A ‘Removal’ Type of Negative
JIEUN JOE AND CHUNGMIN LEE
Seoul National University, Seoul, Korea
This paper aims to introduce a new class of negative predicates, removal predicates, which have their own syntactic and semantic behavior, and to scrutinize their lexical properties and thus broaden the research of negative predicates. Interestingly, removal predicates and their lexical antonyms, generation predicates, have opposite characteristics in syntax and semantics. In this paper, we show that there are two kinds of removal predicates according to their negative force. One type of removal predicates, such as absence of, devoid of,
free from and sterile, which we call absence-state predicates, is
monotone decreasing and moreover anti-additive. And the other type of removal predicates, such as disappear, turn off and destroy, which we call removal-process predicates,is weaker in negativity, failing to
be monotone-decreasing, than the absence-state predicates.
In Section 2, we propose the semantics of removal predicates in contrast with generation predicates. Removal predicates are universal and negative, whereas generation predicates are existential and positive. To show the universal nature of removal predicates we resort
1We thank the NPI Project (KRF '99-'01grant) group including Daeho Chung, Seungho thNam and Ed Keenan for discussion and the audience of the 10 J/K Conference at UCLA including Ora Matushansky for comments and questions.
to strong/weak readings in donkey anaphora. Our findings on the negative nature of removal predicates reveal: (a) monotone-decreasingness and anti-additivity of absence-state predicates; and (b) implicature suspension of removal-process predicates.
In Section 3, we discuss the syntax of removal predicates. We deal with the cases in which any is licensed by predicates such as
disappear in preverbal position. Interestingly, any is not licensed by
didn’t appear, even though it seems more negative than disappear.
In Section 4, we discuss morphological characteristics of removal predicates. It is demonstrated that negative morphemes such as indiscriminately in English, -peri-‘throw away; get through’ in
Korean and -tesimau- ‘throw away; get through’ in Japanese further
support our claim that removal predicates are negative. Throughout the paper, we use the British National Corpus (henceforth BNC) to support our claim.
2. Semantics of Removal Predicates
2.1. Two Kinds of Removal Predicates
In this Section, we define a removal type of negative predicates on semantic grounds. Removal predicates are predicates whose meanings involve absence-state or removal-process. An absence-state predicate denotes a simple event and is not decomposable, whereas a removal-process predicate can be decomposed into a process subevent and a result-state subevent.
In the narrow sense of the term, remove, eliminate and
disappear are removal predicates, but aspectual verbs such as finish
and end, though broader in scope, may be included. Removal predicates are universal and negative. The lexical antonyms of these are generation predicates, whose meanings involve presence-state or the process of some entity’s coming into being. In this paper, we divide removal predicates according to the observation (a) whether the predicate is monotone decreasing; (b) whether the predicate has implicature suspension; and (c) whether the predicate licenses any. (K
stands for Korean examples and J for Japanese ones.) Observe:
(1) Absence-state predicates:
absence of, clear of, be devoid of, free from, be freed of,
independence from, be innocent of, regardless of, (up to here
from Hoeksema and Klein 1995), sterile, barren of, arid,
(2) Removal-process predicates:
a. Implicature suspension and licensing any in object/oblique
position: remove, destroy, get rid of, cut off, blot out, erase,
cancel, break down, annihilate, ttey-(K)/hanare-(J) 'remove',
chwuisoha-(K)/torike-(J) ‘cancel’, kkunh-(K)/tewoki- ‘cut’,
‘sever’(a relationship)(J), agehana- (the head) (J) ‘cut off’
b. Implicature suspension and licensing any in subject position:
disappear, vanish, perish, saraci-(K)/kie-(J)'disappear'
c. Implicature suspension and licensing any in both subject and
object position: end, break (middle verbs)
d. Implicature suspension and not licensing free choice any except
in a sugtrigging context: turn off, kku-(K)/ke-(J)'turn off',
desalinize, dehydrate (we adopt a unified theory of any (Lee
In the next section, we provide the semantics of two kinds of removal predicates: absence-state predicates and removal-process predicates.
