Biased questions: Experimental results and theoretical modelling

Biased questions Experimental results and theoretical modelling Edited by Tue Trinh Anton Benz Daniel Goodhue Kazuko Yatsushiro Manfred Krifka language science press Topics at the Grammar-Discourse Interface 11 Topics at the Grammar-Discourse Interface Editors: Philippa Cook (University of Göttingen), Anke Holler (University of Göttingen), Cathrine Fabricius-Hansen (University of Oslo) In this series: 1. Song, Sanghoun. Modeling information structure in a cross-linguistic perspective. 2. Müller, Sonja. Distribution und Interpretation von Modalpartikel-Kombinationen. 3. Bueno Holle, Juan José. Information structure in Isthmus Zapotec narrative and conversation. 4. Parikh, Prashant. Communication and content. 5. Balogh, Kata, Anja Latrouite & Robert D. Van Valin‚ Jr. (eds.) Nominal anchoring: Specificity, definiteness and article systems across languages. 6. Næss, Åshild, Anna Margetts & Yvonne Treis (eds.). Demonstratives in discourse. 7. Gergel, Remus & Jonathan Watkins (eds.). Quantification and scales in change. 8. Nakagawa, Natsuko. Information structure in spoken Japanese: Particles, word order, and intonation. 9. Same, Fahime. Referring expression generation in context: Combining linguistic and computational approaches. 10. Lemke, Robin, Lisa Schäfer & Ingo Reich (eds.). Information structure and information theory. 11. Trinh, Tue, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.). Biased questions: Experimental results and theoretical modelling. ISSN: 2567-3335 Biased questions Experimental results and theoretical modelling Edited by Tue Trinh Anton Benz Daniel Goodhue Kazuko Yatsushiro Manfred Krifka language science press Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.). 2025. Biased questions: Experimental results and theoretical modelling (Topics at the Grammar-Discourse Interface 11). Berlin: Language Science Press. This title can be downloaded at: http://langsci-press.org/catalog/book/487 © 2025, the authors Published under the Creative Commons Attribution 4.0 Licence (CC BY 4.0): http://creativecommons.org/licenses/by/4.0/ ISBN: 978-3-96110-540-3 (Digital) 978-3-98554-160-7 (Hardcover) ISSN: 2567-3335 DOI: 10.5281/zenodo.17084327 Source code available from www.github.com/langsci/487 Errata: paperhive.org/documents/remote?type=langsci&id=487 Cover and concept of design: Ulrike Harbort Typesetting: Yan Li Proofreading: Carla Bombi Fonts: Libertinus, Arimo, DejaVu Sans Mono, Source Han Serif ZH Typesetting software: XƎL A T EX Language Science Press Scharnweberstraße 10 10247 Berlin, Germany http://langsci-press.org [email protected] Storage and cataloguing done by FU Berlin Contents Preface iii Introduction Manfred Krifka & Tue Trinh v 1 Modalization and bias in questions Anastasia Giannakidou & Alda Mari 1 2 Evidential bias across clause types Beste Kamali 37 3 The contribution of intonation in the conveyance of question bias Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio 73 4 Negative polar questions in Russian: Question bias and question concern Sophie Repp & Ljudmila Geist 101 5 Bias in tag questions Cory Bill & Todor Koev 141 6 Contextual bias and the landscape of Mandarin polar questions Yurie Hara & Mengxi Yuan 175 7 What can Cantonese sentence-final particles tell us about rhetorical questions? Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung 207 8 A note on bias and polarity in Vietnamese Tue Trinh 241 9 Psycholinguistic processing tasks and the study of question bias E Jamieson & Vinicius Macuch Silva 261 Contents 10 Marking the type of speaker bias: Hungarian nem-e interrogatives Beáta Gyuris 293 11 Children’s acquisition of English “high” negation: A window into the logic and composition of bias in questions Rebecca Woods & Tom Roeper 331 12 Everything that rises must converge: Toward a unified account of inquisitive and assertive rising declaratives Daniel Goodhue 371 Index 402 ii Preface The twelve articles in this volume and the introduction developed from talks that were presented at the workshop Biased Questions: Experimental Results & Theoretical Modelling carried out at the Leibniz-Zentrum Allgemeine Sprachwissenschaft (ZAS) in Berlin on February 4–5, 2021. This was the third workshop of the project SPAGAD: Speech Acts in Grammar and Discourse, funded by European Union’s Horizon 2020 Research and Innovation programme under grant agreement no. 787929. Due to the Covid19 epidemic, it had to be carried out as a virtual workshop. The editors thank Yan Li for help with the preparation of this volume and Sebastian Nordhoff for assistance from the side of Language Science Press. Introduction Manfred Krifka Leibniz-Zentrum Allgemeine Sprachwissenschaft (ZAS), Humboldt-Universität zu Berlin Tue Trinh Leibniz-Zentrum Allgemeine Sprachwissenschaft 1 Biased questions: An overview According to the standard analysis of question meaning, a question denotes a set of propositions, i.e. those that answer the question. Polar questions, the subject matter of this volume, are taken by many theorists to denote a set of their positive and negative answers (Hamblin 1973, Groenendijk & Stokhof 1984, Ciardelli et al. 2019).1For example, (1) denotes the set of propositions in (2). (1) Is it raining? (2) {it is raining, ¬it is raining} A common assumption among such theories of polar questions is that the alternatives proposed by the question are balanced, that is, the propositions in the set are equally ranked. This is represented visually in Figure 1. 𝜙 ¬𝜙 Figure 1: Balanced question But as we all know, sometimes questions are biased. For example, the speaker might take one answer to be more likely than another. This leads to an imbalance among the propositions, as in Figure 2. 1There are non-trivial differences in how these theories of question semantics arrive at the set of propositions in (2), but for now we abstract away from these differences. Manfred Krifka & Tue Trinh. 2025. Introduction. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions: Experimental results and theoretical modelling, v–xxiv. Berlin: Language Science Press. DOI: 10.5281/zenodo.17158174 Manfred Krifka & Tue Trinh The first experimental work on this topic is done by Roelofsen et al. (2013). This is a rating experiment, which contrasts prior speaker’s belief (SB) with positive, neutral, and negative contextual evidence (CE). The experiment yields interesting results about the different uses of positive, high negation, and low negation questions as presented in Figure 3. Figure 2: Averaged naturalness ratings for experiment 1. 2.3. Results The results of experiment 1 are presented numerically in Table 1 and are visualized in Figure 2. The ratings for the wh-fillers are listed in Table 2. The results show that PPQs are most natural in neutral CE contexts with neutral or positive SB. They are less natural in negative CE contexts, and even worse in positive CE contexts. Positive SB increases the naturalness of PPQs in neutral and negative CE contexts, while it decreases it in positive CE contexts. Negative SB does exactly the opposite: it decreases the naturalness of PPQs in neutral and negative CE contexts, while it increases it in positive CE contexts. Finally, if CE and SB are not neutral, felicity of the PPQ increases when CE and SB are contrasting. There is a significant interaction between SB and CE. A detailed statistical analysis of the results can be found in Appendix A. PPQs Neg CE Pos CE Neut CE Total Neg SB 5.09 (0.76) 5.16 (0.2) 3.37 (0.29) 4.54 (.97) Pos SB 3.24 (1.08) 5.95 (0.29) 1.41 (0.2) 3.54 (2.02) Neut SB 4.72 (1.1) 5.51 (0.36) 1.87 (0.46) 4.04 (1.74) Total 4.35 (1.24) 5.54 (.43) 2.22 (.91) 4.04 (1.66) Table 1: Averaged naturalness ratings plus standard deviations for experiment 1. wh-fillers Good Bad Positive 1.91 (1.59) 6.49 (1.24) Negative 3.06 (2.19) 6.75 (0.91) Total 2.48 (2.00) 6.62 (1.09) Table 2: Averaged naturalness ratings plus standard deviations for the wh-fillers. Positive and negative questions in discourse 459 (a) PPQs Figure 3: Averaged naturalness ratings for experiment 2. We conducted an experiment testing the hypotheses in (13) and (14). 3.2. Method Participants were recruited using AMT, as explained in section 2.2. The target items were cartoons with dialogues identical to those of experiment 1, but this time with an HNPQ in the third picture of the cartoon (e.g., Didn’t she get a cat?). The fillers consisted of the target items of experiment 1 and 3, as well as 164 additional fillers with wh-questions, which are described in section 2.2 together with additional details about the procedure. 3.3. Results The results of experiment 2 are presented numerically in Table 3 and are visualized in Figure 3.We see that HNPQs are completely natural in negative and neutral CE contexts with positive SB. They are less natural with neutral SB, and even less so with negative SB. Finally, they are completely unnatural in positive CE contexts, no matter what the SB level is. We found a significant interaction between SB and CE. A detailed statistical analysis of the results is provided in Appendix B. HNPQs Neg CE Pos CE Neut CE Total Neg SB 4.66 (0.28) 6.19 (0.16) 4.03 (0.58) 4.96 (1) Pos SB 1.87 (0.29) 6.1 (0.19) 1.76 (0.29) 3.24 (2.09) Neut SB 3.79 (0.39) 6.17 (0.17) 2.86 (0.16) 4.27 (1.54) Total 3.44 (1.24) 6.15 (.17) 2.88 (1.02) 4.16 (1.7) Table 3: Averaged naturalness ratings plus standard deviations for experiment 2. 462 F. ROELOFSEN,N.VENHUIZEN AND G. WEIDMAN SASSOON (b) HNPQ Figure 4: Averaged naturalness ratings for experiment 3. LNPQs Neg CE Pos CE Neut CE Total Neg SB 4.85 (0.63) 6.19 (0.16) 4.78 (0.2) 5.27 (.76) Pos SB 2.33 (0.23) 6.36 (0.18) 3.18 (0.3) 3.96 (1.8) Neut SB 4.19 (0.47) 6.26 (0.21) 3.93 (0.3) 4.79 (1.12) Total 3.79 (1.18) 6.27 (.19) 3.96 (.72) 4.67 (1.39) Table 4: Averaged naturalness ratings plus standard deviations for experiment 3. We found a significant interaction between SB and CE. A detailed statistical analysis of the results is provided in Appendix C. 4.4. Discussion Contextual evidence. The results do not support hypothesis (17): LNPQs are more natural in negative CE contexts than in positive CE contexts; however, they are not significantly more natural in negative CE contexts than in neutral CE contexts. Only if we restrict ourselves to contexts with positive SB, hypothesis (17) is supported. The pattern that we found for LNPQs is strikingly similar to the pattern we found for HNPQs. There is a difference in neutral CE contexts, where HNPQs are more natural than LNPQs, but the overall pictures do not differ as much as expected. Speaker belief. The results clearly reject hypothesis (18), which says that LNPQs are only natural in contexts with negative SB. In fact, we see that LNPQs are least natural in negative SB contexts. Again, the pattern that we find is similar to that of HNPQs: positive SB >> neutral SB >> negative SB where only positive SB results in complete acceptability. Under the assumption that LNPQs require Positive and negative questions in discourse 465 (c) LNPQ Figure 3: Roelofsen et al.’s (2013) experimental results These uses were described in terms of various preference rules. An example of such a rule is “ask only if needed”. According to this rule, there is no need to ask any question if speaker’s belief and contextual evidence coincide. Another rule is “avoid reversing processes”, which says that it would be strange to ask whether ¬𝑝is true when the speaker’s belief or the contextual evidence supports 𝑝. Yet another rule is “use the least marked form”, which would prefer a positive question to a negative or an alternative question, for example. We note that these rules resemble constraints in bidirectional optimality theory (OT) in pragmatics, hence raise the question whether bidirectional OT should be employed in analyzing biased questions. We believe this perspective is promising. It has become consensus to assume three levels – positive, neutral, negative – for both contextual evidence and prior speaker’s belief, or “evidential bias” and xii Introduction “epistemic bias”, which are terms proposed by Sudo (2013) which have gained some currency. We could let SB stand for prior speaker’s belief and CE for contextual evidence, and use +,0and −to represent the three levels positive, neutral, and negative. Bias profiles of questions could then be represented more succinctly. For example, [SB−, CE+] is a question with negative prior speaker’s belief and positive contextual evidence, and [SB0+, CE−] would be a question with neutral or positive prior speaker’s belief and negative contextual evidence. If we adopt the convention of placing the value for SB to the left of that for CE, the profiles can be represented even more economically. Thus, [SB−, CE+] can be shortened to [−/+] and [SB0+, CE−] to [0+/−]. Domaneschi et al. (2017) present further experimental results. Tests were conducted on high negation questions (HiNQ), low negation questions (LowNQ), positive question (PosQ), and positive question with really (ReallyPosQ). The object languages were English and German. Participants were given a selecting task, where they had to pronounce their option. The main results are presented in Figure 4. A recurring question is whether syntactically high negation questions (HPQs) are semantically ambiguous. Ladd (1981) claims that they are; Sailor (2012), AnderBois (2019), Goodhue (2022) all argue that they are not. It should be pointed out, however, that Sailor’s data can be accounted for by assuming that for English, HPQs are ambiguous while NQs are not, and for German, the opposite is the case. (28) English a. Isn’t there a train in the early morning? ¬𝑄𝑝/𝑄¬𝑝 b. Is there no train in the early morning? 𝑄¬𝑝 (29) German a. Gibt es hier nicht einen Zug am Morgen? ¬𝑄𝑝 b. Gibt es hier keinen Zug am Morgen? 𝑄¬𝑝/¬𝑄𝑝 How can we explain this difference between English and German? Suppose we say that in the [SB+, CE−] scenario, the wide-scope negation reading is the best. If we then assume that in German this wide-scope interpretation can also be expressed by a syntactically low negation, we can explain why there is more uses of this option in German. The interesting issue here, therefore, is whether there are different readings of high negation questions and low negation questions. Sudo (2013) looks at the Japanese counterparts of positive questions (PQs), high negation questions (HQs), and low negation questions (NQs). xiii Manfred Krifka & Tue Trinh Domaneschi et al: Bias in polar questions Art. 26, page 15 of 28 nottheuniquemostpreferredchoice.Onepossibleexplanationisthatthepreposedreally wasinterpretedasadiscourseparticle,whichwasproducedinordertosignalinterestin theproposition.Ifthatwasthecase,wewouldexpectdifferentprosodicrealizationsof the really-PosQsinthiscondition,inwhichitwasunexpected,andinthe¬p/pcondition, inwhichitwasexpected. Totestthispost-hocexplanation,weanalyzedasubsetofthedataprosodically:we selectedeightcasesthathadasimilarsyntacticandword-prosodicstructure(Do you like beer?, Do you have a bike?, Do you have a tent?, Do you have dogs?, Do we have a printer?, Do you have a backpack?, Do you like flying? and Have you bought the ticket?)andclassified theintonationalrealizationofreally,thedistributionofpitchaccentsandthenucleartune (lastpitchaccentandboundarytone)inthepolarquestion(seeTable14). Table14suggestssomedifferencesintherealizationofreally-PosQswhenproduced intheexpected¬p/pconditionascomparedtotheneutral/pcondition.Fortheword really,therearethreetypicalrealizations:aplainrise,afallandafall-rise.Theredonot Figure 4: Percentage of selected questions across condition in English. Whiskers represent the 95% confidence interval. preferred choice average value [95% CI] analysis by subjects analysis by items neutral/neutral PosQ 63% [56;70] t1(41)= 3.7, p < 0.001 t2(25) = 2.0, p <= 0.052 neutral/¬pLowNQ 59% [50;68] t1(41) = 2.1, p < 0.05 t2(29) = 2.4, p < 0.03 neutral/p PosQ 54% [45;63] t1(41) = 1,0, p > 0.3 t2(25) = 1.3, p > 0.19 ¬p/p really-PosQ 60% [51;69] t1(41) = 2.3, p = 0.02 t2(29) = 3.2, p < 0.005 p/neutral HighNQ 65% [55;74] t1(41) = 3.1, p < 0.005 t2(29) = 2.8, p < 0.01 p/¬pHighNQ 67% [59;75] t1(41) = 4.6, p < 0.0001 t2(29) = 4.2, p =< 0.0005 Table 13: Average percentage of the most frequently chosen responses for each condition in English, together with the 95% CI (averaged by subjects) and results of one-sample t-test of the by-subjects and by-items analysis. Domaneschi et al: Bias in polar questions Art. 26, page 19 of 28 flight?’, Gibt es eine Mensa? ‘IsthereaMensa?’, Hast du ein Fahrrad? ‘Doyouhaveabike?’, Hast du ein Auto? ‘Doyouhaveacar?’, Hast du einen Ausweis? ‘Doyouhaveapassport?’, Hast du eine Garage? ‘Doyouhaveagarage?’). Theresultsoftheprosodicanalysis(seeTable17)showedthatwirklich ‘really’ was producedwithahigh-risingtuneinallofthecases(unlikeinEnglish,wheretherewas Figure 5: Percentage of selected questions across condition. Whiskers represent the 95% confidence interval. preferred choice average value analysis by subjects analysis by items neutral/neutral PosQ 59% [52;67] t1(41) = 2.7, p < 0.01 t2(29) = 2.1, p = 0.04 neutral/¬pLowNQ 69% [63;76] t1(41) = 6.4 p < 0.0001 t2(29) = 5.2, p < 0.0001 neutral/p PosQ 51% [43;60] t1(41) = 0.3, p > 0.7 t2(29) = 0.5, p > 0.6 ¬p/p really-PosQ 65% [57;73] t1(41) = 3.8, p < 0.0005 t2(29) = 2.7, p = 0.01 p/neutral HighNQ 69% [61;77] t1(41) = 5.1, p < 0.0001 t2(29) = 4.5, p < 0.0001 p/¬pHighNQ 58% [50;66] t1(41) = 2.0, p = 0.05 t2(29) = 2.2, p = 0.03 Table 16: Average percentage of the most frequently chosen responses for each condition in German, together with the 95% CI (averaged by subjects) and results of one-sample t-test of the by-subjects and by-items analysis. ¬p/p (N = 25) neutral/p (N=13) accent on non-finite verb 10 (40%) 1 (8%) rise ending in a high plateau (H-%) 11 (44%) 7 (54%) high rise (H-^H%) 11 (44%) 6 (46%) low rise (L-H%) 1 (4%) 0 nuclear fall (H* L-%) 2 (8%) 0 Table 17: Absolute and relative number of accents on the finite verb and the nuclear tunes of the question in conditions ¬p/p and neutral/p. Figure 4: Domaneschi et al.’s (2017) experimental results xiv Introduction (30) Mary-ga Mary-NOM kita came {∅/ no / desho} ‘Did Mary come’ PQ (31) doko-ka where-ka nihon-shoku Japanese-food nai? not.exist ‘Isn’t there some Japanese restaurant?’ HQ (32) doko-mo where-mo nihon-shoku Japanese-food nai? not.exist ‘Isn’t there any Japanese restaurant?’ NQ Sudo treats epistemic and evidential bias, at different levels, as features which can be combined in different ways as presented in Table 3. Gyuris (2017) looks at three different types of polar questions in Hungarian as presented in Table 4. Table 3: Sudo’s (2013) taxonomy Question type Epistemic Evidential Short PQ none not negative –0+/0+ HQ positive not positive +/–0 NQ positive neutral +/0 Table 4: Gyuris’s (2017) taxonomy -e-interrogative ∧-interrogative ∧-declarative Neutral information question, (11) 3 3 7 Grounding question, (12) 7%3 Indirect Request 7 3 7 Indirect offer, (14) 3 3 7 Conversation starter, (15) 7 3 7 Pedagogical question 3 3 7 Monological question 3 3 7 Exam question, (18) 3 3 7 Rhetorical question, (19) 3 3 7 Gärtner & Gyuris (2017) look at possible bias profiles (see Table 5). There are 7×7=49possible SB/CE combinations. When we consider the three types of questions (PQ, HQ, NQ), we end up with 73×73=117649possibilities, which is xv Manfred Krifka & Tue Trinh Table 5: Gärtner & Gyuris’s (2017) bias profiles Question type Speaker belief Current evidence ePQ 0 ∧PQ 0, for some speakers – ∧DPQ + eHQ + 0 ∧HQ + –0 ∧NQ + – ∧DNQ – an astonishing number due to combinatorial explosion. But one can reduce the number of possibilities by certain general principles. For example, a principle of Markedness could favor non-negated questions over negated ones so that they are also used for neutral questions (cf. also Trinh 2014: 230). Let us now turn to the topic of NPIs in questions (Ladusaw 1979, Kadmon & Landman 1993, Krifka 1995, Van Rooij 2003). It has been observed that NPIs do not occur in declarative questions. Since declarative questions presumably have a [−0+/+] profile, in case of rising contour, or a [−/+] profile, in case of incredulity contour, questions with NPIs must have a [*/−0] profile. Questions with strong idiomatic NPIs (i.e. minimizers) appear to have a [−/−0] profile, meaning their use requires that the speaker not believe and the context not have evidence for the prejacent. (33) a. Did John do anything to help? b. Did John lift a finger to help? Asher & Reese (2005) account for NPIs in questions by assuming that such questions actually come with a negated assertion, which is what licenses the NPI. Another way to explain the felicity of NPIs in questions is by appealing to the fact that a question denotes a set containing a proposition and its negation, and it is the negated proposition which licenses the NPIs. The unacceptability of NPIs in declarative questions can then be explained by saying that these questions are monopolar: they denote sets containing only one proposition. Guerzoni & Sharvit (2014) account for the distribution of NPIs in questions by claiming that many questions contain covert disjunction with a covert negation that licenses the NPI. On this view, it could be argued that declarative questions fail to license NPIs because they lack covert disjunction and negation. xvi Introduction Krifka (1995) and Van Rooij (2003) propose that NPIs in questions create a more equal distribution of the likelihood of both options by way of widening the meaning of the NP complement. Thus, potatoes in any potatoes would denote a superset of potatoes in some potatoes. Strong NPIs would increase the chance of the positive answer being true, and this is a strategy for showing preference for the negative answer. It is, however, not clear how to represent the bias which comes about by way of (strong) NPIs with the features that we have discussed. We believe that a distinction has to be made within the category of contextual evidence (CE). Specifically, we need to differentiate between CE which is factual and CE which is infered from what the addressee says. (34) Factual a. A: Coming in with a dripping raincoat b. B: Is it really raining outside? (35) Addressee’s belief a. A: The rain is bothering me! b. B: Is it really raining outside? This leads us to the question of when an assertion is biased. Presumably, an assertion is biased if the speaker’s belief supports the asserted proposition, there is no contextual evidence against it, but the addressee does not believe it yet. We can enrich our feature notation by placing the addressee’s belief in brackets. (36) It is raining [+/0+] or [+/0+(−0)] (37) It isn’t raining [−/−0] or [−/−0(0+)] (38) Really, it is raining [+/0+(−)] Other relevant topics include in-situ wh-question (Biezma 2009), assertions with question tags, and miratives (Delancey 1997, Torres Bustamante 2012). Last but not least, we would like to emphasize the importance of prosody. In this connection, Bartels (1999) should receive mention. We would draw attention to the discussion on “falling” declarative questions, i.e. those that are not marked by rising contour. Other works on prosody and bias include Grice & Savino (2003), Kügler (2004), and Arnhold et al. (2021). xvii Manfred Krifka & Tue Trinh 2 The contributions to this volume In their chapter Biased questions and modal ranking, Alda Mari and Anastasia Giannakidou point out that questions share a semantic feature with epistemic modals: They are non-veridical in the sense that they do not entail that their prejacent proposition 𝑝is true. The authors suggest that questions with negative and positive bias lie on a continuum between regular questions and assertions, and that they share this property with weak and strong epistemic modals that modify assertions. Both questions and epistemically modified statements require that the modal base of the speaker contains both 𝑝and ¬𝑝worlds (the “nonveridicality axiom of modals and questions”). But questions and epistemically modified statements differ insofar as only the latter have truth conditions and can be said to be true or false. The authors discuss three empirical domains. First, reallyquestions are analyzed as marking a genuine interest of the speaker and having a negative bias. They provide an account of really in which this adverb points to a different ranking of propositions between the assumptions of the speaker and evidence of the context, with reference to experimental work on wirklich in German. They show that negatively biased questions ask for stronger confirmation; a response like I think so is felt to be insufficient. The second domain are questions with high negation and low negation with a positive bias, which are related to assertions with the strong modal must that also expresses a positive bias. The third domain are reflective or conjectural questions marked with weak epistemic modals, such as might. They observe that the modals must be weak, and argue that they widen the modal base, enlarging the range of possibilities, thus unearthing remoter ways in which a proposition may be true. As for assertions, the authors suggest that they are only added into the common ground when not modalized. In her chapter, Evidential bias across clause types, Beste Kamali compares English rising declarative questions, known to express a positive bias towards their proposition, with a polar question type in Turkish. In this language, polar questions are always marked by a particle mi that is attached to a constituent of the sentence, then often marking the focus of the question, or after the (typically final) finite verb of the sentence, indicating a polar question without focus. However, when mi attaches to the direct object, focus may project to the whole sentence. There are subtle differences between such sentences and sentences where mi is attached to the final verb. In particular, the object-mi sentences have a reading that has a similar bias to English rising declarative questions, insofar as they also require a positive evidential bias. However, a careful examination in a battery of tests reveals interesting differences: First, object-mi questions are classixviii Introduction fied as “questions” in the object language, different from English (cf. One question remains: Did Ali make dinner? /#Ali made dinner?). They do not allow for modal adverbials, different from English (cf. You certainly made dinner?). And they can be embedded by rogative predicates, again different from English (*She asked Ali made dinner?). Hence, object-mi questions achieve their positive bias in ways quite different from English declarative questions. According to the proposed analysis, they both are monopolar, in contrast to regular polar questions and verb-final mi-questions, which are analyzed as bipolar. Their differences derive from the fact that they are syntactically and semantically different clause types. The author also draws in evidence from Hungarian and Japanese that show similar question types as Turkish object-mi questions. In their chapter, The contribution of intonation in the conveyance of question bias, Riccardo Orrico, Cristel Portes, Mariapaola D’Imperio provide a comprehensive overview of the role of intonation in the expression of question bias. They begin with a review of the literature on intonation patterns and the different dimensions of meaning they can express. This review is followed by a discussion of two recent experimental studies conducted by the authors themselves. Throughout the chapter, the authors address three main questions about the relationship between phonetic cues and speaker meaning: the kinds of meanings that can be expressed by intonation, the intonational cues that convey these meanings, and the degree of variation within the linguistic population. After a general overview, they provide a detailed comparison of the means used by Italian and French speakers to express question bias. They argue that the way in which meaning is encoded in intonation patterns is not universal. Their study shows that different languages have their own specific cues to express specific meanings. In their chapter, Negative Polar Questions in Russian: Question bias and question concern, Sophie Repp and Ljudmila Geist study the appropriateness conditions of yes–no questions containing the particles razve and neuželi, both of which can be translated as really. In their analysis, they distinguish between the bias profile of a question and the question concern. The bias profile is defined by the question’s epistemic and evidential bias (following Sudo 2013), where the former refers to the speaker’s prior beliefs and the latter to contextual evidence that may or may not conflict with the speaker’s beliefs. By question concern they refer to the purpose of checking or rejecting the prejacent of a question, or its negation. Repp and Geist assume that the appropriateness of different types of polar questions depends on their bias profile and their concern. They demonstrate the usefulness of this asumption for polar questions in English, and then extend the analysis to polar questions in Russian with or without particles razve and neuželi. Repp and xix Manfred Krifka & Tue Trinh Geist show that polar questions with razve and neuželi do not differ in their bias profile but in their ability to check the truth of the proposition which is favoured by the question’s epistemic bias. Repp and Geist support their analysis by a corpus study and two experimental investigations. They further provide a semantic explanation for the difference between razve and neuželi in terms of their consistency with inner and outer negation. In their chapter, Bias in Tag Questions, Corey Bill and Todor Koev study constructions consisting of a VP-eliptical yes/no question which is “tagged” onto an “anchor” declarative sentence of the opposite polarity, as exemplified by it’s raining, isn’t it? and it’s not raining, is it?. It has been observed that tag questions give rise to the inference that the speaker is “biased” toward the anchor, i.e. that she believes the proposition it expresses to be true. Bill and Koev propose to describe such biases in terms of two parameters: (i) whether they are weak or strong, and (ii) whether they are optional or obligatory. They devise diagnostics to test these distinctions, and advance an analysis to derive the observations from syntactic and phonological properties of tag questions. In their chapter, Contextual Bias and the Landscape of Mandarin Polar Questions, Yurie Hara and Mengxi Yuan discuss ma-questions and A-not-A questions in Mandarin Chinese. They present the following observations: (i) positive maquestions can be used in either neutral or positively biased contexts; (ii) negative ma-questions can only be used in negatively biased contexts; (iii) A-not-A questions can only be used in neutral contexts. They then argue that it is contextual bias, not speaker’s bias, which is instrumental for an analysis of this distribution. The concept of “contextual bias” is defined in terms of subjective probability and Farkas & Bruce’s (2010) Table model. In their chapter, What can Cantonese sentence-final particles tell us about rhetorical questions?, Angelika Kiss, Roger Yu-Hsiang Lo, and Justin R. Leung discuss three kinds of questions: (i) information seeking questions (ISQ), (ii) rhetorical question with empty set answers (RQ–), and (iii) rhetorical questions with nonempty set answers (RQ+). It uses the framework of inquisitive semantics to argue that these are in fact natural classes in terms of informativity and speaker’s commitment. Moreoever, ISQ and RQ– make up a subclass in this three membered group. An perception experiment is then presented whose result shows that the semantic distinctions are also reflected prosodically in Cantonese, specifically in distinctions with respect pitch contour and length of the final sentence particle. In his chapter, A note on bias and polarity in Vietnamese, Tue Trinh discusses the distribution of two types of NPIs across two types of polar questions. NPIs in Vietnamese come in two morphological variants, one simple and one complex. Among polar questions in this language, Trinh distinguishes between yes/no xx Introduction questions and agreement questions. The observation is that complex NPIs are acceptable in yes/no questions but deviant in agreement questions, and in yes/ no questions, complex NPIs give rise to negative bias while simple NPIs do not. The analysis Trinh proposes for this fact assumes that complex NPIs require a covert EVEN in the structure, and that agreement questions contain a covert evidential marker. Much of the study on biased questions has been based on the introspective intuition from different languages. In this chapter, Psycholinguistic processing tasks and the study of question bias, E Jamieson and Vinicius Macuch Silva provide another dimension: how can we investigate the issues related to biased questions from an experimental perspective? The chapter discusses different theoretical approaches to biased questions, an overview of psycholinguistic investigations on this topic, as well as the contribution of experimental data to the ongoing discussion and understanding of polar questions. In addition, Jamieson and Silva provide concrete suggestions and guidance regarding the aspects of experimental set-up, providing important resources for future researches. Hungarian interrogatives are marked with a special suffix. In her chapter, Marking the type of speaker bias: Hungarian nem-e interrogatives, Gyuris examines a particle nem-e, which consists of a negation and an interrogative marker. This particle has not been investigated before, and naturally, Gyuris is the first to describe the different characteristics that the particle exhibits, and to a semantic account. Gyuris first discusses the meaning contribution and the distribution of the canonical question particle -e, showing the interaction between its meaning and biases. She then discusses the data on nem-e, and reveals that their meaning and distributional pattern reveal cross-dialectal variation: nem-e gives rise to the outside negation reading but not inside negation reading, and is incompatible with different types of biases. After identifying different uses of nem-e, Gyuris provides analyses of the particle, which predicts and accounts for their distribution. The chapter is a welcome addition to our understanding of biased questions. In their chapter, Children’s acquisition of English “high” negation: A window into the logic and composition of bias in questions, Rebecca Woods and Thomas Roeper investigate the production of English nuclear negative tag structures and negative questions, produced by children and adults. They argue that the structures of negative tags and negative questions aredistinct, the former being simple speech acts that are complex at the clausal level. The negative questions, on the other hand, involve an interrogative clause, scoped over by metalinguistic negation and a question operator. Woods and Roeper provide a new type of evidence xxi Anastasia Giannakidou & Alda Mari Biased questions, then, are not mere requests for information but rely on the speaker’s prior doxastic and epistemic assumptions and expectations.2In this respect, they lie on a continuum between questions and assertions: they ask whether 𝑝but also discriminate towards 𝑝, or its negation. Notice that in the tag question (2b) we actually do have a hybrid declarative and interrogative sentence. Importantly, the bias can also famously be cancelled with an answer of the unexpected polarity. Bias is a choice that the speaker makes based on their assumptions and contextual knowledge but it is not a common ground presupposition, as we will emphasize, hence it can be objected to. In the case of positive bias, the speaker seems to rank 𝑝as more likely, hence they are more epistemically committed to it; in the negative bias, ¬𝑝is considered more likely and there is less epistemic commitment to 𝑝. Speaker commitment is a notion that plays a key role in modality (see Giannakidou & Mari 2021a), and a handy way to refer to the judgment of an epistemic agent towards the veridicality of 𝑝. In our view, commitment is private (we thus differ for instance from Krifka 2015 and Geurts 2019 who consider commitment a public attitude). Giannakidou (2013) connected questions to modal sentences (which she called inquisitive assertions), and Giannakidou & Mari (2018b, 2021a,b) establish a parallel in modality between neutral possibility modals and biased ones (necessity modals): with possibility modals, the speaker is in a nonveridical information state where 𝑝and its negation are equally open possibilities, but with a necessity modal such as MUST there is bias towards the prejacent 𝑝. Consider: (4) a. Ernie is a vegetarian. b. Ernie must be a vegetarian. c. Ernie may/might be a vegetarian. d. Is Ernie a vegetarian? Of the sentences above, only the unmodalized assertion conveys the information that Ernie is a vegetarian, and adds it to the common ground. Only this one is, then, an assertion of 𝑝. The modal sentences, just like the plain question, reveal uncertainty, which we represent as a nonveridical epistemic state, that is to say, a modal base partitioned between Ernie being a vegetarian and not being a vegetarian. The speaker, in choosing to utter modalized sentences, just like with 2There is also contextual bias relating to evidence available in the context, i.e., the common ground (evidential bias, Büring & Gunlogson 2000; Romero & Han 2004; Sudo 2013; Northrup 2014; Domaneschi et al. 2017), or answer bias (Krifka 2015; Malamud & Stephenson 2015). We will not discuss common ground bias here; for more details on factors determining belief formation see Giannakidou & Mari (2021a,b) 4 1 Modalization and bias in questions questions, takes a nonveridical stance (Giannakidou & Mari 2021a, Giannakidou & Mari 2021b). The nonveridical stance can be neutral, as with neutral information question and possibility modals, or biased when the speaker chooses to use MUST. This parallelism between possibility modals and unbiased questions on the one hand, and necessity modal MUST and biased questions on the other will play an important role in our analysis. While we maintain that the categorical difference between modal assertions and questions lays in the first having and the second lacking truth conditions, we also acknowledge that both convey nonveridical states with two alternatives: 𝑝and its negation. Nonveridicality thus “allows us to see that the distinction between assertion and question is not as basic as we thought”, Giannakidou (2013) states, and this, she continues, “seems to support an approach to meaning as semantically non-dichotomous” (Giannakidou 2013 : 117). In the end, what seems to matter is whether a sentence presents the epistemic agent with one or more possibilities about the world, i.e. whether it reflects a homogeneous or non-homogenous (partitioned) epistemic space. “Superficially, this appears to correspond to the contrast between assertion vs. question. However, the more fundamental distinction is between a partitioned or not epistemic space” (Giannakidou 2013: 126). In this paper, we expand on the idea that the distinction between assertion and question is not categorical semantically, and use the phenomenon of bias as a testing ground to establish a parallel between the structure of questions and the structure of modality. We argue that questions and possibility modals share an important piece of meaning: they express nonveridical equilibrium by not discriminating between the two alternatives 𝑝and ¬𝑝; they are by default egalitarian. When a bias device is used, the device manipulates the equilibrium by introducing a ranking function. This is grammaticalized in modality with necessity modals such as MUST, SHOULD and their equivalents (Giannakidou & Mari 2016, 2018b, 2021a,b) – and the expressor of the ranking function can be an overt adverb such as She must definitely be a doctor, or a covert one. We will argue, similarly, that bias inducing devices in questions contribute or manipulate a ranking. The different effects, positive or negative, are due to the lexical contribution of the bias inducing device as we will show. The discussion proceeds as follows. In Section 2 we outline the framework of modality we will be using with the bias inducing ranking function 𝒪. In Section 3 and 4 we will propose that biased questions are equivalent to biased modal conjectures in containing a meta-evaluation function 𝒪which ranks the two possibilities, 𝑝and its negation. Specifically in Section 3 we derive negative bias with the adverb REALLY, and, in Section 4, we discuss positive bias with negation 5 Anastasia Giannakidou & Alda Mari proposing that negation provides the ranking function. The discourse function of a biased interrogative is still that of a question, but its definedness conditions are those of a biased modal verb. Finally, in Section 5, we discuss questions with possibility modals and particles and identify reflectiveness as the inverse of bias, offering an analysis within the modal framework we have established. 