Varieties of Polar Question Bias: Lessons from Vietnamese

Abstract

This paper describes the use conditions of different polar question constructions in Vietnamese and their consequences for the description and analysis of polar question bias. We argue that the behavior of questions with the final particle à highlights the utility and relevance of the theoretical notion of projected bias for describing polar question bias, distinct from and in addition to original bias based on the speaker’s prior beliefs and contextual bias based on evidence available in the context. We also argue that some but not all bias requirements may be described as due to pragmatic competition between different question forms.

1. Introduction

Languages frequently have more than one morphosyntactic strategy for polar question formation. In particular, many polar question constructions are limited in their use to situations where the propositions p and $\neg p$ do not have equal standing, hence reflecting some form of bias between these two alternatives. Question bias is not simply a matter of privileging one possibility in $\{p,\neg p\}$ over the other; for instance, particular constructions may make reference to the speaker’s prior beliefs (Ladd, 1981, et seq) or to evidence available from the conversational context (Büring & Gunlogson, 2000, et seq). See Bill and Koev (2023) and Romero (2024) for two recent overviews, focused on the facts of English polar question forms and the history of their analysis.

In this short paper, we contribute to this literature by describing the bias profiles of several polar question constructions in Vietnamese. We first introduce the morphosyntax of polar questions in Vietnamese in Section 2 and then present the bias profiles of several such constructions in Section 3. We then discuss two consequences of our study that have broader relevance for the description and analysis of question bias:. First, in Section 4, we discuss the use conditions of à particle questions, which highlights the importance of Bill and Koev’s (2023) notion of projected bias, which reflects the speaker’s current expectation for how the question will be resolved. Second, in Section 5 and Section 6, we argue that some but not all of the bias requirements of Vietnamese polar question constructions may be derived from pragmatic competition. We then conclude in Section 7.

2. Polar Question Constructions in Vietnamese

We begin by briefly introducing the range of Vietnamese polar question constructions that we will investigate here. Note that we here take a broad, functionally motivated definition of polar questions as those constructions which conventionally elicit an addressee commitment towards the prejacent proposition p or to its negation $\neg p$, without reference to syntactic clause typing distinctions.

Broadly speaking, polar question constructions in Vietnamese can be placed in two categories, motivated both by their syntax and pragmatics. What we will call standard polar questions involve a negative marker in a sentence-final position, in contrast to the preverbal position of standard negation; see (1). The negative marker (không or chưa) must agree with the aspect of the clause, which itself cannot include negation. An affirmative auxiliary có frequently precedes the verb phrase in không questions.1

(1)	Standard polar questions:	(based on Trinh, 2005, p. 30)
	Nó    có   đọc  sách  không? he/she aff read book neg ‘Does he/she read books?’	b. Nó    đã   đọc  sách  chưa? he/she pfv read book neg.pfv ‘Has he/she read books (yet)?’

The other class of polar question constructions all involve the addition of an invariant and aspectually unrestricted sentence-final expression. These sentence-final expressions can be one of a number of monosyllabic sentence-final particles (2), which we describe further below, or the expression (có) phải không (3), which can also be used as an independent polar question meaning ‘Is it right?’. We refer to the former as particle polar questions and, following prior work (e.g., Nguyen, 1997, p. 240; Trinh, 2005; Duffield, 2013), refer to the latter as tag questions. Unlike in standard polar questions, the main clause in particle polar questions and tag questions may include negation and has no restriction on its aspectual specification.

(2)	Particle polar questions:
	Nó    đọc   sách { á / à / hả / chăng / ư / sao / nhỉ / … }? he/she read book   á   à   hả    … ≈ ‘Does he/she read books?’
(3)	Tag question:
	Nó    đọc   sách  phải     không? he/she read book correct neg ≈ ‘He/she reads books, does not he/she?’

As indicated in (2) above, there are many polar question particles, some of which are associated with particular dialects.2 In our discussion below, we will concentrate on the uses of á, à, and hả, bolded in (2), as well as the tag question with phải không as in (3). These are the forms for which the speakers that we consulted (mostly speakers of Northern varieties) and the second author (from Hanoi, and hence, also Northern) could most comfortably give clear judgments of felicity or infelicity in context, and the discussion of their use conditions will suffice to motivate our theoretical consequences. Particle polar questions and tag questions are all associated with some form of bias, unlike the standard polar question, which is in some sense neutral, as we describe in detail below.