2.2. Negative Nature of Removal Predicates
In this Section, we discuss negative nature of removal predicates. As mentioned in the previous section, absence-state predicates such as be devoid of, sterile and barren are monotone decreasing and anti-
additive, based on Ladusaw (1996). And removal-process predicates such as disappear and turn off are neither monotone decreasing nor
nonveridical in a strict sense. However, removal-process predicates are negative and thus necessarily yield implicature suspension and sometimes license any, while their antonyms, generation predicates, do not license it at all and do not yield implicature suspension. 2.2.1. Monotone Decreasingness and Anti-additivity: Absence-
Absence-state predicates given in (1) in the last section, are monotone decreasing and anti-additive. Consider the definition of monotone decreasingness and anti-additivity from Ladusaw (1996):
(3) a. If A and B are two Boolean algebras, the function f from A into
B is polarity reversing/monotone decreasing iff any a1, a2 in A,
if a1?a2, then f(a2)?f( a1)
b. A functor f is anti-additive iff f (x；y) = f (x);f(y).
On this definition, absence-state predicates such as be devoid of are
monotone decreasing. But removal-process predicates such as destroy
and eliminate are not monotone decreasing. Compare (4) and (5):
(4) |a spelling error |?|an error|
a. John’s term paper is devoid of an error. ？
b. John’s term paper is devoid of a spelling error.
(5) |a spelling error |?|an error|
a. John eliminated an error in the paper. --/-->
b. John eliminated a spelling error in the paper.
(4a) entails (4b), whereas (5a) does not entail (5b). That is, if John’s
paper is devoid of an error, naturally it means there is no spelling error. However, if John eliminated an error, the error could be a spelling
error, a citation error or another kind of error. Thus, be devoid of is
monotone decreasing and eliminate is not monotone decreasing based
on the definition (3).
However, Hoeksema and Klein (1995:153) treat these two predicates, eliminate and be devoid of, as monotone decreasing. Their
claim is based on the fact that eliminate as well as be devoid of
licenses any as shown in (6). In (6a) any is licensed by be devoid of
and in (6b) it is licensed by eliminate. And in (6c,d) any is licensed by
destroy. Still they are not monotone-decreasing. Observe:
(6) a. He was devoid of any of the normal human weaknesses such as
fear or self-doubt. (from BNC)
b. It is likely too that the priest tidied up and eliminated any traces
there may have been distorted thinking or language, as he almost
certainly corrected any theological mistakes, for his own safety.
c. The death of Wordsworth's brother John in the spring of 1805
destroyed any remaining illusions. (from BNC)
d. But the worsening weather conditions as we cycled destroyed any
possibility of us seeing such stunning views. (from BNC)
Even though removal-process predicates such as eliminate and destroy
license any as in (6b, c, d), their negative force is weaker than
monotone decreasing, as demonstrated earlier. Consider (7).
(7) |a red-blocked house|?|a house|
a. John destroyed a house.
b. John destroyed a red-blocked house.
Example (7a) does not entail (7b). That is, in (7a) it is possible that John destroyed a white-blocked house or yellow-blocked house. Therefore, just like eliminate in (5), destroy (or even remove) is not
monotone decreasing, although such verbs become more negative when they are used in their extended mental or abstract senses, as in (6d). It further shows that licensing any is not a sufficient condition
but a necessary condition of monotone decreasingness.
Returning to absence-state predicates, whose negative force is stronger than monotone decreasing, the level of negativity differs. As shown in (3), anti-additivity is defined as follows:
(8) A functor f is anti-additive iff f (x；y) = f (x);f(y).
