key in (12a) is that the fronted phrase in the Horn clause is taken, under
the sharply nonstandard assumptions about NPIs argued for in CP (2014) and
Collins & Postal (2017a,b), Collins, Postal & Yevudey (2017), Collins & Postal (in
press), to contain an instance of NEG. (12a) is also the underlying structure
for the following example of Negative Inversion:
(13) I think that never before have the media played such a major role.
In (13), NEG does not undergo raising from the embedded clause to the main
clause. As a consequence, [NEG ever before] is spelled out as never before.
The most obvious consequences of the view of Horn clauses represented
by (12) are, first, that the view eliminates the mystery of how Horn clauses
can be special cases of Negative Inversion. That is, very strict conditions,
10

early access
Disentangling
partially illustrated in (11), govern the choice of phrases frontable under
Negative Inversion; see CP (2014: Chapters 13 and 14). But the presence of the
posited NEG as part of the fronted NPI phrase in (12a) guarantees satisfaction
of those conditions. Second, since Horn clauses are, under the proposed
analysis, instances of Negative Inversion, it follows without special statement
that constraints Q mentioned above hold for Horn clauses exactly as for
corresponding uncontroversial Negative Inversion clauses.
Evidently, the analysis of Horn clauses as Negative Inversion clauses
in (12a) creates a new potential mystery. Namely, why is the posited NEG
not overtly present in the extracted phrase in the Horn clause in (4)? The
obvious answer, basically given in Horn (1975)and elaborated at length
in CP (2014: Chapters 13 and 14), is that the relevant NEG has raised into
the main clause, whose defining predicate is a CNRP. That idea accounts
for the generalization Horn offered that true Horn clauses (i.e non-quasi-
Horn clauses) can only occur as complements of CNRPs. In effect then, the
proper analysis of Horn clauses as based on phrases fronted under Negative
Inversion is only feasible in combination with syntactic NEG raising, needed
to account for the lack of overt presence of the posited NEG in cases like
(4)/(12a). The fact that such an analysis simultaneously accounts for the fact
that Horn clause structures manifest the strong readings obviously further
supports the assumptions made.
The proposals just gone over also determine (i) that Horn clauses can
only be complement clauses (because occurrence as main clauses would not
allow the existence of an environment accounting for the covert status of the
unpronounced NEG) and (ii) that Horn clauses can occur only subordinate to
negative main clauses (because NEG has raised from the embedded clause).
The fundamental assumption of CP (2014), that syntactic NEG raising has no
semantic consequences, accounts for the fact that (4) is identical in meaning
to I think that never before have the media played such a major role in a
kidnapping.
Summarizing, the logic of our account of Horn clauses only makes sense
under a view which recognizes that some instances of NEG scope fixing
involve syntactic NEG raising. Only that permits an account of why the NEG
which relates the apparently non-negative fronted form in cases like (4) to the
negative form in (13) appears to occur in the main clause. Even granting that
an approach like that due to Bartsch (1973)correctly captures the semantic
facts associated with simple NEG scope fixing examples like (1a), it has never
been shown that it can explicate key properties of Horn clause examples.
11

early access
Chris Collins, Paul M. Postal
4Strong readings where syntactic NEG raising cannot exist
4.1Remarks
Compare Bartsch’s (1973)proposal to the conclusion of CP (2014). On the
basis of her ability to reduce relations like those in (1) to logical inference
based on an appeal to the EMA, Bartsch denied that NEG scope fixing facts
represented a syntactic phenomenon. In contrast, given evidence including
the Horn clause observations in Section 3, CP (2014) concluded that NEG
scope fixing had to be viewed as a syntactic phenomenon. Apparently, one
view had to be right and the other wrong.
Since the distinct approaches each appear to have some motivation, one
might fear the existence of a paradox. But such a theoretically unhappy
conclusion can be avoided. Arguments motivating Bartsch’s approach involve
one class of cases, while those supporting a syntactic view involve a distinct
class of cases. Logically, there is no contradiction in claiming both approaches
could be correct for at least partially different NEG scope fixing subdomains.
This raises a possibility apparently not previously entertained by either
side of the disagreement over the syntactic vs. semantic/pragmatic character
of NEG scope fixing in general. Namely, there are two distinct phenomena at
issue, one syntactic, the other semantic/pragmatic. We now argue directly
for this view.
4.2of the opinion
The material claimed in CP (2014) to support the need for syntactic NEG
raising in a range of cases all involved clauses based on main clause CNRPs,
e.g. think. Further, our claim that Bartsch’s approach fails to account for the
evidence about Horn clauses sketched in Section 3also references CNRPs,
the very predicates for which Bartsch’s approach seems adequate in simple
cases like (1a).
CP (2014) argued that over a range of data of different types, the properties
of clauses based on CNRPs, predicates generally yielding strong readings,
require appeal to syntactic NEG raising. But we failed to raise two other
related questions. First, are there predicates supporting the existence of
syntactic NEG raising which do not permit the formation of strong readings?
Second, are there predicates which permit the formation of strong readings
but which can be argued (by the same evidence types in CP (2014)), not to
permit syntactic NEG raising?
12

