As the name suggests, the N400 is a negative-going waveform that peaks at approximately 400 ms, with a primarily centro-posterior scalp distribution. The amplitude of the N400 evoked by an incoming word indexes the degree to which that word's semantic features match semantic features that have been pre-activated in the context at the time of encounter (Lau et al., 2008; Kutas and Federmeier, 2011; Kuperberg, 2013). The term “pre-activation” has often been associated with active prediction of specific lexical items, but Xiang et al. (2016) use it to refer more generally to the activation of relevant semantic features, regardless of whether active prediction or expectation of the upcoming word is at work.

Lau et al. (2008) offer a very lucid discussion of how the N400 can be used to reflect semantic effects related to “anomaly” and “expectation”—both relevant for NPI licensing as we saw at the beginning of the paper. The N400 response involves the presentation of a congruent or incongruent word before a word target (such as “coffee–tea” or “chair–tea”). A “semantically supportive” context elicits a response of smaller amplitude in the 300–500 ms interval, and although the effects of sentential context on the N400 response may be bigger in magnitude, collectively, we refer to this modulation as the “N400 effect.” Lau et al. state further that the amplitude of the N400 response “is modulated not only by the degree of anomaly per se, but also by predictability. [emphasis ours]. A less expected sentence endings generate a larger N400 [emphases ours] response than highly expected ones, even when both endings are semantically congruent (for example, “I like my coffee with cream and honey” would generate a larger N400 response than “I like my coffee with cream and sugar”) (Lau et al., 2008 p. 921).

In the context of NPI licensing, the (negative or nonveridical) licenser establishes an expectancy that, given the above, predicts an N400 response on the NPI. Expectancy and predictability explain why lexical anomalies (“I like my coffee with cream and sugar/socks,” “Apples grow on noses”) typically show the effect, but it would be erroneous to state that N400 indexes merely semantic anomaly (as it has indeed been stated in the past). The N400 is expected to show up in more patterns of incongruence, as indeed is the case (see Lau et al. for data and specific references). For NPIs, the presence of the N400 will be seen in this light, in particular indexing semantic matching. Following Giannakidou (1997) and Giannakidou and Quer (2013), it is reasonable to assume that the NPI contains a lexical polarity feature that marks it as a dependent item sensitive to negation. The following is a definition for NPI after Giannakidou and Quer (2013):

In the denotation, the NPI contains a non-deictic dependent variable in need of binding.

We can also think of the dependent variable as a variable that cannot introduce a discourse referent (or, cannot be closed by text level existential closure, as suggested in Giannakidou, 1998). Such a variable won't be able to get a value from the context, unlike non-dependent variables that can, and will always appear to be “narrow scope”: its distribution will be constrained in contexts where there is an operator it can be bound by. The presence of a dependent variable thus underlies the very essence of NPI-hood.

The presence of a dependent variable creates limited distribution. The dependent variable class includes NPIs—but also non-polarity variables such as reflexive pronouns, traces, distributivity markers (reduplicated indefinites in Hungarian; Farkas, 1998), the temporal variable of the subjunctive mood (“temporal” polarity in Giannakidou, 2009), and as recently argued in Grano (2011), subjects of exhaustive control verbs such as try, manage, etc. This framework imposes an isomorphism between semantics (dependent variable that cannot remain free) and morphosyntax (a dependent variable being a distinct syntactic object from a non-dependent variable). Being a distinct syntactic object means, by licensing, that the NPI has a polarity feature. This polarity feature POL is subject to a matching requirement (Agree) with the licenser. In other words, the non-deictic variable dependency is lexically encoded in the POL feature. Polarity features have been implied for NPIs since the early days (e.g., Klima's +affective feature, Giannakidou, 1997 sensitivity features, Zeijlstra's NEG feature; Chierchia's +σ feature is within the same spirit)⁶.