Depending on this definition, absence-state predicates such as be
devoid of is anti-additive. Consider (9):
(9) a. John’s term paper is devoid of a spelling error or a citation error.
b. John’s term paper is devoid of a spelling error AND is devoid of
Examples like (9a) and (9b) above entail one another. That is, if it is true that John’s term paper is devoid of a spelling error or a citation error, it is also true that John’s term paper is devoid of a spelling error
and is devoid of a citation error, and vice versa. Therefore, absence-state predicates are monotone decreasing and anti-additive. In the next section, we show the negative force of removal-process predicates. 2.2.2. Implicature Suspension: Removal-Process Predicates
Removal-process predicates given in (2) are neither monotone decreasing nor nonveridical. However, they yield implicature suspension, whereas their lexical antonyms do not have such a semantic characteristic. Observe the relevant examples on implicature suspension from Chierchia (2000):
(10) a. Every student who takes a written test or makes an oral
presentation will pass.
b. Expectation: a student that does both passes.
(suspension of exclusion implicature.)
Chierchia (2000) points out that any licensing contexts as in (10a) can
suspend implicature. That is, potential implicature ‘not both A and B’
is suspended as in (10b). And Horn (1989) indicates that the computation of scalar implicatures appears to be inhibited not only by negation but also generally in ‘negation like’ monotone decreasing contexts such as doubt. Our removal predicates can also suspend
implicature ‘not both A and B’ and take a role as a ‘negation-like’
context, whereas generation predicates cannot suspend such an implicature. Consider the following Korean examples in (11) and Japanese examples in (12):
(11) a. Haksayng-tul-un ppippi-na handphone-ul kke-t-ta.
Students-PL-TOP-beepers-or cellular phones-ACC turn off-PST-DEC
‘Students turned off beepers or cellular phones.’
b. Haksayng-tul-un ppippi-na handphone-ul khy-e-tta.
Students-PL-TOP-beepers-or cellular phones-ACC turn on-PST-DEC
‘Students turned on beepers or cellular phones.’
(12) a.Gakusei-tachi-wa pokeberu-ya keitaidenwa-no suichi-wo kesi-ta.
Students-PL-TOP-pager-or cellular phones-POS-switch-ACC turn off-PST
‘Students turned off beepers or cellular phones.’
b.Gakusei-tachi-wa pokeberu-ya keitaidenwa-no suichi-wo ire-ta.
Students-PL-TOP-pager-or cellular phones-POS-switch-ACC turn on-PST
‘Students turned on beepers or cellular phones.’
In (11a) and (12a), even if some students with both beepers and cellular phones turned off both of them, suspending the implicature 'not both,' the sentences are quite appropriate. However, in (11b) and (12b) even if some students with both beepers and cellular phones turned on only one kind of them, the sentences can be appropriate. Such an asymmetry occurs to all pairs of removal-process predicates and their lexical antonyms. The negative force of removal-process predicates is substantiated in the same line of Horn (1989), Chierchia (2000) and Chungmin Lee (2000). Consider (13), which also suspends implicature:
(13) a. Recently, the measles or chicken pox disappeared from the
b. Recently, the measles or chicken pox appeared in the schools.
In (13a), if some schools did not have measles but had chicken pox, and the other schools did not have chicken pox but had measles, the sentence is not appropriate. In the same situation, however, (13b) is. 2.2.3. (Non)veridicality: Removal-Process Predicates
One of the interesting properties of removal-process predicates is that they are not nonveridical in its strict sense. In the previous section, we demonstrate that removal-process predicates are negative and suspend implicature, but are not monotone decreasing. Giannakidou (1998:116) notes that negative verbs in Greek such as arnume ‘deny’
and apagorevo ‘forbid’ are nonveridical. However, different from other
negative predicates, removal-process predicates are not nonveridical. That is, if Jane denies that she saw Paul, this does not entail that she did not see Paul nor does it imply that she saw Paul, of course. Yet, in the case of removal-process predicates such as disappear and destroy,
if any rumors about his past disappeared, this implies that there are no rumors about his past left. So, disappear is veridical in a sense, with
no overt operator. Consider the following definition of
(non)veridicality from Chungmin Lee (1999), originated from Zwarts (1995).
