Neuropsychologia 51 (2013) 132–141

Contents lists available at SciVerse ScienceDirect

Neuropsychologia

0028-39

http://d

n Corr

Groning

Tel.: þ3

E-m

journal homepage: www.elsevier.com/locate/neuropsychologia

Processing polarity: ERP evidence for differences between positive andnegative polarity

Anna Yurchenko c, Dirk-Bart den Ouden d, Jack Hoeksema a, Olga Dragoy e,f,John C.J. Hoeks a,b, Laurie A. Stowe a,b,n

a Center for Language and Cognition, University of Groningen, The Netherlandsb Neuroimaging Center, University of Groningen, The Netherlandsc Moscow State University, Russiad Department of Communication Sciences and Disorders, University of South Carolina, USAe Moscow Research Institute of Psychiatry, Russiaf National Research University Higher School of Economics, Russia

a r t i c l e i n f o

Article history:

Received 4 June 2012

Received in revised form

27 September 2012

Accepted 24 October 2012Available online 7 November 2012

Keywords:

Language comprehension

ERPs

Negative polarity positive polarity

Processing of negation

32/$ - see front matter & 2012 Elsevier Ltd. A

x.doi.org/10.1016/j.neuropsychologia.2012.10

esponding author at: Center for Language

en, Postbus 716, 9700 AS Groningen, The Neth

1 503636627; fax: þ31 503636637.

ail address: l.a.stowe@rug.nl (L.A. Stowe).

a b s t r a c t

The goal of the present study was to investigate event-related potential (ERP) responses to Dutch

negative and positive polarity adverbs of degree presented in licensed and unlicensed contexts with

negative and affirmative particles directly preceding the polarity item. To control for effects of the

processing of negation as such, neutral adverbs were also presented in negative and affirmative

contexts. The results did not show any significant effect of negation for the non-polar adverbs, allowing

context effects for polarity items to be interpreted as being due to the appropriateness of the context.

Negative polarity violations elicited an N400 response that might reflect the lack of semantic congruity

of the negative polarity item in an affirmative context. In contrast, processing positive polarity items in

context of negation resulted in a positive effect resembling the P600, which may be considered as a

marker of a different sort of integration difficulty caused by violation of licensing conditions and/or a

search for a licensor in the wider discourse context. The study presented here is the first to show an

unambiguous dissociation between responses to negative and positive polarity violations. This

dissociation argues for different mechanisms underlying the processing of these two types of polarity;

we propose that positive polarity items are sensitive to wider discourse context, while negative polarity

items are more sensitive to local lexical context.

& 2012 Elsevier Ltd. All rights reserved.

1. Introduction

The goal of this study was to investigate the way in which twosorts of polarity items are processed. Most languages have certainwords which can only be used in a negative context: negativepolarity items (henceforth: NPIs). A classic example is ever in English,which is generally not acceptable in a positive context (e.g., nJohn was

ever in Pasadena). There is a large range of other elements (otherexamples are lift a finger, give a damn) which show the samecharacteristic. The second type of polarity item consists of wordswhich cannot occur in a negative context, like already (e.g., nJohn is

not already here): positive polarity items (PPIs).At first sight these two constructions seem similar, i.e., parallel to

each other. However, there are both theoretical and experimental

ll rights reserved.

.028

and Cognition, University of

erlands.

arguments against the apparent parallelism (see Sections 1.1 and1.2). Most important for the current study, if NPIs and PPIs areprocessed in substantially different ways, it supports the claim thatthe constructions have different underlying representations andfunctions. Saddy, Drenhaus and Frisch (2004) used event-relatedpotentials (ERPs) to show differences in processing. NPIs in a positivecontext showed a negative deflection in the ERP waveform between300 and 500 ms relative to when they occurred in an appropriate,negative context. Violation of PPIs by a negative context showed alater positivity. The suggestion that the two sorts of polarity itemsshow substantive differences in processing is interesting, but thedissociation has not been replicated, and has in fact been partiallychallenged by Drenhaus, Beim Graben, Saddy and Frisch (2006).

The interpretation of these results is also complicated by theeffects of negation itself. Negation is known to make processingmore difficult (Clark & Chase, 1972; Just & Carpenter, 1971;Wason, 1961; see Horn, 1989, Chapter 3 for an overview of thisliterature), which may have masked or increased the effects ofpolarity in either type of sentence. Additionally, the polarity

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141 133

processing was investigated using relatively complex licensingstructures, which may have influenced the results. This complex-ity limits the generalizability of the results of Saddy et al. (2004).In the current study, we will investigate if the same dissociationholds in a simpler structure, controlling for the effects of negation.

1.1. Linguistic background

NPIs have been studied since the early days of generativegrammar. Early work on these items took a syntactic approach,using features to mark elements capable of licensing NPIs (Klima,1964; see also Progovac, 1993, 1994). However, it has generallybeen agreed since Baker (1970), Fauconnier (1978) and Ladusaw(1979) that the licensing of polarity items is essentially based onmeaning. Consider minimal pairs such as the following:

(1) a. nJones admitted that he had ever been to prison.b. Jones denied that he had ever been to prison.

Syntactically, the two sentences do not differ. The ill-formednessof the first sentence is due to the positive character of the verbadmit, whereas the well-formedness of the second sentence is to beattributed to the negative character of the verb deny, required by theNPI ever. This is a semantic difference, due to the differences inlexical meaning of the two verbs. Much of the literature on NPIs isdevoted to the statement of semantic licensing conditions (cf.Fauconnier, 1978; Gajewski, 2011, Giannakidou, 1998; Hoeksema,2000; Krifka, 1995; Ladusaw, 1979, Linebarger, 1987; van derWouden, 1997; von Fintel, 1999; Zwarts, 1998). As noted bySaddy et al. (2004), if the licensing of NPIs is indeed a semanticmatter, there should be implications for ERP research. In particular,they predicted that an N400 effect, classically regarded as a markerof semantic integration problems, should be found in cases ofviolations of negative polarity licensing conditions.

PPIs, such as English some and already, have received ratherless attention in the literature than NPIs. Although in some senseparallel to NPIs, they differ in not having specific licensors, butrather cannot co-occur with negative elements, which serve asanti-licensors and are in general the licensors of NPIs. However,the presence of negation is not always enough to make a PPIunacceptable. Baker (1970) pointed out that while such itemsmay not normally appear in negative contexts, they are accep-table in double negation contexts (see also Szabolcsi, 2004)

(2) a. nFred is not still drunk.b. I cannot believe Fred is not still drunk.

Thus, a PPI in a locally negative, but globally affirmative contextis acceptable, which is likely to make the processing of PPIs some-what complex. They are even acceptable in some purely negativecontexts. For example, negation in higher clauses does not lead tothe same ungrammaticality for PPIs in a dependent clause as directclause-mate negation (Ladusaw 1979). Compare example (2c):

(2) c. I cannot believe Fred is still drunk.

Seuren (1976, 1985), Horn (1989, 2001) and Israel (1996,2011) made the important observation that emphatic denial inwhich part of an assertion is echoed in its denial likewise allowsfor PPIs. Compare (2a) and (3b):

1 Progovac (1994) draws a parallel between polarity licensing and binding

(3) a. Fred is still drunk, isn’t he?

b. No, Fred ISN’T still drunk.

theory, which makes a distinction between elements which must be locally bound,

and referring expressions which cannot be locally bound, or in other terms,

depend on global context. We will not discuss this point now, but see Harris,

Wexler and Holcomb (2000) for a discussion of binding theory, using ERPs.