Finally, we note that standard polar questions are the only one of these polar question forms that can be embedded.3 See the contrast between examples (4a,b) below, as well as Phan (2024, pp. 169–170) for additional supporting data. We will return to this difference again in Section 5.

(4)	Only standard polar questions can be embedded:  (based on Trinh, 2005, p. 31)
	Tôi muốn biết     [ nó  có  đọc   sách  không ]. I     want   know   he/she aff read book neg ‘I want to  know whether he/she reads books.’ * Tôi muốn biết    [ nó        đọc   sách { á / à / hả / … / phải  không } ]. I     want   know   he/she read book   á   à   hả     correct neg

3. Varieties of Question Bias

We now discuss different notions of question bias and demonstrate their relevance for describing the use conditions of different polar question particles in Vietnamese. Prior work on question bias in English and other languages has established the need to distinguish between bias due to the speaker’s initial epistemic state versus bias due to evidence from the conversational context. We follow Domanesci et al. (2017), Bill and Koev (2023) and Romero (2024) in referring to these as original (speaker) bias and contextual (evidence) bias, respectively. We introduce each in turn in Section 3.1 and Section 3.2, adopting the definitions provided by B&K, which are designed to facilitate their comparison and joint discussion. We then provide an interim summary in Section 3.3.

Bill and Koev (2023) additionally introduced a third notion, which they called projected bias. We will introduce this notion and argue for its relevance in Section 4 below.

3.1. Original Bias

The first notion of bias we introduce, original bias, reflects the speaker’s prior beliefs or knowledge regarding the status of the proposition p. The relevance of this type of bias has been established in works such as Ladd (1981).

(5)	Original bias	(definition from B&K, p. 4)
	A question Q uttered in a context c conveys original bias for a possible answer p iff Q indicates that the prior epistemic state of the speaker in c supports p over $\neg p$.

We can systematically test for sensitivity to original bias by explicitly manipulating the presence or absence of reasons for the speaker’s prior belief in minimally contrasting situations, as in S1–S3 below. For ease of presentation, we present these situations here in English and underline crucial points of manipulation.

(S1)	Situation 1: No bias
	You are sitting in a windowless room with no information about current weather conditions. You call a co-worker in a room that has a window to check on the weather.
(S2)	Situation 2: Original bias for $\mathit{p}$
	You are sitting in a windowless room. A while ago, a friend told you on the phone that it was rainy outside. You call a co-worker in a room that has a window to check.
(S3)	Situation 3: Original bias for $\neg \mathit{p}$
	You are sitting in a windowless room. A while ago, a friend told you on the phone that it was sunny outside. You call a co-worker in a room that has a window to check.

The effect of original bias is reflected in the felicity and infelicity of different question forms, as in (6–7). As we see below, the standard polar question must be neutral with respect to original bias, whereas the à and hả particle questions and the phải không tag question require positive original bias.4 (We also introduce á particle questions below).

(6)	Standard question requires no bias:
	Trời (có) đang mưa không? it    aff prog  rain  neg ≈ ‘Is it raining?’	S1 ✓ S2 #  S3 #
(7)	Original bias for $\mathit{p}$ licenses à and hả particle questions and tag questions:
	Trời đang mưa {à / hả / phải    không }? it    prog rain   à   hả  correct neg ≈ ‘Is it raining?’	S1 #  S2 ✓ S3 #

3.2. Contextual Bias

A speaker might also be positively or negatively disposed towards a proposition based on shared evidence that is available in the present conversational context, as discussed in, e.g., Büring and Gunlogson (2000). We call this contextual bias:

(8)	Contextual bias	(definition from B&K, p. 4)
	A question Q uttered in a context c conveys contextual bias for a possible answer p iff Q indicates that some piece of evidence that has just become available in c supports p over $\neg p$.