With NPIs that are more narrowly sensitive to negation, such as e.g., n-words, or the Dutch NPI hoeven “need,” it is reasonable to assume that the abstract lexical semantic feature is not POL but [+Neg] (see Lin, 2015 for some recent discussion on the acquisition of this feature and its contrast with acquisition of broader NPIs such as any). In the literature on n-words it is very common to assume [+NEG] (Zeijlstra, 2004; Giannakidou and Zeijlstra, 2017 for an overview). In any case, POL and NEG would be the lexical indexing of the NPI-dependency in the grammatical representation of the NPI word. During the incremental comprehension of a sentence, if NEG or POL are compositionally derived prior to encountering the NPI (i.e., in the licenser), that should lead to a reduced N400 on the NPI. This hypothesis relies on the fact that the amplitude of the N400 evoked by an incoming word indexes the degree to which that word's semantic features match semantic features that have been preactivated by its context at the time of encounter (Lau et al., 2008; Kutas and Federmeier, 2011; Kuperberg, 2013). The term “pre-activation” has often been associated with active prediction of specific lexical items. (Xiang et al., 2016) use it in a more neutral sense to refer to the activation of relevant semantic features, regardless of whether active prediction or expectation of the upcoming word is at work, ahead of encountering the full linguistic input. In the context of NPI licensing, it is reasonable to assume that, during the incremental comprehension of a sentence, if a semantic [+NEG] feature or [+POL] feature is compositionally derived prior to encountering the NPI, that should lead to a reduced N400 on the NPI word.

In a series of studies (Saddy et al., 2004; Drenhaus et al., 2005, 2006, 2007), a reduced N400 with a central maximum was found on the German NPI jemals (“ever”) when it was licensed by negation, compared to the ungrammatical counterpart when jemals was not licensed. Similar N400 effects were also found for Dutch (Yurchenko et al., 2013) and English NPIs (Shao and Neville, 1998), a MEG study by Tesan et al. (2012), Xiang et al. (2009), and Xiang et al. (2016). The studies differed considerably in their aims, materials and experimental designs, behavioral task, and test language; but the N400 finding is in line with the idea that NPI licensing involves a semantically driven dependency and suggests that the semantic requirement of the NPIs involves some kind of lexical/morphological feature matching.

Interestingly, another study by Steinhauer et al. (2010) did not find an N400 difference between licensed and unlicensed ever⁷; but a crucial difference between their study and the others mentioned above is that Steinhauer et al. had a larger set of licensors in their stimuli, including various negative licensors such as not, without, rarely/hardly, and also licensers that are not negative per se, but non-veridical, such as every, before, whether, and yes-no questions. It is possible that negative features are only present with classically negative expressions (no, no one, not), and that the varying degree of negativity (or no negativity at all, e.g., with a non-veridical non-negative licenser) causes the N400 effect to be reduced. In Steinhauer et al. the effect could have been watered down by using both negative and non-negative licensors. This, by itself, of course raises the question of how to trace the judgment in the case of non-negative licensers, and this is something that needs to be studied. Overall, what we want to say here is that reduced N400 can be plausibly viewed as a correlate of the underlying lexical semantic matching dependency between the NPI and the licenser that produces an expectancy in the sense of Lau et al. (2008). The range of data confirming this is solid enough to be able to render the N400 effect a predictor. The weakened N400 with non-negative licensers observed in Steinhauer et al. may be a reflection of a matching between an NPI a non-negative licenser, either, as we said, because the non-negative licensers lack [NEG], or because it might have only [+POL]. If the NPI is [+NEG], but the licenser is only [+POL], this is a weaker match. In other words, there appears to be a hierarchy of strength of these lexical features.

At the same time, the presence of P600 effect with NPIs reflects syntactic integration, and the consistent presence of P600 allows us to treat these two components as tied to syntactico-semantic processing.

The majority of the NPI studies mentioned above also reported a posteriorly distributed P600 late positivity effect, which was larger for unlicensed NPIs than for licensed ones⁸. The P600 effect is typically associated with syntactic processing and syntactic complexity, and is reliably elicited by syntactic errors (Osterhout and Holcomb, 1992; Hagoort et al., 1993; see also Osterhout et al., 1994; Kaan et al., 2000; Phillips et al., 2005), or grammatical but syntactically complex constructions (Osterhout et al., 1994; Kaan et al., 2000; Phillips et al., 2005; Gouvêa et al., 2010). Although the precise functional interpretation of the P600 is yet to be determined, a broad generalization that has emerged is that it reflects costs associated with a processing stage in which information from different sources is integrated into one coherent representation. We will therefore take P600 to be an indicator of integration cost (Friederici and Weissenborn, 2007; Kuperberg, 2007; Bornkessel-Schlesewsky and Schlesewsky, 2008; Van Petten and Luka, 2012). Increased P600 amplitudes signal the detection of an integration error or difficulty⁹.

In the particular context of NPI licensing, multiple streams of information—syntactic, semantic, and in some cases pragmatic, as we mention next—are recruited to construct a grammatical representation that can license NPIs. In an ungrammatical sentence that does not license NPIs, the comprehension system fails to integrate an NPI into the current grammatical representation, and therefore produces a large P600¹⁰.

