There is a large body of research on the meaning of single bare wh-questions. In this study we focus on who-questions in German (wer ‘who’), where plurality is unmarked,2 just like in the English equivalent who. Under the so-called “mention-all” reading (Groenendijk & Stockhof 1984), the answer to the wh-question is exhaustive, i.e. true and complete. The answer has to list all individuals satisfying the predicate in a given world or situation. This is also referred to as a universal or maximal interpretation (Caponigro et al. 2012). For example, if the judge in a legal trial asks me “Who came by car?”, and I know that three individuals were in a car and nobody else, I have to name all three individuals and nobody else, unless I want to tell a lie or choose not to cooperate. Giving the name of only one person, or giving the name of two out of the three individuals, are both non-exhaustive answers. As they are incomplete, albeit true, answers, in this context they would violate the exhaustivity requirement imposed by the legal context.

Besides the “mention-all” reading, single bare wh-questions have a “mention-some” reading.3 If I am hosting a party and one of the guests asks “Who came by car?”, because she is looking for a ride home, as the host I will most likely respond by mentioning one or two but not all guests who came by car, i.e. I will answer the question non-exhaustively.

There is consensus that these two readings exist. Different accounts, however, have been proposed as to the linguistic nature of the exhaustivity information triggered by unembedded bare wh-questions. Semantic approaches (i.a. Karttunen 1977; Groenendijk & Stockhof 1984; Nishigauchi 1999; Nelken & Shan 2004; Schulz & Roeper 2011) and pragmatic approaches (i.a. Van Rooij 2003; Schulz & Van Rooij 2006; Zimmermann 2007a) have postulated different mechanisms to account for the nature of exhaustivity information in wh-questions.

Semantic analyses propose that exhaustivity/non-exhaustivity, or in other terms universal/existential interpretations (Reich 1997), are a matter of ambiguity between the exhaustive and the non-exhaustive interpretations (e.g., Beck & Rullmann 1999). In some approaches the exhaustive response (“mention-all” reading) is argued to be the default (e.g., Schulz & Roeper 2011). This is in line with Nishigauchi’s (1999) assumption that English wh-expressions have a covert universal quantifier every. On the semantic view, the exhaustive and the non-exhaustive interpretations are inherent grammatical properties of question meaning. The exhaustive meaning involves universal quantification over the individuals in a contextually given domain and the non-exhaustive meaning involves existential quantification over an individual (Nelken & Shan 2004; Roeper et al. 2007). Depending on the situation, one of the two semantic representations is chosen. We refer to this analysis of exhaustivity in wh-questions as semantic.

In contrast to semantic accounts, in pragmatic accounts exhaustivity is not part of the semantic meaning of a single wh-question (e.g., Van Rooij 2003; Schulz & Van Rooij 2006; Zimmermann 2007a). Taking Van Rooij’s account (2003), exhaustivity inference is a pragmatic mechanism triggering the smallest set of candidates that gives optimal relevance depending on the context. The question “Who came by car?”, for example, is semantically underspecified and the inference denotes the optimal answer (i.e. a certain number of people) satisfying the predicate in question by considering the context of the person asking for this information. Van Rooij (2003) derives relevance as a decision problem by taking into account the goal of the information exchange. He incorporates the decision problem of the questioner as a contextual parameter that resolves a question. The answer should be useful to resolve a question. Crucially, resolvedness is tied to the goals of the questioner. Under this account, the “mention-some” reading can be derived if all true possible answers are equally useful for the questioner.

Previous empirical research has shown that exhaustivity violations of semantic nature, such as with the focus particle only, where exhaustivity is part of its truth-functional meaning, can be discerned from those of non-truth-functional nature based on the perceived degree of violation (Drenhaus et al. 2011, see also Section 1.2). Exhaustivity violations of non-truth-functional nature are perceived as less severe than exhaustivity violations of semantic (truth-functional) nature. Severity can be operationalized as the response strength of the rejection on the Likert-scale.