An NPI in a contradiction parallel to (3b), however, would notbe saved by negation in a context sentence parallel to (3a), again

suggesting that PPIs and NPIs are not really parallel to eachother.1 From these examples it seems clear that the processorshould deal with local negation combined with a PPI by attempt-ing to establish whether the context is globally positive or not.Such a search would make use of different processes than thoserequired for the integration of an NPI with local negation. We willreturn to this point below.

Many types of expression can be NPIs or PPIs. Most commonlydiscussed are indefinite pronouns, such as any and some, but theliterature shows that the range of polarity-sensitive expressions isvery wide and diverse (Hoeksema, 2010; Ladusaw, 1979; Sailer &Richter, 2002; van der Wouden, 1997). Since our goal in this paperwas to compare the processing of NPIs with PPIs, and both withpolarity-neutral items, we had to find items from each category withsimilar syntactic and lexico-semantic properties, allowing them toappear in similar types of sentences. Since Dutch has NPI, PPI andneutral adverbs of degree, we decided to select our items from thisclass of expressions, using the adverbs bijster ‘all that’, bijzonder

‘particularly’, and nogal ‘rather’, as in (4): the translations make useof equivalent polarity expressions to make the effect of presence orabsence of negation more transparent for speakers of English.

(4) a. De hond is n(niet) bijster slim. NPIThe dog is not all-that smartThe dog is n(not) all that smart.

b. De hond is (nniet) nogal slim. PPIThe dog is (not) rather smartThe dog is (nnot) rather smart.

c. De hond is (niet) bijzonder slim. NeutralThe dog is (not) particularly smartThe dog is (not) particularly smart.

The three items are not entirely on a par in terms of frequencyand length. Bijster in particular is less frequent than nogal andbijzonder. This has been pointed out as a weakness of somerelated studies (e.g., Shao & Neville, 1998). Given the lack of anyNPI degree adverbs with a higher frequency, it was nonethelessthe best choice under the circumstances. Since the experimentuses a within-item comparison between negative and positivecontexts, this should not cause a problem.

1.2. Previous experimental research

Previous ERP studies by Saddy et al. (2004), Drenhaus, Frischand Saddy (2005), and Drenhaus et al. (2006) also made use ofadverbial elements in their experimental material, namely theGerman adverbs jemals ‘ever’, an NPI, and durchaus ‘absolutely’, aPPI. There were no neutral conditions in their designs, whichwould have allowed examination of the effects of negation per se.The data from these studies showed a positivity for violations ofPPIs, and for NPIs only a negativity that was similar to the N400generally found for semantically incongruous words. Saddy et al.suggested that the results are consistent with the use of verydifferent processes for the two types of polarity item.

The authors suggest that the presence of the positivity, whichis similar to the P600 reported as a response to a number ofsyntactic violations, may indicate that for PPIs constituency ismore relevant than for NPIs. However, it is not the case that theP600 is associated with syntactic processing only. A number ofauthors have argued for a more general cognitive process which

2 A reviewer notes that an additional difference between (6) and (7) is that

niet is antimorphic (in the sense of Zwarts, 1998), whereas geen is anti-additive, a

weaker type of negation. Hoeksema and Den Ouden (2005) anticipated this point,

and noted that judgments get significantly better when the negative element is

part of a floating quantifier in the VP

(i) De keurmeesters waren geen van allen bijster tevreden.

The inspectors were none of them very content

‘The inspectors were none of them very content

’This suggests to us that the main issue is not so much strength of negation, but

proximity of negation. Corpus data also point in this direction. Of 18 cases of

bijster licensed by geen in the corpus used by Hoeksema and Den Ouden, 17

involved an occurrence of geen in the same VP, and only one external to the VP,

as in (7).3 This does not appear to be the case to the same degree for English rather/

somewhat, which are equally bad in contexts like (6) and (7), according to a

reviewer. We have no immediate explanation for this difference between English

and Dutch. Note however that the ‘‘better’’ version in Dutch is not considered

acceptable, merely less unacceptable.

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141134

may apply to reprocessing or attention (Coulson, King & Kutas,1998; Gunter, Stowe & Mulder, 1997) or conflict resolution(Kuperberg, 2007). Additionally, more recent research has shownthat pragmatic processes are likely to produce late positivitiessimilar to the P600 as well (Burkhardt, 2007; Kaan, Dallas, &Barkley, 2006; see Brouwer, Fitz & Hoeks, 2012, for a recentoverview of the occurrence and possible interpretations of theP600). Since it is clear that the global context plays a significantrole in the processing of PPIs, this seems a more straightforwardexplanation for the positivity seen in response to PPIs. Note also thatan interpretation of the positivity elicited by PPI violations as apragmatic positivity is compatible with the view that the function ofthe P600 reflects a more general cognitive process than syntacticprocessing or even linguistic processing more generally.

Drenhaus et al. (2005) showed that unlicensed NPIs couldelicit a positivity as well, when preceded by a negative element ina position which does not license the NPI, a so-called ‘intrusion’,as in sentences like (5)

(5) nThe boys who read no books for the school assign-ment had any idea what the teacher was asking.

In a reanalysis of the data reported by Saddy et al. (2004),Drenhaus et al. (2006) attempted to show that NPIs, even withoutintrusions, are processed in the same way as PPIs. With a reanalysiswhich they argued to be more sensitive, there was a significantP600 also for NPIs, although it was significantly shorter and lessstrong. If it is true that both NPI and PPI violations elicit the sameeffect, this argues against the claim that there is a qualitativedifference in the processing of the two types of polarity item,despite the dependence of PPIs on a wider pragmatic context,unless we make the assumption that there are several kinds ofP600 which we cannot distinguish clearly. This is clearly anundesirable step in argumentation, as it is essentially unfalsifiable.

One factor that complicates the interpretation of the results ofboth Saddy et al. (2004) and Drenhaus et al. (2006) is the lack of acontrol for negation per se, which may muddy the waters. It iswell-known that negative sentences are more difficult thanaffirmative sentences (Clark & Chase, 1972; Horn, 1989; Just &Carpenter, 1971; Wason, 1961). This could well cause a deflectionin negated relative to affirmative sentences. The effect of negationwould combine with the effects of violation differently for NPIs(where the violation is affirmative) than for PPIs (where theviolation is negated). Ludtke, Friedrich, de Filippis and Kaup(2008) have reported that negation of subject noun phrases,which were used by Saddy and colleagues, causes an apparentlylong-lasting negative deflection in the ERP waveform (up until1400 ms after the presentation of the negated noun). If a nega-tivity similar to that found by Ludtke et al. occurred in theGerman studies and lasted long enough, it could have an impacton the processing of the polarity items. Indeed, a pilot studywhich we carried out using sentences similar to those used in thestudy of Saddy et al. suggested that this might be the case. Theneutral control described above in (4c) is therefore necessary inorder to separate the effects of the presence or absence ofnegation from the effects of violation of polarity.

Since the original work by Saddy et al. (2004), there have beenseveral follow-up ERP studies of polarity licensing for otherlanguages, e.g., Steinhauer, Drury, Portner, Walenski andUllman, 2010; Xiang, Dillon and Phillips (2009) on English, andVespignani, Panizza, Zandomeneghi and Job (2009) on Italian. Wewill return to some of these in the discussion rather than discussthem extensively at this point, because these studies examinedaspects of the processing of NPIs only.