Crucially, sometimes NPIs are sanctioned pragmatically, by implicit negation. Giannakidou (2006) calls this phenomenon rescuing and it is very clear with emotive verbs:

(21) a. John was surprised that Mary has any friends.

In these sentences, any appears without a negative or nonveridical licenser; the emotive verb (be surprised, be lucky, regret), arguably, has a positive thus veridical presupposition that the complement is true. These are, at best, cases of mixed veridicality (Giannakidou and Mari, 2018), but some NPIs, as we see, appear nevertheless (see Baker, 1970; Linebarger, 1987 for earlier data). Giannakidou shows, on the other hand, that the corresponding Greek NPIs are ungrammatical:

(22) ^*Metaniosa pou ipa tipota.

As mixed licensers, emotive verbs are highly variable, as confirmed by a recent paper (Duffley and Larrivée, 2015) where it is shown that the appearance of any is indeed limited with emotives. Giannakidou argues that with emotives we don't have licensing proper but rescuing of the NPI by accessing implicit, i.e., not asserted, negation.

Rescuing is proposed by Giannakidou a secondary mode of licensing that relies on pragmatic inferencing from the global context, which includes the presuppositions and implicatures of the sentence. Horn (2002) further structures global pragmatics with assertoric inertia: one component becomes assertorically inert, and another becomes salient. If the salient component contains negation, NPIs will be licensed. Emotive verbs give rise to a negative inference that has been characterized as an implicature (Linebarger, 1987), or presupposition (Baker, 1970; Giannakidou, 2006, 2016). In the case of emotives, then, the negative meaning arises not from a logical property of the emotive verb—which would render a NEG or POL feature possible—but from implicit negation (regret implicates or presupposes that I wish I didn't). With rescuing, the NPI needs to access the pragmatic level of representation.

In agreement with the rescuing idea, processing literature treats the appearance of NPIs with emotives as non-canonical, and uses labels such as illusory effect (Xiang et al., 2009, 2013), and erroneous pragmatic licensing (Yanilmaz and Drury, 2018). Giannakidou (1998) calls rescuing indirect licensing; Xiang et al. (2016) further study how licensing proper differs from non-canonical licensing in online computation. They look at the P600 effect. If the P600 indexes the integration effort with which an NPI is licensed, it provides a useful tool to examine whether or not negation in the pragmatics is treated by the comprehension system as an equally viable licenser.

Combining observations from the N400 and the P600 time windows, Xiang et al. state that they can construct a complete picture as to when and how negation is computed and used for grammatical purposes. The N400 reveals information about whether a negative meaning is established incrementally in context; the P600 assesses whether negative meaning, if available, can be immediately adopted to serve the grammatical function of NPI licensing.

Xiang et al. in their experiment 1 found the following. All (no, few, only) conditions that contain a legitimate licensor showed, expectedly, a qualitatively similar N400 reduction during the 300–400 ms time window at the critical NPI, relative to the unlicensed condition. They also by and large showed a reduced anterior negativity compared to the unlicensed condition during the late 700–900 ms time window. However, during the P600 time window, although conditions under no, few, and only showed qualitatively similar patterns involving a smaller P600 amplitude relative to the unlicensed condition, the emotive predicate condition yielded a P600 as large as the P600 in the unlicensed condition. This result sets rescuing with emotive predicates apart from licensing proper, and supports the thesis that rescuing does not involve syntax-semantic integration (as that would predict no P600). Trying to access negation in the pragmatics produces the effect. Giannakidou (1998, 2006) is the only currently available theory that predicts this fact, as it is the one that posits two qualitative distinct modes of licensing (licensing proper which involves syntax-semantics integration, vs. rescuing with involves pragmatics).

Let us note that Xiang et al. (2016) consider the possibility that the P600 effect may be due to fact that with emotives the NPI is embedded in a different clause, unlike in the other conditions where licenser and NPI are in the same clause (No student said anything vs. Maria regrets that she said anything). This is also a question raised by a reviewer. Xiang et al. therefore designed a second, self-paced reading experiment aiming to examine whether the ERP pattern could be replicated in a different paradigm, and also to assesses whether the additional processing cost found on the emotive condition is due to its NPI licensing properties or to the other possible sources of processing complexity (such as e.g., embedding). They conducted a 2 × 5-design self-paced reading study, the results of which rule out the possibility that the observed effects among the NPI conditions should be attributed to independent structural or contextual differences among different conditions. Of course, caution needs to be taken in drawing parallel relations between ERP and self-paced reading results; but the fact that the same costs, with the same relative timing, are observed in the NPI conditions from both the ERP data and the reading time data suggests that they are comparable measures to examine the online processing of NPI comprehension (see Xiang et al., 2016 for the experimental details).