Our two large-scale offline questionnaire studies explore the linguistic nature of exhaustivity in bare single wh-questions by comparing these to related structures that only license exhaustive answers (wer alles ‘who all’, plural definite descriptions) and to a structure that can have both an exhaustive and a non-exhaustive answer (einige ‘some’).

This section outlines structures that have been argued to be similar to or to differ from bare wh-questions regarding exhaustivity: wer alles (‘who all’), wer so alles (‘who all’), the plural definite description triggered by die, and the scalar quantifier einige (‘some’).

Most closely related to bare wh-questions are overtly marked wer alles-questions (‘who all’). It has been suggested that alles (‘all’) has its own semantics and directly adds to propositional content of the question (Zimmermann 2007a; b), resulting in a weakly exhaustive interpretation, i.e., referring to the conjunction of all propositions that are true (Beck & Rullmann 1999: 288). An exhaustive list answer to the question “Who all is fishing?” is a case of weak exhaustivity, because we say who is fishing but not who is not fishing. Strong exhaustivity, which would entail that we also know who in a given discourse is not satisfying the predicate in question, has been argued to be triggered in wh-questions embedded under know (Groenendijk & Stokhof 1984). For example, the propositions “John knows who was at the party” and “Mary was not at the party” license the inference that “John knows that Mary was not at the party”.

Besides wer alles, plural definite descriptions contained in a restrictive relative clause served as our second semantic condition for the following reasons. In the statement “The people she invited were students”, for example, the plural definite description the people denotes all the individuals out of the set of relevant individuals who satisfy the predicate “being invited”. Link (1983) proposes a semantic analysis of plural definite descriptions that uses a mereological fusion operator: [|the NP_pl|] = σx [NP(x)], meaning ‘the sum of all x that are NP’. A non-exhaustive interpretation would hence constitute a semantic violation of the exhaustivity requirement of the determiner phrase (DP) the people (Sharvy 1980; Link 1983; Schwarz 2013). Note, however, that other approaches suggest that plural definite descriptions are underspecified and that maximality (i.e. exhaustivity) is triggered pragmatically (e.g., Sauerland et al. 2005; Malamud 2012). In this vein, plural definite descriptions can license a non-maximal interpretation, for example, in the sentence “The windows are open”, which may refer only to a subset of windows (Malamud 2012: 11). Because of the attested uncertainty about the semantic or pragmatic nature of maximality in plural definites and their possible non-exhaustive interpretation, we used plural definite NPs modified by restrictive relative clauses, as in “Name the people who are fishing in the garden”. This sentence represents a clear condition of semantic exhaustivity, because the predicate of the relative clause restricts the set of individuals to the maximum by modifying the noun (Partee 1975). The semantic mechanism is such that the restrictive relative clause, which denotes the set of contextually relevant individuals who are satisfying the predicate, combines with the set of individuals that the plural definite description designates by predicate modification (Montague 1973; Heim & Kratzer 1998).

The structure with einige (‘some’) was chosen because, just like wer, it has two interpretations. Wer has an exhaustive and a non-exhaustive interpretation, and some has the readings: ‘some but not all’ and ‘some and in fact all’. ‘Some but not all’ is called the scalar implicature of “some”. Since Grice (1989), this pragmatically enriched inference ‘not all’ has been analyzed to result from reasoning about the intention of the speaker’s utterance: the speaker chose “some” instead of “all” deliberately to indicate that the latter does not hold. Imagine someone asks “Have the students passed the exam?”, and the answer is “Some (of them) have.” In this case the scalar implicature is clearly activated, and “some” does not have the reading ‘some and in fact all the students’. In contrast, the sentence “Eating some of these pills will kill you” is clearly compatible with the reading ‘eating all of the pills will kill you’. In this case, the logical-semantic meaning of “some” (‘at least one and in fact all’) is activated.4