One last point about the investigation of polarity processingpertains to the potential effect of the position in which negation

occurs. Hoeksema and den Ouden (2005) did an experiment inwhich they asked participants to provide acceptability judgments(using a 5 point scale) for the three items bijster, bijzonder, and nogal

in a variety of contexts. One of the surprising findings of thisexperiment was that there appears to be an effect of type and/orposition of the licensor for both bijster and nogal. Cases where bijster

is directly preceded by niet were judged as better than cases where aVP-internal occurrence of bijster was licensed by a negative elementsuch as geen ‘no’ in the subject noun phrase, although neither wasconsidered truly unacceptable. Example (6) was judged significantlybetter (at po0.05) than example (7)2

(6) Een van hen was niet bijster succesvol als advocaat.One of them was not at-all successful as lawyerOne of them was not at all successful as a lawyer.

(7) Geen van de leerlingen was bijster geschikt voorde taak.None of the students was at-all suited for the taskNone of the students was at all suited for the task.

Mirror-image differences were found between occurrences ofthe PPI nogal in contexts similar to (6) and (7), but in the oppositedirection: direct negation by niet ‘not’ was considered signifi-cantly worse (at po0.001) than negation involving geen ‘no’,although both were considered truly acceptable.3 This impliesthat the more distant negation employed by Saddy and colleagues(Drenhaus et al., 2005, 2006; Saddy et al., 2004) may haveunderestimated the effect of violation for NPIs or overestimatedit for PPIs, making the comparison troublesome.

1.3. Present study

The overall goal of the present study was to see if the dissocia-tion between processing of PPIs and NPIs reported by Saddy et al.(2004) would hold when some important changes were made in theexperimental design, as we have pointed out in the previoussections. First, given Hoeksema and den Ouden’s (2005) findingthat the position or type of negation in the context mattered morethan anticipated for judgments of acceptability, it seemed prudentnot to vary the syntactic context for the adverbs in our ERPexperiment, but to keep them fixed, using the configuration thatyielded the clearest distinction in acceptability between NPIs andPPIs, namely the one exemplified in (6) above, with clause-matenegation directly preceding the element of interest. To create apositive context with the same structure and length as its negativecounterpart, we used various positive adverbs, such as ook ‘also’,and wel, a marker of emphatic affirmation (comparable to Englishstressed auxiliaries: Fred does not like asparagus but he DOES like

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141 135

broccoli). These adverbs may appear in the exact same position asniet, allowing us to compare sentences on a word-by-word basis.

Second, as previously discussed, it was important to disen-tangle any possible effects of negation per se from the effects ofviolation in negative or affirmative sentences. Last, an issue whichwe have not touched on in the preceding discussion was thechoice of task. In most of the studies which have been carried outto date, the participants were asked to make a grammaticality oracceptability decision after reading the target sentences. Thestrength of the ERP methodology is that this sort of task is not,in principle, necessary. In purely behavioral studies a response isnecessary in order to measure accuracy or decision time. WithERPs, we see how the brain responds to the input more directly.We also see the response to the conscious decision process, whichis not necessarily the same for acceptable and unacceptablesentences (see Dimitrova, Stowe, Redeker & Hoeks, 2012, for adiscussion of task effects on ERP patterns). For this reason wechose to ask subjects to concentrate on comprehension. To makesure that they did so, one third of the sentences were followed bycomprehension questions. Thus, the responses we see are those tonatural sentence processing, in which the primary goal is simplyto understand what the sentence means.

2. Materials and methods

2.1. Participants

32 native Dutch speakers participated in the experiment. Six of them were

excluded from further analysis due to excessive artefacts in their EEG data. The

remaining 26 ranged from 18 to 26 years old (mean age 21), 15 female, 11 male. All

of them were right handed, with no known neurological impairment and had normal

hearing and normal or corrected to normal vision. All participants signed an informed

consent in accordance with the Declaration of Helsinki in a procedure approved by

the local medical ethics committee. They were paid for their participation.

2.2. Materials

We constructed 40 correct sentences with a negative polarity adverb (bijster),

40 correct sentences with a positive polarity adverb (nogal), and 40 negated

sentences with a neutral adverb congruent with both negative and affirmative

contexts (bijzonder). For each sentence an alternative version was constructed (in

which negative became affirmative and vice versa), creating in total 120 pairs of

sentences. Thus, we employed a 3 (adverb type)�2 (negative vs. affirmative)

design. Examples (8)–(10) illustrate the six experimental conditions: negative

polarity in negative/affirmative context (correct/incorrect, example (8)), positive

polarity in affirmative/negative context (correct/incorrect, example (9)), and

neutral polarity in negative/affirmative context (correct/correct, example (10)).

The target adverbs are marked in bold.

(8) Harold kon heel ergHarold can3SG.PAST really very

maar/en zijn handschrift wasbut/and his handwriting be3SG.PAST

Harold could draw really very beautifully but/and his handw

(9) Het beleid tegen terrorismethe policy against terrorism

en/maar de partijen zijn echt/nniet

and/but the partyPL be

The policy against terrorism has to be accentuated and/but th

(10) Het pak was bedruktthe suit be3SG.PAST printPART.P

en/maar het was niet/wel

and/but it be3SG.PAST not/actua

The suit was printed in a strange pattern and/but it was not/

Experimental sentences consisted of an introductory clause and a coordinate

clause which included the polarity item. In the negative context the target adverb

was preceded by the negative particle niet ‘not’; in the affirmative context one of

six affirmative particles (wel ‘indeed’, echt ‘really’, ook ‘besides, as well’, toch

‘however’, toen ‘then’, or altijd ‘always’) preceded the target word. The target

adverb was followed by an adjective or a past participle which in 96% of the target

sentences was the last word of the sentence. In 4% of the items the adjective or the

past participle was followed by an obligatory complement. So that the connection

between clauses would make sense for both negative and positive experimental

sentences, the coordinating conjunction was varied between en ‘and’ and maar

‘but’; both conjunctions were presented in all six conditions on each list to prevent

predictability of negation or polarity type.

In order to avoid repetition effects, the 120 pairs of experimental sentences

(negated or affirmative) were assigned to two lists, so that each participant saw

only one version of each experimental item and saw an equal number of items in

each condition. Each list thus included 20 experimental items in each of the six

conditions. 120 fillers of various types were added to the experimental sentences in

each list to distract attention from the target sentences. The order was pseudo-

randomized so that each condition appeared evenly spread across the list in an

unpredictable sequence of conditions and fillers. The lists were then divided into

five blocks of 48 sentences each. The ratio of the number of fillers to the number of

experimental sentences was kept constant across blocks (1:1). Each block included

the same number of experimental items for each condition. Thirteen participants (8

female, 5 male) were assigned to list1 and 13 participants (7 female, 6 male) to list2.

2.3. Procedure

The experiment was run in a comfortably lit sound-attenuated room. Participants

were seated at a distance of 80–90 cm from the computer screen. Programming and

presentation of the experiment were done using E-prime (Psychology Software Tools

Inc., 2001). The sentences were presented word-by-word, in the center of the screen, in

black on a white background. At the beginning of each sentence a cross ‘‘þ ’’ appeared

for 500 ms to ensure that participants’ eyes were fixated at the center of the screen

when the initial word was presented. Each word appeared for 240 ms, followed by an

interstimulus interval of the same duration. At the end of each sentence the participant

was given the opportunity to blink when a row of asterisks was presented in the

middle of the screen for 1750 ms. Participants were instructed to read sentences for

comprehension. To check their attention, 33% of the sentences, both experimental and

filler, were followed by yes/no questions. Participants were instructed to answer by

pressing the keyboard buttons ‘‘1’’ (Yes) or ‘‘2’’ (No). A set of six training sentences was

followed by the five experimental blocks, each lasting 8–10 minutes, with short breaks

in between. The whole experiment lasted about an hour.