What are the overall conclusions from this discussion? We suggest the following:

Experimental, specifically ERP, methodologies are useful in establishing physical, quantitative correlates of interface intuitions that can serve as criteria for distinguishing aspects of the linguistic judgment.

Let us remind again that the N400 is not an index of felicity (as one would think, for instance by reading only, Shao and Neville, 1998), but of predictability, expectancy and matching. Overall, the ERP methodology allowed us to disentangle the key aspects of grammatical judgment with NPI licensing, in ways that would have been impossible with mere intuition. This, we believe, is a promising result—sufficient in itself to generate and enhance interest in pursuing these methodologies further for such types of phenomena.

Before we move on to different kinds of experiments, we want to offer a few comments on why, we think, ERP methodology is useful for NPI licensing, or, in other words, why tracking processing in time matters for this type of phenomenon. As it became clear, NPI licensing is a long-distance dependency: it requires ability of the processor to integrate material that may not be locally adjacent (No student saw anything). Other well-known long distance dependencies are wh-dependencies (Who did Ariadne see t?), and the antecedent—anaphor relations (Ariadne likes talking to herself) that we talked about at the beginning. In all these cases, e.g., upon encountering the NPI, the trace, or the anaphor, the processor must assess the structure that has already been processed. This raises the question of how structured representations are encoded in memory, and how representations are retrieved to extract information. Hierarchically structured representations must be tracked during language processing in order for the parser to “accurately single out grammatically licit antecedents, and representations of structure in memory must be organized in such a way that retrieval operations can make appropriate decisions about acceptable or unacceptable targets” (Xiang et al., 2009, p. 40). This entails that observing processing in time is an effective technique for assessing long-distance phenomena—and although the three phenomena mentioned here are distinct, they all involve explicit reference to previously processed lexical items and structure, hence they benefit from tracking processing on time. ERP methodology can thus provide a secure take, we believe, on this type of syntax-semantics integration.

We move on now to different methodology. We show that mere acceptability judgment tasks can also be useful in revealing more sharpened intuitions with NPIs.

It has been a common observation that Greek exhibits a difference between two variants of NPIs distinguished by “emphatic accent” (Veloudis, 1982; Giannakidou, 1997 et seq., Tsimpli and Roussou, 1996); upper-case indicates the obligatory presence of prosodic prominence in a phrasal context. The emphatic NPI is interpreted as an n-word and participates in negative concord (Giannakidou, 1998, 2000), i.e., it requires negation and cannot appear in questions, unlike non-emphatic NPIs:

Giannakidou and Yoon (2016) present a number of arguments showing that emphatic NPI gives rise to intensified, scalar negation, akin to anybody at all (38c), while the non-emphatic NPI (previously encountered) is non-scalar. To understand the contrast, consider the following scenario (from Giannakidou and Yoon, 2016):

(37) Context: Maria is supposed to read some articles this week for Semantics 2, of which only one is required (the others are optional). Maria is notoriously late in doing her readings, usually doing the minimum. Her friend Ariadne asks the day before class:

The non-emphatic NPI in (37a), in contrast to the emphatic one in (37b), is infelicitous. By using the at-least phrase in the question, the question forces a scalar reading (the required article is the most likely to read, or the least likely to ignore). The non-emphatic NPI is an odd device in the scalar context, while the emphatic NPI is fine. It is useful to see the parallel with any: any with devices such as at all differs from bare any, which can be used in statements that are rather neutral (see Duffley and Larrivée, 2012 for recent discussion).

Duffley and Larrivée (2012, p. 30) conclude that: “a good number of common uses of any are not amenable to a scalar interpretation at all,” as we can see in the examples above. In Greek NPIs, the at all intensification happens purely prosodically.