early access
Disentangling
To answer the first question, one would need a predicate which does
not participate in equivalences such as (1a,b) above, but for which there is
evidence of syntactic NEG raising. Collins & Postal in press argue that the
CU-predicates discovered by Horn 2014bare such predicates. We leave aside
such cases here.
Data bearing on the second question was in fact touched on four decades
ago:
(14)Horn (1978:212)
??It is not my opinion that he will recover until he prays to St. Anselm.
Horn remarked:
(15)
“I find strict NPIs all but impossible embedded under such negated
nominals. . . ”
We agree and would affix a ‘*’ to (14) and to parallel strict NPI examples:
(16) a. * It is not my opinion that he arrived until Friday.
b. *It is not my opinion that he has visited Lourdes in years.
Equally parallel facts are found for the alternate forms:
(17) a. * I am not of the opinion that he arrived until Friday.
b. *I am not of the opinion that he has visited Lourdes in years.
And of course, the possibility of strict NPIs like these in complement clauses
not containing local licensers was one of the arguments in CP (2014) for the
existence of syntactic NEG raising.
But correlated with the facts in (14)-(17), it turns out that expressions like
‘It is not my opinion’/’I am not of the opinion’ permit strong readings. That
is, on one reading, (18a) does justify the inference to (18b) while (19a) justifies
the inference to (19b):
(18) a. It is not my opinion that Mars can be colonized.
b. It is my opinion that Mars cannot be colonized.
13

early access
Chris Collins, Paul M. Postal
(19) a. I am not of the opinion that Mars can be colonized.
b. I am of the opinion that Mars cannot be colonized.
Example (19a), for instance, has a strong reading equivalent to (19b) no less
than (1a) has one equivalent to (1b).
Significantly, the Horn clause phenomenon behaves exactly like strict NPIs
with respect to expressions like ‘It is not my opinion that. . . .’/’I am not of
the opinion that’. They are impossible, yielding contrasts like:
(20) a. I don’t believe that at any time did he commit perjury.
b. *It is not my opinion that at any time did he commit perjury.
c. *I am not of the opinion that at any time did he commit perjury.
An immediate inference from data like (16)-(20) is that the distribution
of strong readings does not fully track the distribution of phenomena like
strict NPIs and Horn clauses, phenomena which support the existence of
syntactic NEG raising. It follows that to the extent that strong readings justify
an EMA approach, the latter mechanism cannot provide any insight into the
distribution of strict NPIs and Horn clauses. More formally, (21) is just false:
(21)
An expression V motivates the posit of an EMA if and only if V is an
expression whose complement clauses K allow Horn clauses and strict
NPIs lacking any overt licenser internal to K.
The falsity of (21) strengthens the argument in CP (2014) that semantic/
pragmatic factors cannot be the basis for the strict NPI and Horn clause facts,
domains which argue for the reality of syntactic NEG raising.
Arguably then, the properties of expressions like I am not of the opinion
argue that neither semantic/pragmatic nor syntactic exclusionist views can
be correct. In current terms, NEG scope fixing is not a unitary phenomenon.
Some of it is a function of syntactic NEG raising, some of it of something
else, plausibly something like Bartsch’s EM inferences.
4.3Island cases
The claimed demonstration in CP (2014) that part of the NEG scope fixing do-
main involves syntactic NEG raising depended heavily on extensive evidence
that strict NPI and Horn clause distribution is sensitive to clausal island
boundaries. Such boundaries are in particular determined by topicalized
clauses, illustrated in:
14