2.4. EEG recordings

The EEG was recorded using the Refa8 amplifier (TMS International) at a

sample rate of 125 Hz with a digital anti-aliasing low-pass filter of 35 Hz and

average reference. Measurements were recorded using 64 electrodes mounted in

an elastic cap (Electro-Cap International), according to the expanded International

10–20 system, plus two mastoid electrodes. The horizontal electro-oculogram

(EOG) was monitored with electrodes situated at the outer canthi of the eyes; the

vertical EOG was measured with electrodes above and below the left eye. The

ground electrode was placed on the sternum. Impedances were kept below 10 kO.

mooi tekenenbeautifully drawINF

niet/nook bijster leesbaarnot/nalso at all legible

riting was not/also at all legible.

moet worden aangescherptmust3SG.PRES be accentuatePART.PASS

nogal eensgezind.really/nnot rather unanimous

e parties are really/not rather unanimous.

met een rare print

ASS with a strange print

bijzonder chic.lly particularly chic

actually very chic.

Fig. 1. Waveforms showing the responses to negated sentences (dotted lines) and affirmative sentences (black lines) for the period of the baseline (�620 to 0 ms.) and

from the onset of the adverbs to 1220 ms after the presentation of the target word. Unacceptable conditions are marked with thicker lines. Panel A shows the response for

neutral adjectives; panel B shows the results for NPIs; and panel C shows the results for PPIs. Significant effects are labeled, Panel A: neutral targets in affirmative (black

line) and negative (dotted line) contexts, Panel B: negative Polarity targets in ungrammatical affirmative (black line) and negative (dotted line) contexts, Panel C: positive

Polarity targes in affirmative (black line) and ungrammatical negative (dotted line) contexts.

4 There were four participants with 4 electrodes left in the frontal ROIs. All 26

participants had 5 electrodes in the central ROIs. One participant had 4 electrodes

in the parietal ROIs.

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141136

2.5. Data analysis

The EEG-data were re-referenced offline to the average of the mastoid

electrodes and filtered with a high-pass filter at 0.01 Hz. The initial continuous

EEG signal was then segmented into epochs beginning 720 ms before the target

adverb and continuing for 1550 ms. thereafter, resulting in 120 segments of six

types. This epoch included the entire word preceding the target (niet or affirmative

particle), the target itself and the following word (adjective or participle), most of

which were final, plus some additional time.

Correction of blinks and horizontal eye movements were done individually for

each participant using the Gratton and Coles algorithm (Gratton, Coles & Dolchin,

1983). After ocular correction, trials containing other artifacts were detected

automatically and excluded from the analysis. Six participants were excluded from

further analysis because of excessive artefacts on the electrodes that were included

in the analysis reported below. For the remaining 26 participants, of the electrodes

which were included in the analysis, 2.24% of the epochs were excluded from the

analysis due to artefacts. No individual participant lost more than 15% of these data.

ERP waveforms were aligned using the 720 ms pre-stimulus interval as a

baseline. This is longer than usual; we chose it because the word (480 ms)

immediately preceding the target word differed between negative and affirmative

conditions. Our goal was to be able to determine if there were baseline problems

due to this difference and to counteract them. The entire baseline is thus included

in the figures below. After calculating the baseline, single participant averages

were computed for each electrode in each condition. Three midline regions of

interest (ROIs), frontal (Fpz, AFz, Fz) central (FCz, Cz, CPz) and posterior (Pz, POz, Oz)

were used in the analysis as well as six lateral ROIs, frontal left (Fp1, AF3, AF7,

F3, F5), frontal right (Fp2, AF4, AF8, F4, F6), central left (FC3, C1, C3, C5, CP3), central

right (FC4, C2, C4, C6, CP4), posterior left (P1, P3, P5, PO3, O1), and posterior right

(P2, P4, P6, PO4, O2). The midline and lateral ROIs were calculated as an average of

the electrodes included. Inclusion criteria for the participants were that for each

midline ROI all three electrodes after artifact rejection retained no less than 17 trials

(out of 20) in each condition, and that for lateral ROIs at least four electrodes in a

group met these criteria. If one electrode was excluded from a ROI due to too few

trials, the corresponding electrode from the homologous ROI was also excluded.4

Electrode within the ROI was thus not a factor; an average value of ERP amplitude

was computed for each region of interest per participant and this value was

included in the analysis.

ERP effects were analyzed using repeated measures ANOVAs for three time

windows: 300–500 ms, 500–800 ms, and 800–1100 ms. Analysis of both midline

and lateral groups included the factors target word Polarity (negative vs. positive

vs. neutral) and the factor Negation (negative vs. affirmative). In addition, for

analysis of the scalp distribution the factors Posteriority with three levels (frontal,

central, and posterior) and Hemisphere with two levels (left vs. right) were

included. Hemisphere was only included for the analysis of the lateral ROIs.

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141 137

Our main focus is on the extent to which the different levels of Polarity

respond differently to negative and affirmative contexts, thus we will report

significant interactions of these two effects (in conjunction with the scalp

distribution factors). We will not consider main effects of Polarity, since there

are clear differences between the lexical items which were tested, as to frequency

or length in syllables. Negation, on the other hand, is of interest, since one of the

issues we are concerned with here is the extent to which the results reported by

Saddy and colleagues (Drenhaus et al., 2005, 2006; Saddy et al., 2004) could be

attributed to the processing of negation. Therefore, we will describe this main

effect and its interactions with other factors if present. Since the scalp distribution

factors are only of interest if they interact with the experimental factors,

indicating the scalp distribution of these effects, we will only report relevant

interactions of these factors. When the assumption of sphericity is violated, the

Greenhouse-Geisser correction will be applied; we report original degrees of

freedom and corrected p-values. In addition to the main analysis, post hoc

analyses of the three levels of the factor Polarity (neutral, negative, and positive)

will be reported to reveal the nature of interactions between Polarity and

Negation. Effects are taken to be significant at po0.05; effects which approach

significance (po0.1) will also be reported for future reference, but not interpreted.

3. Results

3.1. Behavioral data

The analysis of behavioral data showed that all the partici-pants correctly answered at least 85% of the questions(Mean¼93.4%, s.d.¼3.16). That suggests that they were readingthe sentences attentively and comprehending the sense.

3.2. ERP data

As can be seen in the Fig. 1, there is a clear effect of negative vs.affirmative context for both the NPIs and the PPIs. However, thisdifference appears to start earlier for the NPIs, and go on for alonger time for the PPIs. The neutral items, on the other hand,show no difference between negated and affirmative versions.These visually apparent differences are supported by the statis-tical analyses which we carried out.

3.2.1. 300–500 ms time window

In this time window, at the lateral electrodes, Polarity interactedwith Negation and Posteriority (F(4,100)¼3.336, p¼0.036). Thisinteraction reflects the fact that for only the NPIs there was asignificant interaction of Negation with Posteriority (F(1,25)¼4.229,p¼0.045), and the main effect of Negation also approached sig-nificance (F(1,25)¼3.306, p¼0.081). There was no significant differ-ence between negative and affirmative forms for the neutralsentences (main effect and interactions with scalp distribution allF-valueso1), nor for the PPIs (all p-values40.1). The interaction ofPolarity with Negation was also apparent at the midline, although itfailed to reach significance (F(2,50)¼2.789, p¼0.071), with NPIsagain showing a stronger effect of Negation than the other two.