The prosodic distinction remained a theoretical generalization for many years since Veloudis' and Giannakidou's initial observations—but Chatzikonstantinou (2016) conducted production and comprehension experiments suggesting that the difference is observed empirically. For the purpose of this paper, we consider one of his production experiments. 30 native speakers (15 male, 15 female) of Greek were recruited aged from 22 to 55 (mean age 32). All of them were born in Greece and had completed at least the 12 obligatory years of school education while some of them had a higher education degree. The task was administered individually to each participant on a computer screen, and included slides with the scalar (such as (37) above) or non-scalar contexts. Each slide consisted of the context and a target sentence in bold font. The sentences were written in the Greek alphabet, and the whole information fitted in one slide. Each participant was instructed to first read the context, try to get a good understanding of it. There was no time limitation for this and participants were told that they can read the text as many times as they want. Upon this, the instructions guided them to read the target sentence aloud as if it was a kind of summary or continuation of what has been narrated in the context. Participants were also informed that during the whole process that their voice would be audio recorded. In each session, the experimenter was present and assisted with the procedure.

The scalar and non-scalar contexts presented the images we see in Graphemes 1, 2 (from Chatzikonstantinou, 2016). The contours are different, the most notable difference being the Low plateau within which the non-scalar NPI is realized. There is a High peak in the beginning of the utterance and then a deaccentuation till the end. The alignment of the High peak—here aligned with the S in the beginning of the utterance—varied as often it was aligned with the negative marker bearing a more typical negative contour. No intermediate phrase was observed as no pauses during the utterance were perceivable.

Sentential contours are distinct in the two paradigms. The pitch contour (F-0) looks quite different: the emphatic is associated with a L+H^* (the H^* is aligned with the stressed syllable) and then a fall—but the non-emphatic has a flat intonation (and also the part before and after it).

Chatzikonstantinou investigated further the F0 and run a two way Anova (Scalarity (scalar, non-scalar) x Tonal Target (/e/ and /n/) contained in kanenas [the bolded characters are the tonal targets)]. There was a significant main effect of scalarity on the pitch value produced, F_{(1, 83)} = 104,097, p < 0.001. There was also a main effect of tonal target F_{(1, 83)} = 18,859, p < 0.001 and an interaction effect between scalarity and tonal target F = 17.917, p < 0.001. The result suggests that Pitch is a robust acoustic cue that differentiates between a scalar and a non-scalar NPI.

Finally, duration measures were taken from /e/ and /n/ again and ran a two way Anova. The results show that there was a significant main effect of scalarity on the duration F_{(1, 83)} = 51,283, p < 0.001 which suggests that it makes a difference whether you are a scalar or a non-scalar NPI. A marginally significant effect on tonal target was also found [F_{(1, 83)} = 3,964, p < 0.05]. There was no interaction between the two factors (F = 1,621, p = 207).

To sum up, NPIs can be scalar and non-scalar, and the difference surfaces in prosodic properties; for more extensive discussion see Chatzikonstantinou's thesis, chapter 3. The important conclusion here is that the theoretical postulate of two prosodic profiles for Greek NPIs, which has been a mere theoretical statement for about 30 years, is actually confirmed by experimental data. This carries significant promise as one further explores the syntax-semantics and prosody interaction. We find next a similar pattern about the role of prosody in bringing about scalar and non-scalar focus in Basque.

In Basque, the ordinary additive particle, ere, is used to express both a simple additive, non-scalar value (akin to English too/also) and a scalar additive value (akin to English even). In fact, ere is the only particle available in Basque to produce either simple additives or scalar additives (as opposed to other languages that have different items in the lexicon for different readings, e.g., too/also and even in English, también and incluso in Spanish, aussi and meme in French, etc.). Thus, in Basque, a string like (40), with the same lexical items and the same word order can obtain either a simple additive reading or a scalar additive reading:

(40) Mikel ere joan da.

At a first look, it would seem then that sentences containing the particle ere are completely ambiguous between the simple and the scalar additive interpretations. However, Etxeberria and Irurtzun (2015) show that prosody (the placement of the nuclear stress) is the key factor for teasing apart the two readings in (40). In order to verify the effect that prosody plays in disambiguating the simple and the scalar additive readings of ere, Etxeberria and Irurtzun designed two experiments: (i) a production experiment which aimed to test the prosodic patterns associated to each of the readings and, (ii) a perception experiment, a sentence-comprehension task where subjects had to judge the possible readings of utterances with the additive particle with varying prosodic patterns.

In the production experiment, they asked native speakers of Basque to utter pairs of identical strings corresponding to simple additive and scalar additive interpretations after presenting them a context (via written text) that clearly favored one of the two possible interpretations. They made use of three different strings, and two conditions per string, “Simple” and “Scalar,” and all of them contained the same syllable in the accented positions in the element preceding the particle ere (/ru/) and the verb following it (/di/). All participants read the same set of sentences. One of the strings they used is exemplified below between brackets “< >.”