early access
Disentangling
(22) a. They did not believe that Barbara won some race.
b. That Barbara won some race, they did not believe.
That the distribution of strict NPIs and Horn clauses is subject to such island
boundaries is illustrated in (23):
(23) a. They did not believe that Barbara had seen her husband in weeks.
b. *That Barbara had seen her husband in weeks, they did not believe.
(24) a. They did not believe that any of the kangaroos had Tod eaten.
b. *That any of the kangaroos had Tod eaten, they did not believe.
Clearly, syntactic NEG raising cannot function out of a topicalized clause.
Suppose then, as an exclusionist syntactic view of NEG scope fixing would
have it, that all strong readings are a function of syntactic NEG raising. It
would follow that such readings are impossible in analogs of cases like (23b)
and (24b). But is that the case? Consider (see Zeijlstra 2017 for related data):
(25) a.
Even after carefully watching most of the election returns, Stella did
not believe that Hillary had lost.
b.
Even after carefully watching most of the election returns, that
Hillary had lost, Stella did not believe.
While (25b) may be stylistically dispreferred, we find the inference from
these examples to the proposition that Stella believed Hillary had not lost
to be no more possible in (25a) than in the island case in (25b). The strong
reading could depend on syntactic NEG raising in (25b) only if the NEG had
raised across the clausal island boundary. But (23b) and (24b) show that to be
impossible.
Consider too:
(26) a.
I have considered the infinity issue, reconsidered it and discussed
it with leading experts. And that infinite sets exist, I just do not
believe.
b. And I just do not believe that infinite sets exist.
To us, it seems that (26a) allows a strong reading to the same extent as (26b).
The conclusion then is that the strong reading in (26a) is not a function of
syntactic NEG raising. Therefore, since some aspects of the NEG scope fixing
phenomenon are due to NEG raising and some are not, neither an exclusion-
ist semantic/pragmatic approach nor an exclusionist syntactic approach is
15

early access
Chris Collins, Paul M. Postal
viable. At least two different sorts of things are involved in the full range of
NEG scope fixing cases.
4.4only
In this section we consider sentences whose subjects are of the form [only
DP]. We show that these sentences do not involve syntactic NEG raising, but
do involve EM inferences. Therefore, they provide another example where
the phenomena handled by syntactic NEG raising and those handled by EM
inferences are dissociated.
Consider the following example:
(27) Only Carol thinks that it is raining.
This sentence can be characterized in terms of its presupposition and truth
conditions as follows:
(28) a. Presupposition: Carol thinks that it is raining.
b. Truth Conditions:
Nobody other than Carol thinks that it is
raining.
Now, consider whether there is an EM inference in (27). If there is no EM
inference, then the people different from Carol have no opinion, perhaps
because they have not thought about the issue or, having thought about it,
have reached no conclusion. But if there is an EM inference, (27) is equivalent
to (29d) (via the inference steps in (29a-d)):
(29) a. Only Carol thinks that it is raining.
b. Nobody other than Carol thinks that
it is raining.
Truth Conditions of (a)
c. Everybody other than Carol does not
think it is raining.
By Logic
d. Everybody other than Carol thinks it
is not raining.
By EMA
It is clear that (27) does have the EMA inference interpretation in (29d) (as well
as the non-EMA interpretation). But even though (27) has that interpretation,
it arguably is not due to NEG raising. For example, sentences parallel to (27)
preclude strict NPIs and Horn Clauses.
16