3.2.2. 500–800 ms time window

In this time window, we found a significant interaction betweenPolarity, Negation and Hemisphere (F(2, 50)¼5.766, p¼0.006). Again,neutral items showed no effect of negation (all p-values40.2). Thisinteraction resulted from a significant main effect of Negation forthe PPIs consisting of increased positivity in the negated version(F(1,25)¼17.292, po0.001), while NPIs showed a significant interac-tion between Negation and Hemisphere (F(1,25)¼13.485, p¼0.001),as the increased negativity for the unlicensed NPI was stronger overthe right hemisphere, accounting for the interaction with Hemi-sphere. There was also a main effect of Negation (F(1,25)¼6.559,p¼0.017). The midline analysis also showed a significant interactionof Polarity and Negation (F(2,50)¼5.500, p¼0.007).

3.2.3. 800–1100 ms time window

In this time window, the interaction between Polarity andNegation failed to reach significance (F(2,50)¼2.581, p¼0.086), norwere there significant interactions with topographic factors (p-values40.1). The observed tendency is presumably due to aremaining effect of negation which was apparent only for PPIs(F(1,25)¼5.378, p¼0.029). The midline effects were even weaker,although in the same direction.

To summarize, the presence or absence of negation in thepreceding context evoked a significantly different ERP pattern inpolarity but not neutral items. In the 300–500 ms time window,the effects of negation were significant only for NPIs (increasednegativity for the affirmative context). In the time window 500–800 ms, the effects were significant for both NPIs and PPIs;however, the effects for the violations differed in sign and hadsignificantly different scalp distributions: a broadly distributed

positivity for violation of the PPIs by local negation and alateralized negativity for the violation of NPIs by the lack ofnegation. The effects appear to continue somewhat longer forthe PPIs, into the 800–1100 ms time window, although theinteraction does not reach significance.

4. Discussion

In the present study, we investigated the processing ofnegative and positive polarity adverbs in Dutch using the ERPmethod. The goal was to see if we could replicate the dissociationbetween processes involved in the processing of PPIs and NPIsreported by Saddy et al. (2004). We used non-polar adverbs innegative and affirmative contexts to determine the effect thatnegation as such has on the processing of the following material.This is a concern because negation is known to make processingmore difficult, whether or not polarity is present (Clark & Chase,1972; Horn, 1989; Just & Carpenter, 1971; Wason, 1961). It canthus be expected to have an effect in addition to the processing ofpolarity and may obscure the effects of polarity per se. Addition-ally, we wanted to investigate the effects of polarity in the moststraightforward negative context because the structural positionor type of negation in the preceding context appears to modifythe strength of the violation and its perceived grammaticality(Hoeksema & den Ouden, 2005). Using this improved design, wefound a clear dissociation between the processing of PPIs and theprocessing of NPIs. This dissociation is independent of effects ofnegation, which were, somewhat surprisingly, minimal at thetarget adverb. We will discuss this result first and then return tothe processing of polarity items.

4.1. Processing negation

We did not find any effect of negation in the neutral condition,which is somewhat surprising, since negation is commonly knownto tax the processing system. However, previous ERP studies havealso not uniformly found an effect of negation. Nieuwland andKuperberg (2008) found no immediate effect of negation inmaterials similar to ours. Steinhauer et al. (2010) did not findany differences following a mixed bag of negative licensors forneutral control sentences, although there appeared to be a non-significant trend toward a central negativity. The well-knownprocessing difficulties for negated propositions have generally beenfound in studies testing indirect effects of negation in the mentalrepresentation of the information encoded in the proposition. Theclassic studies examined verification times or question answering,rather than the processing of the negated sentence itself (Clark &Chase, 1972; Just & Carpenter, 1971; Wason, 1961; see Horn, 1989,Chapter 3 for a review of these results). Accessing information from

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141138

a negated statement leads to longer response times in both of thesetasks. That does not necessarily mean that the difficulty is localizedin time to the negative element itself but rather may have more todo with the overall representation, so that the effects may show uplater or may not be time-locked to any particular event, which isnecessary to get an ERP effect.

However, as we mentioned in the introduction, Ludtke et al.(2008) did find a negative deflection in the ERP waveforms followingnegation within subject noun phrases like those used in Saddy et al.(2004) study. This suggests that at least this sort of negation isassociated with increased processing cost. Negation in the currentsentences may have simply been easier than negating noun phrases.The materials used by Ludtke et al. (2008) also differed in that thesubject was sentence final. Thus an alternative explanation of whythey found an effect may be that the effect of negation appliesglobally to the entire proposition rather than locally and is only seenat the end of the sentence. If this is true, an effect might haveoccurred later in our sentences at a time point that we did notexamine, although the waveforms in the figures do extend to the finalword and beyond for most sentences. The exact circumstancesunder which negation has an immediate effect thus require furtherinvestigation.

Of more immediate importance, the lack of an effect ofnegation in the neutral control sentences implies that the effectsof these two contexts on the processing of NPIs and PPIs have tobe attributed to the effects of integrating polarity items into theaffirmative or negative context.

4.2. Dissociating the processing of NPIs and PPIs

The difference between the processing of NPIs and PPIs is clear.In the 300 to 500 ms time window, the effects of negation weresignificant only for NPIs. In the following time window from 500to 800 ms, there were significant effects of violation for both theNPIs and PPIs, but these reflect different processes, because theeffect of violation for NPIs is negative, and the effect of violationfor PPIs is positive. In addition, these effects had different scalpdistributions. The effects appear to continue somewhat longer forthe PPIs, into the 800 to 1100 ms time window, although theinteraction was only marginally significant. We will discuss thetwo significant time windows separately.

4.2.1. N400 time window

The time period from 300 to 500 ms after presentation of aword is referred to as the N400 time window, because there istypically a negative deflection during this time window whichpeaks about 400 ms after presentation of a (visual) word. Theamplitude of this component is affected by various factors whichdetermine the activation of the word being read and/or the easewith which it can be integrated into the preceding context (forreview see Kutas & Federmeier, 2011).

The NPIs showed a significant difference between negative andaffirmative contexts in this time window. NPIs in an affirmativecontext (incorrect) compared to the negative context (correct)elicited a negativity with a central-posterior scalp distribution,which is characteristic of the N400. Saddy et al. (2004) found anegative deflection for NPIs in an affirmative context too, as didDrenhaus et al. (2005, 2006), Drenhaus, Blaszczak and Schutte(2007), Shao and Neville (1998) and Tesan, Johnson and Crain(2012). As pointed out in the introduction, the licensing condi-tions for NPIs are primarily semantic and are difficult to clearlydelimit. Although there has to be something ‘‘negative’’ in thesentential context, the source of the negation can arise from thecombinatorial semantics of the entire sentential context: it can beVP negation with a particle as in the current experiment, a wordwhich incorporates negation, like deny, the combination of chance

with fat or slim from elsewhere in the clause or even thepossibility that the proposition is not true, as in yes/no questions,or is not yet true, as with before. The position of the licensor hassome limitations, which are frequently stated syntactically: thelicensor (or part of the licensor as in ‘‘The chance that he will everget out of prison is pretty slim/ngood’’) has to be higher in thesyntactic tree. When there is simply no negation to license theNPI, the violation concerns the semantics only. Thus, it is notsurprising to see an N400 effect, as this component is extremelysensitive to the semantic congruity of a word to its context.