Generally speaking, the choice between the implicature and the logico-semantic reading of “some” primarily depends on observing the maxims of Cooperation and Quantity that guide conversational efficacy (Grice 1989). However, other factors can also play a role in the generation of implicatures, i.a., enriched experimental instructions (Papafragou & Musolino 2003; Guasti et al. 2005), the task context (Politzer-Ahles & Fiorentino 2013; Degen & Goodman 2014; Degen & Tanenhaus 2015), the grammatical structure (Hartshorne et al. 2015), the methodological design (Degen & Tanenhaus 2011; 2015), individual background factors, such as unique interpretive preferences, cognitive measures (Dieussaert et al. 2011; Marty & Chemla 2013; Antoniou et al. 2016), bilingualism (Syrett et al. 2017a; b; Dupuy et al. 2018), L2 acquisition (Miller et al. 2015; Snape & Hosoi 2018), neurological conditions (Kasher et al. 1999; Pastor-Cerezuela et al. 2018), socio-economic status (Wilson 2017), socio-pragmatic abilities (Nieuwland et al. 2010; Zhao et al. 2015; Barbet & Thierry 2016) or theory of mind capacity to infer the intention of the speaker (Cummings 2015). In light of these potential factors influencing the derivation of implicatures, we selected our sample of participants as homogenous as possible, recruiting only native speakers of German who are university students. We elaborate on the task-related factors in Section 5.

Importantly, given our experimental scenarios (see. Section 2.1.2 and Section 3.1.2), for wh-questions the non-exhaustive answer is considered a violation, because as a partial answer it is under-informative, while for einige the exhaustive response is considered a violation, because it is over-informative. Note that in this paper, we focus on the weakly exhaustive meaning of wh-questions.

Assuming that exhaustivity is specified semantically as proposed i.a. by Schulz & Roeper (2011), in the specific experimental set-up in our study a bare wh-question such as “Who is fishing in the pond?” would hence be tantamount to asking “Who all is fishing in the pond?” or “Name everybody who is fishing in the pond”. Therefore, we assume that infelicitous, non-exhaustive answers to bare wh-questions would pattern with non-exhaustive answers in the wer alles-condition and in the plural definite description condition.

In the case of einige, one of the meanings represents the semantic one (‘some and in fact all’), and the other one the pragmatic one (‘some but not all’) (Horn 1972; Grice 1975; 1989). This is in contrast to wh-questions, where, as mentioned above, both the non-exhaustive and the exhaustive meaning are argued to be represented semantically (Nelken & Shan 2004). We expected both non-exhaustive and exhaustive responses to einige to be perceived as more felicitous than non-exhaustive responses in the critical conditions of wer, wer alles, and the plural definite description.

Exhaustivity has been studied in adults using acceptability judgements (Drenhaus et al. 2011), picture-sentence-verification experiments (e.g., Gerőcs et al. 2014) as well as neurophysiological and questionnaire experiments (Drenhaus et al. 2011). Drenhaus et al. (2011) investigated the nature of exhaustivity in German it-clefts comparing them to only-foci using an offline acceptability judgment task with a 1–6 Likert-scale complemented by an ERP study (Event Related Potentials).

Drenhaus et al. (2011) explored whether the nature and source of exhaustiveness effects with it-clefts are semantic (i.e. truth-functional) in nature or whether they are non-truth-functional (i.e. as a presupposition or a generalized conversational implicature). To this end, they first compared exhaustivity violations in it-clefts and only-foci in a questionnaire study. Examples for their test sentences are given in the examples (1) and (2) below (Drenhaus et al. 2011: 7):

The participants were asked to rate the acceptability of sentences such as (1) and (2). Exhaustivity violation of it-clefts (Mean rating = 2.8) differed significantly from that of only-focus (Mean rating = 3.7). Only-focus triggers exhaustivity semantically as “part of the asserted truth-functional content of the utterance” (Drenhaus et al. 2011: 2), whereas for it-clefts, the question arises whether the exhaustiveness effect is derived truth-functionally via the propositional content as well. The authors found a mean difference of 0.9 on a scale of 1–6 between mean acceptability ratings of exhaustivity violations of only-foci and it-clefts. They concluded that given this difference, exhaustivity in it-clefts – unlike in only-foci – is not a truth-functional phenomenon. Importantly, the comparison of the extent of violations across the two structures allowed them to draw inferences about the linguistic nature of it-clefts. Note that there was no non-truth-functional, i.e. pragmatic, control condition in the experiment to compare it-clefts to clearly non-truth-functional violations.