Let O be a monadic sentential operator. O is said to be veridical
just in case Op p is logically valid. If O is not veridical, then O is
nonveridical. E.g., ‘it seems,’ ‘it is possible’, ‘Sue hopes.’ Truth-
functional connectives are likewise defined. E.g., in p and q, both
the p- and q- positions are veridical; in p or q, and p if q, both the
p-and q-positions are nonveridical.
Based on the definition, our removal-process predicates are weakly negative but are veridical. Consider:
(15) a. The death of Wordsworth's brother John in the spring of 1805
destroyed any remaining illusions. (from BNC)
b. Any rumors/doubt about his past disappeared.
In (15a), if it is true that the death of Wordsworth's brother John destroyed any remaining illusions, then it is implied that there were no illusions left. In (15b), if it is true that any rumors/doubt about his past disappeared, then it is also implied that there remained no rumors or doubt.
2.2.4. Licensing any in object/oblique/subject positions
As shown in previous section, removal predicates are divided into two types: (a) absence-state predicates; and (b) removal-process
predicates. Absence-state predicates such as be devoid of and immune
license any in object/oblique/subject position as in (16)l. Consider the examples in (16) from BNC:
(16) Any licensing in oblique position: Absence-state predicates
a. His face was devoid of any warmth or humor.
b. The rest of the room was barren of any furniture.
c. A problem with making recordings direct from electronic
instruments is that they are totally free from any natural
In addition, some removal-process predicates such as destroy,
disappear and cut off license any in object or subject positions. In most cases, if removal-process predicates license any, it denotes an abstract
or psychological stuff such as illusion and doubt, with almost
monotone decreasing force. However, other removal-process predicates such as turn off do not license any (except in a subtrigging context).
Removal-process predicates such as disappear, destroy and break
license any as in (17) but not their lexical antonyms, as shown in (17):
(17) Any licensing in subject/object position: Removal-process predicates
a. Any rumors about his past disappeared. (/*appeared.)
b. I urge the Minister to drop this idea and cancel (/*make) any
tenders he may have called for the construction of such a cruel
detention center. (from BNC)
c. The Government buys very little from South Africa and should,
in fact, have ended (/*begun) any purchase from that country
long ago. (from BNC)
d. The death of Wordsworth's brother John in the spring of 1805
destroyed (/*built) any remaining illusions. (from BNC)
However, generation predicates such as appear can license any in a
subtrigging context, as in (18). In fact, almost every predicate can license any in a subtrigging context. (For a detailed discussion, see Section 3.)
(18) Any student who passed the entrance exam appeared (at the party).
Note that other removal-process predicates such as turn off as well as
its lexical antonym turn on do not license any except in a subtrigging
context, as in (19):
(19) a. *Jane turned off any lamps in the building.
b. *Jane turned on any lamps in the building.
c. Any lamps in the building were turned off/on.
Compared to absence-state predicates, removal-process predicates do not always license any. Additionally, the negative force of these removal-process predicates is weaker than absence-state predicates.
In Korean as well as in Japanese, removal-process predicates such as saraci-(K)/kie-(J) ‘disappear’ do not license strong NPIs such as te
isang (K)/koreijyo (J) ‘any more’, as in (20), and predicates such as
phokiha-(K)/yame-(J) ‘give up’ are anti-additive and thus license the
strong NPI te isang (K)/koreijyo(J) ‘any more’, as in (21) below:
(20) a.* LA -ey-nun te isang kkamagui-tul-i saracie-ssta. (Korean)
b.* LA-ni-wa koreijyo karasu-tachi-ga kie-tta. (Japanese)
‘ In LA any more crows disappeared.’
(21) a. Jane-un te isang nonmwun-ul sseki-rul phokihay-ssta (Korean)
Jane-TOP-anymore-paper-ACC write-ACC-give up-PST
b. Jane-wa koreijyo ronbun-wo kakukoto-wo yame-ta (Japanese)
‘Jane gave up writing paper anymore.’