Nevertheless, several studies on NPIs have not found anysignificant difference between the affirmative and negative con-texts in the N400 time window (Drenhaus et al., 2007; Steinhaueret al., 2010; Xiang et al., 2009). Several of these studies usedrelatively complex structures, with distant licensors. Recall thatnative speakers judged sentences containing subject NP negation(geen) as less acceptable than niet (‘‘not’’), which occurs immedi-ately before the NPI (Hoeksema & den Ouden, 2005). If thisdifference is typical across languages, even the correct sentencemay elicit a larger negativity because the NPI is relativelyunexpected or difficult to integrate. This would have the effectof diminishing the difference relative to the violation sentences,with no significant N400 effect. This explanation is consistentwith data presented by Drenhaus et al. (2005, 2007) andSteinhauer et al. (2010). In their studies, the position and natureof the licensor or of the specific NPI have strong effects on the sizeof the N400 effect. Particularly Drenhaus et al. (2005) found thatthe wh-question licensor elicited a larger negativity on the NPIthan overt negation of the subject noun phrase.

In contrast to NPIs, processing of PPIs showed no evidencewhatsoever of an N400 effect in the current experiment, suggestingthat the two types of polarity are dealt with through differentprocesses. This result is to some extent contrary to Saddy et al.(2004), who found that there was a larger N400 for negativesentences (incorrect) than for affirmative sentences (correct). Thiswas explained as being due to a difficulty in processing similar tothat found for the NPIs. Clearly that difficulty is expressed as adifferent response in the current study. Note that there is not evena trend in PPIs toward a negativity for the PPIs; any difference inthis time window is in the opposite direction: a more positiveresponse to the incorrect sentences, although this is not significant.

There are two differences between the sentences which wereused in the Saddy et al. (2004) experiment and those which weused. First, in our study, the negation was placed immediatelybefore the target polarity items. As already discussed, PPIs withimmediately preceding negation are regarded as less acceptablethan the longer distance negation in subject noun phrases used bySaddy et al., and this may have altered the way in which readersresponded to some extent. According to the account of Saddy et al.(2004), it would be expected that this increased unacceptabilitywould cause a stronger N400 effect. It appears rather that the clearunacceptability led comprehenders to engage in one process only,which elicits the positive response most clearly seen in the latertime window.

A second difference between the two studies concerns thetasks which were used. The comprehension task which we used isless prone to elicit experiment-specific strategies than the taskemployed by Saddy et al. Although this is a possible explanation,the N400 response is in general quite robust during experimentsin which participants are simply asked to read for comprehension,so this appears to be a less likely explanation.

4.2.2. P600 time window

From 500 ms after the presentation of a word which does not fitinto the preceding context, the most frequently reported response

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141 139

is a positive deflection generally called the P600 due to its polarityand approximate latency. This effect was originally reported forsyntactic violations, i.e., words which do not allow the compre-hender to construct a grammatical structure (Osterhout &Holcomb, 1992). However, it has since been demonstrated that apositivity is also present when comprehenders are forced to choosea less common alternative structure for a temporarily ambiguoussequence of words (garden path sentences: Osterhout, Holcomb &Swinney, 1994) and when sentences are syntactically more com-plex (Kaan, Harris, Gibson & Holcomb, 2000). Although at firstregarded as a primarily syntactic response, many researcherscurrently regard the P600 as a reflex of a much more generalprocess (e.g., Coulson et al., 1998; Gunter et al., 1997). More recentresearch has shown that P600-like effects are sometimes elicitedby discourse processing problems (Burkhardt, 2007; Kaan et al.,2006) and a range of other integration difficulties which are notlinguistically characterized as syntactic violations (Hoeks, Stowe &Doedens, 2004; Kuperberg, 2007; see Brouwer et al., 2012, for anoverview).

Both NPIs and PPIs showed significant effects in the early partof this time window (500–700), but these responses were clearlydistinct from each other. The PPIs showed something whichresembles the classic P600, while the NPIs showed a negativityto the violation sentences continuing from the previous timewindow. The distribution of the positivity elicited by PPIs in anegative context is similar to the P600, a broadly distributedeffect. The only other study that we are aware of which hasexamined PPIs, Saddy et al. (2004), also found that positivepolarity violations were characterized by a P600 effect. Saddyet al., following the original view that the P600 primarily reflectssyntactic processing, suggested that this effect might reflect anattempt at syntactic repair. Sentences like (2a) can be describedas grammatically incorrect because the positive polarity adverb isin the scope of negation. That is, positive polarity violation maycause an attempt to repair the sentence so that the positivepolarity adverb is out of the scope of the negative particle.

However, as discussed in the introduction, syntactic restric-tions do not seem to be central to the licensing of PPIs (Baker,1970; Hoeksema, 2010; Horn, 1989, 2001; Israel, 1996, 2011;Szabolcsi, 2004: Seuren, 1976, 1985). Rather a local negativity canbe quite acceptable if the entire message is affirmative as in (2b)and (3b), reproduced here for convenience as (11 a and b),

(11) a. *(I cannot believe that) Fred is not still drunk.b. *(A: Fred is still drunk.

B: No,) Fred is NOT still drunk.

The negated clause is unacceptable out of context, but giventhat the rest of the sentence or discourse context makes theproposition semantically affirmative, it is fine. Given these facts,when confronted with local negation and a PPI the comprehendershould search a wider context to construct a globally positiverepresentation. This suggests that a discourse-based strategy isthe most productive in normal processing. Since a number ofrecent studies have suggested that discourse processes may alsoproduce P600-like effects (Burkhardt, 2007; Kaan et al., 2006),this seems to be a more likely explanation of the P600 seen inboth the current experiment and the experiment by Saddy et al.(2004). Local negation is not regarded as semantically incorrect, itis merely regarded as a reason to invoke a wider search for anappropriate context.

A syntactic repair strategy would have been more plausible forthe NPIs, as these can readily be described as having syntacticlimitations on the position of the licensor. However, no P600effect was found for negative polarity adverbs in affirmativecontexts in the current study. The absence of any positivity in

this time window suggests that the integration problem fornegative polarity violations does not lead to such a syntax-based process. However, a P600 effect was found for NPIs inseveral other studies in which there is a potential negativelicensor that is not in a syntactically appropriate position tolicense the NPI (Drenhaus et al., 2005, 2006). The P600 in thesecases may indeed reflect an attempt at syntactic repair.

To what extent then, is it possible to argue for completelydifferent processing for the two sorts of polarity items? Finding aP600 effect for NPIs is not restricted to cases where syntacticreanalysis of an illusory licensor may occur. In contrast to theclear dissociation between the two types of polarity items seenhere, there are several studies showing P600 effects for NPIs (Shao& Neville, 1998; Steinhauer et al., 2010; Xiang et al., 2009). Saddyet al. (2004) also reported no P600 effect for NPI violations,although Drenhaus et al. (2006) carried out a reanalysis suggest-ing that there may have been a small but significant positivity,completed earlier than that elicited by PPIs. These results mightsuggest that there is a positivity, which in our study is simply notstrong enough to be seen as a statistically significant effect.Drenhaus et al. (2006) explicitly argue for this sort of explanationin their own data. However, the predicted positivity does notsimply fail to reach significance in our study; there is a statisti-cally significant negativity instead.