Drawing on this difference between truth-functional and non-truth-functional violations, Drenhaus and colleagues then developed an online experiment, using ERPs, to confirm the qualitative nature of the difference revealed in their questionnaire study. ERPs are the measured brain responses to specific sensory stimuli (for a review on ERPs and language-processing, see Kaan 2007). We can distinguish different ERP waveforms. N400, one of the best-studied language-related ERP waveforms, is a negative-going wave 300–600 ms post-stimulus. It was first described by Kutas & Hillyard (1980) as a marker of semantic incongruity, as illustrated by the last word in the sentence “He took a sip from the transmitter” (example in Kutas & Hillyard 1980: 203). N400 has also been found for contextually-induced failed expectations and it is linked to the strength of lexical associations (for a review of N400, see Lau et al. 2008).

Another important effect related to language processing is the so-called P600. This is a relatively late (between around 500–1000 ms after the stimulus) positive deflection in the EEG signal, which was originally analyzed as a syntactic component (Hagoort et al. 1999; Friederici & Weissenborn 2007; Gouvea et al. 2010). It emerges in native speakers as a result of syntactic violation, syntactic complexity, syntactic repair/reanalysis or an unexpected syntactic structure (Osterhout & Holcomb 1992). Both N400 and P600 have been shown to be elicited by a broader range of phenomena (see e.g., Van Herten et al. 2005; Politzer-Ahles et al. 2013; Brouwer et al. 2017). P600, for example, has also been found in logical-semantic violations of licensing of negative polarity items in German (Drenhaus et al. 2005; 2006). Furthermore, even though truth-value violations are semantic violations, N400 has not been found in truth-value violations with semantically related words such as “A robin is not a bird” (Fischler et al. 1983). Based on these two findings, Drenhaus et al. (2011) expected a P600 to exhaustivity violations if exhaustivity is “a conventionalized part of the meaning of these structures” (Drenhaus et al. 2011: 6). In line with these predictions, they found an N400 effect for non-exhaustive it-clefts but not for exhaustive it-clefts, as well as a P600 effect for non-exhaustive only-sentences but not for exhaustive only-sentences. These findings support the conclusion based on the offline measure. For our study, we used Drenhaus et al.’s (2011) line of reasoning and compare the extent of perceived infelicity in several conditions to see with which other structures wh-questions pattern.

In Experiments 1 and 2, we employed two types of exhaustivity violations, modelled as answer types, across different structures: non-exhaustive (i.e. naming several but not all) and singleton exhaustivity violations. The labels refer to the number of individuals listed in the answer: in the case of the singleton only one of several individuals is mentioned, in the case of the non-exhaustive more than one but not all individuals satisfying the predicate are mentioned. The singleton condition was added to provide participants with varied response types. We are aware that a singleton answer in case of the plural definite descriptions constitutes a double violation: it is an exhaustivity violation and a violation of number agreement. As we were primarily interested in the non-exhaustive condition, which does not violate number agreement, we carried out planned comparisons between structures for the singleton and for the non-exhaustive level of violation separately.

We created pictures of garden scenes and sentence stimuli, which were embedded in an explicit task context (see Section 2). The NP im Garten (‘in the garden’) was added to all verbal test stimuli to make explicit that the wh-question is asked about the specific picture only. Furthermore, we explained in the instruction that both the person asking the question and the one giving the response are looking at the same picture scene, i.e. share common knowledge.