2 Nonveridical structure and bias inducing meta-evaluation Modal expressions in human languages – modal verbs, adverbs, particles – are common devices to reflect the speaker’s judgement towards the truth of a proposition. Almost all analyses of modality assume that non-alethic modal expressions as a class are nonveridical, i.e., they reflect epistemic states that do not entail that the proposition is true (Kratzer 1977, 1986, 1991, Giannakidou 1997, 1998, 1999, 2013, Condoravdi 2002, Portner 2009, Beaver & Frazee 2016, Giannakidou & Mari 2016, 2018b, 2021a,b, Lassiter 2016; von Fintel & Gillies 2010 being a notable exception). In the modal framework of Giannakidou & Mari (2016, 2018b, 2021a,b) (henceforth GM) the function of epistemic modal expressions is to convey the nonveridical attitude of the speaker: upon hearing or reading a modal sentence, the audience understands that the speaker cannot be fully committed to the truth of the propositional content of the sentence. Modal expressions are therefore characterized as anti-knowledge markers in GM: a speaker cannot use a modal if she knows 𝑝to be true, e.g. if I see rain falling I cannot say It might be raining, or It must be raining. Likewise, when a speaker asks a question they are in a state of not knowing. Both questions and modals, then, convey epistemically nonveridical spaces, they have this common logical basis (see also Sherman 2018). What does it mean for an expression to be veridical or nonveridical? Consider the following declarative sentences: (5) a. It is raining. b. It must be raining. c. It may/might be raining. Let us call the tensed unmodalized sentences bare. In asserting a bare sentence the speaker is saying something that they know or believe to be true – they are veridically committed to it. GM call this the Veridicality Principle of co-operative communication, and it follows from abiding by Gricean Quality, which is fundamental to co-operative conversation: 6 1 Modalization and bias in questions (6) Principle of veridicality for cooperative communication (Giannakidou & Mari 2021a,b) A sentence S can be asserted by a speaker A if and only if A is veridically committed to the content 𝑝of S (i.e., if and only if A knows or believes 𝑝 to be true). Thus, upon hearing an unmodalized sentence the hearer understands that the speaker knows, or has grounds to believe that it is raining. The epistemic attitude of the speaker can therefore be thought of as a veridical state. For a given judge 𝑖(by default with epistemic modals, the speaker): (7) Veridical epistemic state An epistemic state M(𝑖)is veridical with respect to a proposition 𝑝iff: ∀𝑤[𝑤∈𝑀(𝑖)→𝑝(𝑤)]. A veridical state entails 𝑝, thus conveying epistemic (or doxastic, or mixed) commitment of 𝑖to 𝑝.3But when a speaker chooses to modalize, just like when she choses to pose a question, she is uncertain about 𝑝. The epistemic state now is partitioned into 𝑝and ¬𝑝: (8) Nonveridical epistemic state An epistemic state (a set of worlds) 𝑀(𝑖)relative to an epistemic agent 𝑖is nonveridical with respect to a proposition 𝑝iff 𝑀(𝑖)is partitioned into 𝑝 and not ¬𝑝worlds. In a nonveridical state 𝑀(𝑖),𝑝and not 𝑝are open possibilities. The partition can be neutral, as in the case of equilibrium, or discriminating as is the case with bias inducing necessity modals (cf. infra). Both modals and questions are evaluated with respect to nonveridical modal spaces 𝑀(𝑖), where 𝑖is by default the speaker.4Giannakidou & Mari (2016) formulate nonveridicality as a definedness 3As we mentioned already, our notion of commitment is private and subjective, i.e., it represents the set of propositions held by individual linguistic agents (e.g., the speaker, the subject of the attitude in embedding) which Giannakidou (1997) and Giannakidou & Mari (2016, 2018b, 2021a,b) call individual anchors, recycling a term from Farkas (1985); in our definition above the anchor is called judge. Our commitment therefore differs from Krifka’s which is “modelled as a set of propositions, containing the propositions that are publicly shared by the participants” (Krifka 2015: 328–329). Krifka’s commitment corresponds to common ground assumptions, and the goal of speech acts is to “change a commitment state”. Speaker commitment in our view can be entirely private, even solipsistic (Giannakidou & Mari 2016, 2018a,b, 2021a,b), as with verbs of belief, imagination, dreaming and pretending. 4Both questions and modals can be anchored to the addressee or a third party, and we return to the interrogative flip later in the paper. We assume that, by default, 𝑖is the speaker. 7 Anastasia Giannakidou & Alda Mari condition of modals in the form of the Nonveridicality Axiom – which appears to be also a definedness condition of questions: (9) Nonveridicality Axiom of modals and questions a. modal(𝑀(𝑖))(𝑝)can be defined if and only if the modal base 𝑀(𝑖)is nonveridical, i.e. only if 𝑀(𝑖)contains 𝑝and ¬𝑝worlds. b. ques(𝑀(𝑖))(𝑝)can be defined if and only if the modal base 𝑀(𝑖)is nonveridical, i.e. only if 𝑀(𝑖)contains 𝑝and ¬𝑝worlds. The Nonveridicality Axiom is a definedness condition which lexically encodes a speaker presupposition in the Stalnakerian sense of modals and questions. We will revisit the presuppositional nature of bias and offer more comments in Section 3. Now consider the difference between possibility and necessity modals. (10) a. Ariadne might be at the party. b. Ariadne must be at the party. In both cases, the speaker is uncertain about Ariadne being at the party, and leaves open the possibility that she might not be. The uncertainty is in nonveridical equilibrium with MIGHT: Ariadne being at the party is considered a mere possibility, and the speaker has no reason to believe Ariadne being in the party is closer to what is the case than Ariadne not being at the party. Possibility modals are thus very much like information seeking questions in this regard: in both cases the speaker is in a state of true uncertainty and entertains two alternatives equally.5While questions lack truth conditions and are therefore distinct from modal assertions, it is important to understand that the common semantic core between neutral questions and possibility statements is the nonveridical state: they both generate alternatives – and presuppose equilibrium where the speaker entertains two possible alternatives that are not ranked. On the other hand, when a necessity modal such as MUST/SHOULD is used, the equilibrium is manipulated towards the prejacent being considered more likely by the speaker. GM coin the term biased modals for necessity modals: the speaker is positively biased in favor of the prejacent proposition, though they still are not veridically committed to it. Modal bias reveals an epistemic stance supported by evidence in favor of the proposition, but it does not mean that the speaker knows or believes 𝑝to be true, i.e, they are not doxastically or epistemically committed to it. Modals, according to GM and what von Fintel & 5As we noted earlier, this also holds for the conditional protasis; see Liu et al. (2021) for recent discussion and experimental results. 8 1 Modalization and bias in questions Gillies (2010) call ‘the mantra’, remain indicators that the speaker reasons with uncertainty and that they leave both options, 𝑝and ¬𝑝, open. With biased modals, in addition, the speaker positively discriminates towards 𝑝, she appears to think of 𝑝as a better possibility than ¬𝑝. Bias thus reveals that the two possibilities are ranked. The ranking is done, GM propose, with a metaevaluation function 𝒪 which is always present in a modal structure – and in a nonveridical structure more generally. Let us see what the epistemic MUST modal structure is precisely. First, we have the modal base of epistemic MUST which is nonveridical about the prejacent and contains both 𝑝and ¬𝑝worlds. To derive the truth conditions of MUST the literature assumes (see e.g. Kratzer 1991, Portner 2009, Giannakidou & Mari 2016, 2018b, 2021a,b) that MUST uses a set of propositions 𝒮which describe shared stereotypical/normalcy conditions. The Kratzer/Portner semantics posits an ordering source Best which ranks worlds according to how close they are to the stereotypical ideal. Our account encodes that the modal base is partitioned into stereotypical and non-stereotypical worlds, but we dissociate stereotypicality from ranking. As an epistemic modal, MUST associates with an epistemic modal base 𝑀(𝑖)that contains the worlds compatible with what the speaker knows or believes. 𝑤0is the world of evaluation, by default the actual world: (11) 𝑀(𝑖)(𝑡𝑢)(𝑤0)=𝜆𝑤′(𝑤′is compatible with what is known by the speaker 𝑖 in 𝑤0at 𝑡𝑢)6 Epistemic modality is thus by default subjective, and knowledge changes with time. Epistemic modality is therefore parametric to knowledge at 𝑡𝑢, as is often acknowledged in the literature (see Portner 2009, Hacquard 2006, 2010, Giannakidou & Mari 2016). In the epistemic modal base 𝑀(𝑖)(𝑡𝑢)(𝑤0), we define Ideal𝒮as a function over 𝑀(𝑖)(𝑡𝑢)(𝑤0), still in the spirit of Portner (2009). The output Ideal𝒮is a subset of 𝑀(𝑖)(𝑡𝑢)(𝑤0): (12) Ideal𝒮(𝑀(𝑖)(𝑡𝑢)(𝑤0)) = {𝑤′∈𝑀(𝑖)(𝑡𝑢)(𝑤0)∶∀𝑞∈𝒮⟨𝑤′∈𝑞⟩} So defined, Ideal𝒮delivers the worlds in the modal base in which all the propositions in 𝒮are true. 𝒮is a set of propositions that corresponds to common ground norms.7The truth condition for MUST says that 𝑝is true in the Ideal𝒮 6Our notation 𝑀(𝑖) corresponds to the Kratzerian notation using set intersection ∩𝑓epistemic(𝑤0,𝑖,𝑡𝑢), where this returns the set of worlds compatible with what it is known in 𝑤0 by 𝑖. 7Since only those worlds are considered in which 𝑎𝑙𝑙the propositions in 𝒮are true, the function Ideal𝒮determines a cut-off point. 9 Anastasia Giannakidou & Alda Mari set of 𝑀(𝑖). We assume that, by default, 𝑀(𝑖)is projected at the time of utterance in the actual world: (13) Given a set Ideal𝒮and the utterance time 𝑡𝑢, JMUST (PRES (𝑝))K𝑀,𝑖,𝒮is defined only if 𝑀(𝑖)is nonveridical and is partitioned into Ideal𝒮and ¬Ideal𝒮worlds. If defined, JMUST (PRES (𝑝))K𝑀,𝑖,𝒮= 1 iff ∀𝑤′∈Ideal𝒮∶𝑝(𝑤′,𝑡𝑢) We will now postulate that Ideal𝒮and ¬Ideal𝒮worlds are ranked according to an ordering source 𝒪, which, in the case of MUST, is introduced by a silent adverb, defined as in (14):8 (14) For any Ideal𝒮, J∅K𝒪,M,𝑖,𝒮=𝜆𝑞. Ideal𝒮is a better possibility with respect to ¬Ideal𝒮 relative to 𝑀(𝑖)and 𝒪&𝑞 ∅is a covert adverb that provides a “meta-evaluation” that compares Ideal𝒮to its complement in 𝑀(𝑖). Ideal𝒮worlds are privileged by the ranking function 𝒪 which ranks the Ideal worlds as better possibilities (in the sense of Portner 2009 and Kratzer 1986). The 𝒪function manifests itself in the form of an adverb or a particle. (15) a. John might be sick perhaps. b. John must probably/definitely be sick. c. John must be sick. Such structures are characterized as modal spread in Giannakidou & Mari (2018b), and the syntax is called the modal skeleton. The above are equivalent semantically in our system, with the example in (15c) containing the default null adverb. The modal skeleton always contains the metaevaluation, but with possibility modals 𝒪is empty, hence the equilibrium: (16) ModalP 𝒪:Adverb/particle ModalP Must 𝑀(𝑖) 𝒮 TP 8For more discussion on future shifting and the nonveridicality of MUST, see GM for details. The key observation here is that only in Ideal𝒮worlds is 𝑝true. 10 1 Modalization and bias in questions In the GM system, modal spread is the default modal structure, 𝒪being always part of it. With MUST, the null adverb is akin to probably which induces the intrinsic positive bias of MUST favoring 𝑝worlds as better possibilities (15). Non-biased possibility modals differ from MUST in that they convey equilibrium, which is now defined as follows: (17) Nonveridical equilibrium (Giannakidou & Mari 2018b, 2021a,b) A partitioned (𝑝and ¬𝑝) space 𝑀(𝑖)is in nonveridical equilibrium if the ordering 𝒪is empty. Possibility modals are structures with an empty 𝒪. In discussion of the phenomenon of reflection later where we have possibility adverbials in questions, we will see that 𝒪can also induce widening in the modal base in which case we end up with harder to answer or unanswerable questions. In sum, the epistemic nonveridical structure involves two ingredients: (i) a partitioned modal base 𝑀(𝑖), and (ii) a metaevaluation 𝒪that is either empty (possibility, neutral), or ranks the Ideal𝒮worlds as better possibilities than non-Ideal𝒮 worlds (bias inducing modals). The preference for higher ranking of Ideal𝒮is lexically specified: necessity and possibility modals differ lexically in that higher ranking of Ideal𝒮is only a feature of the former. The availability of 𝒪ranking is not limited to modal expressions but we expect to find it whenever we have a partitioned nonveridical (𝑝,¬𝑝) structure as is the case with questions. We now proceed to show how this analytical framework explains positive and negative bias in questions. We start with REALLY. Our basic idea is that the presence of bias in questions indicates, just like with modal verbs, the presence of a ranking function 𝒪. The expressor of 𝒪can be an adverb such as really, as we argue next – but negation can also function as 𝒪. The commonality between questions and modals is that they both convey partitioned nonveridical epistemic spaces which can be ranked and have bias, or can be left unranked in the egalitarian state of equilibrium. Negation can function as 𝒪when it is used as a focus sensitive operator. 3 Deriving negative bias with REALLY-questions Questions with the adverb really, as mentioned earlier, are known tn the literature to express negative bias. We give below the English example and its Greek and Italian equivalents: (18) Is Ernie really a communist? 11 Anastasia Giannakidou & Alda Mari (19) Ine cop.3sg o the Ernie Ernie pragmati/st’ alithia truly/really kommunistis? a.communist ‘Is Ernie really a communist?’ (20) Ernie Ernie è cop.3sg davvero really un indef comunista? communist ‘Is Ernie really a communist?’ By using REALLY (really,pragmati,st’alithia,davvero) the speaker seems to be genuinely interested in knowing if Ernie is a communist, like when one utters I want to know whether Ernie is really a communist. In addition, as we said at the beginning, in choosing to use REALLY the speaker intends to show that she has negative bias, i.e., she has some doubt that Ernie is a communist (see also a recent discussion of the REALLY effect in Bill & Koev 2025 and more references therein). With a REALLY question, as a reviewer suggests, the speaker raises the stakes for the addressee to give a truthful answer because her doubt about the asked content is enhanced. The negative bias of wirklich (the German counterpart of really) is experimentally confirmed in Liu et al. (2021) in questions and conditionals – and it appeared to be one of the most solid experimental findings. Romero & Han (2004) propose an influential account of the adverb really as verum which we consider after we first lay out our proposal. 9 Bias, as we said earlier, is the destruction of the egalitarian state of equilibrium in a positive or negative direction because the speaker has some priors (in the sense of pre-questioning beliefs and expectations) that make them think that Ernie is not a communist is a better possibility than Ernie is a communist. It is for this reason that the stakes are higher for the answerer to show that he is. Importantly, the negative bias is not a belief that the prejacent is not true; it is rather, as with MUST, an indication that the speaker is considering a nonveridical partition and compares the likelihoods of 𝑝and ¬𝑝. Unlike with MUST, the favored proposition now seems to be the negative one. Bias is anchored to the speaker, who ranks the two possibilities prior to asking a question; it can thus best be understood as a speaker-anchored definedness condition on asking the question. For Stalnaker (1978), and this is important to note, presuppositions are preconditions that need to be satisfied before the common ground can be updated; hence they are requirements on the speaker’s 9The really effect occurs also in constituent questions, as in Who did really go to Berlin?, alternative questions (e.g. Did he really go to Berlin, or not?), and in declarative questions e.g. He really went to Berlin?. 12 1 Modalization and bias in questions knowledge, not on the common ground.10 We ask here how bias comes about with REALLY expressions. Part of the puzzle is also that REALLY expressions, while positive, they bias towards a negative answer. Let us observe further that the negative bias of REALLY holds across nonveridical constructions, including conditionals (21a) and imperatives (21b). (21) a. If John really studied very hard, he will pass the exam easily. Speaker’s assumption prior to the question: John most likely did not study very hard. b. Really, close the door! Speaker’s assumption prior to the question: the addressee resists closing the door. None of the existing accounts explains this generalized REALLY effect. The prior assumptions could be the speaker’s private assumptions but they could also be contextual, as the two types of information intersect. In some cases the prior could be merely a contextual challenge prior to assertion, as in the following example: (22) A: I know, John told you that he will study very hard for the exam, but I am still worried. B: If John really studied very hard, he will pass the exam easily. And if he says he studies hard, I believe him. In this example, B is entertaining A’s uncertainty about whether John studied hard, while also asserting his own favoring position tilting towards the positive. This is consistent with the nature of bias which is speaker-anchored. (We thank a reviewer for bringing up this case). In positive veridical assertions, REALLY again is anaphoric to prior assumptions, but now in a positive direction. Observe the following example from Romero & Han (2004): (23) a. ? I am sure I am tired. b. I really am tired. 10von Fintel (2008) and subsequent literature in the context change potential tradition think of the presuppositional component of the meaning of a sentence as being a requirement on the information state it is used to update. “Since the information state a sentence is used to update in the ideal case is the common ground, the presuppositional requirements are imposed on the common ground” (von Fintel 2008: 5). In effect, then, definedness conditions can be understood as common ground presuppositions, or as in the Stalnakerian sense which effectively makes no distinction between felicity conditions of the speaker and common ground. 13 Anastasia Giannakidou & Alda Mari (iv) decision based accounts Van Rooy & Šafářová (2003); (v) Evidence-based accounts (Büring & Gunlogson 2000, Sudo 2013, Roelofsen & Farkas 2015, Goodhue 2018), stemming from the work of Ladd (1981b). It would be impossible to render justice here to the whole literature, and for the purpose of this paper we only focus here on the interrelations between MUST, Focus-neg and bias; for extended recent discussions see Krifka (2017), Larrivée & Mari (2022). Larrivée & Mari (2022) establish a series of correlations between high Focusneg in questions and MUST, pretty much in the sprit we outlined for REALLY. They argue that Focus-neg and MUST share important similarities, namely (i) they are nonveridical, and (ii) they convey the speaker’s prior that 𝑝is likely. However, they are in complementary distribution in evidential contexts. MUST is felicitous in contexts that are compatible with 𝑝, whereas negative questions are felicitous in contexts that are incompatible with 𝑝(called ‘negative evidence’ by Büring & Gunlogson 2000, see also Sudo 2013). The following examples are from Larrivée & Mari (2022). (39) John looked pretty happy coming back from school. a. – Il he doit must avoir have réussi succeded son his examen exam de of math. math ‘He must have passed the big maths test.’ b. – #N’a-t-il not.have-1sg-he pas neg réussi succeeded son his examen exam de of math? math ‘Didn’t he pass the big maths test?’ (40) John looked pretty down coming back from school. a. – #Il he doit must avoir have réussi succeded son his examen exam de of math. math ‘He must have passed the big maths test.’ b. – N’a-t-il not.have-1sg-he pas neg réussi succeeded son his examen exam de of math? math ‘Didn’t he pass the big maths test?’ We argue here that the high Focus-neg introduces a metaevaluation which, like MUST, produces positive bias. The bias is positive because the negation particle functions here as a modal adverb in focus. The negation particle in a question is forced to this meaning for two reasons. First, just like with REALLY, bias inducing negation bears focus and functions as a focus particle. (Low negation, by contrast, is de-accented and functions normally as the expected proposition negating function). Second, being a focus particle means that (i) there is a set of 20 1 Modalization and bias in questions alternatives in the discourse, and (ii) in questions, the relevant alternative will have to be ¬𝑝. If ¬𝑝is the alternative Focus-neg, which needs to express new information, cannot not function as negation. Focus-neg thus has no contribution other than the ranking presupposition. As we noted for REALLY, the focus adverb (here, neg) requires an alternative contextual assumption of the opposite polarity to make sense, i.e., for it to be a reasonable alternative. In other words, the opposite polarity effect follows from the anaphoricity of focus and the requirement that the focused statement provide new information with respect to the alternative. (41) Jques Focus-neg (PRES (𝑝))K𝒪,M,𝑖is defined only if (i) the modal base 𝑀(𝑖)is nonveridical and partitioned into {𝑝,¬𝑝}worlds. (ii) 𝑝worlds are more preferable than ¬𝑝worlds Jques Focus-neg (PRES (𝑝))K𝒪,M,𝑖={𝑝,¬𝑝} Since a question does not assert ¬𝑝the contribution of Focus-neg as 𝒪arises as a meaning reanalysis. Just like with REALLY, we can assume that neg is in the 𝒪position in the question skeleton – recall that that was also the position of REALLY: (42) CP ques whether1 NonveridicalP 𝒪:Focus-neg NonveridicalP ... 𝑡1𝑀(𝑖) 𝒮 TP Putting the effect of REALLY and negation together, then, we must say that their content is only presuppositional. Thus, our account explains their contribution and the polarity reversal in a very simple way by acknowledging their status as focus operators, and allowing an extended syntactic structure in questions parallel to the (independently motivated) one for modals with a position for 𝒪. No other account that we know of offers an explanation of these facts with such simplicity and no additional ad hoc assumptions. 21 Anastasia Giannakidou & Alda Mari Before closing, we want to go back to the strength of veridical commitment, and note that modals and questions mirror each other in terms of strength. In our theory, commitment with modals proceeds according to the following scale: (43) Scale of veridical commitment (Giannakidou & Mari 2016, 2018b, 2021a,b) 〈𝑝, MUST 𝑝, MIGHT 𝑝, *〉; where 𝑖is the speaker, 𝑝conveys full commitment of 𝑖to 𝑝; MUST 𝑝 conveys partial commitment of 𝑖to 𝑝, and MIGHT 𝑝conveys trivial commitment of 𝑖to 𝑝. No modal expression conveys negative commitment. * in (43) indicates that no epistemic modal can convey negative commitment (see Ernst 2009, Homer 2015, Giannakidou & Mari 2018b as for how this relates to the neg-raising property of universal epistemic modals).12 With REALLY and Focus-neg, the commitment, which is a presupposition rather than an assertion, proceeds as a mirror image: (44) Scale of commitment in question presuppositions: 〈𝑝, FOCUS NEGATION 𝑝, MIGHT 𝑝, REALLY 𝑝〉; where 𝑖is the speaker, 𝑝conveys full commitment of 𝑖to 𝑝; HIGH NEGATION 𝑝conveys partial commitment of 𝑖to 𝑝, MIGHT 𝑝conveys trivial commitment and REALLY conveys negative commitment of 𝑖to 𝑝. Like MUST and MIGHT, Focus-neg and REALLY questions are nonveridical. Focus-neg and MUST are equivalent in terms of function and commitment in questions, and likewise REALLY contributed ranking but is negatively biased for the reasons we explained earlier. Why modal assertions and questions are mirror images of commitment is a question that we cannot answer fully; but within the GM system the contrast can be simply due to the fact that, in questions, REALLY and negation are forced to function in an alternative way given their opposite polarity anaphoric property due to focus. We now complete our exploration of the relation between questions and modality by considering questions where modal particles, including a different kind of negation, are used. The particles, we will show, are equivalent to unbiased possibility modals rather than MUST, and do not rank the two alternative propositions. Rather, they widen the options and create vagueness. The effect we observe is 12Cross-linguistically, attitudes like ‘doubt’ can instead convey negative commitment. As far as we are aware, at least in Indo-European languages, there is no modal verb/auxiliary equivalent. 22 1 Modalization and bias in questions an enhancement of equilibrium, not its shrinking (as is the case with bias) – with the opposite effect of creating more uncertainty. This operation, which we call ‘reflection’, targets the modal base. 5 The reflective question: enhancing uncertainty In this section, we consider the use of possibility modal particles and verbs in questions, resulting in the question now becoming vague and potentially unanswerable. We call such questions reflective. In the literature the term ‘conjectural’ has also been used (see Littell et al. 2010, Matthewson 2010, Eckardt & Beltrama 2019, Frana & Rawlins 2019 a.o.). The reflective question is well documented for a number of languages, including Greek, Japanese and Korean (Kang & Yoon 2018, 2019). In discussing it along with bias, we view the two phenomena as pragmatic duals. 5.1 Modal particles in questions: The data For a long time, it was thought that epistemic modals do not occur in questions (Coates 1983, Drubig 2001, Jackendoff 1972, Leech 1971, McDowell 1987). Jackendoff, specifically, claimed that while may can either be interpreted deontically or epistemically in a declarative sentence (John may leave early tonight), it can only be interpreted deontically in a question (May John leave early tonight?). Yet Ernst (2009) presented examples with modal adverbs in questions (Is she possibly the murderer?), and Hacquard & Wellwood (2012) offer corpus data with possibility modals in questions. The following examples are from that work: (45) a. With the owners and the players on opposite sides philosophically and economically, what might they talk about at the next bargaining session? b. Might he be blackballed by all institutions of higher learning? c. What might the Grizzlies have been like if their leading scorer and rebounder, 6-foot-10 center Brent Smith, had not missed his third straight game because of a sprained ankle? The authors conclude that epistemic modals of possibility are very natural in questions. In Greek, we find possibility modal verbs and particles in questions: the socalled evaluative subjunctive (Giannakidou 2017), and the particles taxa, mipos, arage ‘maybe/possibly’, the latter two only used in questions, as we see below 23 Anastasia Giannakidou & Alda Mari with examples from Giannakidou (2017) (see earlier discussions in Giannakidou 2009, Rouchota 1994): (46) Pjos Who irthe came.3sg sto to-the party? party ‘Who came to the party?’ (47) Pjos Who 𝑛𝑎 subj irthe came.3sg sto to-the party? party ‘Who might have come to the party?’ (48) Pjos Who 𝑏𝑜𝑟𝑖 might / / *prepi must na subj irthe came.3sg sto to-the party? party ‘Who might/#must have come to the party?’ (49) (𝑁𝑎) subjV tou him milise talked-3sg (arage/mipos/taxa)? particle ‘Might she have talked to him?’ (50) Tou him milise talked-3sg (#arage/#mipos/taxa). particle ‘She talked to him (as-if).’ (51) Tou him milise? talked-3sg ‘Did she talk to him?’ Here we observe the subjunctive, the possibility modal bori and the particles arage/mipos/taxa in questions (both polar and wh-questions). As we can see, the particles can spread in a fashion reminiscent of modal spread with modal verbs and adverbs (He may possibly be here tonight). The MUST modal is excluded as can be seen in (49), indicating that we are not dealing with a bias related phenomenon. In the declarative context, only taxa can be used, which means literally ‘as if, allegedly’ and casts doubt on the truth of the proposition is attaches to (see Giannakidou 2022 for more details on the Greek particles; Ifantidou 2001). The particles correspond to might, as indicated in the English translations. Giannakidou (2017) argues that the particle question differs from the plain one in being open-ended, vague, and primarily self-addressed. It does not require an actual or full answer; for instance, Pu na evala ta gialia mou? ‘Where might I have put my glasses?’ is a question that one poses to herself in a wondering, reflective manner without expectation of knowing the answer. Bare questions cannot be used this way. (In fact, first person unmodalized questions are quite uncommon). 24 1 Modalization and bias in questions This open-endedness is found also in Japanese and Korean discussed recently in Kang & Yoon (2018, 2019), and with subjunctive questions in Salish (Matthewson’s (2010) ‘conjectural’ questions). English data from Ernst (2009), Hacquard & Wellwood (2012) illustrate the same reflective character for English questions with possibility modals: (52) Is she possibly/*probably a spy? (53) Might/*must she be a spy? These have been described as ‘weaker’ questions. Kang & Yoon (2018, 2019) in their discussion of a similar use of the particle nka in Korean note that by using nka, the speaker “reflects on her own background assumptions and is not simply requesting information from the addressee”. Nka-questions, they argue, are feigned monologues, i.e., the speaker says something as if it were a monologue without expecting an answer necessarily. Because of the monologic nature of the utterance, it does not necessarily obligate the hearer to respond, and while there may be differences between the particles and across languages, these observations hold also for the Greek and English particle questions above. The self-reflective character of particle questions is further evidenced by the fact that directly addressing the hearer is odd. (54) # Ti what na subj efages ate.2sg arage araje xthes? xthes ‘What might you have eaten yesterday?’ (55) Ti what efages ate.2sg xthes? xthes ‘What did you eat yesterday?’ (56) # Na subj efages ate.2sg araje araje xthes? xthes ‘Might you have eaten yesterday?’ (57) Efages ate.2sg xthes? xthes ‘Did you eat yesterday?’ For the Greek particles, Giannakidou (2022) calls this ‘anti-addressee’ effect. Cross-linguistically, the effect is also observed in Korean (Kang & Yoon 2019). This effect has been discussed in the literature by appealing to constraints on the ‘interrogative flip’, that is to say the phenomenon whereby questions containing 25 Anastasia Giannakidou & Alda Mari an evidential target the addressee mental state (a.o. Bhadra 2017). Extending the view that evidentials enhance the interrogative flip to epistemic modals, one could argue (see e.g. Eckardt & Beltrama 2019) that what the modal or particle is questioning is the addressee’s knowledge, and that a modal question presupposes lack of knowledge on the part of the addressee. Mari (2021) shows that existential epistemic modal questions (58) – which include Italian future questions (59–60) – are by default self addressed (see also Eckardt & Beltrama 2019; pace Ippolito & Farkas 2022). (58) Dove where possono might essere be i the miei my occhiali glasses ? ‘Where might my glasses be?’ (59) Dove where saranno be.fut.3pl i the miei my occhiali glasses ? ‘Where might my glasses be?’ (60) Sarà be.fut.3pl a at casa? home ‘Might he be home?’ Such questions are akin to questions with forse which is the possibility adverb ‘maybe’ in Italian:13 (61) a. È Is a at casa? home ‘Is he at home?’ b. È Is forse maybe a at casa? home ‘Is he maybe at home?’ 13The Italian future has been considered as an epistemic modal (see Pietrandrea 2005, Mari 2009, Giannakidou & Mari 2018a, Baranzini & Mari 2019; a contrario, Mari 2010, Frana & Rawlins 2019; Eckardt & Beltrama 2019 for Italian future as an evidential.) The German modal particle wohl is also reported to have similar use of adding a speculative component (the example below is from Zimmermann 2011 with his translation). Zimmermann (2011) writes: “The question above is not about whether or not Hans has invited Mary, but by using wohl the speaker indicates her awareness that the addressee may not be fully committed to her answer.” (Zimmermann 2011: 2020): (i) Hat has Hans Hans wohl prt Maria Mary eingeladen? invited ’What do you reckon: Has Hans invited Mary?’ 26 1 Modalization and bias in questions This way of understanding the impossibility to be addressee-oriented needs to postulate a device to undo the flip and render the question reflective when needed (see discussion in Frana & Rawlins 2019 and Mari 2021). Our view that these questions are, to begin with, reflective explains right from the bat the impossibility of being addressee-addressed without needing to stipulate any device to undo a putative flip. Characteristically, the Greek negative particle mi(n) can also be used in questions with the same reflective flavor (Chatzopoulou 2018): (62) Min neg eidate saw.2pl ton the Jani? John ‘Did you maybe see John?’ This use of a negative particle, which we can call Modal-neg, remains productive in literary and other registers. Mi(n) is the negation of non-indicative contexts in Greek (Giannakidou 1997, Veloudis 1980), dhen being the negation in indicative sentences. The min negation is the semantic allomorph appearing in subjunctive contexts: (63) *Min/Dhen neg eidate saw.2pl ton the Jani. John ‘You did not see John.’ (64) Na subj min/*dhen neg deite see.2pl ton the Jani. John ‘Do not see John!’ The particle mipos, emerges from a historical path that fused the negation mi with the indicative complementizer pos. Crucially, the reflective negation min is not used for bias: (65) Dhen/*Min neg eidate saw.2pl ton the Jani? John ‘Didn’t you see John?’ (66) Idate saw.2pl ton the Jani, John, etsi so dhen/*min not einai? is ‘You saw John, didn’t you?’ So, in Greek, there is a bias-producing focus negation dhen, and a reflective negation min. The fact that negation behaves like a modal particle speaks again 27 Anastasia Giannakidou & Alda Mari to our earlier point: a seemingly non modal element can acquire a modal function in questions because, these lacking truth conditions, it cannot function in the canonical way.14 Questions with possibility modals are weakened versions of the informationseeking question – weaker in that they involve more uncertainty. Notice that they can be continued by Who knows?, unlike regular information questions: (67) A: Pjos Who (arage) arage-particle.subj na V irthe came-3sg sto to-the party? party Pjos who kseri! knows ‘Who might have come to the party? Who knows!’ (68) Sarà be.fut.3sg a at casa? home Chi who lo that sa! knows ‘Might he be home? Who knows !’ We can thus see that possibility modals (verbs, adverbs, and particles) create a vague question with more uncertainty, harder or even impossible to answer. The intent of the speaker, when choosing to use the possibility modal, is not simply to seek information from an addressee, but rather to signal that the question is hard to answer, that the hearer may not be an authority on giving a true answer, or that there is no addresse at all. Reflections, as we noted earlier, can actually be incompatible with the second person (what we called anti-addressee effect). And, as Kang & Yoon (2019) observe, they have the flavor of a monologue in being directed towards the speaker herself. Let us now proceed to capture the effect of the epistemic modal in the framework we have developed. 5.2 Analysis: modal particles widen the modal base The first analysis we know that addresses the use of of possibility modals in questions is Giannakidou & Mari (2016). They argued that, in Greek, possibility modals and the subjunctive are equivalent, and that, for the MIGHT question, the answer set contains modalized propositions: 14Salish employs negation in questions too; the following example is from Matthewson (2022): (i) A: Nee=hl neg=CN gwila-n=aa? blanket=2sg.ii=Q ‘Do you have a blanket?’ This is a neutral question according to Matthewson, hence this negation seems more akin to a question particle. 28 1 Modalization and bias in questions (69) JWho came to the party?K= {Bill came to the party, Marina came to the party, Ariadne came to the party, Nicholas came to the party, ...} (70) JWho might have come to the party?K= {it is possible that Bill came to the party, it is possible that Marina came to the party, it is possible that Ariadne came to the party, it is possible that Nicholas came to the party, ...} This analysis says that the possibility question is still an information seeking question, but instead of asking about 𝑝, it asks about POSSIBLY 𝑝. While a question with MIGHT indeed may ask about the possibility of 𝑝– like, for instance Is it possible that Ariadne ate? – the reflective effect is more than that. In our current framework which uses metaevaluation, we will argue that the possibility expression enlarges the spectrum of the possibilities. Possibility expressions widen the set of the possibilities considered attracting attention to 𝑝, but seeking 𝑝in a larger set. (71) Ernie is possibly a communist? (72) CP ques whether1 NonveridicalP 𝒪:POSSIBLY NonveridicalP ... 𝑡1𝑀(𝑖) 𝒮 TP (73) Jques POSSIBLY(pres(𝑝))K𝒪,M,𝑖is defined only if (i) 𝑀(𝑖)is nonveridical and partitioned into {𝑝,¬𝑝}worlds. (ii) ∩𝒪⊃𝑀(𝑖) Jques POSSIBLY(pres(𝑝))K𝒪,M,𝑖={𝑝,¬𝑝} The set of possibilities extends beyond the modal base thus making it harder to think of what would be a “correct” answer. This accounts for why reflection questions seem open-ended, vague, and potentially lacking answers. The epistemic state of the speaker entertains a broader set of potential answers, not a narrower one as in the case bias were a potential answer is favored by the speaker’s 29 Anastasia Giannakidou & Alda Mari Repp, Sophie. 2013. Common ground management: Modal particles, illocutionary negation and verum. In Daniel Gutzmann & Hans-Martin Gärtner (eds.), Expressives and beyond: Explorations of conventional non-truth-conditional meaning, 231–274. Leiden: Brill. Roberts, Craige. 1996. Information structure: Towards an integrated formal theory of pragmatics. In Jae Hak Yoon & Andreas Kathol (eds.), OSU Working Papers in Linguistics, vol. 49, 91–136. Columbus: The Ohio State University Department of Linguistics. Roelofsen, Floris & Donka F. Farkas. 2015. Polarity particle responses as a window onto the interpretation of questions and assertions. Language 91(2). 359–414. Romero, Maribel & Chung-Hye Han. 2004. On negative yes/no questions. Linguistics and Philosophy 27(5). 609–658. Rouchota, Villy. 1994. On indefinite descriptions. Journal of Linguistics 30(2). 441– 475. Sadock, Jerrold M. 1971. Queclaratives. Papers from the Regional Meeting of the Chicago Linguistic Society (CLS) 7. 223–231. Sherman, Brett. 2018. Open questions and epistemic necessity. The Philosophical Quarterly 68(273). 819–840. Stalnaker, Robert. 1978. Assertion. In Peter Cole (ed.), Syntax and semantics, vol. 9, 315–332. New York: Academic Press. Sudo, Yasutada. 2013. Biased polar questions in English and Japanese. In Daniel Gutzmann & Hans-Martin Gärtner (eds.), Beyond expressives: Explorations in use-conditional meaning, vol. 28 (Current Research in the Semantics/Pragmatics Interface (CRiSPI)), 275–296. Leiden: Brill. Van Rooy, Robert & Maria Šafářová. 2003. On polar questions. In Robert B. Young & Yuping Zhou (eds.), Semantics and Linguistic Theory (SALT) 13, 292–309. Cornell University, Ithaca: CLC Publications. Veloudis, Ioannis. 1980. Negation in Modern Greek. University of Reading. (Doctoral dissertation). von Fintel, Kai. 2008. What is presupposition accommodation, again? Philosophical Perspectives 22. 137–170. von Fintel, Kai & Anthony S. Gillies. 2010. Must... Stay... Strong! Natural Language Semantics 18. 351–383. Zimmermann, Malte. 2011. Discourse particles. In Klaus von Heusinger, Claudia Maienborn & Paul Portner (eds.), Semantics (HSK 33.2), 2012–2038. Berlin: De Gruyter. 36 Chapter 2 Evidential bias across clause types Beste Kamali University of Amsterdam, Institute of Logic, Language, and Computation Declarative (polar) questions such as English rising declaratives are associated with a range of effects including positive evidential bias, which are commonly analyzed as arising from the non-interrogative clause type of these questions (Gunlogson 2003, Farkas & Roelofsen 2017, among others). A detailed comparison between English rising declaratives and Turkish object attachment interrogatives shows that (a) positive evidential bias is not an exclusive feature of declarative questions and (b) in fact a cluster of effects beyond evidential bias is common to rising declaratives and this particular kind of polar interrogative. I argue that both question forms are underlain by the monopolar question meaning in the sense of Krifka (2015), which can yield the evidential bias cluster of effects across clause types. The account helps to disentangle types of biased inferences some may still be rooted in clause type, offers a way of approaching variation in bias paradigms, and makes typological predictions regarding the relationship between monopolar question meaning and clause type. 1 Introduction Some polar question forms trigger pragmatic inferences that go beyond the simple truth conditional meaning associated with polar interrogatives, globally called bias. The English rising declarative, a particularly well-researched declarative question, presents a prime case of biased question. In (1), if we first observe the use in context, while the polar interrogative felicitously conveys an unbiased hope to find jumper cables, the declarative question infelicitously appears to presume that such a hope is warranted. Further, certain lexical and grammatical elements such as an adverb of complete ignorance like by chance require an unbiased context and are infelicitous in a rising declarative. Beste Kamali. 2025. Evidential bias across clause types. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions: Experimental results and theoretical modelling, 37–71. Berlin: Language Science Press. DOI: 10.5281/zenodo.17158178 Beste Kamali (1) To passerby in a parking lot Excuse me, but my battery’s dead … a. Do you (by chance) have jumper cables? b. # You’ve (by chance) got jumper cables? (Gunlogson 2003) English rising declaratives are also known for their requirement of public evidence in favor of the content proposition of the question (Gunlogson 2003). This notion has come to be known as contextual bias or evidential bias (Sudo 2013). In (2a), the speaker infers that it is raining on the basis of the contextual evidence of wet coat, and the rising declarative is completely fitting. Whereas in contexts without such evidence (2b) or those with counterevidence (2c), a rising declarative is not felicitous. Hence, the rising declarative is said to bring about inferences of positive evidential bias.1 (2) a. [Ev. bias +] My friend enters my windowless office in a dripping wet coat. It’s raining? / Is it raining? b. [Ev. bias 0] Speakers are on the phone, no related utterance has been made. # It’s raining? / Is it raining? c. [Ev. bias -] My friend enters my windowless office in visibly dry summer clothes. # It’s raining? / Is it raining? (2) also shows that in contrast to the rising declarative, the English polar interrogative is felicitous in all three contexts, hence it is termed evidentially neutral or evidentially unbiased. The English polar interrogative appears to be unrestricted and unmarked in numerous other respects, including the availability of adverbs of complete ignorance, as seen in (1a). Beyond its bias properties, rising declaratives exhibit a host of other peculiarities, some of which are not straightforwardly relatable to evidential bias or bias in general. (3) summarizes some poignant cases including usage restrictions (infelicity when used as exam questions or as polite requests) and more structural features (the inability to support negative polarity items and to take part in polar alternative questions). The data in (3) come from Huddleston (1994) and Gunlogson (2003). 1Throughout, I will be using Sudo’s (2013) terminology that distinguishes these three evidential conditions as positive, zero, and negative bias conditions. 38 2 Evidential bias across clause types (3) a. # The empty set is a member of itself? Exam question b. # You can pass me the salt? Polite request c. # Anybody’s home? NPI d. # She ordered coffee, or not? Polar Alt Notable analyses of the meaning of English rising declaratives treat the evidential bias inferences as an extension of the generally biased nature of the form, typically only making reference to bias in an unspecified manner. Similarly common across the literature is the centrality of clause type as the root cause of these inferences. In the seminal work of Gunlogson (2003), declaratives are informative and inherently express commitment by their nature. This accounts for the infelicity in examples such as (1b), (3a) and (3b), where a commitment by the addressee cannot be said to be involved. The evidential bias observed in (2) is a consequence of informativeness in addition to commitment. Uninformativeness with respect to the addressee is a necessary condition for an utterance to be interpreted as a question act. Declaratives can be interpreted as a question only if the addressee is already understood to be publicly committed to the content proposition. For this reason, inferences based on evidence need to be public. Farkas & Roelofsen (2017) offer a larger framework of bias and neutrality based on principles of markedness. This account relies heavily on the marked status of question forms such as tags and declarative questions as opposed to the default status of polar interrogatives, argued to result in special discourse effects including bias. For Rudin (2022), who follows the Table framework of Farkas & Bruce (2010), the rising tune found both on the polar interrogative and the rising declarative serves to block the addition of a speaker commitment in the form of the content proposition 𝜙to the discourse commitments and forces the addressee to make one, because Issues must be resolved. What separates the rising declarative and the polar interrogative is that the former, unlike the latter, fails to project a future common ground including ¬𝜙and in combination with broader pragmatic principles, this leads to the inference that the speaker takes the addressee to have a bias towards 𝜙, which subsumes but also goes beyond evidential bias. I introduce in this paper the case of an unmarked polar interrogative from Turkish which triggers inferences of evidential bias without any other associations of bias. The data thus shows that neither declarative clause type nor bias in the general sense is a necessary condition for evidential bias to arise. The necessary condition, I argue, is an underlyingly monopolar meaning in the sense of Krifka (2015). I further show that, despite its interrogative clause type, this form aligns with rising declaratives in a number of ways including all of the features 39 Beste Kamali exemplified in (3), which I dub the Ev+ cluster of effects. Moreover, this cluster is partially observable in evidentially biased forms found in other languages, providing further empirical footing for the approach. For English rising declaratives, the implication of the analysis is that evidential bias and what I will argue to be related effects are driven by the monopolar denotation of these questions, much in the same spirit as Rudin (2022). However, I do not propose to supplant the cumulative understanding of rising declaratives with one based on evidential bias. That would be contrary to the empirical evidence, as we will see below. What I aim to do is to illustrate the viability of a crosslinguistically and cross-structurally valid alternative to the analysis of the empirical presentation of rising declaratives that concerns evidential bias and related effects. The paper continues in Section 2 with the empirical discussion, where Turkish polar questions, and specifically the evidentially biased object attachment questions, are introduced, followed by the introduction of the Ev+ cluster. In Section 3, an account is sketched which relies on the monopolar question meaning proposed by Krifka (2015), which I argue leads to inferences of positive evidential bias and the Ev+ cluster due to its interactions with focus. Section 4 documents differences between the Turkish and the English Ev+ forms and suggests that these may indeed be clause type-based. Section 5 compiles existing findings in unambiguous Ev+ forms in two further languages and illustrates the commonness of the features. Section 6 reconsiders the connection between declarative clause type and Ev+ and suggests that the connection is through the defaultness of the monopolar meaning. Section 7 concludes the paper. 2 The evidentially biased Turkish polar interrogative and the Ev+ cluster Let us delve into the two empirical novelties that necessitate a reconsideration of the connection between evidential bias and clause type and the generalization regarding the family of effects that cluster with evidential bias across clause types. Most of the observations in this section will be based on Turkish polar questions, which I first briefly introduce. 2.1 Turkish polar questions at a glance Polar questions in Turkish are built by the addition of the vowel-harmonic clitic -mI to the corresponding declarative. The clitic is focus-sensitive, therefore it 40 2 Evidential bias across clause types attaches to a focused element (Ladd 1996, Kornfilt 1997). In (4b), a cleft translation is adopted to facilitate the intended focus interpretation, while Turkish polar questions do not show any signs of being clefts or have any other marked morphosyntax other than carrying the clitic -mI (boldfaced throughout). The reader is referred to Kamali & Krifka (2020) for a recent and empirically detailed treatment of focus in Turkish polar questions. (4) a. Ali Ali dün yesterday yemek dinner yap-tı. make-past ‘Ali made dinner yesterday.’ b. Ali Ali mi MI dün yesterday yemek dinner yap-tı? make-past ‘Was it Ali that made dinner yesterday?’ When no particular element is in focus, there are two possible attachment sites (Kamali 2011a,b). The clitic may be found on the discourse-new object, or on the verb. I will call these object attachment and verb attachment questions. Both instances may be found in all-new and out-of-the-blue situations such as (5). Angular brackets signify mutually exclusive potential attachment sites. (5) A, B and Ali are roommates. They meet at the breakfast table for the first time after B’s month-long trip. A: Good morning! I didn’t know you were back. B: Good morning! You were both asleep when I arrived. Tell me … Ali Ali dün yesterday yemek dinner <mi> MI yap-tı make-past <mı>? MI ‘Did Ali make dinner yesterday?’ Both forms are surface-ambiguous with multiple interpretations available due to limits on attachment site among other things. The clitic is attached to the verb if any overt marking in the verbal morphological complex is focused, such as the verb stem or tense. It is also attached to the verbal complex in cases of verum/polarity focus which require no further marking. Note that (5) is a context where none of these focus interpretations is supported without accommodation, which we will notice in various other contexts in due course. I use the term verb attachment only to refer to the broad focus/focusless interpretation of polar questions carrying the clitic on the verbal complex. A similar observation pertains to cases where the clitic is found on the object. Unsurprisingly, this attachment site is used for cases of object focus (6a). Not only 41 Beste Kamali that, VP focus is also expressed this way (6b), suggesting that the correspondence from object to VP to broad focus is regulated by focus projection in the sense of Selkirk (1995). (6) shows this through the example of alternative questions, which in Turkish carry -mI on both disjuncts. Note, again, that these construals (the part of the AltQ up to the disjunction) are not felicitous in (5) without accommodation. So, again, by the term object attachment, I will only be referring to the broad focus construal felicitous in (5). (6) a. Ali Ali [yemek]𝐹 dinner mi MI yap-tı, make-past yoksa or [tatlı]𝐹 dessert mı? MI ‘Did Ali make [dinner]𝐹or [dessert]𝐹?’ b. Ali Ali [𝑉𝑃yemek dinner mi MI yap-tı]𝐹 make-past yoksa or [𝑉𝑃ders lesson mi MI çalış-tı]𝐹? study-past ‘Did Ali [𝑉𝑃make dinner]𝐹or [𝑉𝑃study his lesson]𝐹?’ A consequence of focus projection is that polar questions lacking a discoursenew object will automatically cause the clitic to be on the verb. Hence, data with unergative verbs or those with a discourse-given object (along with the wellknown interactions with differential object marking, cf. Özge 2011) are uninformative for the purposes of this paper and are disregarded. The meaning of the verb attachment and the object attachment options differ in interesting ways. In particular, as we will shortly see, object attachment shares important features with the English rising declarative. We will also be referring to verb attachment for comparison and for a fuller understanding of the paradigm. 2.2 The evidentially biased polar interrogative In (5), both clitic placement options may be possible, but they have different meanings. Object attachment, in particular, implicates the presence of a nonlinguistic cue in favor of the content proposition (first intuited to my knowledge by Göksel & Kerslake 2005). Consider possible continuations to the conversation in (5), which are not interchangeable. (7) A: Yes. Why do you ask? B′: [Continuing from object attachment] When he cooks, he always leaves the kitchen in this kind of mess, that’s why. B′′: [Continuing from verb attachment] I’m starving, that’s why. 42 2 Evidential bias across clause types Indeed, object attachment questions carry positive evidential bias exactly like English rising declaratives. They are licensed in the presence of evidence (8a), and not in the absence of evidence (8b) or in the presence of counterevidence (8c). For this and other resons we will see throughout, object attachment questions and similar forms are translated with an English rising declarative in (8) and elsewhere.2To aide the reader, when a translation of a polar question is itself infelicitous in the given (pragmatic or structural) context, I prefix the judgement in square brackets. I will be referring to positive evidential bias as Ev+ in short. (8) a. [Ev. bias +] Speaker observes tell-tale signs of Ali’s recent cooking. Ali Ali dün yesterday yemek dinner mi MI yap-tı? make-past ‘Ali made dinner yesterday?’ b. [Ev. bias 0] Speakers are on the phone. No related utterance has been made. # Ali dün yemek mi yaptı? ‘[#]Ali made dinner yesterday?’ c. [Ev. bias -] Speaker notices the kitchen looking and smelling exactly as they left it the previous day. # Ali dün yemek mi yaptı? ‘[#]Ali made dinner yesterday?’ From the perspective of what we know about English rising declaratives, we could expect the Ev+ object attachment questions to be declarative questions. This is not correct, however. Both verb attachment questions and object attachment questions are interrogatives. Let us see why. First, contexts of unresolved question typically require a true interrogative, barring declarative questions (Gunlogson 2003) (9b). In Turkish, the clitic -mI is required in such contexts regardless of position, hence both object and verb attachment align with the English polar interrogative and not with with declarative question (9a). (9) a. Bir one soru question cevapsız. answerless Ali Ali yemek dinner <mi> MI yap-tı make-past <mı>? MI ‘One question remains. Did Ali make dinner?’ b. # One question remains. Ali made dinner? 2I save the English polar interrogative to translate verb attachment questions and polar questions where attachment site does not matter for the point being made. As I will explore, neither corresponcence is perfect, but this two-way division is both maximally accurate and, I believe, reader-friendly. 43 Beste Kamali Secondly, interrogatives are incompatible with certain biased adverbs such as adverbs of speaker certainty or evidentiality (Huddleston 1994), whereas declarative questions are fine with them (Gunlogson 2003). In Turkish, neither object nor verb attachment questions may feature these adverbials (10a), again patterning with interrogatives. (10) a. # Müdür manager buna this kuşkusuz/besbelli certainly/evidently izin permission <mi> MI verdi gave <mi>? MI ‘[#]Has the manager certainly/evidently given permission for this?’ b. The manager has certainly/evidently given permission for this? Thirdly, interrogatives, but not declaratives, can be embedded under a rogative predicate. (11a) shows a grammatical embedding under wonder of both verb and object attachment options. (11b) shows that the clitic is obligatory in this complement clause, indicating, again, that both forms are interrogatives. (11) a. Ali Ali yemek dinner <mi> MI yap-tı make-past <mı>, MI merak wonder ed-iyor-um. do-pres-1sg ‘I wonder if Ali made dinner.’ b. * Ali yemek yaptı, merak ediyorum. (‘I wonder if Ali made dinner.’) The equally common nominalized embedding form can also be used to show this, but with a few clarifications. (12a) shows the baseline example where a declarative is embedded under a non-rogative verb, rejecting a rogative verb. When the verb attachment form is nominalized, the clitic has to be replaced by a periphrastic construction expanding the standard nominalization with a V-notV form (12b). The nominalization of the object attachment question, in contrast, retains the clitic on the embedded object while using the standard nominalization (12c) (Kamali 2011b). In neither case does nominalization itself suffice to embed the interrogative clauses, indicating that the periphrasis in (12b) and the clitic in (12c) are interrogative markers. Indeed, neither form accepts a non-rogative matrix predicate and requires a rogative one. The meaning of (12b) and (12c) are for all relevant purposes identical to the meaning of (11a) and (11b). (12) a. Merve Merve [Ali’nin Ali-gen yemek dinner yap-tığ-ın]-ı make-nomin-3sg-acc sanıyor thinks / *soruyor. asks ‘Merve thinks/*asks that Ali made dinner.’ 44 2 Evidential bias across clause types b. Merve Merve [Ali’nin Ali-gen yemek dinner yap-ıp make-conv yap-ma-dığ-ın]-ı make-neg-nomin-3sg-acc soruyor asks / *sanıyor. thinks ‘Merve asks/*thinks if Ali made dinner.’ c. Merve Merve [Ali’nin Ali-gen yemek dinner mi MI yap-tığ-ın]-ı make-nomin-3sg-acc soruyor asks / *sanıyor. thinks ‘Merve asks/*thinks if Ali made dinner.’ So, we see here a polar question form, the Turkish object attachment question, that is restricted to positive evidential bias contexts while being an interrogative through and through. As it is not a declarative, the Ev+ feature of the Turkish object attachment question cannot be due to clause type. The competing form in the paradigm, the verb attachment question, is also an interrogative, so there are no surprises there. Now, even if both object attachment and verb attachment questions are interrogatives, there could be a markedness asymmetry between the two forms. This could put the object attachment form in a marked position and may lead to its associations with bias. When we look at the full paradigm with verb attachment questions, we see that this cannot be the case. Departing from the distribution of polar interrogatives in the English paradigm, the competing form in Turkish, verb attachment, is not unmarked. Recall that the English polar interrogative is in many ways a default. Directly related to our purposes, it is evidentially neutral (2). The Turkish verb attachment question, in contrast, is not a default. We will see several other reflections of this in due course. For now, observe that it is not evidentially neutral (13). Namely, it is banned from positive evidential contexts (13a) (it is Ev-/Ev0, or has “antievidential bias”). As such, it is in complementary distribution with the object attachment form. (As English does not have a form that parallels this distribution, the translations given are approximate.) (13) a. [Ev. bias +] Speaker observes tell-tale signs of Ali’s recent cooking. # Ali Ali dün yesterday yemek dinner yap-tı make-past mı? MI (‘Did Ali make dinner yesterday?’) b. [Ev. bias 0] Speakers are on the phone. No related utterance has been made. Ali dün yemek yaptı mı? ‘Did Ali make dinner yesterday?’ 45 Beste Kamali nopolar question meaning provides the core meaning which leads to positive evidential bias and the Ev+ cluster, while the Turkish verb attachment question stands out as a true bipolar question manifesting the absence of these effects. The account does not make reference to clause type. Polar interrogatives as well as declarative questions may have this underlying meaning (but see Section 6 on why declaratives may have to). Krifka (2015) argues that there are two possible polar question meanings. The bipolar meaning denotes the speaker’s invitation to restrict future commitments in the discourse to either the content proposition 𝜙or its complement ¬𝜙, similar in essence to the standard two-membered set of polar question meaning (Hamblin 1973, Karttunen 1977). These questions may be answered with a conclusive no or a negative fragment and take part in polarity alternative questions. I simplify below. (26) Bipolar question: 𝜙? = Do you commit to 𝜙or ¬𝜙? A: I’m hungry. Did you make dinner (or not)? B: Yes. / No. / (No,) I didn’t. The second meaning is the monopolar meaning, which denotes an invitation to restrict future commitments to the content question itself without its complement.8These questions may not occur with the polar alternative but rather may occur with an open alternative such as or what. Their possible answers are also restricted, making a conclusive no answer degraded and a negative fragment even worse. This is due to the absence of the complement proposition ¬𝜙from the meaning. (27) Monopolar question: 𝜙? = Do you commit to 𝜙? A: It smells delicious. Did you make dinner (or what)? B: Yes. / ?No. / #(No,) I didn’t. Krifka already envisions that English rising declaratives are biased towards the yes answer because they are monopolar. Differently from Krifka, I would like to connect the monopolar meaning to evidential bias in particular, and along with that to the rest of the Ev+, with no association to clause type. 8An earlier proposal of this sort is due to Biezma & Rawlins (2012) where a monopolar propositional content is interpreted with a Q operator. Their account is stronger, though, in postulating this meaning to be the default polar question meaning. Hence, the English polar interrogative as well as presumably the rising declarative would receive a monopolar analysis. 52 2 Evidential bias across clause types 3.1 Rising declaratives and object attachment questions are monopolar First, let us observe that rising declaratives and object attachment questions both behave in a way expected of monopolar questions. A conclusive no answer to a rising declarative or an object attachment question is degraded. A negative predicate fragment is unacceptable. (For contrast, see (30) and (40) for examples of what I take to be non-conclusive no answers.) (28) A: Ali yemek mi yaptı? ‘Ali made dinner?’ B: Evet. yes /?Hayır. no /#(Hayır,) no yap-ma-dı. make-neg-past ‘Yes. / ?No. / #(No,) he didn’t.’ As we see in (23) and (29), neither rising declaratives nor object attachment questions can take part in a polar alternative question. The infelicity of the no answer and the polar alternative are in line with the missing alternative ¬𝜙from the monopolar meaning. (29) # Ali Ali yemek dinner mi MI yap-tı, make-past yoksa or (Ali Ali yemek) dinner yap-ma-dı make-neg-past mı? MI ‘[#]Ali made dinner, or (did he) not?’ Recall that the Turkish object attachment question is a proper polar interrogative, so it should in principle be able to take part in a polar alternative question or be answered with a no in a conclusive manner, which might not be said that easily of the rising declarative. Still, this polar interrogative diverges from the expected pattern in line with the usage restrictions we have observed earlier. Where both Turkish object attachment questions and English rising declaratives do occur is contexts where distinct propositional alternatives are evaluated instead of the polar alternative ¬𝜙, such as 𝜓the cat knocked over the shelves, or Hasan will order pizza etc. We can observe the surfacing of such an alternative, for instance, in the form of a voluntary addition to a no answer as in (30). It is noteworthy that the speakers appear to be spontaneously converging on a Question Under Discussion (QUD) (Roberts 1996) like What happened?. (30) A: Ali yemek mi yaptı? ‘Ali made dinner?’ 53 Beste Kamali B: Hayır, no kedi cat raflar-ı shelves-acc devir-di. knock.over-past ‘No, the cat knocked over the shelves.’ Another example where distinct propositional alternatives surface under monopolar question forms is propositional alternative questions with distinct propositional alternatives (31). Here, again, 𝜙and 𝜓are contrasted. This is in stark contrast to the unavailability of polar alternative questions contrasting 𝜙and ¬𝜙 given in (29). (31) Ali Ali yemek dinner mi MI yap-tı, make-past yoksa or Hasan Hasan pizza pizza mı MI söyle-yecek? order-fut ‘Did Ali make dinner or will Hasan order pizza?’ ‘Ali made dinner or Hasan will order pizza? (Which one is it?)’ (31) suggests that a monopolar question with content proposition 𝜙can project broad focus alternatives similar to an assertion of the proposition 𝜙. Indeed, the same 𝜙in an assertion (32) and a rising declarative (33) may be cued by what happened, a trigger of broad focus alternatives. (33) would be expressed by object attachment in Turkish, skipped here for reasons of space. (32) A: What happened? B: The shelves collapsed. Other broad focus alternatives not picked up by the speaker: {Somebody broke the window, officemate fell off the ladder while changing a lightbulb, the loudspeakers malfunctioned …} (33) A: What happened? The shelves collapsed? Other broad focus alternatives not picked up by the speaker: {Somebody broke the window, officemate fell off the ladder while changing a lightbulb, the loudspeakers malfunctioned …} So, both English rising declaratives and Turkish object attachment questions, the difference in their clause types notwithstanding, can be argued to exhibit the monopolar question meaning.9This meaning lacks the polar alternative but allows the generation of broad focus alternatives via focus. In contrast, as we will see below, bipolar questions only lead to the two polar alternatives. One consequence of the monopolar/bipolar division is that the polar alternative ¬𝜙 and broad focus alternatives {𝜓,𝜋…} do not mix. 9Rudin (2022) also argues English rising declaratives to be monopolar. 54 2 Evidential bias across clause types 3.2 Verb attachment questions are bipolar Equipped with this idea, let us consider bipolar questions. English does not have an unambiguous bipolar question form. The English polar interrogative may be found in an alternative question with distinct propositional alternatives as in (31) as well as a polar alternative question as in (29), see (34). (34) a. Did Ali make dinner or will Hasan order pizza? b. Did Ali make dinner, or not? But Turkish verb attachment questions, complementary to object attachment questions in a range of features reviewed earlier, behave perfectly like unambiguous bipolar questions. In terms of the felicity of a no answer and the polar alternative, we see that verb attachment supplies the correct form. (35) a. A: Ali Ali yemek dinner yaptı make-past mı? MI ‘Did Ali make dinner?’ b. B: Evet. yes / Hayır. no / (Hayır,) no yap-ma-dı. make-neg-past ‘Yes. / No. / (No,) he didn’t.’ (36) Ali Ali yemek dinner yap-tı make-past mı, MI yoksa or (Ali (Ali yemek) dinner) yap-ma-dı make-neg-past mı? MI ‘Did Ali make dinner, or (did he) not?’ Conversely, verb attachment is infelicitous in cases relying on a distinct propositional alternative like (31), illustrating the unavailability of focus alternatives when bipolar polar alternatives are in question (37). As with the object attachment question, this form is a proper polar interrogative and should not have a problem being in a proposition-level alternative question, but in fact it does. (37) # Ali Ali yemek dinner yap-tı make-past mı, MI yoksa or Hasan Hasan pizza pizza söyle-yecek order-fut mi? MI (‘Did Ali make dinner, or will Hasan order pizza?’) Similarly, (38) shows the closest verb attachment equivalent of the discourse in (30). Unlike object attachment, with verb attachment there is a clear judgement that B’s no answer resolves the QUD. An attempted continuation parallel to the corresponding object attachment example is infelicitous (38b). One may utter this proposition next in the discourse, but it must be separated from the previous QUD with a full prosodic and topical reset (38c). 55 Beste Kamali (38) a. A: Ali yemek yaptı mı? ‘Did Ali make dinner?’ b. B: # Hayır, kedi rafları devirdi. (‘No, the cat knocked over the shelves.’) c. B′: Hayır. Fark ettin mi? Kedi rafları devirdi. ‘No. Have you noticed? The cat knocked over the shelves.’ Based on these tests, verb attachment questions prove to be a good candidate for a true bipolar question. Conversely, object attachment questions and rising declaratives both demonstrate features expected of monopolar questions. Between the two meanings, not only the presence of the polar alternative ¬𝜙, but also the concomitant availability of distinct focus alternatives varies. (39) Bipolar question ({𝜙,¬𝜙}) : Verb attachment question (Turkish) Monopolar question (𝜙) : Object attachment question (Turkish) Rising declarative (English) With the proposed distinction, on the one hand the abundantly clear complementary distribution between the two forms in Turkish can be derived in a straightforward way. This is in contrast to English, where the polar interrogative is effectively ambiguous between the two meanings.10 On the other hand, the division collapses the two Ev+ forms together and provides a common underlying reason for the cluster of effects in their monopolar meaning. 3.3 How to derive the Ev+ Cluster? I will now turn to relating the monopolar question meaning to the Ev+ cluster. This will be a proof of concept rather than a fully fleshed out analysis. Here is the idea in a nutshell: The monopolar meaning, not the bipolar meaning, provides the necessary conditions for the Ev+ cluster. It does not provide sufficient conditions for all of the effects, however, among them positive evidential bias. This particular inference can be said to arise from an accommodation of the QUD in the absence linguistic cues to establish one. Fist, the necessary conditions. We have seen in Section 2 that rising declaratives and object attachment questions both work perfectly when the content proposition is a try-out: including an inference based on public evidence, guess, or perception of the speaker regarding what the addressee has said. The tried out 10Recall (34). Whether this is due to some hidden structural ambiguity or another process is not clear at the moment, but see Section 6 for some related discussion. 