Removal-process predicates in Korean license a weaker existential NPI form etten- N-i-ra-to ‘any’, as in (22):
(22) a. Etten toshi-i-ra-to pakoyhay-ss-ta. (Korean)
any city -be-DEC-C destroy-PST-DEC
Lit. ‘(They) destroyed any city.’
b.* Etten toshi-ra-to kenselhay-ss-ta.
any city -be-DEC-C destroy-PST-DEC
Lit. ‘(They) constructed any city ’.
Etten- N -i-ra-to (K) ‘any’ is allowed to occur with removal predicates, but is awkward with generation predicates. In a clearer piece of evidence, removal predicates, as universal ones, are not likely to be combined with contrastive topic marker (heretofore CT-marker) -nun,
whereas their lexical antonyms have no such problems. Consider (23):
(23) a. ?? pakoy-/cwukiki-/saraciki-NUN hay-ss-ta
destroy/kill / disappear CT do-PST-DEC
‘ It (is) destroyed/killed/disappeared’
b. kenselhaki-/kwucohaki-/natanaki-NUN hay-ss-ta
construct /rescue / appear CT do-PST-DEC
‘It (is) constructed/rescued/appeared’
In the next section, we show another semantic characteristics of removal predicates: universality.
2.3 Universal Nature of Removal Predicates
In this section, we discuss universal nature of removal predicates as well as existential nature of generation predicates. Krifka (1996) and Yoon (1996) introduce total/partial predicates to explain preferred strong/weak readings in a donkey sentence. We replace total by universal and partial by existential to generalize the phenomena. Consider the following examples from Krifka (1996:140):
(24) a. Every farmer who owned a donkey kept it healthy during the
rainy season.(strong reading)
b. Every farmer who owned a donkey kept it sick during the rainy
season. (weak reading)
In the above examples, healthy is a total predicate and sick is a partial
predicate. When the given predicate is total, the sentence is true if it involves almost all donkeys. However, when the predicate is partial, it is acceptable, even when the event involves ‘some’ of the donkeys with the anaphor it. The same judgment is valid for our removal and
generation predicates. Removal predicates are total and, in our terms, universal, while generation ones are partial and existential. Consider:
(25) a. Every student who owned an error in his essay obliterated it.
b. Every student who owned a reward added it to his resume.
(26) a. Every student who owned a lamp turned it off.
b. Every student who owned a lamp turned it on.
As in (25a) and (26a), removal predicates such as obliterate and turn
off show strong readings, while as in (25b) generation predicates such as add and turn on show weak readings. Thus, in the donkey sentence, if the main predicate is a removal predicate, the E-type pronoun it
exhibits a strong reading, whereas if the main predicate is a generation predicate, the E-type pronoun it exhibits a weak reading. That is, in the situation of (25a), the sentence is expected to be true if every student obliterated any kind of errors in his paper, while in the situation of (25b) the sentence is expected to be true even if some students added only remarkable records, but not all of them. Interestingly, the following pair of aspectual predicates shares this strong/weak contrast:
(27) a. Every student who owned a comic book finished (reading) it.
b. Every student who owned a comic book began (to read) it.
E-type pronoun it in (27a) has a strong reading as with removal predicates, whereas (27b) has a weak reading like generation predicates. Ter meulen (1995:32) tried to show that aspectual verbs such as finish and end are monotone decreasing, while the aspectual verbs such as start and begin are monotone increasing. However, in the
strict sense of monotonicity, aspectual verbs such as finish and end are
not monotone decreasing. Observe:
(28) |a spelling test |?|a test|
a. Jane ended a test. --/-->
b. Jane ended a spelling test.
2(28a) does not entail (28b).
3. Syntax of Removal Predicates
In this Section, we discuss the syntax of removal predicates. We analyze the cases in which any is licensed by removal predicates such
as disappear in the preverbal position. Our analysis is based on the two
theoretical assumptions: (a) any reconstruction at LF; (b) Unaccusative
3.1. Asymmetries in NPI-licensing
Consider the following examples in (29):
(29) a. Any rumors/doubts about his past disappeared.
2 We owe Tim Stowell for his observation that finish does not seem to have any monotone-decreasingness effects.