A further point of concern is that the structures in our conditions inevitably differ regarding the overtness of the exhaustivity marking, as for instance in wer alles vs. wer. The lexical exhaustivity marker alles may highlight exhaustivity more than the covert exhaustivity operator in the unmarked condition wer, and this may result in statistically significant differences given our large sample sizes. Since statistical significance is influenced by the number of observations, significant comparisons could be observed even though differences are small. Generally speaking, large-scale studies may lead to statistically significant results without indicating linguistically meaningful differences (see Lin et al. 2013; Angst et al. 2017).

To address this concern, we operationalized linguistic meaningfulness using the MID. Observed means in the same neighborhood, i.e. pairwise differences within the bounds of MID, are regarded as linguistically not meaningful differences, while observed mean differences exceeding the MID-threshold are considered linguistically meaningful differences. In Section 4 we describe the statistical analysis methods using MIDs and discuss the extent of observed differences.

Table 2 illustrates the estimated means and medians of scores per Sentence Type and Answer Type with standard deviations (SD), standard errors (SE) and 95% confidence intervals of the mean (CI). Estimations are computed from 536 items (4 items per condition per participant * 134 participants).

Median scores in all exhaustive conditions were 6, in the non-exhaustive conditions 4 with one exception (einige), and 2 and 3 for the wer alles and wer so alles singleton conditions, respectively. SD inspection based on Table 2 shows that scores associated to infelicitous responses are more spread than the ratings of felicitous responses. Figure 2 illustrates the distribution of responses visually in the critical Sentence Type conditions as a function of Answer Type.

Figure 2 shows that the rating task leads to similar variance on the conditions, that is, the effect of different interpretations and strategies are constant across the conditions. In other words, the conditions wer, wer alles, wer so alles and plural definite descriptions (die) are comparable in terms of the distribution of responses along the scale. However, there is a visible cloud of jitter around score 6 in the wer-condition. In 107 of 134 cases (80%) the mean within-participant difference between the non-exhaustive wer- and non-exhaustive die-conditions is smaller than 0.25 (with wer being more felicitous), while the group-level mean difference between these conditions (aggregated over participants) is 0.366.

A CLMM with Sentence Type using only the set of non-exhaustive responses was computed. List as a random factor did not increase explanatory power of the model as revealed by model comparisons. The model fitted with Sentence Type yielded a better model than the intercept-only model, as measured by the decrease of at least two points in the AIC values (LR = 4.4339, df = 1, p = 0.035). This shows that Sentence Type is a significant predictor (β = 0.1294, se = 0.0612, Z = 2.114, p = 0.0345), that is, the type of the question affected plural non-exhaustive responses, causing a variation over the Sentence Type conditions. Results from pairwise comparisons between Sentence Type conditions will be presented in Section 4.1 in order to be able to directly compare them to those from Experiment 2.

Table 4 illustrates the estimated means and medians of scores per Sentence Type and Answer Type with standard deviations (SD), standard errors (SE) and 95% confidence intervals of the mean (CI). Estimations are computed from 921 items (3 items per condition per participant * 307 participants).

Median scores for the exhaustive responses were 6, as in Experiment 1, and the non-exhaustive responses to wer alles, and to plural definite descriptions received a median score of 4, whereas einige received a median score of 6 and wer a median score of 5. Responses in the singleton conditions all received a median score of 2, much lower than the ratings of the non-exhaustive responses. SD inspection based on Table 4 shows that scores associated to infelicitous responses are more spread than the ratings to felicitous responses, just like in Experiment 1. Figure 4 illustrates the distribution of responses visually in the critical Sentence Type conditions as a function of Answer Type.

A CLMM with Sentence Type, Answer Type and their interaction was computed. List as a random factor did not increase explanatory power of the model as revealed by model comparisons. The best-fitting random-structure contained participant and item as random factors.

First, an intercept-only model was built. Then, Answer Type, Sentence Type and their interaction term were entered into CLMM analyses in a stepwise manner. The best-fitting model included Answer Type (p < 0.001), Sentence Type (p < 0.001) and their interaction (p < 0.001) as fixed factors, indicating that all these three terms were significant predictors of felicity ratings.