56 2 Evidential bias across clause types possibility is one among many. In contrast, contexts that exclude these two forms and in fact require the bipolar verb attachment form in Turkish are like exams. There is a right answer and a wrong answer which is the logical complement, instead of numerous possibilities. This difference, I suggest, stems from the difference in the sets of alternatives supported by the two meanings. There are two facets to this, which I will not try to reduce to one in this paper (but presume to be reducible): the first is the possibility of unlimited propositional focus alternatives {𝜙,𝜓,𝜋…} afforded by the focus semantic value of monopolar questions. This fits the function of trying out, because when venturing a guess or making an inference about the true state of affairs behind my friend’s wet coat or the mess in the kitchen, I am picking one possibility among many. Likewise, when I ask a confirmation question or echo the addressee’s last statement in surprise or sarcasm, I am picking what I possibly misheard, or pretend to have misheard, as the one utterance by the addressee which could have been any number of similar utterances. Hence, the set against which the content proposition is evaluated is the set of focus alternatives. Let us take the wet coat scenario with the rising declarative as the unambiguously monopolar form. The dialogue in (40) does not automatically come to an end if an initial guess is wrong. It will be continued with another, perhaps less likely guess (40c).11 It is in fact odd to stop guessing if the perceived question under discussion has not been resolved while the dialogue is still continuing (40d). (40) A’s friend enters their windowless office in a dripping wet coat. a. A: It’s raining? b. B: Nope. c. A2: Oh dear, someone activated the sprinklers on your floor again? d. A2′: ??OK! The second facet is the restricted alternatives of the bipolar question, and the restriction, specifically, to the two polar alternatives {𝜙,¬𝜙}. This makes for a very poor choice of form for a try-out, especially in the presence of a monopolar form in the paradigm. One piece of evidence that polar alternatives are not evoked in Ev+ contexts is found in the absence of polar alternative questions in such contexts. (41) shows that polar alternative questions are infelicitous in Ev+ contexts.12 11A2 indicates A’s second turn in conversation responding to B. 12A reviewer points out the felicity of It’s not sunny? in this context. This suggests that negative propositions can be part of the set of focus alternatives of a monopolar question without their logical complement (this time the positive It’s sunny?) being involved. Unfortunately space does not allow me to expand on this interesting lead. 57 Beste Kamali (41) My friend enters my windowless office in a dripping wet coat. #Is it raining, or not? / #Is it, or is it not, raining? / # It’s raining, or not? If we consider from this point of view a typical use that rejects our monopolar forms but goes very well with our bipolar form, the restrictions start making sense also in that direction. Take an exam question or a polite request. If met with the answer no, the exam is graded 0 and the request is denied. The questioner cannot attempt to request a response to the same question/request with a question from a different angle parallel to making a second guess after a failed one. The question is resolved. Whether yes or no, one answer excludes the other and becomes immediately conclusive because 𝜙and ¬𝜙are the only alternatives on the table. A clear manifestation of the underlying bipolar meaning in contexts such as court questions and indirect questions is the common use of polar alternatives when necessary pragmatic conditions are met, such as in (42). (42) a. Court hearing Did you or did you not tell him that if we were going to attack you’d let him know?13 b. Threat Will you or will you not turn that damn radio off? What about the sufficient conditions to create positive evidential bias? I suggest that contexts of positive evidential bias constitute a particular kind of trying out, namely one where no QUD has been established and hence the speakers must consult the physical environment to construct one. They must do so because expressions with open focus alternatives must be licensed in a QUD that limits the space of salient alternatives (Roberts 1996, Biezma & Rawlins 2012). In this line of thinking, the bias arises as a pragmatic inference based on the knowledge that both interlocutors must converge on a QUD. Let me start with how linguistic cues delimit the alternatives in some English declarative questions and Turkish object attachment questions. The most straightforward case is an alternative question with distinct propositional alternatives as in (31), repeated as (43). Even though each alternative is in the form of a monopolar question, the resulting alternative question does not lead to any inferences of bias, nor any evidence needs to be involved. This is because each of the exactly two relevant alternatives are overtly included in the form. The QUD is just that, or something like which of the following is the case. 13https://www.washingtonexaminer.com/policy/defense-national-security/milley-under-fire-iwould-never-tip-off-any-enemy (Last accessed 12 June 2025.) 58 2 Evidential bias across clause types (43) Ali Ali yemek dinner mi MI yap-tı, make-past yoksa or Hasan Hasan pizza pizza mı MI söyle-yecek? order-fut ‘Ali made dinner or Hasan will order pizza? (Which one is it?)’ A guessing challenge is slightly more complicated, but alternatives may nevertheless be construed relatively easily. Consider the rising declarative/object attachment question in (44) in response to the guessing challenge Guess why I’m so happy today. Here, the alternatives would be delimited by the knowledge of all possible events that might make the challenger happy, such as winning the lottery or passing an exam. There is no need for particular evidence towards a marriage proposal, because the challenge is simply to pick a likely state of affairs out of many potential ones, which is exactly what the monopolar form delivers. As in (43), there is no urge to look into the physical environment. (44) Your friend said “Guess why I’m so happy today?”. You respond. Sevgili-n lover-poss.2sg evlenme marriage mi MI teklif et-ti? propose-past ‘Your girlfriend proposed?’ Now consider the same question in a context of positive evidence noticed privately by A, as in (45). A wants to know whether B’s girlfriend proposed, using the same monopolar form as in (44), hence indicating that he is considering multiple states of affairs for the true answer. B may be looking at her computer all the while and there may have been no prior exchange about a marriage proposal. Not being aware of A’s attention on the ring, the monopolar question is puzzling to B. A must have in mind a QUD with multiple alternatives, to which B has no immediate access. But, assuming A is being cooperative, B must be able to construct a congruent QUD that matches A’s. B looks to the physical context for clues and realizes that her ring is a novelty that could lead to the hypothesis that B received a marriage proposal. B may then utter the maximally congruent (45b) or (45c). Alternatively, B may not try to match A’s QUD or fail to do so, in which case she could answer as in (45d). (45) A notices his friend B is wearing what looks like an engagement ring. a. A: Sevgili-n lover-2sg evlenme marriage mi mi teklif et-ti? propose-past ‘Your girlfriend proposed?’ b. B: Yes! Cool ring, huh? c. B′: No. You mean my grandma’s ring who just passed away? d. B′′: Yes, who told you?/No, why? 59 Beste Kamali Notice that effects of this QUD congruence emerge irrespective of whether the true answer to the question at hand is yes as in (45b) or no as in (45c). Maximal question/answer congruence with the monopolar form is achieved only when the QUD is also addressed. Hence, yes and no alone are perceived to be odd unless the evidence in question is public. QUD is a promising notion to connect the monopolar meaning to the biased inference based on evidence.14 Monopolar questions are unambiguously Ev+, that is, they evoke a need to refer to the physical environment to fully make sense, as a result of the pressure to construct a QUD based on the non-linguistic context in the absence of other clues revealing the nature of the relevant alternatives. This way, the requirement of public evidence which Büring & Gunlogson (2000) and Gunlogson (2003) address at length would have a pragmatic source, namely based on a matching QUD construal by both parties. Putting together everything said in this section, I sketched an account that connects English declarative questions with Turkish object attachment interrogatives around the cluster of Ev+ features and across clause types, successfully excluding Turkish verb attachment interrogatives. Both English rising declaratives and Turkish object attachment questions are monopolar. They can support the generation of broad focus propositional alternatives while excluding the polar alternative. The monopolar meaning underlies the Ev+ cluster of effects, because the monopolar form allows the speaker to formulate a question where one of many possibly true states of affairs is picked. Inferences of positive evidential bias arise when interlocutors have to decide on a QUD that restricts that set of states of affairs. Turkish verb attachment questions are bipolar questions and denote the pair of polar alternatives only. Hence this form is employed in cases where only the content proposition and its negation represent potential states of affairs, presenting the complement set of effects to those of the Ev+ cluster. One case has been left out, which I will touch upon briefly before closing this section. Regarding negative polarity/concord licensing, our empirical excursion 14A reviewer suggests that a polar question with narrow focus such as (4b) or (i) may be subsumed in the monopolar analysis following Krifka (2014). It becomes an interesting question, then, to what extent they are evidentially biased and whether this would fall out of the proposal given here. I leave this topic to future research, but note that empirically there may be an inference of evidential bias associated with (i). (i) Hey, did [LEA]𝐹win the tournament? a. Felicitous in [Ev. bias +] I see Lea fans celebrating. b. ?? in [Ev. bias 0] Speakers are on the phone, no related utterance has been made. c. Infelicitous in [Ev. bias -] I see Luke fans celebrating. 60 2 Evidential bias across clause types has suggested that it is a bipolar or non-monopolar feature. This could be because bipolar forms have a morphosyntactic representation of the negation in their meaning (related to the polar alternative). Our sketch analysis so far cannot say much more on this topic. 4 Residual clause type bias? In this paper, data demanded the generation of an account that can predict evidential bias effects along with the Ev+ cluster both in declarative and interrogative polar questions. Thus the proposed account does not rely on clause type and its consequences in deriving evidential bias. However, widely observed characteristics of English rising declaratives indicating a general bias concerning the addressee’s epistemic state are not accounted for. In this account, such effects must come from different sources than the proposed monopolar meaning, so the possibility arises that rising declaratives are biased in multiple ways and the residual non-evidential inferences of bias are in fact due to the clause type of the rising declarative. In this section, I will describe the ways in which our two monopolar forms differ from one another and suggest that these cases may in fact provide the key examples to address this residual bias. Much of the rising declarative data discussed comes from Gunlogson (2003). We have seen in examples (9–11) some of the ways in which the Turkish object attachment question and the English rising declarative are different. These are well-known differences between English polar interrogatives and rising declaratives, which we then used to argue that Turkish object attachment questions must be interrogatives. One case was the possibility of embedding under rogative predicates, which may be a direct morphosyntactic effect of the clause type, say, if the clause type is represented in the features of the C head, as commonly assumed. We have also seen that a discourse reference to an unresolved question could only be made with an interrogative. This requires a finer understanding, but one could still hypothesize that the formal qualities of the declarative do not fit the purpose. An intriguing possibility is that the intonation that marks the declarative as a question is only available when it is delivered within the question act. The last piece of data was the unavailability of adverbs like epistemic and evidential adverbs in polar interrogatives. I leave it open why this might be the case, but the interaction between clause type and such adverbials is empirically confirmed (the reader is referred to Krifka 2021 for a suggestion). There are further, less structural, differences between the monopolar object attachment interrogatives and rising declaratives. Let us start with the case of 61 Beste Kamali we have been entertaining, the monopolar meaning must be barred from arising in verb attachment questions. Kamali (in review) attributes this to a failure of focus projection à la Selkirk (1995) because main prominence is not on its default position. The versatility of English polar interrogatives, noted in their evidential neutrality and the lack of distinct forms in polar and propositional alternative questions, implies that they command the union of effects, meaning both meanings are available at the same time. How exactly this ambiguity is encoded is an open question, but under the line of thought pursued here, it may be hypothesized that (part of) the English interrogative syntax drives the bipolar meaning whereas the monopolar meaning ensues by default. 7 Conclusion I have argued that not just evidential bias but rather the entire Ev+ cluster of features result from the monopolar question meaning and not from clause type. The sketched approach empirically solidifies the monoversus bi-polar question dichotomy and their coexistence in grammar, and opens new avenues in approaching the problem of form-meaning correspondence in polar questions. One implication of the approach was the emergence of a residual bias inference that appears to indeed be clause-type-dependent, once evidential bias proper is disentangled from it. Another was the crosslinguistic commonness of an unambiguous monopolar question form sharing a number of the Ev+ cluster features. And finally, I have suggested that the monopolar meaning may correspond to declarative questions more commonly if it is the default. All in all, if correct, this approach points to a a more refined relationship between polar question form and meaning enriched by two universally available polar question meanings and regulated by diverse language-specific factors. Abbreviations 1/2/3pl 1/2/3 plural person 1/2/3sg 1/2/3 singular person acc accusative aor aorist com commitative conv converbial cop copula dat dative fut future gen genitive loc locative neg negation nomin nominalizer past past pfx prefix poss possessive pres present top topic 68 2 Evidential bias across clause types Acknowledgements This work is funded by the European Union Marie Skłodowska Curie Actions grant EPOQ-101067203 to Beste Kamali. I cordially thank Manfred Krifka for many sessions leading to the ideas in this paper. I also thank Maria Aloni, Beata Gyuris, Craige Roberts, Floris Roelofsen for helpful discussions, Deniz Rudin for comments on a pre-final draft, and two anonymous reviewers for their helpful feedback. References Bhatt, Rajesh & Veneeta Dayal. 2020. Polar question particles: Hindi-Urdu kya. Natural Language & Linguistic Theory 38. 1115–1144. DOI: 10.1007/s11049-02009464-0. Biezma, María & Kyle Rawlins. 2012. Responding to alternative and polar questions. Linguistics and Philosophy 35(5). 361–406. Büring, Daniel & Christine Gunlogson. 2000. Aren’t positive and negative polar questions the same? Unpublished manuscript retrieved from semanticsarchive, UCLA & UCSC. http://hdl.handle.net/1802/1432. Farkas, Donka F. & Kim Bruce. 2010. On reacting to assertions and polar questions. Journal of Semantics 27. 81–118. DOI: 10.1093/jos/ffp010. Farkas, Donka F. & Floris Roelofsen. 2017. Division of labor in the interpretation of declaratives and interrogatives. Journal of Semantics 34(2). 237–289. DOI: 10.1093/jos/ffw012. Giannakidou, Anastasia. 2000. Negative... concord? Natural Language & Linguistic Theory 18(3). 457–523. Göksel, Aslı & Celia Kerslake. 2005. Turkish: A comprehensive grammar. London: Routledge. Gunlogson, Christine. 2003. True to form: Rising and falling declaratives as questions in English. New York: Routledge. Gyuris, Beáta. 2017. New perspectives on bias in polar questions: A study of Hungarian -e.International Review of Pragmatics 9. 1–50. DOI: 10.1163/1877310900000003. Hamblin, Charles L. 1973. Questions in Montague English. Foundations of Language: International Journal of Language and Philosophy 10. 41–53. Hara, Yurie & Mengxi Yuan. 2025. Contextual bias and the landscape of Mandarin polar questions. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions, 179–208. Berlin: Language Science Press. 69 Beste Kamali Huddleston, Rodney. 1994. The contrast between interrogatives and questions. Journal of Linguistics 30. 411–439. Kamali, Beste. 2011a. The question particle in Turkish: Consequences for the interfaces. In Mary Byram Washburn, Chuoying Ouyang Sarah Ouwayda, Bin Yin, Canan Ipek, Lisa Marston & Aaron Walker (eds.), West Coast Conference of Formal Linguistics (WCCFL) 28. Kamali, Beste. 2011b. Topics at the PF interface of Turkish. Harvard University. (Doctoral dissertation). Kamali, Beste. in review. Motivating a duality of primary polar question meanings. Ms. ZAS. https://lingbuzz.net/lingbuzz/005477/current.pdf. Kamali, Beste & Manfred Krifka. 2020. Focus and contrastive topic in questions and answers, with particular reference to Turkish. Theoretical Linguistics 46. 1–71. Kamali, Beste & Takanobu Nakamura. 2024. Toward a crosslinguistically viable account of evidential bias. Presented at Polar Questions Across Languages (POQAL 12.4.2024). https : / / drive . google . com / file / d / 1BD5phHn2cINQFvyMw6IDzbvce0d1tnyZ/view. Kamali, Beste & Hedde Zeijlstra. 2024. Negative dependencies in Turkish. Languages 9(11). DOI: 10.3390/languages9110342. Karttunen, Lauri. 1977. Syntax and semantics of questions. Linguistics and Philosophy 1. 3–44. Kornfilt, Jaklin. 1997. Turkish. London: Routledge. Krifka, Manfred. 2014. Focus and contrastive topics in question and answer acts. Ms. HU Berlin. Krifka, Manfred. 2015. Bias in commitment space semantics: Declarative questions, negated questions, and question tags. In Sarah D’Antonio, Mary Moroney & Carol-Rose Little (eds.), Semantics and Linguistic Theory (SALT) 25, 328–345. Ithaca: CLC Publications. Krifka, Manfred. 2021. Modelling questions in commitment spaces. In Moritz Cordes (ed.), Asking and answering, 63–95. Tübingen: Narr. Kuno, Masakazu. 2008. Negation, focus, and negative concord in Japanese. Toronto Working Papers in Linguistics 28. 195–211. Ladd, D. Robert. 1996. Intonational phonology. Cambridge: Cambridge University Press. Özge, Umut. 2011. Turkish indefinites and accusative marking. In Andrew Simpson (ed.), Workshop on Altaic Formal Linguistics WAFL 7, 253–267. Cambridge: MIT Working Papers in Linguistics. 70 2 Evidential bias across clause types Roberts, Craige. 1996. Information structure: Towards an integrated formal theory of pragmatics. In Jae Hak Yoon & Andreas Kathol (eds.), OSU Working Papers in Linguistics, vol. 49, 91–136. Columbus: The Ohio State University Department of Linguistics. Rudin, Deniz. 2022. Intonational commitments. Journal of Semantics 39(2). 339– 383. DOI: 10.1093/jos/ffac002. Selkirk, Elisabeth. 1995. Sentence prosody: Intonation, stress and phrasing. In John A. Goldsmith (ed.), The handbook of phonological theory, 550–569. Cambridge: Blackwell. Sudo, Yasutada. 2013. Biased polar questions in English and Japanese. In Daniel Gutzmann & Hans-Martin Gärtner (eds.), Beyond expressives: Explorations in use-conditional meaning, vol. 28 (Current Research in the Semantics/Pragmatics Interface (CRiSPI)), 275–296. Leiden: Brill. Watanabe, Akira. 2004. The genesis of negative concord: Syntax and morphology of negative doubling. Linguistic Inquiry 35(4). 559–612. Ye, Shumian. 2021. From maximality to bias: Biased A-not-A questions in Mandarin Chinese. In Semantics and Linguistic Theory (SALT) 30, 355–375. 71 Chapter 3 The contribution of intonation in the conveyance of question bias Riccardo Orrico Radboud University, Nijmegen Cristel Portes Aix Marseille Université, Laboratoire Parole et Langage Mariapaola D’Imperio Aix Marseille Université, Laboratoire Parole et Langage, CNRS The aim of this chapter is to address the issue of how meanings related to bias in polar questions are conveyed by intonation. We present an overview of previous studies conducted on two Romance languages, Italian and French. The review of these studies is used to show how intonation can be used in different languages to signal the presence and the polarity of question bias, as well as the source of information from which the bias stems. We also point out that future research on this topic should deal with, among other things, how gradient phonetic information maps onto meanings and the role of individual variability. 1 Introduction A growing body of research has been dedicated to the discursive role of intonation, taken to be involved in the conveyance of meanings related to the attitude of discourse participants relative to the expressed proposition. Starting from Pierrehumbert & Hirschberg (1990), several intonational meaning models have been proposed, with the idea that intonation can express commitment or beliefs on the part of the speaker and, at the same time, project such beliefs/commitments to Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio. 2025. The contribution of intonation in the conveyance of question bias. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions: Experimental results and theoretical modelling, 73–100. Berlin: Language Science Press. DOI: 10.5281/zenodo.17158180 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio the interlocutor. Hence, the present chapter aims at presenting an overview on the role of intonation in encoding and decoding polar question meaning, reporting on recent studies that have addressed this issue in two Romance languages, Italian and French. According to standard semantic theories (e.g., Hamblin 1973), declaratives denote a proposition and function as a means for the speaker to assert something, by making a commitment (among others Krifka 2017). Interrogatives, on the other hand, denote a set of propositions and are used by the speaker to lead the addressee to engage in an assertion. Certain types of questions, however, are used to also assert something (in addition to eliciting information) and, as a consequence, they convey a bias about the information asked (Dayal 2017). Such questions are typically referred to as biased questions. In the case of a biased question, the questioner is not simply requesting information from a position of complete ignorance, but is also expressing an expectation about what the information might be. Formal models of the semantics and pragmatics of questions have actually addressed such phenomena. For instance, Asher & Reese (2007) provide a formal proposal in which biased questions are accounted for as complex speech acts, containing both a question and an assertion. Similarly, Malamud & Stephenson (2015) analyse polar questions by proposing that they project speaker or addressee commitment to the truth of one of the propositions. These formalizations provide an understanding of question bias as an object of complex nature, which can be analysed along different dimensions: i) polarity of the bias (and conflictual information between speaker and addressee), ii) degree/strength of (tentative) commitment proposed by the speaker, iii) source of information on the basis of which the speaker proposes commitment. Languages can express these dimensions of bias using different strategies, among which the most investigated are negative questions (Ladd 1981, Krifka 2017, Arnhold et al. 2021), tag questions (Ladd 1981, Asher & Reese 2007, Malamud & Stephenson 2015) and presence of polarity items or other lexical markers (Romero & Han 2004, Asher & Reese 2007, Frana & Rawlins 2019a,b). The role of intonation in conveying information about question bias has also been recently recognized, though a comprehensive account of its contribution (especially in isolation) is still lacking. First, the link between intonation and question bias has mainly been investigated in interaction with other linguistic levels (e.g. morpho-syntactic properties of utterances). However, in languages like Italian, in which a strong role is assigned to intonation in the definition of the discourse function of utterances (e.g., the absence of specific syntactic cues for questionhood), recent studies have unveiled that intonation alone can be responsible for the conveyance of question biases. Past research on intonational 74 3 The contribution of intonation in the conveyance of question bias meaning has unveiled that intonation can be used not only to signal the presence of bias, but also to specify its polarity (i.e. towards which proposition the speaker is biased) and degree (how strong the bias is). The way in which types of biases can be intonationally conveyed is related to elements that are responsible for encoding these meanings. Most of the studies addressing the role of prosody in modulating the meaning of utterances have argued/assumed that phonological elements of intonation (i.e. pitch accents and boundary tones) are the main cues involved in the expression of these meanings. In this framework, general prosodic properties of phonetic nature have typically been linked to the expression of paralinguistic meanings (see, for example, the treatment of pitch range by Liberman & Pierrehumbert 1984). Nevertheless, in spite of the fact that published work on this issue is still rather scarce, experimental evidence has shown that the relationship between prosody and discourse-related meanings is far more complex, with several cues beyond phonological elements having a role in the definition of the semantics and pragmatics of utterances, e.g. pitch range, voice quality, and temporal information (see also Section 2). Finally, recent experimental investigations have also unveiled the role of extralinguistic information of speakers and listeners in modulating the way bias is perceived. These results have implications on linguistic theory, both on the pragmatic/semantic and the prosodic side. On the one hand, these results support an idea of question bias as a gradient feature of utterances rather than a categorical one, as often treated in formal models; on the other, they also shed light on the way intonation is exploited to convey meaning. Building on these premises, this chapter aims to characterise the general role of intonation in the expression of question bias in two Romance languages, i.e., Italian and French, trying to address three main issues: i) what kind of meanings are concerned with bias in questions and can be conveyed by prosodic means?, ii) which are the elements of intonation that are responsible for conveying meanings related to question bias?, and iii) is the relationship between intonational cues and pragmatic/semantic meanings stable within a linguistic community? The remainder of this chapter is organized as follows: Section 2 will deal with the general role of intonation in the conveyance of meaning; next, Section 3 and Section 4 will review some recent findings on the role of intonation in two selected Romance languages (Italian and French). Finally, Section 5 will discuss the implication of these findings for linguistic theory and methodologies. 75 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio 2 Intonational meaning Very different conceptions of intonation meaning have been proposed in the literature (see Portes & Beyssade 2015, Prieto 2015, Westera et al. 2020 for recent overviews). Some phoneticians conceive it as a direct link between phonetic parameters of prosody and communicative functions (Local 2003, Xu 2005), while phonologists assume that the relation of prosody to meaning is mediated by phonological units such as global tunes (Gussenhoven 1984), individual tones (Pierrehumbert & Hirschberg 1990), or intermediate constructs like pitch accents and boundary tones (Steedman 2007). The latter approaches have the advantage of being more compatible with semantic theories of meaning and of being engaging in a fruitful dialogue with them as evidenced by various contributions to this book. In this trend of research, it is now well established that, beyond information structural meanings, intonation also conveys meanings related to epistemicity (the commitment of the speaker/addressee towards the content of the utterance: Bartels 1999, Gunlogson 2003), evidentiality (the source of information: speaker, addressee or else; Gunlogson 2008, Escandell-Vidal 2017), dialogue negotiation (agreement/disagreement: Steedman 2007) and mirativity (unexpected information: DeLancey 2001, Rett & Murray 2013, Celle et al. 2018) For instance, Bartels (1999) proposes that, in English pitch contours, the Lphrasal tone is a morpheme which conveys “assertiveness” in questions as well as in statements, by indicating not only the speaker’s commitment, but also by instructing the addressee to publicly commit herself to the proposition at stake. Gunlogson (2003) argues that, while commitment to the propositional content is expressed by the declarative syntax of the utterance, intonation is responsible for the attribution of that commitment to the speaker (falling contour) or the addressee (rising contour). A very similar conception is presented in Portes & Beyssade (2015), who propose that, in French, L% signals that the speaker assumes responsibility for the content of the utterance whereas H% signals that the speaker delegates this responsibility to the addressee. Moreover, in her 2008 paper, Gunlogson reformulates the meaning of rising intonation as indicating that the commitment of the speaker is contingent on the addressee’s ratification, which gives a rising declarative its questionhood. This new formulation relies on the more fundamental principle that commitments have a source, in that one of the participants obviously has more independent evidence supporting the content than the other. Escandell-Vidal (2017) also places evidentiality at the centre of her conception of intonational meaning. Modelling the meaning of three different prosodic contours in Castilian Spanish polar interrogatives, she proposes 76 3 The contribution of intonation in the conveyance of question bias a common Q feature conveying unspecified polarity. While the meaning of the L+H* H% low rise is modelled by Q alone, the meaning of the H* H% high-rise combines Q with the “Self” modality of the evidential feature Evid, and the meaning of the L+¡H* L% rise-fall combines Q with the “Other” modality of Evid. The role of intonation in the negotiation dimension of dialogue can also be approached through the relevance of the addressee’s reaction as a separate feature, such as the notion proposed by Beyssade & Marandin (2007, 2006). Specifically, the authors propose that, in the meaning of tags or intonation meaning, the feature call on the addressee refers to a request from the speaker concerning the addressee’s next move. For instance, a declarative followed by the tag n’estce pas (‘isn’t it’) in French requires an answer as the next move instead of an acknowledgement. Moreover, intonation may signal participants’ agreement or disagreement. For instance, in Steedman’s (2007) semantics of English intonation, some nuclear pitch accents mark consensual information (L+H*, H* and H*+L), while others can mark potentially conflictual information (L*+H, L* and H+L*). Building on this idea, Portes & Beyssade (2015) further proposed that pitch accent choice may convey different degrees of disagreement, H+L* indicating a stronger disagreement than H*+L in French. Finally, the difference between participants’ commitments may relate to the mirative dimension of intonational meaning. According to Celle et al. (2018), the concept of mirativity was introduced by DeLancey, as “the linguistic marking of an utterance as conveying information which is new or unexpected to the speaker” (DeLancey 2001: 369–370). In other words, mirativity refers to the situation in which an interlocutor is surprised by a piece of information that s/he received, though, differently from incredulity, the speaker is not conveying that s/he does not believe that the proposition holds. In fact, the mirative attitude should be considered a dimension that is clearly separated from incredulity, in which the interlocutor shows a negative bias towards (i.e. does not believe) a proposition. Though it is conveyed by morphology in languages such as Lhasa, Tibetan or Cheyenne, mirativity is claimed to be marked by intonation in English, as shown by the contrast between the following utterances (cited by Celle et al. 2018: 219 from Rett & Murray 2013: 455): (1) a. John arrived on time. b. (Wow,) John arrived on time! It is still unclear and a matter of lively debates how to semantically/pragmatically model these different aspects of intonational meaning (and of other various linguistic forms) as parts of a unified concept of commitment or as separated/ 77 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio Recent studies investigating the issue on both the production and perception sides have found that both tune choice (pitch accent and boundary type) and nuclear accent placement are involved in the conveyance of the epistemic position of the speaker towards the utterance. Despite data from production, showing high levels of inter-speaker variability, some patterns were found to correlate with specific pragmatic conditions. For example, the analysis of distribution reported in Orrico et al. (2019) shows that a L+H* HL-L% contour is mainly employed in the counter-expectation condition (i.e., the speaker is rejecting some information provided by the addressee). As for perception, an experiment recently reported in Orrico & D’Imperio (2022) was designed with the aim of testing whether and how phonologically different tunes could be interpreted by SI listeners as mapping onto different degrees of speaker certainty towards the expected answer. The tunes investigated were the rise-fall-rise tune with the earlypeaked pitch accent (L+H* HL-H%) and the two rise falls, differentiated on the basis of the pitch accent (i.e. L*+H HL-L% and L+H* HL-L%). Additionally, the L*+H H(*)L-L% tune was included in the experimental design to test whether the pitch accent placement could play a role in the conveyance of these meanings. 45 SI listeners took part in the experiment. They were asked to listen to questions out of context and to rate them by assigning a score (from 0 to 100) according to the degree of perceived certainty of the speaker for a specific answer to the question. Results show that specific cues within question tunes are exploited by SI listeners to extract bias information. Specifically, the L+H* HL-H% tune was found to be rated significantly higher in speaker certainty than L+H* HL-H%, while the L*+H HL-L% was globally rated the lowest in certainty score. Additionally, the tune with a nuclear pitch accent placed earlier in the utterance (L*+H H(*)L-L%) received the highest certainty score. Finally, a further element of variation across question tunes is the pitch excursion within the pitch accent. In fact, data from Salerno Italian showthat questions can have at least two different distributions as far as the excursion of the pitch accent in nuclear position is concerned, as shown in Figure 3. In both examples reported in Figure 3, the nuclear pitch accent produced by the speaker is an early rise (L+H*), though with different pitch span values correlating with specific pragmatic information of the context they were uttered in. More specifically, Orrico et al. (2019) reported that instances of L+H* nuclear accents with a narrower pitch span were typically employed to encode counterexpectational questions, i.e., when the speaker is conveying his/her incredulity towards the information contained in the question. A parallel investigation was conducted to perceptually test the role of pitch span within the nuclear region, as reported in Orrico & D’Imperio (2020). The 84 3 The contribution of intonation in the conveyance of question bias Figure 3: Examples of L+H* with different span values. The figure shows the question utterances Milena lo vuole amaro? (Does Milena take it [coffee] black?), on the left, and Loredana un ingegnere? (Loredana an engineer?), on the right, uttered by the same male speaker but with a narrower pitch span (adapted from Orrico et al. 2019). experimental setting included a collection of the same judgment as in Orrico & D’Imperio (2022), that is the assignment of certainty scores by SI listeners, though the manipulated variable was pitch span. 45 participants took part in the experiment; they were asked to rate a set of 18 experimental stimuli, which were artificially created by manipulating the height of the L+H* nuclear accent into three different steps and the height of the boundary tone in six different height steps. Results show (Figure 4) that the wider the span in both pitch accent and boundary tone, the higher the certainty score assigned to the tune. However, while for pitch accent the three different span levels were rated as all conveying a different degree of speaker certainty, the same was not found for boundary tones, given that only the lowest two steps were classified as conveying a significantly different score. In other words, while boundary tones were found to have a categorical effect on the perceived bias of a polar question (corresponding to the two phonological L% and H% boundaries), different span levels within pitch accent were found to gradually map onto the pragmatic level. 