Situation 4 below is modeled after contexts discussed in Gunlogson (2001, pp. 95–96), parallel to S1 above in providing no original bias but with a manipulation providing contextual bias for p. S4 licenses the use of à and hả particle questions, but not the standard question or tag question, as indicated in (9) below.5

(S4)	Situation 4: Contextual bias for $\mathit{p}$
	You are sitting in a windowless room. You see your co-worker enter the office wearing a wet raincoat.
(9)	Felicity judgments for question forms in S4, with contextual bias for $\mathit{p}$:
	Trời đang mưa { #không / ✓à / ✓hả / #phải  không }? it      prog  rain      neg           à      hả     correct neg ≈ ‘Is it raining?’

This difference in felicity between the à and hả particle questions and the phải không tag question, which all pattern together with respect to original bias (7), establishes the relevance of contextual bias in addition to original bias.6 Note that we will also show in Section 4 below that à and hả particle questions do not always pattern together.

For completeness, we demonstrate that none of the question forms considered here are licensed in a situation that provides negative contextual bias, in S5, just as with S3 above, which provides negative original bias. In such situations, speakers may form a polar question with a different prejacent instead. Note that a red, sweating face indicates a hot, sunny day, and therefore, no rain in the shared background of our speakers.

(S5)	Situation 5: Contextual bias for $\neg \mathit{p}$
	You are sitting in a windowless room. You see your co-worker enter the office with a red sweating face.

A conversational context might also support both original and contextual biases, in particular pointing to opposite conclusions. The last particle question we describe here, with final á, is specifically licensed in such situations, but not in situations providing only original or contextual bias alone:

(S6)	Situation 6: Original bias for $\neg \mathit{p}$ and contextual bias for $\mathit{p}$
	You are sitting in a windowless room. A while ago, a friend told you on the phone that it was sunny outside. You see your co-worker enter the office with a wet raincoat.
(10)	Á question requires both original bias for $\neg \mathit{p}$ and contextual bias for $\mathit{p}$:
	Trời đang mưa á? it      prog  rain  á ≈ ‘Is it raining?’	S1 #  S2 #  S3 #  S4 #  S5 #  S6 ✓

3.3. Interim Summary

Having introduced the notions of original and contextual bias, as well as methods for their contextual manipulation, we now expand on the data above by considering nine types of situations, which systematically cross positive, negative, and no bias of these two types.7 See Figure 1 below. We report the felicity of each polar question type discussed in this section in each of these nine situations in Figure 2 below. These figures give a visual representation of the bias profile associated with each polar question form, which we also summarize below each figure.

One generalization that these figures help make apparent is that no polar question construction with prejacent p is felicitous in a situation where there is only negative bias, whether original or contextual; see the right, bottom, and bottom right cells. Practically, speakers use the prejacent’s negation or yet another proposition to form polar questions in such situations. That is, if there is original or contextual bias against the proposition ‘it’s raining’, a speaker may form a polar question based on a prejacent such as ‘it’s sunny’ instead, or with the negation ‘it isn’t raining’ as the prejacent in the case of particle and tag questions. There is also a similar asymmetry in the use conditions of á, which specifically reflects contextual evidence for p when p is originally unexpected (bottom left cell)—which we may succinctly describe as expressing “surprise” towards p—but is infelicitous when original and contextual bias conflict in the opposite way (top right cell).

4. Projected Bias and À Questions

In addition to the notions of original and contextual bias, Bill and Koev (2023) introduced a third notion that they called projected bias, as defined in (11). In this section, we will discuss this third notion of question bias and argue for its relevance in the description of à particle questions.

(11)	Projected bias	(definition from B&K, p. 4)
	A question Q uttered in a context c conveys projected bias for a possible answer p iff Q indicates that the speaker in c expects answer p rather than $\neg p$.

As B&K discussed, projected bias has a somewhat different status from both original and contextual bias above, both in its status within the literature and as a conceptual notion. First, the idea of projected bias has appeared as a component of particular theoretical analyses of certain bias requirements (as in Krifka, 2015; Malamud & Stephenson, 2015; and Jeong, 2018), rather than as a descriptive notion itself. Second and relatedly, the notion of projected bias is different in being “forward-looking” (that is, “forward-looking as it is about the answer that is expected to be chosen by the addressee,” B&K p. 4), and in requiring reference to a mental model of the discourse, unlike both original and contextual bias, which may be described solely in terms of the speaker’s own epistemic state without reference to the discourse itself.