It is noteworthy that the singleton violations in the wer-condition are far more spread than those in the condition with plural definite descriptions or the wer alles-conditions. This effect in the wer-singleton condition is reflected by a very high between participant variance in the median ratings (Mean = 2.94, Median = 2, Mode = 2, Variance = 1.925, N = 307 participants). This indicates that the singleton responses were rated very differently between participants, while the mean within-participant SD of ratings was low (1.42). The high between-participant variance reflects that around 20% of the participants consistently accepted the singleton violation as a legitimate answer (median rating of 5 or 6), while half of the respondents rejected it systematically, yielding a median rating of 1 or 2 (for the distribution of median responses in the wer-singleton condition, see Appendix C).

For 169 out of 307 participants (55%) the mean difference between the non-exhaustive wer-condition and the non-exhaustive condition with plural definite descriptions is smaller than 0.25 (with non-exhaustive responses to wer rated as more felicitous), while the group-level mean difference between these conditions is 0.324, comparable to the difference in Experiment 1.

Table 5 summarizes the MID estimates using the distribution-based techniques including standard deviation criteria and effect sizes, and the significance values of the comparisons. Multiple comparisons between experimental conditions were based on the best-fitting CLMMs.

For non-exhaustive answers, the statistically significant differences between wer – wer so alles, wer – plural definite descriptions and wer so alles – plural definite descriptions did not yield linguistically meaningful differences, since they did not reach the MID-threshold (see Nature of difference, Table 5). Effect size statistics corroborated these results.

In contrast, for non-exhaustive answers the statistically significant differences between wer – einige, einige – plural definite descriptions and einige – wer so alles revealed linguistically meaningful differences. For the singleton answers, the statistically significant difference between wer so alles – wer alles again was not linguistically meaningful.

Table 6 summarizes the MID estimates for Experiment 2 in the same way as for Experiment 1. For the non-exhaustive answers, the statistically significant differences between wer – wer alles and wer – die (plural definite descriptions) did not yield linguistically meaningful differences, since they did not reach the threshold for MID by the one-third SD criterion. Also, Cliff’s δ effect sizes of –017 (wer-wer alles) and 0.16 (wer-die) for non-exhaustive answers are comparable to the Cliff’s δ effect size of 0.18 for the wer-die comparison in Experiment 1 (non-exhaustive answer).

For non-exhaustive answers, the difference between wer alles – die did not even show a statistically significant difference. For non-exhaustive answers, the statistically significant differences between wer – einige, einige – die and einige – wer alles all revealed linguistically meaningful differences. For the singleton answer, the statistically significant differences between wer alles – einige, wer – wer alles and wer – die proved to be linguistically meaningful. For the singleton answer, the differences between the comparisons einige – wer and wer alles – die were not significant.

Judgments in the conditions wer alles and plural definite descriptions were found to be comparable despite their differences in form. In Experiment 2, wer alles and plural definite descriptions did not show any differences, neither statistically nor regarding the MID, in judgments of exhaustive, non-exhaustive and even singleton answers, providing empirical evidence for the assumption that the two structures share crucial semantic properties.

Judgments in the conditions wer alles and einige were different for both singleton and non-exhaustive answers in Experiment 2. We found non-trivial, i.e., linguistically meaningful, differences in both singleton and non-exhaustive conditions: a singleton answer to wer alles was considerably less acceptable than a singleton answer to einige. Likewise, a non-exhaustive answer to wer alles was considerably less acceptable than a non-exhaustive answer to einige. This indicates a difference in linguistic nature in these two structures.