3.2 French Given that linguists’ awareness that prosody is a crucial dimension of grammar is relatively recent, an obvious issue in the study of intonational meaning is how it is conceived. In a paper about the compositionality of intonational meaning, Portes & Beyssade (2015) argue that the contribution of prosody to utterance meaning depends not only on how the interface between prosody and morphosyntax is conceived, but also, and more crucially, on how the notion of prosodic form and the dimensions of meaning are conceptualized. In the present 85 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio Figure 4: Certainty score assigned to different span values of the L+H* pitch accent (left) and boundary tones (right) (Orrico & D’Imperio 2020). section we present several papers on the meaning of French intonation, which contribute to a better understanding of these questions. First, Portes & Beyssade (2015) propose that prosody is formally structured by a developed phonological model. Specifically, a more detailed and fine-grained semantic description of prosodic meaning is allowed and required only when compositionality is proposed. This is the reason why more sophisticated accounts of intonation meaning have been proposed in phonological approaches to intonation (Gussenhoven 1984, 2014, Pierrehumbert & Hirschberg 1990, Steedman 2007; among others). Adopting such an intonational approach, Portes et al. (2014) developed a detailed semantics in order to account for the meaning of four French intonational contours (see Figure 5 below). On the basis of a theory initiated by Beyssade & Marandin (2007), Portes et al. (2014) proposed that the meaning of the four contours shown above can be accounted for by using three distinctive features, i.e., speaker commitment,attitude attribution to the addressee (does the addressee agree or disagree with the speaker?) and call on addressee’s next move (a verbal answer, an acknowledgement, another behaviour?). If intonational contours can make public speaker’s projections about beliefs attributed to the addressee and the addressee’s next move, the latter should dialectically reflect these projections of the speaker in reacting to them. Specifically, in the above study participants were asked to match a declarative sentence bearing one of the four contours with one of four possible meanings summarized in Table 1. Results showed that participants very robustly matched the intonation they heard with the hypothesized reaction for three of the contours (but not for the 86 3 The contribution of intonation in the conveyance of question bias Table 1: Proposed intonation contour meanings designed using the three features proposed by Beyssade & Marandin (2007), and associated reactions (from Portes et al. 2014). Contour Pototypical Meaning Features Reaction L*L% S commits her-/himself to the truth of 𝑝J’en prends note S signals that (s)he anticipates no disagreement from ‘I get it’ A (about 𝑝) A acknowledged that S has no disagreement S proposes to A to update CG with 𝑝about 𝑝and that 𝑝can be added to the CG H*H% S signals that (s)he doesn’t commit to 𝑝J’en sais rien S signals that (s)he attributes a belief about 𝑝to A (that ‘I’ve no idea’ 𝑝or ¬𝑝)Contradicts S’s attribution to A S proposes to A to commit to 𝑝or to commit to ¬𝑝 H*L% S commits her-/himself to 𝑝Tu dois avoir raison S signals that (s)he anticipates a disagreement from A: ‘I guess you’re right’ S signals that (s)he attributes to A the belief that ¬𝑝 Acknowledgement of S’s commitment S proposes to A to update CG with 𝑝Remaining potential disagreement of A H+!H*H% S signals that (s)he does not commit to 𝑝Si, si, je t’assure S anticipates a disagreement from A: S signals that ‘No, really, it’s true’ (s)he attributes to A the belief that 𝑝) Comments both on S’s disagreement and S proposes to A either to commit to 𝑝or to commit on the confirmed commitment of A to ¬𝑝 87 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio Figure 5: Four renderings of the utterance Jules a engagé un ingénieur ‘Jules committed an engineer’ with the four intonational phrase final contours (adapted from Portes et al. 2014: 18): a) L*L%, b) H*H%, c) H*L% and d) H+!H*H%. H*L%), indicating a promising way to investigate fine grained intonational meaning. As for H*L%, Portes & Reyle (2014) reports a corpus investigation dedicated to its meaning, possibly more sensitive to context and better perceived in presence of a real potentially opposing addressee. One of these four French intonation contours, the rise-fall-rise H+!H*H%, was also investigated in a production study (Michelas et al. 2016). In this study, participants had to read a declarative question matched with a negatively biased context eliciting incredulity or matched with a neutral context eliciting an unbiased yes-no question. The questions matching the negatively biased context were produced with the H+!H* pitch accent in 62% of the cases and with H* in the remaining 38%. Conversely, the questions matching the neutral context were produced with H* in 97% of the cases, with 3% of H+!H*. This result demonstrates that the negative bias was mainly conveyed by the choice of the pitch accent. If we relate this result to the proposal by Portes et al. (2014), it appears that the semantic feature attribution of intention (agreement/disagreement) should be conveyed by pitch accent choice in French, as proposed by Portes & Beyssade (2015). Therefore, the other two features (i.e. speaker commitment and call on addressee), could be jointly conveyed by boundary tones. An important observation is that, in the negatively biased context, H+!H* was followed by a rise attributed to an H% boundary tone in 69,4% of the productions. However, the remaining 30,6% exhibited a fall that has been interpreted as a L%. This variability may in fact be explained by a truncation of the H% boundary tone due to lack of segmental space. This hypothesis remains to be further investigated. If confirmed, this would be a strong piece of evidence that phonological constraints, internal to the intonational system of a language, play an important role in observed variability. 88 3 The contribution of intonation in the conveyance of question bias A more striking variability is the production of 38% H*H% patterns in the negatively biased context. One possible explanation is that the mere rise expresses mirativity (surprise) in that context, while the H+!H*H% contours rather express incredulity. Indeed, a surprise meaning may also convey a negative speaker bias even if it is to a lesser degree. One may further speculate that there could be an additional prosodic distinction affecting H*H% depending on whether it appears in the neutral context or in the negatively biased context: a pitch range variation would certainly be a good candidate. This nuance of meaning hence remains to be better studied in French and other languages. To conclude this section, several studies on French intonational meaning in the last two decades have contributed to explore an approach where detailed accounts both of intonational form in a phonological approach and of intonational meaning in a semantic perspective allow to give evidence in favour of a possible compositionality of intonational meaning. Such an approach deserves to be refined and enriched by a better understanding of the observed variation, an issue that will be developed in the following section. 4 Variability in intonational meaning A further point to be addressed is the role of individual variability in both the encoding and decoding of intonational meaning. Within the last decade, the issue of variability has become central in phonetic studies, both at the segmental and at the prosodic level. The sources of such variability can be related to two main factors, i.e., environmental and cognitive factors (Kidd et al. 2018). More specifically, speech has been found to vary as a function of gender and sexual orientation, socio-economic status, exposure to specific language input as well as cognitive (affecting also neurotypical populations) and emotional states (see Pierrehumbert 2016 and Yu & Zellou 2019 for a review of these issues). Variability in the use of intonation has been found both in production and perception. With specific reference to Italian, the study by Gili Fivela and colleagues (Gili Fivela et al. 2015) investigated intonation production in several cities and described the general situation as a “mixing of patterns”, reporting that despite the interand intra-variety variability, similar patterns are easily found across Italy, also in geographically distant cities. The authors attribute these findings to factors of socio-political nature, which have an effect that is larger than the individual speaker, e.g. internal migration and the influence of the language of media broadcastings. Recent studies have linked intonational variation in production to individual-specific choices. Great importance has been given to the 89 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio role of exposure to non-native dialects: while the distribution of polar question tunes can be linked to specific meanings, individual speakers are found to behave very differently from each other as a function of the input they were exposed to throughout their lives, e.g. by having a non-SI parent (Orrico et al. 2019, 2020). The impact of linguistic input on linguistic behaviour at the individual level has also been observed in perception. Both Orrico & D’Imperio (2020) and Orrico & D’Imperio (2022) have considered the role of past linguistic experience in the attribution of bias-related meanings conveyed by SI question tunes. Taken together, these studies show that the effect of having been exposed for a long time to non-native input (either a foreign language or a non-native dialect of the same language) has a major impact on the way listeners extrapolate tune information. Crucially, the two studies show that the effect is visible at different levels, both at a coarse-grained level – i.e., by asking which phonological elements are considered to primarily convey the bias-related meaning – and at a finer phonetic level – i.e., how gradual variation in pitch span affects the interpretation of question meaning. Interestingly, the general picture yielded by these investigations is that exposure affects the extent to which listeners rely on specific tonal cues to interpret meanings. Indeed, Orrico & D’Imperio (2022) reported that the exposed population discarded the role of boundary tones and relied only on pitch accents. Similarly, Orrico & D’Imperio (2020) reported that manipulations in the span of pitch accents were interpreted in a coarser way by exposed listeners. Similar investigations have also been carried out for French. A recent study by Portes & German (2019) has in fact addressed the issue of how listeners manage intonation-meaning mappings in case of exposure to different regional varieties. They reported a study in which a rise-fall tune in French was interpreted not only on the basis of the context in which it is produced, but also of the presence of extralinguistic cues to the specific geographical location. Note that this rise-fall tune is ambiguous since it is used in Continental French to encode a statement and in Corsican French to encode a polar question. In their model, tune-meaning mappings are stored in the listeners’ mental representation in association with a contextual label (in this case, Continental or Corsican French), in a way that contextual cues can determine which mapping is relevant in the situation and therefore activated. Crucially, the extentto which the contextual cue is functional to the correct interpretation of the meaning is strictly linked to the amount of relative exposure to the two varieties, with the greatest advantage predicted for listeners with a balanced exposure to the two systems (German & Portes 2020). A further source of variation of the tune-meaning mapping concers listeners’ cognitive states. Recent research has pointed out that the degree of cognitive and emotional empathy determines the way in which intonational meaning is 90 3 The contribution of intonation in the conveyance of question bias processed. In an eye-tracking study, Esteve-Gibert et al. (2020) show that the interpretation of ambiguous utterances by French listeners varies as a function of the empathy skills of the listeners: while listeners with low empathy skills show a weaker reliance on prosodic information leading to the disambiguation of lexical items, listeners with higher empathy skills attended more closely to prosodic information and showed a higher sensitivity to the possible meanings conveyed by the tonal contour. Similarly, Orrico & D’Imperio (2020) introduced the degree of empathy skills in their experimental setting, reporting that empathic listeners were more attentive to fine tonal modifications and to the way they are used to encode meanings than the low-empathy group. Interestingly, empathy was also found to affect the role of exposure to other linguistic systems. Specifically, Orrico & D’Imperio (2020) found that the perceptual change as a result to exposure to non-native language was strongest in high-empathy listeners. These results provide an impression of the extreme sensitivity of intonational meaning to a number of different factors, which should be taken into account when conducting research on this issue. 5 Which dimensions of bias in questions are conveyed by intonation? This chapter has reported a discussion on the general role of intonation in the conveyance of information linked to question bias. The idea is that the speaker aims at signalling that s/he is not completely ignorant with respect to the propositional content of the utterance. The findings reported here also allow to characterize the dimensions over which intonation operates in the expression of these meanings. First, intonation serves the role of signalling that the speaker has his/her own beliefs with respect of the question being asked and, additionally, it allows for finer specifications of the type and strength of the belief as well as how they are positioned relative the interlocutor’s beliefs. Intonation is hence also involved in the conveyance of the presence of speaker bias, through the expression of speaker commitment. This function of intonation are attributed to the type of pitch accent in nuclear position in Italian, with the L+H* early rise conveying that the speaker is committing to the surface proposition within the question, as opposed to the L*+H late rise, which signals that the speaker is not making a commitment toward the truth of the proposition. On the contrary, French exploits boundary tones to encode this meaning, with L% boundaries being linked to the expression of speaker commitment. Moreover, both languages make use of intonational cues 91 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio to encode negative bias, by signalling that some piece of information is in contrast with what the speaker previously believed to be true. On one hand, this meaning is assigned to pitch accent type in French, specifically through the use of a H+!H*. On the other hand, Salerno Italian conveys this meaning by modulating the pitch span within the L+H* rise, with a narrower span in case of negative bias. Moreover, the pitch span feature is also exploited by Salerno Italian to encode information about the relative strength of speaker commitment: a wider excursion would allow signalling that the expectation for a positive answer to the question is significantly higher. Finally, we show that intonation was also involved in the attribution of commitment from the speaker to the addressee, which is again encoded through the type of pitch accent in French. As for Italian, an impressionistic analysis of the data collected suggests that the meaning of commitment attribution might be conveyed by boundary tones. More specifically, the presence of a H% boundary in questions might signal that the speaker is attributing the addressee the role of source of the commitment and, therefore, the commitment expressed by the speaker is only tentative and dependent on the addressee’s ratification. Taken together, the findings reported here for both Italian and French allow us to make general considerations about intonational meaning. On the one hand, intonational meaning is not universal. Specifically, although intonation might have a similar role in different languages, the way in which specific meanings are encoded within an intonation contour is language-specific. Moreover, both French and Italian would extract similar functions from intonation as related to the expression of meanings relative to question bias, i.e., signalling the presence of bias, specifying the position of the speaker relative to the proposition and to the addressee and, possibly, defining the source of the commitment. Nevertheless, the way in which the meanings are grammaticalized within the two languages appears to be different. Note that differences might also be found within different varieties of the same language. Recall, for example, the differences in the meaning assigned to the rise-fall in Continental and Corsican French. Furthermore, lack of universality is also found as far as fine phonetic modifications are concerned. For instance, the incredulity conveyed by narrower pitch span in Salerno Italian is not found in all other varieties of the same language. In Bari Italian, for example, the same meaning is conveyed through a wider span within the pitch accent (Savino & Grice 2011). The link between pitch range, information and pragmatic meanings, as also reported in Section 2, has been reported in several languages. Among these, Savino & Grice (2011) and Crocco & Badan (2016) reported the role of pitch excursion in modulating the meaning of a question in two varieties of Italian. Hirschberg & 92 3 The contribution of intonation in the conveyance of question bias Ward (1992) also found that information relative to pitch range was instrumental in determining the difference between an incredulous and an uncertain reading of the rise-fall-rise contour in English. Further, Vanrell (2011) and Borràs-Comes et al. (2014) showed that , in Catalan, a different scaling of the peak within the pitch accent is a crucial cue for discriminating across linguistic categories, both in production and perception. Both the above studies as well as the findings reported here for Salerno Italian allow to clarify yet another controversial issue of intonational meaning, specifically which are the tonal elements responsible for meaning conveyance. While information relative to pitch range and level beyond the two phonological discrete tones (L and H) is often discarded both in the analysis of intonation and its meaning, these studies point out that gradual information relative to pitch height might actually be involved in the expression of pragmatic meanings and therefore deserve to be acknowledged and accounted for in intonational meaning models. Nevertheless, the way in which this type of information should be integrated in existing models is not yet proposed here, especially considering that most of the models proposed seem to advocate for the absence of a direct link between phonetic information and pragmatics/semantics, given phonology mediation. One possibility, which appears to be compatible with findings in Orrico & D’Imperio (2020) for Salerno Italian, is that pitch span manipulation is not meaningful per se, though it modulates the core meaning of the pitch accent (or the tonal discrete element it modifies). For instance, in the case of L+H*, while its core meaning would be committing the speaker to some proposition, a pitch span modification would modulate the speaker confidence to that commitment and, in this specific case, the degree of bias. An intonational meaning model allowing for these types of gradual modification of meaning would, in turn, require that also the pragmatic side should acquire a finer granularity. Finally, the data reported above suggest that the way intonation is linked to pragmatic meanings is extremely sensitive to various sources of variation, making the investigation of these issues both a theoretical and a methodological challenge. On the theoretical side, one major implication of this issue is relative to how we even understand meanings conveyed through intonation. As reported above, two studies that address this issue are Portes & German (2019) and German & Portes (2020) reported in Section 4, in which the authors propose an exemplarbased model in which intonational contours are stored with contextual-specific meanings and retrieved when needed during communication. In general, very little has been done as far as processing of intonation in variable contexts is concerned. Some experimental studies have been conducted (among others Kurumada et al. 2014, Roettger & Rimland 2020), which, together with experimen93 Riccardo Orrico, Cristel Portes & Mariapaola D’Imperio Savino, Michelina. 1997. Il ruolo dell’intonazione nell’interazione comunicativa: Analisi strumentale delle domande polari in un corpus di dialoghi spontanei (varietà di Bari) [The role of intonation in communicative interaction: Instrumental analysis of polar questions in a corpus of spontaneous dialogues (Bari variety)]. Università/Politecnico di Bari. (Doctoral dissertation). Savino, Michelina. 2012. The intonation of polar questions in Italian: Where is the rise? Journal of the International Phonetic Association 42(1). 23–48. Savino, Michelina & Martine Grice. 2011. The perception of negative bias in Bari Italian questions. In Sónia Frota, Gorka Elordieta & Pilar Prieto (eds.), Prosodic categories: Production, perception and comprehension, 187–206. Dordrecht: Springer Netherlands. DOI: 10.1007/978-94-007-0137-3_8. Steedman, Mark. 2007. Information-structural semantics for English intonation. In Chungmin Lee, Matthew Gordon & Daniel Büring (eds.), Topic and focus: Cross-linguistic perspectives on meaning and intonation, 245–264. Dordrecht: Springer Netherlands. DOI: 10.1007/978-1-4020-4796-1_13. Vanrell, Maria del Mar. 2011. The phonological relevance of tonal scaling in the intonational grammar of Catalan. Universitat Autònoma de Barcelona. (Doctoral dissertation). Warren, Paul. 2017. The interpretation of prosodic variability in the context of accompanying sociophonetic cues. Laboratory Phonology 8(1). 1–21. DOI: 10. 5334/labphon.92. Westera, Matthijs, Daniel Goodhue & Carlos Gussenhoven. 2020. Meanings of tones and tunes. In Carlos Gussenhoven & Aoju Chen (eds.), The Oxford handbook of language prosody. Oxford: Oxford University Press. DOI: 10 . 1093 / oxfordhb/9780198832232.013.29. Xu, Yi. 2005. Speech melody as articulatorily implemented communicative functions. Speech Communication 46(3). 220–251. DOI: 10.1016/j.specom.2005.02. 014. Yu, Alan C. L. & Georgia Zellou. 2019. Individual differences in language processing: Phonology. Annual Review of Linguistics 5. 131–150. 100 Chapter 4 Negative polar questions in Russian: Question bias and question concern Sophie Repp University of Cologne Ljudmila Geist University of Stuttgart We explore the contextual appropriateness conditions of yes-no-questions in Russian, focusing on negative questions with the particles razve and neuželi, in comparison to questions without a particle, providing corpus evidence and experimental evidence from two acceptability judgement studies. We argue that both razvequestions and neuželi-questions are felicitous when there is a conflict between the speaker’s original assumption that a positive proposition is true (epistemic bias for p), and contextual evidence indicating the falsity of p(evidential bias for ¬p). The differences between razve-questions and neuželi-questions are about what we call the question concern. The question concern comprises the different goals that a speaker pursues by asking a question, such as to double-check the epistemic bias or the evidential bias, as well as the strength of the conflict between the epistemic or evidential bias that is signaled by the speaker together with the speaker’s concomitant psychological state. We argue that razve-questions can double-check epistemic or evidential bias, whereas neuželi-questions can only double-check and express disbelief in the evidential bias. The latter difference correlates with the availability of outer and inner negation readings (Ladd 1981): razve-questions can have both readings, neuželi-questions only have inner negation readings. We explain this difference in terms of the illocutionary operators verum and falsum, where outer negation corresponds to falsum and inner negation to verum scoping over propositional negation (Repp 2009). We argue that neuželi denotes verum, which cannot occur with falsum in the same question, whereas razve is a propositional semantic operator, which can occur in the scope of verum or falsum. Sophie Repp & Ljudmila Geist. 2025. Negative polar questions in Russian: Question bias and question concern. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions: Experimental results and theoretical modelling, 101–139. Berlin: Language Science Press. DOI: 10.5281/zenodo.17158182 Sophie Repp & Ljudmila Geist 1 Introduction Questions expecting a yes or no answer come in different forms. In English or German, for instance, a {p, ¬p} question may have interrogative or declarative syntax. The syntactic form has an influence on the contextual appropriateness of the question. It affects for example whether or not there must be evidence pointing towards the truth of por ¬p in the immediate context. It may also impose requirements on the speaker’s previous beliefs or current expectations regarding the truth of por ¬p. Similar requirements have been found to be triggered by the presence of a negative marker in a question, modulated by the marker’s form and syntactic position. Furthermore, in various languages particles have been shown to influence the contextual requirements for yes-no-questions. Finally, prosody has been shown to be important. In recent years, there have been increased efforts to find empirically adequate descriptions of the contextual appropriateness conditions of yes-no-questions in several languages, and various theoretical proposals have been made on how to account for these conditions (e.g. Ladd 1981, Büring & Gunlogson 2000, Gunlogson 2003, 2008, Van Rooij 2003, Romero & Han 2004, Šafářová 2005, Repp 2009, 2013, AnderBois 2011, 2019, Northrup 2014, Trinh & Crnič 2011, Trinh 2014, Malamud & Stephenson 2015, Romero 2015, Seeliger 2015, 2019, Domaneschi et al. 2017, Farkas & Roelofsen 2017, Krifka 2015, 2017, Gärtner & Gyuris 2017, Westera 2017, 2018, Goodhue 2018, 2019, 2021, Jeong 2018, Rudin 2018, Seeliger & Repp 2018, Silk 2020, Romero 2020, Arnhold et al. 2021).1 The central term in the discussion of the contextual appropriateness conditions of yes-no-questions is question bias. This term intuitively is used to express that a question may be non-neutral: it may be biased, for instance because the speaker has particular expectations regarding the true answer. As suggested above, a question can be biased in various ways, and various bias types have been suggested. A fruitful systematic division was proposed by Sudo (2013), who distinguishes between epistemic bias and evidential bias.Epistemic bias refers to the speaker having particular beliefs, desires or expectations with respect to the truth of 𝑝/¬𝑝. Thus the term subsumes various speaker bias types beyond those based on beliefs, contrary to what the term epistemic might suggest. In this paper, we will use the term epistemic bias for the speaker’s beliefs only.2Evidential bias, the other major bias type, is the bias contributed by the contextual evidence. It has become clear in the discussion of question bias that there is no such thing as a truly neutral question. All questions have what has been termed a bias 1The final draft of this paper was submitted in 2022. Therefore, there is no reference to literature beyond that date. This holds for all literature in this paper. 2Other speaker bias subtypes would then be, e.g., bouletic and deontic bias. 102 4 Negative polar questions in Russian: Question bias and question concern profile (Gärtner & Gyuris 2017). The bias profile describes the exact requirements concerning epistemic bias and evidential bias for the felicitous use of a question with a certain form. The bias profile of a seemingly neutral question – a simple positive question with interrogative syntax in English like Is the cafeteria open? – contains the requirement that there be no contextual evidence against the proposition that “surfaces” in the question (the prejacent of the question). For instance, we cannot ask Is the cafeteria open? if our colleague, who hates microwave food and only eats it when the cafeteria is closed, comes into our office with a steaming hot package of microwave food: this would be evidence against the proposition the cafeteria is open. In addition to the bias profile, questions may differ with respect to what exactly the speaker wishes to achieve by asking the question, apart from finding out about the truth of 𝑝or ¬𝑝. For instance, Ladd (1981) argued that a negative interrogative question may be used to double-check – in modern parlance – the evidential bias, i.e. what might be true in view of the contextual evidence, or the epistemic bias, i.e. what might be true in view of what the speaker had assumed before perceiving the contextual evidence. This difference is reflected in the difference between the inner negation reading vs. outer negation reading of a yes-no question. Consider the question Isn’t the cafeteria open? This question is felicitous when there is contextual evidence for ¬𝑝(the cafeteria is not open) and where the speaker had assumed that 𝑝, among other contexts. In the cafeteria scenario discussed above, the evidence for ¬𝑝could be our colleague eating microwave food. If our assumptions about the usual opening times of the cafeteria support the truth of 𝑝(the cafeteria is open), we can ask Isn’t the cafeteria open? felicitously. With an inner negation reading, the question double-checks the evidential bias ¬𝑝, i.e., a negative proposition. With an outer negation reading, it double-checks the epistemic bias 𝑝, i.e. a positive proposition (see Section 2 for more details). Which bias a speaker wishes to double-check may depend on various factors. The disputability of the actual evidence may play a role: Can the evidence be interpreted in a different way than what first impressions suggest, or can it not? For instance, our colleague might have a very urgent task to finish and has no time to go to the cafeteria. Another factor is the willingness of the speaker to accept the evidence. For example, if our colleague is not eating microwave food but is holding a coffee cup with the cafeteria stamp on it, there is evidence for the proposition the cafeteria is open. We might not want to accept the implications of this evidence because we had resorted to microwave food ourselves, thinking that the cafeteria is closed. If that is the case we could ask the rejecting declarative question Surely the cafeteria isn’t open?! (Seeliger 2015, 2019, Seeliger & Repp 103 Sophie Repp & Ljudmila Geist 2018). Thus, irritation or indignation resulting from a conflict between evidential and epistemic bias might play a role for the communication of expressive meaning components in a question. We will call the set of shades of meaning that concern (a) the goal of the question – to double-check the evidential vs. the epistemic bias, or to reject (implications of) the evidence –, and (b) the communication of psychological states like indignation or irritation the concern of the question. Question concern is thus about what the speaker wishes to achieve by asking the question, beyond finding out what is the truth. The bias profile, in contrast, is about restrictions on the context in terms of situational evidence for 𝑝/¬𝑝and about speaker-internal assumptions about the truth of 𝑝/¬𝑝. We propose to keep question bias and question concern apart in the descriptive framework for polar questions, which often is not done in investigations of yes-no-questions. We also propose to keep apart types of speaker bias. Recall that we reserve the term epistemic bias for speaker belief. The motivation for such fine-grained distinctions is that subtle shades of meaning might be relevant for the contextual felicity or infelicity of a question. Mixing them might blur the view on appropriateness conditions. For instance, expectations concerning truth might be quite different from previous beliefs, if there is very convincing evidence. Furthermore, questions might not only be incredulous but also confirmative (Jeong 2018, Rudin 2018, Goodhue 2021). Also, the goal of a negative interrogative question might not be to seek information at all but to make a suggestion, e.g. Don’t you want to come in? (Romero & Han 2004, AnderBois 2019). Importantly, question concern and question bias may interact in specific ways for different question types, for instance a suggestion might have a different bias profile than a double-checking question with the same syntactic form. Overall, we believe that working with precise distinctions will help finding the correct generalizations regarding the contextual appropriateness conditions for yes-no-questions. The goal of this paper is to explore question bias and particularly question concern for negative yes-no-questions in Russian. Russian is a language where yes-no-questions have a declarative syntax. It is a negative concord language, and the negative marker is a syntactic head directly preceding the verb. Russian is known to use particles to express subtle shades of meaning in questions. We focus on two particles that seem to feature in translations of English negative yesno-questions with an interrogative and/or declarative syntax: razve and neuželi, both roughly meaning ‘really’. We will argue that questions with these particles have the same bias profile but different question concerns. The paper is structured as follows. In Section 2, we illustrate the contextual appropriateness conditions (bias, concern) for yes-no-questions with different for104 4 Negative polar questions in Russian: Question bias and question concern mal characteristics in English, to set the stage for our discussion of Russian. In Section 3, we discuss the appropriateness conditions of Russian questions with the particles razve and neuželi. We discuss observations from previous literature, which focuses on positive questions, and we present new corpus evidence and intuitive judgements for negative questions with and without particle, probing for potential inner and outer negation readings of negative questions. Section 4 and Section 5 present experimental evidence from two contextual appropriateness acceptability experiments, which tested hypotheses we developed on the basis of the discussion in Section 3. Section 6 puts forth a theoretical proposal for the meaning of razve and neuželi in Russian negative questions. 2 Question form, question bias and question concern In this section, we discuss the contextual appropriateness conditions for a variety of English yes-no-questions, which shows that these questions differ in their bias profile as well as in their question concern. As we will see in Section 3, the shades of meaning expressed by the Russian particles razve and neuželi overlap with different question forms in English. We will continue using our cafeteria example from the previous section and consider two types of context where there is a conflict between the evidential bias and the epistemic bias, as is illustrated in (1) and (2). Other context types will be discussed in Section 4. In the scenario in (1), there is evidence for the cafeteria not being open yet (¬𝑝), whereas the speaker, Amy, had assumed that the cafeteria is open already (𝑝). In (2), there is evidence for the cafeteria being open already (𝑝), whereas Amy had assumed that it is not open yet (¬𝑝). Table 1 and Table 2 list seven question types: positive and negative interrogatives (a–c), positive and negative ‘simple’ declaratives, i.e. without an adverb or particle (d– e), and positive and negative rejecting declaratives, with the adverb surely (f–g). As these tables show, it is always either the positive or the negative question that is felicitous in each context. In what follows, we will discuss the different question types in these two contexts. (1) Scenario (i): Evidential bias: ¬𝑝; Epistemic bias: 𝑝 Question: {the cafeteria is open, the cafeteria is not open} Ben, who hates microwave food, walks into the office with a microwave dish, which suggests that the cafeteria, which has a staff shortage and has reduced opening hours, is not open yet. Amy had assumed that the cafeteria is already open. Amy asks ... (Table 1) 105 Sophie Repp & Ljudmila Geist Table 1: Question forms (types) and question concern in English for Scenario (i) in (1). Example Question type Question concern a. # Is the cafeteria already open? positive interrogative - b. Isn’t the cafeteria open yet? negative interrogative with NPI checks ¬𝑝 c. Isn’t the cafeteria already open? negative interrogative with PPI checks 𝑝 d. # The cafeteria is already open? positive declarative - e. The cafeteria isn’t open yet? negative declarative with NPI checks ¬𝑝 f. Surely the cafeteria is already open?!a positive rejecting declarative rejects ¬𝑝 g. # Surely the cafeteria isn’t already open? negative rejecting declarative - aAdditional context for (f): Amy was going to get some food from the cafeteria in a couple of minutes. (2) Scenario (ii): Evidential bias: 𝑝; Epistemic bias: ¬𝑝 Question: {the cafeteria is open, the cafeteria is not open} Ben walks into the office with a plate full of food that looks like it is from the cafeteria. Amy had assumed that the cafeteria is not open yet. Amy asks ... (Table 2) Starting with the interrogative questions (a–c), note that for the negative interrogatives we have chosen examples with so-called preposed negation (the clitic n’t attaches to the clause-initial verb: Isn’t the cafeteria...), which in contrast to questions with non-preposed negation require that there be an epistemic bias for 𝑝(Romero & Han 2004). Furthermore, one of the negative interrogatives contains a negative polarity item (NPI, yet) and one contains a positive polarity item (PPI, already). As Table 1 shows, negative questions with a NPI vs. PPI double-check the truth of different propositions. The question with the NPI double-checks ¬𝑝, which is the evidential bias, whereas the question with the PPI double-checks 𝑝, which is the epistemic bias (Ladd 1981, Romero & Han 2004). Thus, these questions have different question concerns. We already saw in the introduction that 106 4 Negative polar questions in Russian: Question bias and question concern Table 2: Question forms (types) and question concern in English for Scenario (ii) in (2). Example Question type Question concern a. Is the cafeteria already open? positive interrogative checks 𝑝a b. # Isn’t the cafeteria open yet? negative interrogative with NPI - c. # Isn’t the cafeteria already open? negative interrogative with PPI - d. The cafeteria is already open? positive declarative checks 𝑝 e. # The cafeteria isn’t open yet? negative declarative with NPI - f. # Surely the cafeteria is already open?! positive rejecting declarative - g. Surely the cafeteria isn’t already open?b negative rejecting declarative rejects 𝑝 aPositive interrogatives can also be asked out of the blue. It is due to the context that the question in (a) checks 𝑝because that context in Scenario (ii) provides evidence for 𝑝. bAdditional context for (g): Amy believes that the cafeteria is not open yet because she is meeting her friend, who works there, in two minutes, so she has an interest in the cafeteria being closed. negative interrogatives can have different question concerns. We see now that this difference seems to become “visible” when the question contains polaritysensitive items.3Since in a question with a PPI the negation does not seem to anti-license the PPI, this negation has been called outer negation, whereas the negation that licenses NPIs is inner negation (Ladd 1981, Romero & Han 2004). The idea is that outer negation is “too far out” to anti-license the PPI, whereas inner negation is not. Also recall from Section 1 that an outer negation question 3AnderBois (2011, 2019), who doubts the existence of the inner-outer negation ambiguity, suggests that polarity items are no reliable indicator because they are licensed in a variety of contexts, and that inner negation readings are like outer negation readings with a NPI. We note here that there are other indicators of the ambiguity beyond polarity items, such as the syntactic form of the negation marker (Büring & Gunlogson 2000, Arnhold et al. 2021). We will also argue further below that for Russian the ambiguity is reflected in restrictions on question particles. 