early access
Disentangling
(30) a. * Only Carol thinks that Mike has seen his mother in years.
b. *Only Carol thinks that Mike will get here until midnight.
c. *Only Carol thinks that ever before has Mike been arrested.
Of course, one could claim that there is in fact NEG raising, and that (30a-c) are
ruled out in some other manner (perhaps related to the fact that the negation
only appears in the assertion, not the presupposition). However, there is no
reason to appeal to further conditions to block cases like (30). Under the
approach in the current paper, the sentences in (30) are unacceptable because
they do not involve NEG raising. And the interpretations of (27) are not due
to syntactic NEG raising, but rather to EM inferences.
Paradigms involving only phrases make the case that EM inferences do
exist independently of syntactic NEG raising. They thus support our view
that no exclusionist approach to NEG scope fixing can be correct.
5Equivalences revisited
The basic claim in the early literature on NEG raising, e.g. in Fillmore 1963,
is that syntactic NEG raising accounted for the semantic equivalence of
sentences like (31a,b):
(31) a. Stephanie doesn’t think that it is raining.
b. Stephanie thinks that it is not raining.
The logic of the account runs as follows. One structure for (31a) involves
syntactic NEG raising, as in (32):
(32) Stephanie does NEG1think that it is <NEG1> raining.
On the standard syntactic analysis elaborated in CP (2014), the NEG is in-
terpreted in its original (embedded clause) position, but pronounced in its
matrix clause position. Therefore, on syntactic structure (32), (31a) has the
interpretation of (31b).
However, having now clarified that EM inferences exist independently of
syntactic NEG raising, (31a) could be interpreted as equivalent to (31b) for a
distinct reason. On the Bartschian account of NEG scope fixing, (31a) can only
have structure (33), with no NEG in the embedded clause:
17

early access
Chris Collins, Paul M. Postal
(33) Stephanie does NEG1think that it is raining.
But under such a view, structure (33) can have the same interpretation as
(31b) because of the EM inference.
And if, as we have argued, one is motivated to posit a conception of NEG
scope fixing with both syntactic NEG raising and EM inferences, then nothing
known prevents (31a) from having ambiguously either structure (32) or (33)
and from being equivalent to (31b) with either one.
More precisely, (31a) may have at least three separate analyses:
(34) a.
The matrix NEG is associated with neither syntactic NEG raising nor
any EM inference.
b.
The matrix NEG is associated with an EM inference but with no NEG
raising.
c.
There is no underlying matrix NEG but there is syntactic NEG raising.
In this case, there can be no EM inferences since the main clause is
semantically positive.
We showed in CP (2014:180) that case (35a) permits no strict NPIs (highlighted)
in the embedded clause, effectively illustrating (34c):
(35) a. * I don’t think that Vincent knows jackshitAabout physics,
because I have never heard of him.
b. *I don’t think that Marilyn has seen her mother in ages,
because I don’t know Marilyn.
These are unacceptable in the framework of CP (2014) because the highlighted
strict NPIs signal the presence of syntactic NEG raising while the because con-
tinuations signal that the matrix clauses are semantically negative. Therefore,
the pre-continuation part of (35a) has the following structure:
(36) I do NEG1think that Vincent knows [<NEG1> jackshitA] about physics.
Since the matrix clause is then, at the level of semantic interpretation, a posi-
tive clause, the continuation because I have never heard of him is anomalous.
The anomaly exists because (35a) jointly implies both that I have thought
about Vincent and that I have never heard of him.
While we have shown that for various cases (e.g. those involving of the
opinion phrases, topic islands and only phrases), EM inferences can be genuine
without any concomitant syntactic NEG raising, the status of cases like (31a)
remains undetermined. This holds since no direct evidence exists that such
18

early access
Disentangling
simple examples have an analysis free of syntactic NEG raising but permit
the EM inference yielding reading (31b). The possibility remains open since
no known evidence supports the contrary view either.
6Conclusion
This paper has expanded on previous work, including our own, to argue
that neither an exclusionist syntactic approach (based on NEG raising) nor
an exclusionist pragmatic/semantic approach based on EM inferences can
provide viable accounts of the known domain of NEG scope fixing. We have
thus in effect argued that there are two distinct sorts of phenomena within the
NEG scope fixing domain, although the two subdomains overlap extensively,
e.g. in the complements of CNRPs which are not islands.
Our results can be stated conditionally: even if one subscribes to the CP
(2014) arguments that a syntactic account of NEG scope fixing is required for
some cases, one should recognize that a unitary exclusionist account of the
facts is not viable.
A reviewer raises the question of whether all languages exhibit both
syntactic NEG raising and EM inferences. Even from the heavily syntactic
perspective of CP (2014) there is no reason to believe that every language has
syntactic NEG raising. Languages differ greatly as to the presence of many
syntactic features, e.g. as to whether they have passive constructions, relative
clauses with overt wh-movement, noun incorporation, etc. So there is no
known reason why certain languages could not lack syntactic NEG raising.
However, given the present study, it is entirely possible that such languages
would still admit some EM inferences.
References
Bartsch, Renate. 1973. ‘‘negative transportation” gibt es nicht. Linguistische
Berichte 27(7). 1–7.
Collins, Chris & Paul M. Postal. 2014.Classical NEG raising: an essay on the
syntax of negation. MIT Press. http://www.jstor.org/stable/j.ctt9qf7ct.
Collins, Chris & Paul M. Postal. 2017a. Interclausal NEG raising and the scope
of negation. Glossa: a journal of general linguistics 2(1). 1–29.https://doi.
org/10.5334/gjgl.79.
19