The studies which found a P600 all required participants tomake an active grammaticality decision while reading the sen-tences. It has been shown (Coulson et al., 1998; Kaan & Swaab,2003; McKinnon & Osterhout, 1996) that the grammaticalityjudgment task can increase the amplitude of the P600 potential,particularly if it is difficult to make the decision and may evenintroduce a positivity that is not apparent when no task is carriedout (Osterhout & Mobley, 1995). This means that the ungramma-tical condition is likely to be associated with a positivity, as thedecision depends on considering more possibilities than thegrammatical case. It also suggests that the positivity may beassociated with strategic decision-making processes rather thanthe processes which support normal sentence processing. Evenmore convincing than the effects of the decision per se, instruct-ing participants that ungrammaticalities are totally irrelevant tothe task is likely to eliminate the effect entirely (Hahne &Friederici, 2002), which strongly supports the strategic accountof at least some P600 effects. This seems to be the most likelycandidate to explain the difference between our results and thoseof the studies reporting a P600 effect for NPIs.

However, there is a more interesting alternative, which stemsfrom the fact that the materials used in the previous experimentswere generally more complex than the materials in the presentstudy. The experimental sentences presented by Saddy et al.(2004) contained an embedded relative clause that preceded thetarget polarity items; those of Shao and Neville (1998) containedpolarity items in embedded clauses. Thus, the P600 effectobserved in these studies might reflect additional cost related tothe attempt to integrate the NPIs in the complex context (Kaanet al., 2000), or it might reflect the fact that interpretation of thesecomplex sentences is more difficult in general.

The interpretation of the negativity elicited by the NPIs in thistime window is not entirely clear. It should not be interpreted as astraightforward prolongation of the N400 effect, as it has a rightlateralized scalp distribution rather than a central posteriormaximum, as in the N400 time window. The difference in scalpdistribution indicates that it was not generated by the same brainsources as those active during the previous time window, andthus that the nature of the process is different from that in theN400 time window. This lateralized negativity might be similar tothe right frontal negativity in this time window reported bySteinhauer et al. (2010). They suggested that the effect might

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141140

indicate semantic working memory demands (Mecklinger,Schriefers, Steinhauer & Friederici, 1995; Munte, Schiltz & Kutas,1998). None of the other NPI studies which we are aware of havereported a similar effect.

5. Conclusion

To sum up, our ERP study confirms that NPIs and PPIs areprocessed differently, even 1) in fairly simple contexts, 2) wherethe responses are not muddied by a possible response to negationper se, and 3) with a task which is not likely to encourageconscious decision processes, which may affect the outcome.Positive polarity adverbs in negative context elicit a P600 effectwhile negative polarity adverbs in affirmative context elicit anN400 effect. As can be seen in our discussion of previous results, anumber of factors appear to play a role in the exact pattern ofresponses to these two types of polarity items. Abstracting awayfrom these differences however, the important point is that bothstudies which have compared processing of NPIs and PPIs in thesame participants found that the participants dealt with the twotypes of polarity in substantially different ways. In fact, thecurrent experiment shows an even clearer dissociation betweenthe two than the Saddy et al. (2004) study.

We suggest that the difference between the two types ofpolarity is due to the substantial difference in the kind ofinformation that is relevant to licensing the two types of polarity.For NPIs, negation within a rather limited domain is important,although the form the negation takes can vary rather freely. ForPPIs, on the other hand, the entire discourse context is relevant tojudging whether the PPI is appropriate. If this viewpoint iscorrect, we should be able to find interesting effects of discoursecontext on the processing of PPIs in locally negative contexts.

Many other issues touched on here remain open for furtherresearch. We have already alluded to effects of the structuraldistance of the negative licensor, but equally interesting might bea deeper investigation of differences among individual NPIs (seeinter alia Gajewski, 2011; Giannakidou, 2011; Hoeksema, 2012;Steinhauer et al., 2010; Zwarts, 1998) and of differences amongvarious types of negative contexts, apart from how they affectstructural distance. Alternatively, we have suggested that whenNPIs occur in a complex structure with no local licensor, thesearch through a wider structural domain may elicit a P600response as well. These are all issues for future research.

Acknowledgments

Thanks to Hanneke Loerts and Nienke Meulman for help inconstructing the materials and to Tam Ho for analysis of a pilotversion of the study. We also acknowledge the valuable contribu-tions of two anonymous reviewers.

References

Baker, C. L. (1970). Double negatives. Linguistic Inquiry, 1, 169–186.Brouwer, H., Fitz, H., & Hoeks, J. (2012). Getting real about semantic illusions:

Rethinking the functional role of the P600 in language comprehension. BrainResearch, 1446, 127–143.

Burkhardt, P. (2007). The P600 reflects cost of new information in discoursememory. NeuroReport, 18, 1851–1854.

Clark, H. H., & Chase, W. G. (1972). On the process of comparing sentences againstpictures. Cognitive Psychology, 3, 472–517.

Coulson, S., King, J. W., & Kutas, M. (1998). Expect the unexpected: Event-relatedbrain response to morphosyntactic violations. Language and Cognitive Pro-cesses, 13, 21–58.

Dimitrova, D. V., Stowe, L. A., Redeker, G., & Hoeks, J. C. J. (2012). Less is not more:Neural responses to missing and superfluous accents in context. Journal ofCognitive Neuroscience, 24(12), 2400–2418.

Drenhaus, H., Beim Graben, P., Saddy, D., & Frisch, S. (2006). Diagnosis and repairof negative polarity constructions in the light of symbolic resonance analysis.Brain and Language, 96, 255–268.

Drenhaus, H., Blaszczak, J., & Schutte, J. (2007). Some psycholinguistic commentson NPI licensing. In: E. Puig-Waldmuller (Ed.), Proceedings of Sinn undBedeutung, 11 (pp. 180–193). Barcelona, Spain: Universitat Pompeu Fabra.

Drenhaus, H., Frisch, S., & Saddy, D. (2005). Processing negative polarity items:When negation comes through the backdoor. In: S. Kepser, & M. Reis (Eds.),Linguistic evidence: Empirical, theoretical, and computational perspectives (pp.145–163). Berlin, Germany: Mouton de Gruyter.

Fauconnier, G. (1978). Implication reversal in a natural language. In: F. Guenthner,& S. J. Schmidt (Eds.), Formal semantics and pragmatics for natural languages(pp. 289–301). Dordrecht, The Netherlands: D. Reidel.

Gajewski, J. (2011). Licensing strong NPIs. Natural Language Semantics, 19, 109–148.Giannakidou, A. (1998). Polarity sensitivity as (non)veridical dependency. Amster-

dam: John Benjamins.Giannakidou, Anastasia (2011). Positive polarity items and negative polarity

items: variation, licensing, and compositionality. In: C. Maienborn, K. vonHeusinger, & P. Portner (Eds.), Semantics: An International Handbook of NaturalLanguage Meaning (2nd ed.). Berlin: Mouton de Gruyter.

Gratton, G., Coles, M. G. H., & Dolchin, E. (1983). A new method for off-line removalof ocular artifact. Electroencephalography and Clinical Neurophysiology, 55,468–484.

Gunter, T. C., Stowe, L. A., & Mulder, G. (1997). When syntax meets semantics.Psychophysiology, 34, 660–676.

Hahne, A., & Friederici, A. D. (2002). Differential task effects on semantic andsyntactic processes as revealed by ERPs. Cognitive Brain Research, 13, 339–356.

Harris, T., Wexler, K., & Holcomb, P. (2000). An ERP Investigation of Binding andCoreference. Brain and Language, 75, 313–346.