They measured syllable duration (in ms.), F0 mean and maxima (in Hertz), and intensity mean and maxima (in dB.) in the three syllables, as well as the F0 declination between F0 maxima in syllables /ru/ and /di/. The measurements show a clear difference between strings uttered in the simple condition and strings uttered in the scalar condition in that the stress associated to the element preceding the particle ere in the scalar condition is stronger (in F0 and intensity) than in the simple condition and they argue that this is a signature of their focal nature, since narrow focus is associated to nuclear stress in Basque.

Furthermore, they also show that in the Scalar condition the region following ere displays reduced F0 values in comparison to the Simple condition, which they linked to the well attested effect of postfocal pitch compression (cf. Elordieta, 1997, 2003; Elordieta and Irurtzun, 2009; Irurtzun, 2013; Hualde and Elordieta, 2014). They conclude that speakers associate different prosodic patterns to different interpretations of the same string, which is a remarkable fact because despite the fact that the contexts of utterance were unambiguous enough so that speakers would not convey any differences in their prosodic marking, i.e., despite the fact that the exact interpretation of ere (scalar and non-scalar) could be inferred from the context alone, speakers produce different tunes.

In order to check whether this intonational pattern is enough to convey the intended meaning, they run a perception experiment. For the perception experiment they designed a magnitude-estimation task with the help of a Visual Analogue Scale (VAS) with unambiguous interpretations at both ends (since all Central Basque speakers are bilingual speakers of Spanish and Basque, unambiguous Spanish sentences at both ends were used (with también “also”–Irune también ha aprobado “Irune also passed the exam”– and incluso “even” –Incluso Irune ha aprobado “Even Irune passed the exam”–) (Figure 4).

Participants had to listen to three strings uttered with two different interpretations (simple and additive) in mind, which were taken from the productions of a participant in production experiment. Besides, for the item Irunek ere gainditu du “(Even) Irune (too) passed the exam,” they created an additional pair of test items: Condition Synth1, a manipulation of the item for “Scalar” by stylizing F0, raising the peak of the pitch accent in the subject by 25 Hz, and flattening the postaccentual region (Figure 5), and Condition Synth2, a manipulation of the item for “Scalar” by stylizing F0, raising the peak of the pitch accent in the subject by 50 Hz and flattening the postaccentual region (Figure 6).

These experimental items (the same string with the same lexical items and the same word order in all cases, i.e., Irunek ere gainditu du “Irune also passed it” or “Even Irune passed it”) were offered to the participants without any kind of context and participants had to judge the range of possible interpretations of each utterance in the VAS by cutting the judgment line in two: (i) if they thought that the utterance was ambiguous and that it could equally represent the two readings, subjects were instructed to place the delimiter in the middle of the line; (ii) if they thought that it represented more the reading to the left, but still leaving some plausibility to the reading to the right they should place the delimiter on whichever place they felt on the left; (iii) alternatively, if they judged that the utterance was unambiguous in the other direction, they should place the delimiter more to the right. Subjects were explicitly instructed that they could place the delimiter at any point in the line. Etxeberria and Irurtzun controlled the validity of the technique with completely unambiguous fillers that could only be given one interpretation and hence should be placed at the extreme left or right boundary of the line.

The results of the VAS (from 0 to 100, 0, the value on the leftmost edge, 100 the value on the rightmost edge) show a clearly skewed distribution (Simple M = 12.31, SD = 15.58; Scalar M = 71.88, SD = 26.37). The results are very interesting in that the stronger the accent the interpretation gets more scalar (Synth1 M = 78.47, SD = 28.74 and Synth2 M = 86.88, SD = 17.30). As Figure 6 shows, responses to different conditions show a different behavior, with clearly skewed distributions, significantly so in the cases of conditions Simple, Synth1 and Synth2 (Figure 7).

Thus, the paper by Etxeberria and Irurtzun shows that constructions with ere can vary in their interpretations between the simple and the scalar additivity readings but that these two readings differ depending on where the focal intonation, i.e., the nuclear stress, is placed. As a consequence, the two interpretations that can be obtained in Basque in sentences with ere are not to be considered as genuine ambiguity. In other words, there is a clear correspondence between the nonfocal or focal nature of the element preceding the additive particle ere and interpreting ere the sentence as simple additive or scalar additive. This shows that Basque make use of prosodic properties to disambiguate the scalar or non-scalar interpretations of the additive particle ere.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.