early access
Chris Collins, Paul M. Postal
Collins, Chris & Paul M. Postal. 2017b. NEG raising and Serbo-Croatian NPIs.
Canadian Journal of Linguistics/Revue canadienne de linguistique 62(03).
339–370.https://doi.org/10.1017/cnj.2017.2.
Collins, Chris & Paul M. Postal. in press. Dispelling the cloud of unknowing:
More on the syntactic nature of NEG raising. In Laurence R. Horn &
Kenneth Turner (eds.), Pragmatics, truth and underspecification: Towards
an atlas of meaning (Current Issues in Semantics and Pragmatics (CRiSPI)).
Leiden: Brill.
Collins, Chris, Paul M. Postal & Elvis Yevudey. 2017. Negative polarity items
in Ewe. Journal of Linguistics 54(2). 1–35.https : / / doi . org / 10 .1017 /
S002222671700007X.
Crowley, Paul. 2016. Neg-raising and neg movement. Massachusetts Institute
of Technology.
Fillmore, Charles J. 1963. The position of embedding transformations in a
grammar. Word 19(2). 208–231.
Gajewski, Jon Robert. 2005.Neg-raising: polarity and presupposition. Cam-
bridge, MA: Massachusetts Institute of Technology PhD Thesis. http :
//hdl.handle.net/1721.1/33696.
Horn, Laurence R. 1975. Neg-raising predicates: toward an explanation. In
Robin E. Grossman, L. James San & Timothy J. Vance (eds.), Papers from
the eleventh regional meeting of the chicago linguistic society, vol. 11,
280–294. Chicago, IL: University of Chicago.
Horn, Laurence R. 1978. Remarks on neg-raising. In Peter Cole (ed.), Pragmatics
(Syntax and semantics), 129–220. New York: Academic Press.
Horn, Laurence R. 1989.A natural history of negation (The David Hume
series). Chicago & London: University of Chicago Press.
Horn, Laurence R. 2014a. Foreword: Classical NEG raising: The first 900 years.
In Classical NEG raising: An essay on the syntax of negation, ix–xvii. MIT
Press. http://www.jstor.org/stable/j.ctt9qf7ct.4.
Horn, Laurence R. 2014b. The cloud of unknowing. In Jack Hoeksema & Dicky
Gilbers (eds.), Black book,178–196. Groningen: University of Groningen.
Horn, Laurence R. & Samuel Bayer. 1984. Short-circuited implicature: a nega-
tive contribution. Linguistics and Philosophy 7(4). 397–414.https://doi.
org/10.1007/bf00631074.
McCawley, James D. 1998.The syntactic phenomena of English. Chicago &
London: University of Chicago Press.
20

early access
Disentangling
Romoli, Jacopo. 2013. A scalar implicature-based approach to neg-raising.
Linguistics and Philosophy 36(4). 291–353.https://doi.org/10.1007/s10988-
013-9136-2.
Zeijlstra, Hedde. 2017. Does neg-raising involve neg-raising? Topoi.1–17.
https://doi.org/10.1007/s11245-017-9461-0.
Chris Collins
Department of Linguistics
New York University
New York, NY 10012
cc116@nyu.edu
Paul M. Postal
Department of Linguistics
New York University
New York, NY 10012
pp11@nyu.edu
21