107 Sophie Repp & Ljudmila Geist checks the truth of a positive proposition, whereas an inner negation question checks the truth of a negative proposition. There are various analyses of the difference between outer and inner negation. It has been explained in terms of scope relations between the negation operator and an epistemic conversational operator (Romero & Han 2004), as illocutionary vs. propositional negation (the operator falsum vs. ‘normal’ propositional negation (Repp 2009, 2013), or in terms of scope relations between speech act operators (Reese 2006, Reese 2007, Asher & Reese 2005, Krifka 2015, 2017), see Romero (2020) for a recent review. We will elaborate on some of these theories in Section 6, where we make a theoretical proposal for the meaning contribution of the two Russian particles under investigation in this paper, razve and neuželi. Turning next to simple declarative questions without an adverb (d–e), Table 1 and Table 2 indicate that they are felicitous in the same contexts as the corresponding polar interrogatives, and that negative declaratives seem to have the same question concern as negative interrogatives with inner negation. Negative declaratives cannot host a PPI. A difference between interrogatives and declaratives which we cannot observe in the tables is that declarative questions require evidence in the situation supporting the expressed proposition, which is not the case for interrogatives. Positive interrogatives can be asked totally out of the blue, and negative interrogatives with an outer negation reading do not require evidence either (e.g. Büring & Gunlogson 2000). Conversely, simple declarative questions do not require an epistemic bias for 𝑝or ¬𝑝, in contrast to negative interrogatives, which require an epistemic bias for 𝑝. Declarative questions with the adverb surely (f–g), so-called rejecting questions (Seeliger 2015, 2019, Seeliger & Repp 2018), have a bias profile that seems to be the mirror image of the profile of simple declarative questions, at least to some extent. Table 1 and Table 2 show that the felicity of rejecting declarative questions is reversed in the two scenarios: Positive rejecting questions require evidence for ¬𝑝and previous belief that 𝑝, and negative rejecting questions require evidence for 𝑝and previous belief that ¬𝑝.4Moreover, the speaker of a rejecting question finds it hard to accept that the evidence is true, hence the term rejecting question: the speaker rejects what they perceive. In the table notes, we have suggested specifications of the context scenarios for the rejecting questions, which state that the speaker has a fairly firm belief and/or plans associated with that belief. Thus, rejecting questions do not only have a different bias profile but also a different question concern than simple declarative questions or interrogative questions. They do not (only) double-check a bias, they reject the evidence or the 4Positive rejecting questions are often somewhat marked, see Seeliger (2019) for discussion. 108 4 Negative polar questions in Russian: Question bias and question concern implications of the evidence concerning the truth of 𝑝/¬𝑝, i.e. they essentially express irritation or indignation, whilst still leaving room for the truth as it is suggested by the evidence. The latter aspect makes them questions, rather than rejections. Our short review has shown that yes-no-questions with different forms are felicitous in different contexts, which can systematically be described in terms of bias profile and question concern. We have no space to review the theoretical literature on the contextual appropriateness conditions of questions here. We already mentioned various theoretical proposals for the role of negation above. Approaches that capture bias beyond negation have modelled biases for instance as arising from conventional implicatures associated with a conversational operator (e.g. Romero & Han 2004, also cf. Repp 2009, 2013), from an interplay of a presupposition brought in by a silent operator with certain pragmatic principles (e.g. Trinh 2014, also cf. Seeliger & Repp 2018), from a requirement on contextual backup of contingent commitments (≈bare declarative questions, Gunlogson 2008), from a requirement of placing a metalinguistic issue on the conversational table (Malamud & Stephenson 2015), from intonation signaling the suspension of a Gricean maxim (Westera 2017), or, somewhat more narrowly, from lack of speaker commitment (Rudin 2018, Goodhue 2021), to name just a few. 3 The Russian question particles razve and neuželi As mentioned in Section 1, Russian yes-no-questions have a declarative syntax. Interrogativity is marked prosodically (e.g. Bryzgunova 1980, Odé 1989, Meyer 2004, Meyer & Mleinek 2006). The word order in yes-no-questions may vary but this is mainly determined by information-structural features rather than by sentence type (e.g. King 1994). The particles razve and neuželi occur in sentenceinitial position, and they both roughly mean ‘really’. As we will see in this section, both particles occur in questions whose bias profile contains an evidential bias for the denoted proposition, and an epistemic bias for its complement, which is similar to English simple declarative and interrogative questions, schematically shown in Table 3. We will argue that Russian questions with the particles razve vs. neuželi differ with respect to question concern. With a razve-question, the speaker may double-check the epistemic bias or the evidential bias, partly depending on additional formal features of the question. With a neuželi-question, the speaker may only double-check the evidential bias. The concern of a neuželi-question additionally includes expressing disbelief of the contextual evidence, potentially accompanied by irritation or, in some cases, joy. We first discuss positive questions and then turn to negative questions. 109 Sophie Repp & Ljudmila Geist 3.3 Negative questions with and without particles: Experimental investigations We have argued that razve and neuželi impose different restrictions regarding the concern of a question both in terms of expressive meaning components and in terms of the bias that is checked. An issue that we have not addressed yet is what bias profile Russian negative question without a particle (i.e., “bare” questions) have. Recall that razve and neuželi-questions have the same bias profile as English negative interrogative questions with preposed negation. Now, English interrogatives may also occur with non-preposed negation, as in Is the cafeteria not open? (vs. Isn’t the cafeteria open?). For such interrogatives, it has been suggested that they may appear in “neutral” contexts, i.e., in contexts without evidence for ¬𝑝, and without an epistemic bias for 𝑝(Romero & Han 2004). Given that Russian does not have the option to prepose the negation (the negation invariably attaches to the verbal head), and given that particles indicate a specific bias profile, it may be the case that it is bare negative questions that are felicitous in neutral, “unbiased” contexts. It is usually hard to come by neutral contexts for asking a negative question. The reason is, that usually, one can ask a positive question instead. For instance, instead of asking Is the cafeteria not open? one can easily ask Is the cafeteria closed?. The positive question is less complex and should be preferred unless particular meaning components are to be expressed. Still, we can imagine a situation where two people want to draw up a list of eating places on campus which are not open on a strike day. This would then be a list of negative propositions. To explore the bias profile of bare negative questions and to test the generalizations that we proposed in the previous subsections about the question concern of razveand neuželi-questions, we conducted two acceptability judgement experiments, which we report in Section 4 and Section 5. In Experiment 1 (Section 4), we examined the difference between Russian negative yes-no-questions with vs. without a particle regarding their bias profile. We compared bare questions to razve-questions both in negative list contexts (unbiased contexts) and in contexts with evidence for ¬𝑝and speaker belief in 𝑝(biased contexts). We hypothesized that bare negative questions should be felicitous in unbiased list contexts whereas negative razve-questions should be infelicitous. Conversely, we hypothesized that razve-questions should be felicitous in biased contexts whereas bare questions should be less felicitous/somewhat degraded. We did not expect complete infelicity because negative questions are more complex than positive questions, and the negation in bare negative questions might be taken to indicate a bias even without a particle. Similar observations have been made for English 116 4 Negative polar questions in Russian: Question bias and question concern interrogatives with non-preposed negation, which are not restricted to unbiased contexts (Romero & Han 2004). Still, the use of razve maximizes the expression of the appropriate shades of meaning so that in the biased contexts the more complex form should be preferred. Table 4 summarizes our predictions. Table 4: Predictions for Experiment 1 Unbiased context Biased context negative list evidence: ¬𝑝, epistemic bias: 𝑝 bare negative question 3degraded negative razve-question 7 3 In Experiment 2 (Section 5) we explored the subtle meaning differences between razve and neuželi that we discussed in Section 3.1, applied to negative questions. Specifically, we hypothesized that neuželi-questions signal that the speaker experiences a strong conflict between epistemic bias and evidential bias, whereas razve-questions do not. To test these different degrees of conflict in the experiment, we manipulated on the one hand the source and “strength” of the epistemic bias: reliable and strong (by hypothesis appropriate for neuželiquestions) vs. not so reliable and strong (by hypothesis appropriate for razvequestion). On the other hand, we manipulated the description of the speaker’s psychological state as merely doubtful (razve) vs. truly puzzled (neuželi). Together, these manipulations created contexts that licensed what we will call checking questions, i.e. questions that by hypothesis contain razve rather than neuželi, vs. disbelieving questions, i.e. questions that by hypothesis contain neuželi rather than razve.6 Table 5 summarizes our expectations for Experiment 2. It shows that we do not expect total infelicity if the “wrong” particle is used, but degradedness. As we highlighted in Section 3.2, the meaning differences between the particles are very subtle and readers might accommodate contextual information that is not explicitly given to ‘make sense’ of the use of a particle. In Section 6 we will give further details on how exactly we implemented the contextual features in Experiment 2. 6We do not use the term ‘rejecting question’ despite the similarity between neuželi-questions and rejecting questions in expressing disbelief, because of the different bias profiles. 117 Sophie Repp & Ljudmila Geist Table 5: Predictions for Experiment 2 Context for Context for checking question disbelieving question negative razve-question 3degraded negative neuželi-question degraded 3 To reduce the complexity of the experimental designs, both experiments tested questions with inner negation only (i.e. questions with eščë). We included questions that potentially may receive an outer negation reading (i.e. questions with uže) in the filler items. 4 Experiment 1: No particle vs. razve Experiment 1 explored the felicity of negative yes-no-questions with and without the particle razve. The questions were embedded in two types of contexts. Either the goal of the larger conversation was to produce a list of negative propositions such that the prejacent of the question, ¬𝑝, was one of them and the purpose of the question was to augment the list by one item (= unbiased context), or the purpose of the question was to check the truth of the proposition ¬𝑝in view of a conflict between an epistemic bias for 𝑝and evidential evidence for ¬𝑝(= biased context). 4.1 Method Participants in Experiment 1 read short scenarios, consisting of a scene-setting passage leading up to a conversation between two interlocutors, Dima and Katja, followed by a question asked by Katja. An example is given in (13). In the scenesetting passage, a larger situation is described which involves Dima, Katja and a third person that has or has not carried out a certain action (Julja in (13)). The question that Katja asks is about that action (Julja wrote some leaflets). The experiment had a 2×2design with the factors bias context (neutral vs. biased) and particle (no particle vs. razve). The factor bias context was manipulated in the final sentence of the scene-setting passage. That sentence announces the subsequent conversation and the question that Katja asks, including its purpose, i.e. the concern of the question. In the unbiased context, Dima and Katja’s conversation is about actions that the third person hasn’t carried out yet, 118 4 Negative polar questions in Russian: Question bias and question concern because these are relevant for Dima and Katja’s future plans; the conversation starts with Katja asking a relevant negative question. In the biased context, Katja is described as having a prior belief concerning the state of affairs regarding the third person’s action at issue: Katja believes that the action has been carried out. This belief is in conflict with the contextual evidence, which suggests that the action has not been carried out: In (13) Dima makes some comments that are interpreted by Katja in this way. By asking the question, Katja wants to find out whether 𝑝or ¬𝑝. The factor particle was implemented by the form of the question: It either contained the particle razve or it did not. All questions contained the negative marker ne ‘not’ and the weak NPI eščë ‘yet’, and thus by hypothesis contained inner negation. Hence, Katja double-checks ¬𝑝to find out whether 𝑝 or ¬𝑝. (13) In the village where Dima and Katja live, there is a very old oak tree that is to be cut down. Dima and Katja have organised a citizen’s committee and are opposing the cutting down of the tree. Julja, their friend and environmental activist, helps them prepare leaflets. This afternoon, Dima spoke to Julja and found out that she has already written the text for the leaflets. After lunch, Dima is talking with Katja. Unbiased context: They speculate about what Julja hasn’t done yet to find out how much more time she needs. Biased context: Katja thought the text for the leaflets was ready, but now she isn’t so sure, as Dima has just told her that he wanted to offer Julja help. a. Katja: Razve part Julja Julja eščë yet ne not napisala wrote tekst text dlja for listovok? leaflets [+particle] ‘Has Julja really not written the text for the leaflets yet?’ b. Katja: Julja Julja eščë yet ne not napisala wrote tekst text dlja for listovok? leaflets [-particle] ‘Has Julja not written the text for the leaflets yet?’ The experiment contained 24 lexicalizations in the four conditions just described. In half of these lexicalizations, the action by the third person had actually been carried out, in the other half, it had not been carried out. Dima – and the participants (but not Katja) – always knew what was the actual state of affairs. In addition to the experimental items, there were 24 filler lexicalizations that on the one hand had a balancing function, and on the other hand served as controls. All fillers contained the PPI uže ‘already’, to counterbalance eščë in the 119 Sophie Repp & Ljudmila Geist experimental items. Also for counterbalancing, 75% of the fillers did not contain a negation. Furthermore, 50% of the fillers had the ‘opposite’ bias profile than in the experimental items (i.e. evidential bias 𝑝; epistemic bias ¬𝑝), and the state of affairs was 𝑝; the other 50% had no bias and the state of affairs was ¬𝑝. We implemented these balancing aspects in four filler types. One filler type (nobias.uže.pos), contained as target question a simple positive question presented in a neutral context. This type served as control because there should be no doubt about the high contextual appropriateness of a positive question in an unbiased context. Another non-biased filler type (nobias.li.uže.pos) contained a target question with a clause-initial verb followed by the particle li. We expected this filler type to be fairly acceptable. The opposite-biased-context fillers had as target razve-questions with negation (opp.razve-uže.neg) or without (opp.razveuže.pos). We expected opp.razve-uže.neg questions to be unacceptable because their bias profile is violated, whereas opp.razve-uže.pos questions were expected to be acceptable. These last two filler types served to double-check our core assumptions concerning the bias profile of razve-question (see Table 3 in Section 3), which on the basis of previous literature, we consider uncontroversial. The 48 lexicalizations were distributed over four lists in a Latin square design so that each list contained 24 experimental and 24 filler items. In addition, there were two practice items on each list. The task of the participants was to judge the acceptability of the question in the given context situation on a seven-point scale. The end points of the scale were labelled with soveršenno ne podchodit ‘does not fit at all’ or očen chorošo podchodit ‘fits very well’. For the statistical analysis, these end points were transformed to the numbers 1 and 7, respectively, with the other scale points sitting in between. In addition to giving the acceptability judgment, participants had to verify a statement about the context (e.g. Dima knows that Julja has not yet written the text for the leaflets in (13)), to ensure that they read the contexts carefully. The verification statement was shown to the participants on an individual screen page after they had read the test item and given the acceptability judgment. The experiment was run as a web experiment on soscisurvey.de. 36 participants (27 female, 9 male; mean age: 34.8; age range: 20–56) with Russian as their native language were recruited via prolific.com. Before taking part in the experiment, they gave informed consent. They were paid for participation. 4.2 Results The data from four participants were excluded from the statistical analysis because they did not judge at least 80 per cent of the verification statements cor120 4 Negative polar questions in Russian: Question bias and question concern rectly, because they clicked the same scale point throughout the entire experiment, and/or because they did not give the expected judgements for those fillers which left absolutely no room for degrees of acceptability (simple positive questions). This left the data from 32 participants for analysis. The analyses were conducted by fitting a cumulative link mixed model for ordinal data (R package ordinal, Christensen 2019). Bias context and particle were fixed factors. They were sum-coded. Participant and lexicalization were random factors. The final, maximal model contained random intercepts and slopes for the experimental factors and their interaction. Figure 1 shows the results for the experimental conditions in terms of proportions of rating levels. The median for razve-questions in unbiased contexts is 6, for all other conditions it is 7. Figure 1: Proportion of rating levels for the experimental conditions in Experiment 1. Darker bars reflect greater contextual appropriateness. Table 6 shows the model estimates for the effect of the two experimental factors. There was a main effect of bias context, which was modulated by an interaction of bias context and particle. We resolved the interaction by forming a subset for razve-questions and a subset for bare questions. Razve-questions were judged to be more suitable in biased contexts than in unbiased contexts (b=1.01,SE =0.24,z=4.16,𝑝<0.001), and bare questions were judged to be more suitable in unbiased than in biased contexts (b= -0.38, SE = 0.14, z= -2.71, 𝑝= 0.0017).7 We checked a potential influence of the actual state of affairs, which was balanced across lexicalizations, because knowing what was actually the case might 7Due to convergence issues this model did not contain random slopes for lexicalization. 121 Sophie Repp & Ljudmila Geist Table 6: Model estimates for the experimental factors in Experiment 1. estimate SE z-value p-value bias context 0.32 0.14 2.36 0.02* particle -0.08 0.13 -0.60 0.55 bias context ×particle 0.72 0.16 4.46 <0.001*** have influenced participants’ judgements: Participants might have developed their own bias(es). The analysis revealed that the questions were judged as less suitable when the action at issue had been carried out, which suggests that the knowledge about the actual state of affairs did influence the assessment of the contextual evidence by the participants. Importantly, there was no interaction with the experimental factors. The results for the fillers are given in Figure 2. We can see that, as predicted for the opp.razve-uže.neg fillers, negative razve-question with the PPI uše are not acceptable in a context with the “opposite” bias from the experimental bias contexts, i.e. epistemic bias for ¬𝑝and evidential bias for 𝑝. The median for these questions is 2, which we take to reflect contextual inappropriateness. Razve-questions with uše but without negation (opp.razve-uže.pos) are suitable in such contexts (median: 7). We take these results to confirm our (uncontroversial) core assumptions regarding the bias profile of razve-questions. As the presence of uže rather than eščë should not play a role for the acceptability of these questions – recall that our corpus investigations showed that both uže and eščë are felicitous in razvequestions –, we assume that it is indeed the altered context that is responsible for these results. Questions with the particle li seem to be somewhat degraded in unbiased contexts, although the median is also 7. 4.3 Discussion The results suggest that negative polar questions with the weak NPI eščë but without a question particle are more acceptable in unbiased list contexts than in contexts with an epistemic bias for 𝑝and an evidential bias for ¬𝑝. Negative polar questions with eščë and with the particle razve are more acceptable in the biased contexts than in the unbiased contexts. The results also indicate that the difference between the question forms in the different context types are small. The median judgement for bare questions in biased contexts is the highest one possible. Similarly, the presence of razve in an unbiased list context does not lead 122 4 Negative polar questions in Russian: Question bias and question concern Figure 2: Proportion of rating levels for the four filler types in Experiment 1. to straight contextual inappropriateness: The median judgement is the secondhighest possible. Overall, we interpret the findings from Experiment 1 as supporting our hypothesis that Russian negative yes-no-questions without a particle or with the particle razve have the contextual appropriateness conditions that we suggested. Razve-questions are felicitous in contexts where there is evidence for the prejacent of the question and additionally an epistemic bias for the complement of the prejacent of the question. The question concern is to resolve the conflict between the biases by checking one of them – in questions with eščë, this is the evidential bias. Bare questions are most felicitous in contexts where there are no biases for 𝑝or ¬𝑝. Thus, their contextual appropriateness conditions are more similar to English interrogative questions with non-preposed negation than to English interrogative questions with preposed negation or negative declaratives. Thus, razve changes the bias profile of a negative question, as we had hypothesized, but it seems to do that only to a degree. Regarding the high acceptability of bare questions in biased contexts, our results suggest that a negative question per se – independently of the presence of razve – is indeed a more complex and thus a more marked form than a positive question, and a marked form is likely to be interpreted as signalling a marked meaning, for instance a particular bias. Our results suggest that the additional presence of razve is not strictly necessary to mark a bias, although it is preferred. Furthermore, our results clearly suggest that the correlation of bias signalling and preposing is language-specific: In Russian there is no preposing. Note, however, that even English or German negative questions presented out of context 123 Sophie Repp & Ljudmila Geist can be understood as being biased independently of the position of the negation (cf. Romero & Han 2004). Future research, especially corpus investigations must show in what contexts bare negative questions in Russian are predominantly used and in how far these contexts come with biases. Regarding the results for the acceptability of bare questions in unbiased list contexts, we found that they are highly felicitous, as expected. Interestingly, though, they are not actually judged to be “perfectly” acceptable in the unbiased contexts: only 82 percent received the highest or the second highest scale point (7: 63%; 6: 19%). This contrasts with the bare positive questions in the fillers, 93% percent of which received the highest scale points (7: 75%; 6: 18%). One reason for this finding might be that, as mentioned above, making lists of actions that have not been carried out is, arguably, an unusual activity. A straightforward alternative to making such a list is making a to-do list. It may also be the case, of course, that our experimental manipulation did not make the negative list idea sufficiently prominent for the participants, so that there was room for a biased interpretation of the context. This latter aspect might also be responsible for the fairly high contextual acceptability of razve-questions in the unbiased contexts. Another aspect that might contribute to explaining this finding is that participants may always interpret razve-questions as biased and assume that the speaker will have had a reason to express the bias – due to previous beliefs that are not mentioned in the scene-setting passage but that might be present in the larger conversation. Participants might thus just enrich the context as they see fit. 5 Experiment 2: Razve vs. neuželi In Experiment 2, we explored the idea that questions with the particles razve and neuželi have the same bias profile but signal different concerns of the question, as discussed in Section 3. Specifically, we investigated the source and presumed strength of the biases, which by hypothesis impact the extent of the conflict between the biases as experienced by the speaker, and his/her psychological state (emotional attitude). In Experiment 1, the source for the epistemic bias for 𝑝in razve-questions was left unspecific: Katja in (13) just thought that something was the case, and in other experimental items assumptions, beliefs etc. were mentioned. The source for the evidential bias was an utterance by Dima, which suggested that there might be evidence for ¬𝑝. Thus both biases had rather weak sources and were based on indirect signs rather than direct observation. There was no indication of a major conflict experienced by the person asking the ques124 4 Negative polar questions in Russian: Question bias and question concern tion, Katja. In Experiment 2, we manipulated the above-mentioned ingredients of question concern for razve vs. neuželi-questions. On the basis of our considerations in Section 3, we hypothesized that razvequestions are checking questions, and that neuželi-questions are disbelieving questions (also see Section 3.3). Table 7 summarizes how we implemented these hypotheses by manipulating various parameters in the scene-setting passage. Table 7: Meaning contributions of razve and neuželi in terms of bias source, speaker attitude, and ensuing conflict between speaker preference and likely true state of affairs as assessed by the speaker. razve neuželi checking question disbelieving question conflict between speaker preference and likely true state of affairs weak strong epistemic bias for 𝑝based on one piece of direct evidence in one previous situation based on a general belief that builds on experience gathered in many previous situations evidential bias for ¬𝑝 based on one piece of indirect but plausible evidence based on one piece of fairly unequivocal evidence psychological state doubt, uncertainty irritation, puzzlement, (speaker attitude) disbelief For razve-questions, i.e. checking questions, we assumed that the (psychological) conflict experienced by the speaker is weak (similar to Experiment 1). The source of the epistemic bias is the speaker’s previous observation of a single piece of direct evidence for 𝑝. The evidential bias for ¬𝑝is indirect but potentially plausible. Due to this constellation, the speaker has doubts s/he wishes to dispel. For neuželi-questions, i.e. disbelieving questions, we assumed that the (psychological) conflict the speaker experiences is strong. The epistemic bias for 𝑝stems from a rather firmly held belief that has developed from multiple previous observations, i.e. from general experience in the past. Recall from Section 3 that we do not follow Baranov’s (1986) suggestion that assumptions which are not based 125 Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung Table 2: The pitch and duration of the sentence-final particles aa1and aa3in Lo et al. (2019). Question type Pitch Duration Information-seeking questions (ISQ) low, level short Non-empty set rhetorical questions (RQ+) rising short Empty set rhetorical questions (RQ−) low, level long empty set rhetorical questions, with a low, level pitch and a relatively long duration. The two prosodic cues of pitch and duration were thus shown to play a role in production as they proved to be sufficient to distinguish the three question types prosodically. It remains to be seen, however, whether a three-way distinction is also present in perception, as the relevance of a prosodic cue is only as strong as the extent to which speakers rely on it in language processing. Thus a threeway distinction found in perception would corroborate those accounts that treat the meaning of information-seeking questions, empty set rhetorical questions and non-empty set rhetorical questions as different, that is, Jamieson (2018) and the one offered in this paper, as opposed to accounts that consider only a twoway distinction between information-seeking and rhetorical questions (Biezma & Rawlins 2017). And in case a three-way prosodic distinction is found, the experiment would also test the prediction that the less dependent an utterance type is on the common ground and the more generalizable across contexts, the more strongly the sentence-final particle is associated with a certain prosodic form. 5 Perception experiment We have proposed a semantic account that makes a distinction among two kinds of rhetorical questions (i.e., ones suggesting an empty set answer and ones suggesting a non-empty set answer) and information-seeking questions. This account has received support from production studies where a three-way distinction has been found (Lo et al. 2019, Lo & Kiss 2020). However, for a stronger support, we need to see if the three question types also differ from each other in perception. It has been shown by Kharaman et al. (2019) that German rhetorical questions are distinguished from informationseeking questions in perception, but the type of rhetorical question was not an independent variable (and it is not specified in the paper whether non-empty set 228 7 Cantonese sentence-final particles & rhetorical questions rhetorical questions were used at all). The question of whether the three question types are distinguishable from each other in perception still remains open. In what follows, we present our experiment in which we ask if speakers of Cantonese associate the three question types with distinct prosodic cues. If there is indeed a three-way distinction in meaning, as proposed by us and by Jamieson (2018), we expect a three-way distinction in form to be found in perception as well. On the other hand, if the two types of rhetorical questions are associated with the same prosodic cues, it corroborates the assumption made by the accounts in which rhetorical questions form a natural class based on the obviousness of the answer they suggest and nothing beyond that (Biezma & Rawlins 2017). In this case, rhetorical questions have the same meaning and, as a reviewer suggests, there could be some mechanism ensuring that in case a non-empty answer cannot be found in the common ground, the rhetorical question is by default interpreted as one with an empty set answer. If a three-way distinction is found in perception, however, there is a further prediction following from our account. Namely, speakers associate certain prosodic forms more readily with those question types that are interpreted more independently of the context, because those are the ones that are more likely to conventionalize their prosodic properties. As we have argued in Section 3.3, the interpretation of information-seeking questions and empty set rhetorical questions is less dependent on the particular context of utterance, because of their semantic properties: they are found on the edges of the informative/inquisitive scale for interrogatives (see the rightmost three columns of Table 1), which means that they are closest to being “purely” inquisitive and “purely” informative. Information-seeking questions express the maximum level of speaker ignorance: they are inquisitive and non-informative at the same time. Empty set rhetorical questions are inquisitive, too; however, their informative content is as informative as possible, since it is reduced to a single alternative. Thus the hearer does not need to consult the common ground in either of these cases: for informationseeking questions, all alternatives are elements of the meaning, and for empty set rhetorical questions, the informative content always covers the empty set answer alone, regardless of the domain. But the intended meaning of a non-empty set rhetorical question cannot be determined without making reference to the common ground. Note that this latter prediction, too, follows from the account of Jamieson (2018), who calls non-empty set rhetorical questions “pragmatic” rhetorical questions, for the same reason that it cannot be interpreted in the intended way unless the hearer recognizes the referent in the common ground as intended by the speaker. 229 Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung Lastly, we expect participants to associate the three question types with prosodic forms with which they were produced in Lo et al. (2019) (see Table 2). However, all three question types looked at here can be marked both by aa1and aa3. While in production, we can control which sentence-final particle is presented to the participants, in our perception experiment the stimuli were presented auditorily only. Since we cannot know for sure which sentence-final particle they thought they heard, the expectation that we will find the same forms in perception that we found in production cannot be confidently predicted. 5.1 Stimuli Our stimuli were created from natural utterances produced by a female in her thirties, who is a native speaker of Hong Kong Cantonese and a linguist. The sentences were presented to her in written form, and she was instructed to produce them in a natural way as information-seeking questions. There were 12 sentences, all of which ended with the sentence-final particle aa3(呀). This particle was also used in Lo et al. (2019) for its compatibility with all three relevant question readings (Sybesma & Li 2007: 1760). An example is shown in (31). (31) 有 jau5 have 邊個 bin1go3 who 想 soeng2 want 飲 jam2 drink 咖啡 gaa3fe1 coffee 呀？ aa3? sfp ‘Who wants to drink coffee?’ The speaker read the information-seeking version of each utterance in a soundattenuated booth. The elicited utterances were then manipulated by changing the duration and pitch contour of the sentence-final particle. The duration and pitch dimensions were manipulated orthogonally, with the duration taking on values of 250 ms, 310 ms, 370 ms, 430 ms, 490 ms, and 550 ms, and the pitch contour realized with a rise of 0 Hz, 25 Hz, 50 Hz, or 75 Hz. These values were determined based on the results of Lo et al. (2019), that is, in a way that combining the two prosodic cues, duration and rising in pitch, would include the three contours found in the production experiment (see Table 2). Some of these manipulations are presented schematically in Figure 1. 