In Experiment 1, judgments of non-exhaustive answers to wer patterned with those to plural definite descriptions. In Experiment 2, judgments of non-exhaustive answers to wer patterned with those to plural definite descriptions and with wer alles. The small magnitude of differences between wer, wer alles, and plural definite descriptions in non-exhaustive conditions do not exceed the MID-thresholds in either of the experiments.16

The small magnitude of differences is driven by inter-individual variance in interpretation rather than by systematic within-participant differences which would have caused a reliable group-level difference. Only 20% of the participants in Experiment 1, and 55% of the participants in Experiment 217 distinguished between non-exhaustive answers to wer (more felicitous) and non-exhaustive answers to plural definite descriptions (less felicitous), pointing to inter-individual variation.

Based on these results, we argue that exhaustivity triggered by plural definite descriptions contained in a restrictive relative clause is similar to the exhaustivity requirement of bare wh-questions in “mention-all” contexts. We take these findings to support the semantic analysis of exhaustivity in bare wh-questions as suggested by Nishigauchi (1999) and Schulz & Roeper (2011). In this line of argumentation, bare wh-questions in the “mention-all” reading can be assumed to contain an underlying universal quantifier in a parallel fashion to plural definite descriptions contained in a restrictive relative clause. Note that given the limits of the offline-judgment task, we cannot rule out that some participants interpreted the wh-questions as a “mention-some” request and may have arrived at similar ratings as those participants who interpreted the wh-questions as “mention-all”.18

Unlike for non-exhaustive answers, for singleton answers to wer vs. wer alles and wer vs. plural definite descriptions we found a linguistically meaningful difference of around 0.9. We argue that the difference between wer and wer alles in the singleton condition is caused by the fact that the lexical marker alles encodes the exhaustivity requirement lexically, only allowing “mention-all” readings. As wer is ambiguous between “mention-some” and “mention-all” readings, a singleton answer may reflect a “mention-some” reading and hence result in a higher acceptability. Importantly, 50% of the participants showed a median response of 1 or 2 in this condition, indicating that they interpreted this structure as being equal to wer alles.

As expected, non-exhaustive answers to einige were accepted to a higher degree than the non-exhaustive answers to bare wh-questions in both experiments. This difference was non-trivial and indicates that the same set-size of individuals exerts a different effect on the perceived felicity of the answers to einige and to bare wh-questions. As already mentioned in Section 5, the inclusion of contrastive actions might have influenced the acceptability of either of the response types in the case of einige.

Singleton answers to einige and wer did not reveal any difference, with ratings being equally low. According to the semantic analysis singleton answers to wer should be dispreferred, which was supported by our results. Does this mean that einige received a semantic interpretation? We suggest that the reason for the similar low rating for singleton answers to einige is not the same as for wer: the singleton answer to einige simply represents a violation of number because einige Leute/Menschen (‘some people’) denotes plural, and this is what caused the low ratings.

Evaluating the data for plural definite descriptions, wer and wer alles in non-exhaustive conditions together, we found the magnitude of pair-wise differences between these three structures in the non-exhaustive conditions to be below the Minimal Important Difference limit (MID) in all cases. Crucially, the differences do not pass the strictest one-third SD-criterion for MID (Eton et al. 2004; Farivar et al. 2004) and do not approach the medium threshold for effect size defined by Romano et al. (2006), which suggests that these structures are alike. These parallels between wer and plural definite descriptions and wer alles are expected under a semantic account. Notably, our findings are consistent with recent research in language acquisition. A comprehension study with four-to-six-year-old children did not find any relationship between the mastery of pragmatics of quantification using scalar implicature and mastery of exhaustivity in wh-questions, suggesting that the two are not derived by a common pragmatic process (Foryś et al. 2017).

Turning to the comparison with the scalar quantifier einige, the difference we observed between wer and einige is in line with a semantic account as well. We acknowledge that an alternative explanation exists according to which einige and wer have a similar meaning representation; the differences in mean ratings would then result from the differences in the effect of the determining context on the requirements to exhaust in these conditions. It is not clear to us at this point how this explanation could account for all our data in the same way as the semantic account does. Therefore, given the clear parallels between wer and plural definite descriptions and wer alles we conclude that exhaustivity in wh-questions is of semantic nature.