Hoeks, J. C. J., Stowe, L. A., & Doedens, G. (2004). Seeing words in context: Theinteraction of lexical and sentence level information during reading. CognitiveBrain Research, 19, 59–73.

Hoeksema, J. (2000). Negative polarity items: Triggering, scope and c-command.In: L. R. Horn, & Y. Kato (Eds.), Negation and polarity: Semantic and syntacticperspectives (pp. 115–146). Oxford, England: Oxford University Press.

Hoeksema, J. (2010). Negative and positive polarity items: An investigation of theinterplay of lexical meaning and global conditions on expression. In: L. R. Horn(Ed.), The expression of negation (pp. 187–224). Berlin, Germany: Mouton deGruyter.

Hoeksema, J. (2012). On the natural history of negative polarity items. LinguisticAnalysis, 38(1/2), 3–33.

Hoeksema, J., & den Ouden, D.-B. (2005). Positief- en negatief-polaire bepalingenvan graad: Een empirisch onderzoek. Tabu: Bulletin voor taalwetenschap, 34(3/4), 129–144.

Horn, L. R. (1989). A Natural History of Negation. Chicago: Chicago University Press.Horn, L. R. (2001). Flaubert triggers, squatitive negation and other quirks of

grammar. In: J. Hoeksema, H. Rullmann, V. Sanchez-Valencia, & T. van derWouden (Eds.), Perspectives on Negation and Polarity Items (pp. 173–200).Amsterdam: John Benjamins.

Israel, M. (1996). Polarity sensitivity as lexical semantics. Linguistics and Philoso-phy, 19, 619–666.

Israel, M. (2011). The grammar of polarity: pragmatics, sensitivity, and the logic ofscales. Cambridge: Cambridge University Press.

Just, M. A., & Carpenter, P. A. (1971). Comprehension of negation with quantifica-tion. Journal of Verbal Learning and Verbal Behavior, 10, 244–253.

Kaan, E., Dallas, A. C., & Barkley, C. M. (2006). Processing bare quantifiers indiscourse. Brain Research, 1146, 199–209.

Kaan, E., Harris, A., Gibson, E., & Holcomb, P. J. (2000). The P600 as an index ofsyntactic integration difficulty. Language and Cognitive Processes, 15, 159–201.

Kaan, E., & Swaab, T. Y. (2003). Repair, revision, and complexity in syntacticanalysis: An electrophysiological differentiation. Journal of Cognitive Neu-roscience, 15, 98–110.

Klima, E. S. (1964). Negation in English. In: J. A. Fodor, & J. J. Katz (Eds.), Thestructure of language: Readings in the philosophy of language (pp. 246–323).Englewood Cliffs, N.J.: Prentice-Hall.

Krifka, M. (1995). The semantics and pragmatics of polarity items. LinguisticAnalysis, 25, 209–257.

Kuperberg, G. R. (2007). Neural mechanisms of language comprehension: chal-lenge to syntax. Brain Research, 1146, 23–49.

Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaningin the N400 component of the event-related brain potential (ERP). AnnualReview of Psychology, 62, 621–647.

Ladusaw, W. A. (1979). Polarity sensitivity as inherent scope relations (Doctoraldissertation). Austin, TX: University of Texas at Austin.

Linebarger, M. (1987). Negative polarity and grammatical representation. Linguis-tics and Philosophy, 10, 325–387.

Ludtke, J., Friedrich, C. K., de Filippis, M., & Kaup, B. (2008). Event-related potentialcorrelates of negation in a sentence-picture verification paradigm. Journal ofCognitive Neuroscience, 20, 1355–1370.

McKinnon, R., & Osterhout, L. (1996). Constraints on movement phenomena insentence processing: Evidence from event-related brain potentials. Languageand Cognitive Processes, 11, 495–523.

Mecklinger, A., Schriefers, H., Steinhauer, K., & Friederici, A. D. (1995). Processingrelative clauses varying on syntactic and semantic dimensions: An analysiswith event-related potentials. Memory and Cognition, 23, 477–494.

A. Yurchenko et al. / Neuropsychologia 51 (2013) 132–141 141

Munte, T. F., Schiltz, K., & Kutas, M. (1998). When temporal terms belie conceptualorder. Nature, 395, 71–73.

Nieuwland, M. S., & Kuperberg, G. R. (2008). When the truthis not too hard tohandle: An event-related potentials study on the pragmatics of negation.Psychological Science, 19, 1213–1218.

Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited bysyntactic anomaly. Journal of Memory and Language, 31, 785–804.

Osterhout, L., Holcomb, P. J., & Swinney, D. A. (1994). Brain potentials elicited bygarden-path sentences: evidence of the application of verb information duringparsing. Journal of Experimental Psychology: Learning, Memory, and Cognition,20, 786–803.

Osterhout, L., & Mobley, L. A. (1995). Event-related brain potentials elicited byfailure to agree. Journal of Memory and Language, 34, 739–773.

Progovac, L. (1994). Negative and positive polarity: a binding approach. Cambridge:Cambridge University Press.

Saddy, D., Drenhaus, H., & Frisch, S. (2004). Processing polarity items: Contrastinglicensing costs. Brain and Language, 90, 495–502.

Sailer, M., & Richter, F. (2002). Not for love or money: collocations. In: G. Jager,P. Monachesi, G. Penn, & S. Wintner (Eds.), Proceedings of Formal Grammar 2002(pp. 149–160).

Seuren, P. A. M. (1976). Echo: Een studie in negatie. In: G. Koefoed, & A. Evers(Eds.), Lijnen van taaltheoretisch onderzoek (pp. 160–184). Groningen: TheNetherlands: Tjeenk Willink.

Seuren, P. A. M. (1985). Discourse Semantics. Oxford: Blackwell.

Shao, J., & Neville, H. (1998). Analyzing semantic processing using event-relatedbrain potentials. The Newsletter of the Center for Research in Language, 11(5),3–20.

Steinhauer, K., Drury, J. E., Portner, P., Walenski, M., & Ullman, M. T. (2010). Syntax,concepts, and logic in the temporal dynamics of language comprehension:Evidence from event-related potentials. Neuropsychologia, 48, 1525–1542.

Szabolcsi, A. (2004). Positive polarity–negative polarity. Natural Language andLinguistic Theory, 22, 409–452.

Tesan, G., Johnson, B. W., & Crain, S. (2012). How the brain responds to any: AnMEG study. Brain and Language, 120, 66–72.

van der Wouden, T. (1997). Negative contexts: Collocation, polarity and multiplenegation. London: Routledge.

Vespignani, F., Panizza, D., Zandomeneghi, P., & Job, R. (2009). Never in the wrongplace: An ERP study of unlicensed NPIs in Italian. Proceedings of the 22nd annualCUNY conference on human sentence processing. Davis, CA: University ofCalifornia p. 219.

von Fintel, K. (1999). NPI-licensing, Strawson-entailment, and context-dependency. Journal of Semantics, 16, 97–148.

Wason, P. C. (1961). Response to affirmative and negative binary statements.British Journal of Psychology, 52, 133–142.

Xiang, M., Dillon, B., & Phillips, C. (2009). Illusory licensing effects acrossdependency types: ERP evidence. Brain and Language, 108, 40–55.

Zwarts, F. (1998). Three types of polarity. In: F. Hamm, & E. W. Hinrichs (Eds.),Plurality and quantification (pp. 177–238). Dordrecht: Kluwer.