5.2 Procedure Seventeen native speakers of Cantonese took part in the experiment. In each trial, after listening to the stimulus, the participant completed a three-alternative forced choice task, with each option exemplifying one of the three readings. For 230 7 Cantonese sentence-final particles & rhetorical questions Duration of SFP = 250 (ms) Duration of SFP = 550 (ms) 1.50 1.64 1.77 1.91 2.05 1.50 1.64 1.77 1.91 2.05 200 225 250 275 Time (s) Fundamental frequency (Hz) Figure 1: Schematic representation of some manipulations. instance, upon hearing an utterance like (31), with a prosodically manipulated sentence-final particle, the participant had to pick one of the three options shown in (32) that they thought matched the intonation of the utterance. (32) a. 講嘢嘅人 gong2je5ge3jan4 speaker 唔 m4 not 知道 zi1dou3 know 邊個 bin1go3 who 想 soeng2 want 飲 jam2 drink 咖啡 gaa3fe1 coffee (ISQ) ‘The speaker does not know who wants to drink coffee.’ b. 講嘢嘅人 gong2je5ge3jan4 speaker 認為 jing6wai4 think 冇人 mou5jan4 nobody 想 soeng2 want 飲 jam2 drink 咖啡 gaa3fe1 coffee (RQ−) ‘The speaker thinks nobody wants to drink coffee.’ c. 講嘢嘅人 gong2je5ge3jan4 speaker 已經 ji5ging1 already 知道 zi1dou3 know 邊個 bin1go3 who 想 soeng2 want 飲 jam2 drink 咖啡 gaa3fe1 coffee (RQ+) ‘The speaker already knows who wants to drink coffee.’ 5.3 Analysis Given that there were three possible responses for each trial, we employed a Bayesian multilevel multinomial logistic regression, with a random intercept for each participant but shared fixed effects for all participants (Koster & McElreath 2017). The reference level for the dependent variable was the informationseeking question response. Alongside the intercept, the predictor variables included duration and pitch rise, both of which were z-transformed before entering the analysis. The priors for all coefficients were a normal distribution with a mean of 0 and a standard deviation of 1. For the correlation structure of the 231 Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung random intercept, a zero-mean multivariate normal distribution was used with an exponential distribution with rate 1 and an LKJ distribution with 𝜉=2as the priors for the standard deviation and the correlation matrix associated with the covariance matrix. 5.4 Results Overall, the model confirms that different combinations of pitch contours and duration on the sentence-final particle led to different proportions of informationseeking, empty set, and non-empty set rhetorical question responses. Specifically, the model results suggest that, relative to information-seeking question responses, a longer duration led to both more empty set (95% credible interval [CrI] =[0.33,0.47]) and non-empty set rhetorical question responses (95% CrI =[0.39,0.60]), even though the posterior predictive pattern for non-empty set rhetorical questions does not show a clear trend (see Figure 2). ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Pitch rise = 50 (Hz) Pitch rise = 75 (Hz) Pitch rise = 0 (Hz) Pitch rise = 25 (Hz) 250 310 370 430 490 550 250 310 370 430 490 550 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Duration of SFP (ms) Response proportion Resp. ● ● ● ISQ RQ− RQ+ Figure 2: Response proportions of information-seeking questions (ISQ), empty set rhetorical questions (RQ−) and non-empty set rhetorical questions (RQ+) as a function of manipulated duration. With respect to pitch rise, again a larger pitch rise led to more empty set (95% CrI =[0.66,0.80]) and non-empty set rhetorical question responses (95% CrI = [0.59,0.80]), although the trend for non-empty set rhetorical question responses is not as clear (see Figure 3). 232 7 Cantonese sentence-final particles & rhetorical questions ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Duration = 430 (ms) Duration = 490 (ms) Duration = 550 (ms) Duration = 250 (ms) Duration = 310 (ms) Duration = 370 (ms) 0 25 50 75 0 25 50 75 0 25 50 75 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Pitch rise of SFP(Hz) Response proportion Resp. ● ● ● ISQ RQ− RQ+ Figure 3: Response proportions of information-seeking questions (ISQ), empty set rhetorical questions (RQ−), and non-empty set rhetorical questions (RQ+) as a function of manipulated pitch rise. 5.5 Discussion The results of the perception experiment do not entirely conform the results of the production experiment of Lo et al. (2019). While the prosodic cues participants associated with information-seeking questions were similar to the prosodic cues of this question type in production (a lower, non-rising pitch and a relatively short duration, see Table 2), rhetorical questions showed some novel patterns. Namely, empty set rhetorical questions (RQ−), which were produced with a low and level pitch, were associated with a rising contour in perception. And rhetorical questions with a non-empty set as answer (RQ+), which were produced with a rising pitch, were not clearly associated with any pitch contour, as shown in Figure 2. As for duration, Figure 2 shows us that regardless of pitch, the longer the sentence-final particle is, the more likely speakers are to interpret the utterance as an empty set rhetorical question. This is a contrast that was also found to be significant in production (see Table 2). Cantonese wh-questions with the particle aa3thus provide a further example for the observations made in other languages (see Section 4.1) that boundary tones are not sufficient in characterizing the prosody of information-seeking and rhetorical questions. 233 Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung In general, a three-way distinction between information-seeking questions, empty set and non-empty set rhetorical questions was found, as in production, even if the three particular forms were somewhat different in production and perception. While we do not know why there was only a partial match between the production and perception of the specific pitch contours, the three way contrast found in both cases can be interpreted as a support for there being a three-way contrast in meaning. 6 General discussion The results of our experiment confirm that there is a three-way distinction in form between information-seeking, empty set and non-empty set rhetorical questions. These results thus support our proposal for the semantics of these question types. In addition to that, we posited a conditional prediction that in case there is a three-way distinction, those question types that depend less on the context (i.e., that do not depend on the common ground) in order to be interpreted as intended would be associated with certain prosodic forms more strongly than those that do rely on the common ground. This prediction, too, is borne out: as Figure 2 and Figure 3 show, only information-seeking questions and empty set rhetorical questions were clearly associated with a certain combination of prosodic cues. We argue that this tendency is explained by an inquisitive semantic analysis. We proposed in Section 3.3 that information-seeking questions, non-empty set rhetorical questions and empty set rhetorical questions form a scale in terms of inquisitiveness and informativity. On the inquisitive end of the scale, we find information-seeking questions which are inquisitive and not at all informative. Considering questions only, on the informative end of the scale we find empty set rhetorical questions, which are both inquisitive and informative. And between the two are rhetorical questions with a non-empty answer, which are also both inquisitive and informative, but which are less informative than the ones with an empty set answer. We have argued that these properties map to different levels of context-dependence and that there is an asymmetry between the two types of rhetorical questions. If the suggested answer is the empty set, the addressee receives the issue already resolved and there is no need to consult the common ground. This is not the case if the rhetorical question suggests a non-empty answer, which the addressee can only resolve by consulting the common ground. The intended answer itself is not directly encoded by virtue of the marked form of the utterance; it thereforeremains highly context-dependent, a property that led Jamieson 234 7 Cantonese sentence-final particles & rhetorical questions (2018) to call them “pragmatic” rhetorical questions. Information-seeking questions, which are inquisitive and not informative, are again less dependent on the context compared to non-empty set rhetorical questions in the sense that the addressee does not need the common ground in order to arrive at the intended interpretation of the utterance. In sum, only those utterance types which are relatively context-independent have been associated with a clear prosodic pattern: information-seeking questions and empty set rhetorical questions. The explanation we offer is that the utterance types that can be interpreted more independently from the context are also the ones that are more likely to conventionalize their prosodic cues. And if prosodic cues become conventionalized, it explains why participants associated them more strongly with the relevant question types. The reason why non-empty set rhetorical questions were not associated with a certain intonational pattern is because they convey context-sensitive information, namely, they intend to point to an already known answer. But which member(s) of the domain it points to is not grammatically encoded, simply because there is no straightforward way to do so. As for the mismatch between our production and perception results, we are not in a position to establish straightforward form-meaning mappings for the three wh-question types in Cantonese. The form-meaning relations that we propose here are at a more abstract level, so further research is needed to explore the durational and pitch properties of the sentence-final particle aa3in the three question types. We speculate that the mismatch found between the production and perception results may be due to perceiving the sentence-final particle as aa1, especially in empty set rhetorical questions, even though instead of a high level tone, it got associated with a rising contour, which is closest to the high level tone of this particle in terms of its pitch. Cantonese sentence-final particles usually come in families: the same segmental material can wear multiple tones, each of which alters its meaning. So even though intonation has been shown to completely override the lexical tone found on the last syllable of the utterance, we cannot exclude the possibility that the tonal differences between aa3and aa1 had no effect on the dependent variable. In addition, the mismatch between production and perception may also be due to prosodic cues that were not included here as independent variables. On the one hand, breathy voice has been shown to be a marker of German rhetorical questions in production (Braun et al. 2018), and the role of intensity in either producing or processing the three utterance types remains unexplored. Considering these cues in future perception experiments may therefore be advantageous. On the other hand, the fact that all stimuli were produced from utterances that were 235 Angelika Kiss, Roger Yu-Hsiang Lo & Justin R. Leung meant to be information-seeking questions may also have played an important role. Even though Lo et al. (2019) only found significant differences between the prosodic realization of the sentence-final particles of the three question types, further prosodic cues may have characterized the rest of the utterance, which may have affected perception of our stimuli. As far as the validity of our analysis is concerned, the results of our experiment not only support our analysis but that of Jamieson (2018) as well. In his account, too, “generic” and “pragmatic” rhetorical questions differ in how much the addressee has to rely on the context (namely on the common ground) in order to interpret the question. It is not the goal of this paper to evaluate Jamieson’s account, however; comparing it to the account offered here is left for future work. For now we can only point to one advantage, namely that the “generic” and “pragmatic” nature of empty set and non-empty set rhetorical questions follows from our inquisitive semantic analysis, without having to posit a metavariable in order to derive the meaning of empty set rhetorical questions. 7 Conclusion In this paper, we have argued that when analyzing the meaning of rhetorical questions, the kind of answer they suggest plays an important role in their meaning. Information-seeking questions commit the speaker to all alternatives, empty set rhetorical questions, to the empty set alternative, and non-empty set rhetorical questions, to Α, all alternatives except the empty set one. We presented experimental results that we take as a support to this claim. First, a three-way prosodic distinction has been found between these three question types. Second, participants clearly associated information-seeking and empty set rhetorical questions with certain forms. These utterance types are on the two edges of the informative/inquisitive scale of interrogatives, and as such, they have a meaning that generalizes across contexts, as opposed to non-empty set rhetorical questions, which remain context-sensitive. The question of what the intonational contours of the sentence-final particles in rhetorical questions really look like remains open, along with the question of what other prosodic cues not mentioned here may play a role in shaping the intonation of information-seeking and rhetorical questions in Cantonese. But the claims made here are nevertheless falsifiable and have cross-linguistic relevance, and as such, they invite further work on rhetorical questions. 236 7 Cantonese sentence-final particles & rhetorical questions Acknowledgements We thank Lisa Cheng, Beáta Gyuris, E Jamieson, Zoe Lam, Michela Ippolito, Pilar Prieto, Floris Roelofsen, Tue Trinh and Maxime Tulling; the audiences at International Congress of Phonetic Sciences (2019), the Annual Conference of the Canadian Linguistic Association (2019), Laboratory, Phonology 17, Sinn und Bedeutung 25, LSA95, the workshop Biased Questions: Experimental Results & Theoretical Modelling, and the 4th Forum on Cantonese Linguistics; as well as all our participants. References Abels, Klaus. 2003. Who gives a damn about minimizers in questions? In Semantics and Linguistic Theory (SALT) 8, 1–18. Asher, Nicholas & Brian Reese. 2007. Intonation and discourse: Biased questions. In Shinichiro Ishihara, Stefanie Jannedy & Anne Schwarz (eds.), Interdisciplinary Studies on Information Structure (ISIS) 8 (Working papers of the SFB 632), 1–38. Potsdam: Universitätsverlag. Asu, Eva Liina, Heete Sahkai & Pärtel Lippus. 2020. The prosody of rhetorical and information-seeking questions in Estonian: Preliminary results. In Proceedings of the 10th International Conference on Speech Prosody, 381–384. Banuazizi, Atissa & Cassandre Creswell. 1999. Is that a real question? Final rises, final falls and discourse function in yes-no question intonation. In Chicago Linguistic Society (CLS) 35, 1–14. Chicago. Biezma, María & Kyle Rawlins. 2017. Rhetorical questions: Severing questioning from asking. In Semantics and Linguistic Theory (SALT) 27, 302–322. Braun, Bettina, Nicole Dehé, Jana Neitsch, Daniela Wochner & Katharina Zahner. 2018. The prosody of rhetorical and information-seeking questions in German. Language and Speech 62. 779–807. Caponigro, Ivano & Jon Sprouse. 2007. Rhetorical questions as questions. In Sinn und Bedeutung 11, 121–133. Ciardelli, Ivano, Jeroen Groenendijk & Floris Roelofsen. 2019. Inquisitive Semantics. Oxford: Oxford University Press. Ciardelli, Ivano, Nadine Theiler & Floris Roelofsen. 2017. Composing alternatives. Linguistics and Philosophy 40. 1–36. DOI: 10.1007/s10988-016-9195-2. Clark, Herbert H. 1996. Using language. Cambridge: Cambridge University Press. Dehé, Nicole & Bettina Braun. 2019. The prosody of rhetorical questions in English. English Language and Linguistics 24(4). 1–29. 237 Tue Trinh There is, I believe, a possible functional account of this asymmetry. The account will turn on another fact about agreement questions, namely that it is biased towards the positive answer (Trinh 2010). Suppose the speaker sees John with a copy of The critique of pure reason in his hand. In this context, the agreement question in (5a) is felicitous but the yes/no question in (5b) is not. (5) Context: the speaker sees John with The critique of pure reason in his hand. a. John John đọc read Kant Kant à? a b. # John John có pos đọc read Kant Kant không? neg? In the very same context, the agreement question in (6a) would be infelicitous, while the yes/no question in (6b) would be felicitous.3 (6) Context: the speaker sees John with The critique of pure reason in his hand. a. # John đọc cả Hegel à? John read also Hegel a b. John có đọc cả Hegel không? John pos read also Hegel neg? We learn two things from (5) and (6). First, if there is contextual evidence that 𝜙, a polar question with prejacent 𝜙is only felicitous when it is formulated as an agreement question. Second, if there is no contextual evidence that 𝜙, a polar question with prejacent 𝜙is only felicitous when it is formulated as an yes/no question. In other words, yes/no questions require the context to be “prejacentneutral”, while agreement questions require it to be “prejacent-biased”. I conjecture that the prejacent bias of agreement questions might contribute to the functional pressure on the grammar to have a short answer expressing agreement with the prejacent but no short answer expressing disagreement with the prejacent.4 3Note that if the context is also such that reading Kant entails reading Hegel, (6a) would be fine. This proves the point I am making. I thank a reviewer for pointing this out. 4The idea that answers which agree with the prejacent of the question are preferred by the language system can be found in Roelofsen & Farkas (2015). 244 8 A note on bias and polarity in Vietnamese 1.2 Two types of NPIs Vietnamese is a language that build NPIs from wh-elements. For example, the word ai, as an interrogative pronoun, means ‘who’, but as an NPI, means ‘anyone’. Similarly, gì means either ‘what’ or ‘anything’ (Bruening & Tran 2006). A nonnegated declarative sentence would disambiguate such expressions towards the interrogative reading, while a polar question would disambiguate them towards the NPI reading.5 (7) a. John John đang prog đọc read gì what ‘What is John reading?’ b. ‘Is John reading anything?’ i. John John có pos đang prog đọc read gì anything không neg ii. John John đang prog đọc read gì anything à a This ambiguity extends to which-phrases. The Vietnamese word for which is nào, which combines with singular NPs. As Vietnamese is a classifier language of the East Asian variety, singular number is indicated by a classifier (Chierchia 1998, Trinh 2011). I will gloss the classifier as cl. (8) a. John John đang prog đọc read quyển cl sách book nào which ‘Which book is John reading?’ 5Note that a negated declarative sentence would allow these expressions to be ambiguous between the interrogative and the NPI reading, as exemplified in (i). (i) John John không not gặp met ai who ‘Who did John not meet?’ / ‘John did not meet anyone’ I use the term NPI to describe expressions denoting existential quantifiers whose occurrence is limited to environments that must be characterized semantically as involving negation in some sense. For the purpose of this particular discussion, I will take NPIs to be expressions that can be understood as existential quantifiers in polar questions and under negation, but cannot be so understood in non-negated declarative sentences. Thus, it is possible for something to qualify as an NPI even if its distribution turns out to differ from that of English anything with respect to other environments. I believe this terminological practice is common in the literature, and thank a reviewer for pointing out the need to make this clear. 245 Tue Trinh b. ‘Is John reading any book?’ i. John có đang đọc quyển sách nào không John pos prog read cl book any neg ii. John John đang prog đọc read quyển cl sách book nào any à a In what follows, we will not be concerned with the interrogative reading of wh-phrases. For this reason, I will gloss cl+NP+nao simply as “any NP”. NPIs in Vietnamese come in two morphological variants, simple and complex. Those we just discussed are the simple ones. Complex NPIs are built out of simple NPIs by prefixing the latter with the word bất kỳ (Trinh 2020), which I will gloss as bk. (9) John John có pos đang prog đọc read bất kỳ bk quyển sách nào any book không neg ‘Is John reading any book at all?’ As indicated by the translation in (9), adding bất kỳ to the NPI in Vietnamese has a similar interpretive effect as adding at all to the NPI in English: it gives rise to the inference that the speaker is biased towards the negative answer, in the sense that she has more reasons to think that the negative answer is correct than to think that the positive answer is. In the case of (9), the inference would be that the speaker strongly suspects that John is not reading any book.6Simple NPIs, on the other hand, do not induce such negative bias. The question in (8b-i), for example, does not give rise to any inference about which answer the speaker strongly suspects to be correct. Another difference between simple and complex NPIs pertains to their distribution across the two types of polar questions: whereas simple NPIs are acceptable in both yes/no and agreement questions, as shown by (10a), complex NPIs are acceptable in yes/no questions but give rise to deviance when they occur in agreement questions, as shown by (10b). (10) Intended reading: ‘Is John reading any book at all?’ a. John John có pos đang prog đọc read bất kỳ bk quyển sách nào any book không? neg b. # John John đang prog đọc read bất kỳ bk quyển sách nào any book à? a 6Note that I am describing the effect of at all in canonical, non-negated English yes/no question containing an NPI, as exemplified by the translation of (9). It was pointed out to me that at all can occur in a high negation question, e.g. Isn’t John reading any book at all?, which gives rise to a positive speaker’s bias (Dan Goodhue p.c.). I have nothing to say about this fact. 246 8 A note on bias and polarity in Vietnamese 2 Analysis We have seen that Vietnamese polar questions come in two variants, yes/no questions and agreement questions. Yes/no questions are prejacent-neutral while agreement questions are prejacent-biased. We have also seen that NPIs in Vietnamese come in two variants, simple and complex. Simple NPIs are acceptable in both yes/no and agreement questions. Complex NPIs are acceptable in yes/ no questions but cause deviance in agreement questions. In yes/no questions, complex NPIs give rise to negative bias while simple NPIs do not. The present section will be devoted to an analysis of these facts. 2.1 Introducing whether For the purpose of this paper I will assume a simplified version of the analysis proposed in Hamblin (1973), Karttunen (1977). Let us define two functions. The first is yes, the identity function, and the second is no, the negation function. (11) a. yes =def 𝜆𝑝∈𝐷𝑠𝑡.𝑝 b. no =def 𝜆𝑝∈𝐷𝑠𝑡.¬𝑝 We will say that a function 𝑓of type ⟨𝑠𝑡,𝑠𝑡⟩is a “polarity”, i.e. that 𝑝𝑜𝑙𝑎𝑟𝑖𝑡𝑦(𝑓), if 𝑓is either yes or no. For polar questions, I assume the presence of a (overt or) covert whether (Bennett 1977, Higginbotham 1993, Krifka 2001, Guerzoni & Sharvit 2014). (12) JwhetherK =𝜆𝑄∈𝐷⟨⟨𝑠𝑡,𝑠𝑡⟩,𝑡⟩.𝜆𝑝∈𝐷𝑠𝑡.∃𝑓∈𝐷⟨𝑠𝑡,𝑠𝑡⟩.𝑝𝑜𝑙𝑎𝑟𝑖𝑡𝑦(𝑓)∧𝑝=𝑄(𝑓) The base position of whether is above TP. When it moves, it leaves a trace of type ⟨𝑠𝑡,𝑠𝑡⟩. Predicate abstraction proceeds in the familiar way. (13) 𝜆𝑝.∃𝑓.𝑝𝑜𝑙𝑎𝑟𝑖𝑡𝑦(𝑓)∧𝑝=𝑓(𝜙) whether 𝜆𝑔.𝑔(𝜙) 1𝑡1(𝜙) 𝑡1𝜙 247 Tue Trinh I will assume that at the relevant level of analyis, a yes/no question in Vietnamese whose prejacent is 𝜙has the logical form [whether 𝜙]. Thus, the question in (1b), reproduced below in (14a), has the logical form in (14b), which denotes the set in (14c). (14) a. John John có pos đọc read Kant Kant không? neg b. 𝛼 whether 1𝑡1𝜙 John read Kant c. J𝛼K={yes(John reads Kant), no(John reads Kant)} 2.2 Introducing evid What about agreement questions? Recall that agreement questions are prejacentbiased, in the sense that they require the context to contain evidence for the prejacent. I will adopt the analysis proposed in Trinh (2014) and assume the existence of an evidential marker evid, which is attached to TP and has the following interpretation. (15) Jevid 𝜙K={J𝜙Kif there is contextual evidence that 𝜙 #otherwise Thus, [evid 𝜙]presupposes that there is contextual evidence that 𝜙. I propose that at the relevant level of analysis, the agreement question (16a) has the LF in (16b), and the denotation in (16c). (16) a. John John đọc read Kant Kant à? a b. 𝛼 whether 1𝑡1evid 𝜙 John read Kant 248 8 A note on bias and polarity in Vietnamese c. J𝛼K={yes(evid(John reads Kant)), no(evid(John reads Kant))} Both answers contain [evid(John reads Kant)]as a subconstituent. Thus, both answers presuppose that there is contextual evidence that John reads Kant, which means the question presupposes that there is contextual evidence that John reads Kant.7We thus account for the fact that agreement questions are evidentially biased toward the prejacent. How do we account for the fact that yes/no questions are prejacent-neutral, i.e. that a yes/no question with prejacent 𝜙is infelicitous in contexts where there is evidence that 𝜙? I propose that this effect comes about as an anti-presupposition. I will assume the principle of Maximize Presupposition as a primitive of grammar (Heim 1991).8 (17) Maximize Presupposition (MP) Presuppose as much as possible! Given MP, a yes/no question will be understood as indicating that there is no contextual evidence for the prejacent, since if there were such evidence, the speaker would have used an agreement question instead (Sauerland 2008). 2.3 Introducing even Let us now address the fact that NPIs, both simple and complex, are acceptable in yes/no questions. A well-known fact about whether is that it licenses NPIs. Various attempts have been made to derive this observation (cf. Ladusaw 1979, Krifka 1991, 1995, Van Rooy & Šafářová 2003, Guerzoni & Sharvit 2007, 2014, Nicolae 2015, Roelofsen 2018, Jeong & Roelofsen 2023). For this paper, I will assume it as a primitive. (18) whether licenses NPIs in its scope 7I assume that if all answers to a question have a presupposition then the question itself inherits that presupposition. This follows from the fact that questions are quantificational structures (Heim 1983, 1992, Schlenker 2008). 8A reviewer raises the question whether MP should be considered a principle of rational communication of a Gricean sort which is external to the language faculty rather than one of grammar. I will not attempt to address this question adequately, as that would take us beyond the scope of this paper. What matters is that I use MP without deriving it, not how it is derived. Nevertheless, I would note that Heim (1991) did point out how it would be difficult to derive MP from principles of information exchange. Thus, given contextual, i.e. pragmatic, knowledge, #a sun is shining conveys the exact same amount of information as the sun is shining. It is not clear how to explain the contrast between these sentences in terms of their communicative function. 249 Tue Trinh Given (18), we predict, correctly, that NPIs of both types are acceptable in yes/no questions. However, we also predict, incorrectly, that NPIs of both types are acceptable in agreement questions as well, given our analyis of agreement questions as containing whether. Our task, therefore, is to specify a distinctive grammatical property of complex NPIs which explains the deviance caused by their occurence in agreement questions. Recall a complex NPI consists of a simple NPI plus the element bk. I propose that bk, by itself, has no independent semantics. Rather, it is just the morphological reflex of a c-commanding operator, even.9 (19) bk is the morphological reflex of a c-commanding even in the structure As its name suggests, even has a meaning akin to that of English even. For the purpose of this discussion, we will give even the interpretation in (20). (20) Jeven 𝜙K={J𝜙Kif ∀𝜓∈ALT(𝜙).J𝜙K≤likely J𝜓K #otherwise Thus, [even 𝜙]asserts 𝜙and presupposes that 𝜙is the least likely among the alternatives of 𝜙. I will assume, following many works, that NPIs induce “subdomain” alternatives. Alternatives of sentences containing NPIs are generated by point-wise composition in the familiar way. In their basic meaning, NPIs are just existential quantifiers (Kadmon & Landman 1993, Krifka 1995, Chierchia 2013).10 (21) a. JanyDbookK=𝜆𝑃.∃𝑥.𝑥∈D∩JbookK∧𝑃(𝑥)=‘a book in D’ b. ALT(anyDbook) ={anyD′book |D′⊆D}={a book in D′|D′⊆D} c. ALT(John read anyDbook) ={John reads anyD′book |D′⊆D} Simple NPIs do not come with bk. I will take this to mean that they do not come with even. In polar questions with even, whether can be base-generated either above or below even. Consider the first possibility. 9For similar ideas see Heim (1984), Guerzoni (2004), Crnič (2014), Jeong & Roelofsen (2023). 10Note that as quantifiers, NPIs must QR to be interpretable. I make the standard assumptions that NPIs have narrowest scope, i.e. that they raise to the smallest clause containing them. This means that John read anyDbook has the following LF. (i) anyDbook 𝜆1[John read 𝑡1] I thank a reviewer for pointing out the need to make this clear. 250 8 A note on bias and polarity in Vietnamese (22) 𝛼 whether 1𝑡1even 𝜙 John read anyDbook J𝛼K={yes(even(John read anyDbook)), no(even(John read anyD book))} This configuration results in a yes/no question for which both answers, the positive as well as the negative, have the same unsatisfiable presupposition, induced by the subconstituent in (23). (23) # even(John read anyDbook) Presupposition: John reads a book in D ≤likely John reads a book in D′, for any D′⊆D Given that likelihood respects logical entailments, i.e. that 𝜙≤likely 𝜓if 𝜙⇒𝜓, and given the logical truth that for any D and D′such that D′⊆D, if John reads a book in D′then John reads a book in D but not vice versa, both answers in (22) presuppose that a weaker sentence is less likely than a stronger sentence, which is necessarily false. I will take this fact to mean that such a parse as (22) will be ruled out as deviant by the grammar. Having even scoping above the trace of whether, however, results in a polar question with one felicitous answer, namely the negative. (24) 𝛼 whether 1 even 𝑡1𝜙 John read anyDbook J𝛼K={even(yes(John read anyDbook)), even(no(John read anyDbook))} 251 Tue Trinh The positive answer in this case is equivalent to the positive answer in (22), and is deviant for the same reason, namely because it has a necessarily false presupposition. The negative answer, however, does not have such a presupposition. Let us consider it. (25) even(no(John read anyDbook)) Presupposition: ¬John reads a book in D ≤likely ¬John reads a book in D′, for any D′⊆D Negation is scale-reversing, so for any D′⊆D, if it is not the case that John reads a book in D then it is also not the case that John reads a book in D′, but not vice versa. This means the negative answer in (24) has a trivially true presupposition. We thus see that if even is present in a polar question, it has to be parsed above the base position of whether, and within this parse, only the negative answer is acceptable. This means that polar questions with even have only the negative answer as the one felicitous answer. And because complex NPIs require a c-commanding even, we predict that for polar questions with complex NPIs, only the negative answer is felicitous. Asking a polar question with a complex NPI, then, amounts to presenting the hearer with the negative answer as the only choice. I propose that this is what brings about the inference that the speaker of such a question is biased towards the negative answer (cf. Guerzoni 2004). For concreteness, I will take this inference to be a conversational implicature of the question.11 Let us now come (back) to the question why complex NPIs cause deviance in agreement questions, as evidenced by (10b), reproduced below in (26). (26) # John John đang prog đọc read bất kỳ bk quyển sách nào any book à? a Intended reading: ’Is John reading any book at all?’ Given what we have said, this question will have the parse in (27a) and the denotation in (27b). 11The issue arises, of course, as to how a ‘yes’ answer to such a negatively biased question is still possible. Note that this issue arises in the same way for Guerzoni (2004). One plausible response is to say that a ‘yes’ answer, in this case, requires the accomodation of a slightly different question, namely one without the negative bias. Thus, such an answer is also a move to deny the negative presupposition of the biased question. 252 8 A note on bias and polarity in Vietnamese (27) a. 𝛼 whether 1 even 𝑡1evid 𝜙 John is reading anyDbook b. J𝛼K={even(yes(evid(John is reading anyDbook))), even(no(evid(John is reading anyDbook)))} Let us consider the inferences licensed by this question. Due to the presence of evid, it has the presupposition in (28a). And due to the presence of even, it has the implicature in (28b). (28) Inferences licensed by (27) a. There is contextual evidence that John is reading a book in D b. The speaker strongly suspects that John is not reading a book in D I submit that these two inferences are responsible for the question being perceived as deviant. The reason, I claim, is that a rational speaker cannot both strongly suspect ¬𝜙while at the same time take some fact in the context to be evidence that 𝜙. Thus, if he really strongly suspects that John is not reading a book, the sight of John reading a book would have to be interpreted by him to be evidence that John is pretending to read a book.12 3 Interim summary Polar questions contain a covert whether, which accounts for the intuition that they ask the hearer to confirm a proposition or to confirm its negation. Polar questions in Vietnamese come in two variants, yes/no questions and agreement questions. Agreement questions contain evid, the evidential marker which introduces the presupposition that its prejacent is supported by contextual evidence. Yes/no questions, in contrast, do not contain evid. Given Maximize Presupposition, yes/no questions anti-presuppose that there is contextual evidence for the prejacent. This accounts for the fact that in prejacent-biased contexts, agreement 12I admit that this point needs further explication. I hope to pursue this task in future research. 253 Tue Trinh Menéndez-Benito, Paula. 2010. On universal free choice items. Natural Language Semantics 18(1). 33–64. Nicolae, Andreea. 2015. Questions with NPIs. Natural Language Semantics 23. 21– 76. Roelofsen, Floris. 2018. NPIs in questions. Talk given at NYU Linguistics Colloquium. Roelofsen, Floris & Donka F. Farkas. 2015. Polarity particle responses as a window onto the interpretation of questions and assertions. Language 91(2). 359–414. Sauerland, Uli. 2008. Implicated presuppositions. In Anita Steube (ed.), The discourse potential of underspecified structures, 581–600. Berlin: Mouton de Gruyter. Schlenker, Philippe. 2008. Be articulate: A pragmatic theory of presupposition projection. Theoretical Linguistics 34(3). 157–212. Trinh, Tue. 2005. Aspects of clause structure in Vietnamese. Humboldt-Universität zu Berlin. (MA thesis). Trinh, Tue. 2010. Asking with assertions. Talk given at the South East Asian Linguistics Society SEALS 20, Zürich. Trinh, Tue. 2011. Nominal reference in two classifier languages. In Sinn und Bedeutung 15, 629–644. Trinh, Tue. 2014. How to ask the obvious: A presuppositional account of evidential bias in English yes/no questions. In Luka Crnič & Uli Sauerland (eds.), The art and craft of semantics: A Festschrift for Irene Heim, vol. 2, 227–249. Cambridge: MIT Working Papers in Linguistics. Trinh, Tue. 2020. Bipartite exhaustification: Evidence from Vietnamese. In Dun Deng, Fenrong Liu, Mingming Liu & Dag Westerståhl (eds.), Monotonicity in logic and language, 207–216. Berlin: Springer. Van Rooy, Robert & Maria Šafářová. 2003. On polar questions. In Robert B. Young & Yuping Zhou (eds.), Semantics and Linguistic Theory (SALT) 13, 292–309. Cornell University, Ithaca: CLC Publications. 260 Chapter 9 Psycholinguistic processing tasks and the study of question bias E Jamieson University of York Vinicius Macuch Silva Goethe University Frankfurt The study of biased questions is situated in a rich semantic tradition, based primarily on introspective cross-linguistic work. In this chapter we explore how psycholinguistic processing tasks can complement this existing work and help hone our understanding of the factors that influence question production, and the meanings these questions can carry, in this complex semantic-pragmatic domain. 1 Introduction In recent years, a growing body of literature in formal semantics has sought to understand “non-canonical questions” (Dayal 2017), and particularly “biased questions” – interrogatives in which the questioner has some sort of expectation as to what the answer to the question might be.1The so-called “bias” may originate in an expectation based on, for example, epistemic or deontic information, such as when the questioner is knowledgeable about the state of affairs they are 1Questions with biases have also been of interest to researchers in the interactional linguistic tradition. This research focuses on how questions can reflect stances that users may choose to take in communicative interaction, in particular regarding the likelihood of the states of affairs that the questions refer to (e.g. Heritage & Raymond 2021, Raymond & Heritage 2021), and how responses to these questions may be organised (e.g. Heritage 2012, Lee 2015). Here we focus on what has been established in formal semantics, but acknowledge the links between the work conducted in both fields. E Jamieson & Vinicius Macuch Silva. 2025. Psycholinguistic processing tasks and the study of question bias. In Tue Trinh, Anton Benz, Daniel Goodhue, Kazuko Yatsushiro & Manfred Krifka (eds.), Biased questions: Experimental results and theoretical modelling, 261–291. Berlin: Language Science Press. DOI: 10.5281/zenodo.17158192 E Jamieson & Vinicius Macuch Silva inquiring about or when they use the question as a form of directive. In this paper, we focus exclusively on epistemic biases. The main body of this work on non-canonical questions has relied on introspective judgments to delineate the pragmatic contexts in which these questions can be used, and to postulate how, semantically, the bias is incorporated into the question. More recently, experimental work has begun in this domain, primarily focused on production tasks (Domaneschi et al. 2017). This work has begun to hone the wide array of epistemic and evidential biases that have been claimed to be relevant for biased question production, but a number of questions, including about how those interrogative forms are interpreted, remain. In this paper, we illustrate how psycholinguistic processing experiments can complement the formal semantic research that has characterised biased question research to date. In Section 2, we give a brief overview of the trajectory of the methodologies used to study question bias, from introspective observations to more recent experimental work. In Section 3, we lay out some questions that can be addressed by approaching the topic of question bias through processing tasks, and what can be gained from these kinds of studies. With particular reference to the studies presented in Macuch Silva & Jamieson (2025) and Tian et al. (2021), we show how these questions have begun to be explored. Finally, we discuss some of the challenges of using processing methodologies to investigate a complex phenomenon like question bias, which spans the interfaces of semantics and pragmatics, as well as (morpho)syntax and prosody. We discuss how to design tasks to deal with these challenges, and thus to allow psycholinguistic research to contribute to this growing field. 2 Studying question bias: from introspection to production 2.1 Theoretical developments Negation is a polyfunctional linguistic device which can be used pragmatically for purposes as varied as mitigating the meaning of an assessment (e.g. Fraenkel & Schul 2008, Giora et al. 2005, Krifka 2007) or modulating the acceptability of a request (e.g. Koike 1994). The rich formal semantic literature on question bias begins with an observation about the role of negation in polar questions, more specifically Ladd’s (1981) observation that, in English, a question with a negative surface form can in fact indicate that the questioner has a belief of the positive proposition. This can be seen in (1–2), adapted from Ladd (1981), in which 262 9 Psycholinguistic processing tasks and the study of question bias S uses a negated form of the question to express their belief of the existence of a vegetarian restaurant. (1) S is visiting A, and believes there is a good vegetarian restaurant in town. A: You guys must be starving. Shall we get something to eat? S: Yeah, isn’t there a vegetarian restaurant around here? (2) S is visiting A, and believes there is a good vegetarian restaurant in town. A: We should get something to eat, although I don’t know if there’s anything you’ll like around here. S: Oh, isn’t there a vegetarian restaurant around here? Ladd proposes that there is an “ambiguity” in the usage of the negated question, whereby in situations like (1) it is used with the aim of confirming the questioner’s existing belief of 𝑝, while in situations like (2) it is used with the aim of confirming the evidence the questioner has just received for ¬𝑝in the face of their existing belief. Ladd’s initial observations about (American) English have spawned considerable further research reasoning about the pragmatic conditions that trigger the use of those particular type(s) of biased question, as well as if, and if so, how, the meaning of the question is semantically encoded into particular linguistic forms (Büring & Gunlogson 2000, Van Rooy & Šafářová 2003, Romero & Han 2004, Asher & Reese 2007, AnderBois 2011, Northrup 2014). There has also been interest in how English biased questions are represented in terms of different syntactic forms. Early research focused on questions with syntactically “high” negation - n’t, like those in (1–2), which, despite any potential differences in meaning, have the same form. “Low” negation not (as in (3)) was not assumed to share these characteristics. Instead, low negation questions seemed to be used to genuinely question the truth of the negative proposition ¬𝑝from current contextual evidence, with no involvement of any prior beliefs.2 (3) Is there not a vegetarian restaurant around here? The proposed distinction in the meanings of high and low negation questions has furthermore led to some syntactic proposals that these different negative 2However, some later work incorporated “low negation” questions as having similar bias imports to “high negation” questions (Van Rooy & Šafářová 2003, Asher & Reese 2007). Domaneschi et al. (2017) propose that this may be the case in some contexts in German, but argue that it is not the case for English. 263 E Jamieson & Vinicius Macuch Silva markers -n’t and not occupy different positions in the syntax (Cormack & Smith 2012, Krifka 2015). While much of the work on biased questions has focused on English, further work has indicated that biasing an interrogative, and having different forms to represent this, is also possible in other languages, with accounts for German (Büring & Gunlogson 2000), Japanese (Sudo 2013) and Hungarian (Gyuris 2017) contributing to Gärtner & Gyuris’s (2017) attempt to delimit the possible bias profile space for questions cross-linguistically. This research has shown that biased questions may be marked syntactically, morphologically or prosodically, sometimes in addition to the use of a negation marker. For the remainder of this article, we use the term question form to refer to any linguistic form or realisation which is associated with a bias in the literature, regardless of whether the form itself consists of a morphosyntactic construction, a particle or morpheme, or an intonation pattern. The rich introspective research on biased questions has led to various categorisations of the most important contextual phenomena that might license the use of these interrogatives, as well as the specific question forms associated with particular contexts. However, this has led to conflicting accounts regarding the role of the different factors, as well as the relationship between the semantic/ pragmatic factors and the form of the interrogative (e.g. whether the negation is low or high in English, or whether particular particles are used in Japanese, Sudo 2013). The literature has generally agreed that original speaker bias and contextual evidence are relevant factors for the licensing of these questions, but the ways in which those two types of influence have been understood and grouped together has varied considerably (see Romero 2020 for an overview). More recently, scholars have started gathering empirical data through experimental methodologies in order to tease apart the various proposals on the ways these biases and the linguistic forms that may express them interact. In the next section, we review the results of some of these studies. Though we acknowledge the need to explore this topic cross-linguistically, throughout this paper, we focus on the results of research on English biased questions, as this is where the most work has been done into understanding the interaction between the question form and the pragmatic factors that may license it. 2.2 Experimental tasks A prolific research tradition in phonetics has investigated the realisation of polar questions across different languages, particularly in Romance (e.g. Crocco 2006, Escandell-Vidal 2017, Giordano 2006, Grice & Savino 1997, 2003, 2004, Henriksen 264 9 Psycholinguistic processing tasks and the study of question bias et al. 2016). Plenty of the work in this tradition has not only been experimental in nature but it has also explicitly addressed question bias or speaker certainty and epistemic stance, both in the perception (Armstrong & Prieto 2015, Vanrell et al. 2017, Orrico et al. 2019, Prieto & Borràs-Comes 2018) and production (Armstrong 2017, Vanrell et al. 2014) of polar questions. However, the (morpho)syntactic question forms that have driven formal semantic research into question bias have not been subject to the same quantity of experimental investigation. For morphosyntactic question forms, the first experimental work to be conducted consisted of a series of acceptability judgment tasks investigating which interrogative structures participants find acceptable, given particular combinations of original speaker bias and contextual evidence. For example, Sailor (2013) finds no evidence for the sort of ambiguity of meaning for high negation in American English claimed by Ladd (1981). However, Sailor’s study was small and focused on the acceptability of negative polarity items such as either in a high negation or a low negation construction, rather than the licensing of the interrogative structure itself. While the acceptability of negative polarity items can be extrapolated to the “biased” meaning of the question, the study does not directly address the production (or interpretation) of the question forms in context. Roelofsen et al. (2013) conduct an online acceptability judgment task investigating how natural different polar questions are when presented in various belief and bias contexts in English. Questions were shown as part of short cartoon contexts, with participants rating the items on a scale from 1 (natural) to 7 (unnatural). Roelofsen et al. (2013) found that high negation questions were acceptable in contexts with a preexisting speaker belief for 𝑝, and negative or no contextual evidence against 𝑝– as hypothesised. Perhaps surprisingly, they also found that low negation questions did not behave as differently as predicted to high negation questions, being preferred in contexts with prior speaker belief and negative or no contextual evidence. While the study in Roelofsen et al. (2013) is broader in scope than Sailor (2013) and with a larger participant pool, there were some issues with the contextual evidence and speaker bias presented in their examples. In the negative cases, the evidence relied on a conversational implicature (for example, “Kate got a dog” does not entail that she did not also get a cat), which led to unexpected results (Roelofsen et al. 2013: 460). Secondly, as identified by Domaneschi et al. (2017: 6), the conditions with absence of speaker bias were too neutral. For example, Rose telling Jennifer that “Kate got a dog” out of the blue (without establishing, for example, that Kate prefers cats) renders any polar question somewhat unnatural. Finally, we would point out that, for example, the belief that arises from a friend saying “I am going to get a dog” is different to “I have got a dog”. It is 265 E Jamieson & Vinicius Macuch Silva therefore questionable how strong the beliefs were in the various contexts, and to what extent this may have affected participants’ rating of items. Roelofsen et al. (2013: 466) find a “scale of speaker belief” in their results, and claim that “the neutral and negative [speaker bias] contexts in the experiment suggest the absence of positive speaker belief, but strictly speaking, they do not exclude it”. While this will always, to some extent, be the case, the speaker beliefs – or lack thereof – could have been strengthened. Roelofsen et al.’s (2013) study presents an interesting potential counterpoint to the theoretical literature positing a true semantic difference between high and low negation questions. However, their materials included possible confounds which may have impacted on their results. We thus turn to Domaneschi et al. (2017), which builds on Roelofsen et al. (2013). Domaneschi et al. (2017) conduct lab-based production tasks in both English and German in order to “resolve” the conflicting claims in the literature as to which question forms (in these cases, low vs. high negation) are preferred in which contexts. Here, we discuss results from their English experiment only. In their study, Domaneschi et al. (2017) asked participants to read through a short context that set up a prior belief for the participant, and then presented them with some contextual evidence that either supported, challenged, or was neutral in respect of that belief. Participants then chose one of a list of possible questions to ask, and produced this question into a microphone so that information about prosodic realisations could be recorded. The forms available to participants included positive polar questions (Do you ...), high and low negative polar questions (Didn’t you ... /Did you not ...), as well as positive polar questions prefaced by a surprise marker (Really? Did you ...). Domaneschi et al. (2017) find that participants have strong preferences as to which syntactic form of a question to produce given any one combination of prior belief and contextual evidence. Indeed, while low negation questions (LoNQ) are produced at a rate of 59% in contexts with a neutral bias and negative contextual evidence, high negation questions (HiNQ) are preferred in situations where there is a prior bias for the proposition expressed in the question, being produced at similar rates both in contexts with neutral evidence for the proposition (65%) and in contexts with evidence against the proposition (67%). The results for questions with negation are summarised in Table 1. These results help us to understand what epistemic and evidential profile biased questions can have in English (and German). This study has since been replicated and extended upon by Maro et al. (2021), who find the same factors influencing production choice of biased questions in Italian. However, these studies only begin to examine the complexities of licensing (andunderstanding) question 266 9 Psycholinguistic processing tasks and the study of question bias Table 1: Attested pragmatic profile of polar questions with negation in English, from Domaneschi et al. (2017) speaker bias contextual evidence 𝑝neutral ¬𝑝 𝑝 neutral HiNQ ¬𝑝 HiNQ LoNQ bias. For example, Maro et al. (2021) note possible additional roles for the tense of the verb in the auxiliary construction. Moreover, although Domaneschi et al. (2017) show that there are clear preferences for particular forms to be produced in each context, as summarised in Table 1, the distribution of choices for the remaining forms is not uniform, with certain “non-preferred” forms being produced above chance – indicating that perhaps these distinctions are not so clear cut. For example, in situations constructed so that participants had a prior bias towards the positive proposition 𝑝, and so that the contextual evidence was either neutral or in favour of ¬𝑝, high negation questions were preferred. However, with a prior bias for 𝑝, when the contextual evidence was neutral, the second most preferred form (around 20% of the data) was a positive polar question, while when the contextual evidence was in favour of ¬𝑝, around 25% of the data was a low negation question. So, although there seem to be clear preferences when people are asked to choose between different polar question forms, it is likely that in freer production the data would be considerably noisier, if including (polar) questions as the next relevant discourse move. Furthermore, research by Jamieson (2018) indicates that the likelihood of a high negation question being produced in a given context appears to be stronger in certain contexts than others, despite the same basic bias profiles. Participants in Jamieson’s study were presented with 20 short contexts in a conversation with a friend. These contexts were designed to include a prior speaker belief and either negative or neutral contextual evidence. There were also 30 filler examples. Participants then chose between either a high negation question or a tag question to produce in response to these contexts.3All participants saw the same contexts, but the order was randomised. 3Tag questions are argued to be licensed in the same sort of belief and bias contexts as biased matrix questions (Ladd 1981, Reese & Asher 2006, Malamud & Stephenson 2015, Krifka 2015). 267 E Jamieson & Vinicius Macuch Silva Both of the contexts in (4) and (5) contain a prior belief, and the evidential context is neutral. We might, therefore, expect that high negation questions would be produced at around the same rate in both contexts. However, around 48% of participants chose to produce a high negation question in context (4), while only around 10% of the same participants did so in (5). (4) We are talking about a friend of ours who moved away. You are pretty sure she went to Berlin. You say: ⟶Didn’t she move to Berlin? ⟶She moved to Berlin, didn’t she? (5) A friend has said she is going to pick up her new pet. You are pretty sure it is a dog. You say: ⟶Isn’t she getting a dog? ⟶She’s getting a dog, isn’t she? This suggests that the choice to produce a high negation question may be more probabilistic than deterministic in any given context, even if the speaker has a prior belief of 𝑝. Any number of factors or normative expectations may be influencing the participant’s choice of question at any one given time, including, potentially, other non-epistemic biases. Finally, while data on language production and perception in both the phonetic and semantic traditions provide a strong indicator of overall usage distributions for particular question forms, given biases and contextual information – and thus can tap into intended meaning – language comprehension data may provide a very different picture. By looking at how participants process these questions with biases, we can gain a greater understanding of how the questions are interpreted, and thus speak to ongoing theoretical discussions of how these biases are expressed by the interrogative constructions at hand. In the next section, we therefore explore how to design a processing study to investigate the real-time interpretation of biased questions. We firstly give an overview of related domains where (predictive) processing tasks have been used to explore issues of interpretation, and what sorts of issues surrounding the interpretation of question bias in particular can be addressed using these types of tasks. We then review two studies which have started to explore those issues: Tian et al. (2021) and Macuch Silva & Jamieson (2025). We do not present full results from these studies, but show how the methodologies can enhance the existing literature by investigating the question of whether the information structure of the question can allow participants to predict upcoming material, given a particular context. 268 9 Psycholinguistic processing tasks and the study of question bias Finally, we address the methodological choices that need to be made when designing a processing study to help understand when and how in interpretation biases are set for any given interrogative structure. We continue to focus on English, and thus on how variation in (morpho-) syntactic question forms may allow participants to predict upcoming material, but note that the topics and methodologies discussed in Section 3 would extend to the other information structural cues seen in biased questions in other languages (see Section 2.1). 3 Using processing tasks to explore question bias 3.1 What can processing studies bring to our general understanding of question bias? Language production studies like those discussed in Section 2.2 are a vital part of linguistic research, allowing us to shed a light on how language users plan and realise different sorts of linguistic elements, from individual phonemes to entire utterances. However, despite being deeply interconnected in conversation, the processes of producing and comprehending language are usually studied separately from one another, mostly in order to maximise experimental control and precision of measurement. For instance, given that language users may want to achieve various communicative goals with the bits of language they produce, focusing on how they choose to structure their linguistic signals given goals and (linguistic) contexts which are manipulated experimentally can shed light on the processes and motivations behind specific realisations. Conversely, focusing on how language users parse different types of linguistic signals can shed light on the inferences they draw when processing utterances in real time. Existent models of how sentences or otherwise non-sentential utterances are built up and how meaning is imparted in production therefore only tells us so much about how understanding is achieved in comprehension. In psycholinguistics, comprehension research sets about addressing the question of how language is understood based on the signal that is parsed – possibly including information from non-linguistic sources as well as multiple signaling modalities. A key tenet of modern psycholinguistic research is that language processing is incremental; that is, comprehenders process information as they parse it, one chunk at a time (Altmann & Steedman 1988). While this means that to some extent processing happens on the fly for any individual lexical item, parsing a new word or chunk can both update previous beliefs about what the possible meaning of the signal is (Levy et al. 2009), and, importantly, can help predict 269 E Jamieson & Vinicius Macuch Silva construction for either high or low negation questions in either belief and bias context. However, we do find reliable evidence for a facilitatory effect at the noun (e.g. “cat”) in high negation questions which follow a discourse which has been set up to include a prior belief. Interestingly, this result seems to map onto the descriptive results reported by Tian et al. (2021), who find suggestive evidence that comprehenders have a preference to direct their gaze to images representing 𝑝when hearing the noun in a high negation question. All in all, these results indicate that comprehenders track epistemic information during the processing of negated polar questions, and that they might be sensitive to how that information interacts with the syntactic form of the question, such that they seem to expect specific lexical information downstream in the sentence when processing high negation questions. Ultimately, this might mean that there is no specific semantic meaning associated with the negation marker in a high negation question in English, but rather that the meaning of the auxiliary construction as a whole is inferred pragmatically in context. However, this is only the tip of the iceberg regarding the study of processing biased questions. Firstly, Macuch Silva & Jamieson (2025) only investigate two cells from Table 1; how would low negation questions be processed in contexts with a prior belief but no contradictory evidence, for example? Furthermore, it may be that word reading times are not a fine-grained enough measure of processing for this subtle interface phenomenon. Indeed, at the current level of analysis, the pragmatic expectations of interest are categorical in nature – we expect there to be a preference for a particular question form, but not how much of a preference – which realistically means these expectations may not map well onto real-time processing demands and their acccompanying behavioural correlates, unless there are strong enough pressures in place such that the relevant biases actually translate into measureable expectations. In summary, the studies from Tian et al. (2021) and Macuch Silva & Jamieson (2025) illustrate how we can use processing tasks to help address the major topics of concern in the study of question bias, but there is much more that can be done. In the final section, we therefore reflect on some of the design decisions of these studies and overview how to potentially design new studies to help understand when and how in processing biases are set for any given interrogative structure. 3.3 Designing a processing study to investigate question bias 3.3.1 The task In order to study “online” interpretation processes, it is important to use tasks which measure participants’ reactions and understanding on a moment-by276 9 Psycholinguistic processing tasks and the study of question bias moment basis, either in addition to or possibly without explicitly asking them to make “offline” decisions or judgments – in which various extralinguistic or metalinguistic pressures may affect the outcome of the task. There are a variety of experimental tasks that can be used to investigate online processing, each with their own perks and pitfalls. The two studies we discussed in Section 3.2 have employed two very different types of tasks: a self-paced reading task and a visual world eye tracking task. Here, we outline how to weigh up design choices for these two methodologies in particular, though a myriad of other psycholinguistic methods such as mouse tracking or ERP studies could be used to investigate question bias. 3.3.1.1 Self-paced reading Self-paced reading tasks (Just et al. 1982) allow for relatively cheap studies of psycholinguistic processing based on common computer equipment, making them particularly useful for gathering data both in the laboratory and online. In a standard “moving window” self-paced reading task, participants press a button to read a sentence one word or chunk at a time, the measurement of interest being the time spent between button presses. In general, self-paced reading tasks have been found to be comparable to other, more complex, cognitive measures of psycholinguistic processing (Marsden et al. 2018), which speaks to their usefulness as a method. A number of potential pitfalls arise, however, with using a self-paced reading task to understand a discourse phenomenon such as question bias. Firstly, given that question bias is strongly situated in interactional contexts (e.g. Hennoste et al. 2017, Heritage & Raymond 2021), reading may not be particularly informative of how people actually process biased questions naturalistically. Of course, experimental psycholinguistics in general is not situated in rich interactive settings, and so this may not be a concern specific to this methodology, but the issue should nevertheless be taken into account. Constructing scenarios or (reported) dialogues which set up some form of a discourse context for the test sentence can go a certain way to solving this potential concern, at least under the assumption that one can extrapolate from reading processes to naturalistic comprehension in the first place. A related concern when using self-paced reading, or any reading method for that matter, is that it is not possible to know how participants interpret the prosody of the read sentence. The licensing and interpretation of an interrogative can often be dependent on its prosodic structure (Pierrehumbert & Hirschberg 1990, Grice & Savino 1997, Vanrell et al. 2012, Hedberg et al. 2017), and in polar 277 E Jamieson & Vinicius Macuch Silva questions prosody can be an important factor whether or not intonation itself is the linguistic form said to mark any bias, as discussed earlier in Section 2.2. Regarding negated biased questions where the bias is claimed to be marked primarily morphosyntactically, Domaneschi et al. (2017) find that while participants in their German production study consistently produce high negation questions with a final rise, participants in their English study produce high negation questions around 50% of the time with a final rise, and 50% of the time with a final fall.5Using a self-paced reading study, therefore, does not allow us to understand how prosody might impact the interpretation of a biased question, which could well be processed differently by different participants, depending on which contour they project on reading. The obvious solution to the above issue is to present the stimuli auditorily to participants, thus making it possible to take account of effects of prosody on biased question interpretation. For that, one could retain the same general setup of a self-paced reading study but instead use a self-paced listening task (Ferreira, Henderson, et al. 1996), which follows the same format but with auditory rather than written stimuli. While self-paced listening is a rather underexplored psycholinguistic method, especially compared to self-paced reading, it has been shown to replicate the results of self-paced reading tasks (Ferreira, Anes, et al. 1996, Papadopoulou et al. 2013), and might thus be ideal for cases where written stimuli may not be so appropriate and where using other auditory delivery methods may not be feasible. One potential limitation of the method is that, compared to self-paced reading, self-paced listening might exacerbate even more the unnaturalness of the comprehension process, as it is very much unlike normal listening. Secondly, recording the time it takes for a participant to press a button and read (or listen to) a particular segment in a test sentence may simply not be a fine-grained enough measure of online processing to allow us to determine how comprehenders process a complex semantic-pragmatic construction like a biased question, as discussed in the previous section. For instance, in Macuch Silva & Jamieson (2025), we find small differences in reading times at the noun when comparing high negation questions in both bias conditions. It is likely that the effect size of interest is indeed very small, and so the granularity of a method such as self-paced reading may not allow capturing such subtle effects, which might otherwise be measureable using more fine-grained methods. Next, we discuss some of the trade-offs of using one such higher-resolution method, namely eye tracking. 5Domaneschi et al. (2017) propose that this intonational difference corresponds to the ambiguity in high negation described by Ladd (1981) and shown in examples (1–2). 278 9 Psycholinguistic processing tasks and the study of question bias 3.3.1.2 Eye tracking While self-paced reading still relies on the participant to actively engage in some element of the task beyond simply processing (i.e. pressing a button to move on to the next reading segment), eye tracking studies (Rayner 1978) allow for an even more natural reading process – given that eye movements are an essential part of processing (written) language. In eye tracking studies, eye movements and fixations are recorded as a participant reads through a sentence. Additionally, any “regressions” that the participant makes, such as re-reading a segment, can also be recorded. Where self-paced reading provides a singular measure for each word or chunk in a sentence, eye tracking can thus provide an extremely rich data set containing several measures for even one test item. Eye tracking can rely on the same sort of measurement principles and assumptions as self-paced reading: namely, that when the eye is fixated on a particular target, that target is being considered, and that longer durations spent on any one target indicate that it is proving to be more difficult to process. However, unlike self-paced reading, eye tracking allows for more naturalistic reading processes, such as backtracking and re-reading. In the study of question bias, therefore, it may be possible using an eye tracking experiment to see, when a comprehender encounters the main contentful verb in the question, whether they then regress to the negation marker which they may have previously processed, indicating difficulty in integrating semantic information from verb. However, classical eye tracking studies involving the reading of sentences have the same issues as selfpaced reading tasks in the study of biased questions: that questions are inherently interactional and thus not so often read, and a participant’s interpretation of the sentence prosody cannot be controlled. Visual world paradigms (Huettig et al. 2011) allow researchers to track participants’ eye movements towards particular (referential) targets, as they process auditory stimuli. This method was used to explore processing of different types of question forms in Tian et al. (2021), discussed in Section 3. However, it does not trivially lend itself to investigating discourse situations which involve expectations that might not be referential in nature.6In order to fully understand how biased questions are processed, the visual world paradigm would need to be combined with an effective way to present contextual evidence for the question to the participant. This could perhaps be done through video or audio story telling prior to the presentation of the visual world, or it could involve designing scenar6For research employing eye tracking while reading to investigate how spatial information can inform abstract language processing, see Guerra & Knoeferle (2014, 2017). 279 E Jamieson & Vinicius Macuch Silva ios with epistemic expectations which can be encapsulated in concrete objects or actions as they are traditionally represented in visual world. Both classic eye tracking and visual world paradigms can present researchers with an array of useful data that could help address major questions in the study of question bias, as outlined above. However, crafting useful designs where beliefs and biases can be straightforwardly operationalized might be a challenge for traditional setups. In the next section, we discuss some ways in which materials for such experiments can be designed. 3.3.2 Material design Designing materials for any study of question bias is in itself a complex task, regardless of whether you are conducting a processing study or not. In this section, we focus on particular decisions that are important for processing studies – again looking in particular at self-paced reading and eye tracking – only touching on general decisions that must be made in any experimental study of question bias. 3.3.2.1 Setting up the contexts In order to capture potential interactions between speaker beliefs and contextual evidence in the processing of a biased question, the experimental stimulus must be combined with a discourse context which sets up beliefs and biases (as in e.g. Domaneschi et al. 2017 and Macuch Silva & Jamieson 2025). Ideally, these contexts should be as realistic and naturalistic as possible given the constraints of a controlled laboratory experiment, and importantly they must actually include the various biases that are being set up for the context, at an appropriate “strength” – for example, the biases should not be so strong that the very act of asking a question becomes irrelevant. In Macuch Silva & Jamieson (2025), we conducted an independent task prior to the main data collection in order to collect norms of the experimental discourse contexts, to understand the baseline likelihood of expecting certain epistemic beliefs in those contexts. Participants in the norming study were asked to read a short scenario corresponding to the background contexts that would be used to set up the questions. Participants then answered a question pertaining to the scenario, which paraphrase the relevant negated question that would appear in the main study. An example can be seen in (11). (11) Scenario: Our friend is getting a new pet. You heard from her sister that it would be a cat. 280 9 Psycholinguistic processing tasks and the study of question bias Q: Do you think the person in this scenario is getting a cat? Yes/No/Don’t know For contexts in which a prior speaker belief was to be established, we expected scenarios to show a bias for “yes” responses, while for scenarios which were to establish a belief, a bias against “yes” was expected.7By carrying out a norming study in advance, we can be more certain of ensuring that the beliefs and biases established in the scenarios map to the types of situations that have been detailed previously in the literature, and likewise that any expectations related to particular forms or contexts can be at least coarsely quantified. Yet, as discussed in Lemke et al. (2021), collecting study-specific norms (which are not necessarily generalisable to other data sets) might not be the best way of gauging variability or uncertainty in the expectations of interest; instead, relying on corpus estimates or norms which are open and freely available might be a more viable alternative. 3.3.2.2 Test stimuli It is important that the test stimuli used in any processing study of biased questions meet the standard expectations of any well-conducted experimental paradigm: for example, in terms of appropriate numbers of stimuli, filler items and randomisation or counterbalancing procedures. However, there are other factors that must also be taken into account in a processing study. For instance, in order to take valid measurements of the relevant moments in a processing task, each stimulus must follow the same basic frame so that the critical form is in a comparable position or time-window in every sentence. For example, in a straightforward declarative sentence, all items may follow the structure as in (12). (12) The det boy noun walks verb to prep the det shop. noun If the verb in the example in (12) is the critical form in the stimulus, to which the relevant facilitation or processing difficulties might be attributed, measurements can then be taken directly at the verb as well as at the words immediately following it in the case of reading, or a reasonable time-window around the verb can be normalised across different sentences in the case of unconstrained listening. 7Responses could be either in favour of no or don’t know in the no belief conditions. 281 [Document text truncated for crawler view.]