Concretely, in cases where there is only original or contextual bias, or where both biases are aligned, we may expect projected bias to follow these biases. However, in situations where original and contextual bias conflict (that is, where one suggests p and another suggests $\neg p$), the direction of projected bias is in principle open. Such situations are exactly where we might see subtle contrasts in the acceptability of constructions that are sensitive to projected bias.

We argue that the behavior of Vietnamese à particle questions highlights the value of this notion of projected bias, as they are best described as felicitous when there is projected bias for p. Recall that we have already seen that à particle questions are licensed by positive original bias but not with no bias (see S1 and S2 in (7)), or else by positive contextual bias (as in S4; see (9)). What will be particularly informative is contrasts such as in (12–13) below. Both contexts provide original bias for $\neg p$ and contextual bias for p, but they differ in the speaker’s understanding of the nature of this conflict and its resolution.

(12)	Surprised but willing to believe $\mathit{p}$:
	Minh is a classmate at school. He’s discussed his family before but did not mention any siblings, so you think he’s an only child. Your friend tells you that Minh went to the airport to pick up his (Minh’s) sister. You reply:
	Minh có      chị { ✓à / ✓á / ✓hả }?   Tớ cứ tưởng nó là    con một. Minh have sister     à      á       hả       I    thought   he cop only.child ‘Minh has a sister? I thought he’s an only child.’
(13)	Surprised and unwilling to believe $\mathit{p}$:
	You and Minh have been friends since grade school, so you have known that he’s an only child. But you have been out of town recently and had not heard that Minh’s parents have adopted a girl. Your friend tells you that Minh went to the airport to pick up his (Minh’s) sister. You reply:
	Minh có   chị { #à / ✓á / ✓hả }?   Không thể       nào! Minh have sister     à      á      hả       neg   possible prt ‘Minh has a sister? That’s impossible!’

The à question is felicitous in (12) because, based on the friend’s claim and their general reliability, the speaker is willing to believe that their friend must know something that they do not. The speaker reasons therefore that the addressee would resolve the question of $\{p,\neg p\}$ in the direction of p: that is, that there is projected bias for p, and hence, the à question form is licensed. In contrast, in (13), the speaker has a strong original belief that $\neg p$, and hence, does not expect p to be the answer of $\{p,\neg p\}$. There is then no projected bias for p and so the à question is infelicitous.

Empirically, our description here improves upon previous descriptions of the bias profiles of these polar question forms in Vietnamese. Trinh (2010) notes that à particle questions can be used to question a proposition that was presupposed by another speaker’s utterance but is not common ground. Both contexts in (12–13) satisfy this description, but the à question is not felicitous in (13) (Trinh does not discuss other polar question particles). In addition, some works have described à and hả as equivalent and simply associated with different dialects (see, e.g., Tran, 2009, p. 18), but we see in these examples that their behaviors diverge. The hả question is felicitous in both of (12–13), as it simply tracks the existence of original or contextual bias for p under our description.

The notion of projected bias is also relevant in situations without conflicting sources of bias, but instead reflecting a sensitivity to the predictive power of such information. Consider example (14) below. Here, there is original bias for p and no contextual bias, but there is still perceived uncertainty as to how the addressee will reply. In other words, there is original bias for p but no projected bias for p, and the use of an à particle question here is accordingly degraded.

(14)	Double-checking context:	(based on Rudin, 2018, p. 37; Rudin, 2022, p. 346)
	You and your friend made plans two days ago to get drinks tonight. You have not spoken about it since and worry that the friend may have forgotten.
	Chúng ta vẫn đi tối nay { ??à / ✓hả / ✓phải   không} 1pl  still go tonight      à       hả      correct  neg ‘We’re still on for tonight?’

We can summarize the bias profile of à questions as in Figure 3 below. Projected bias for p may be supported by original or contextual bias for p, which then leads the speaker to expect answer p as the resolution of Q. As the mere presence of original or contextual bias does not guarantee projected bias, we shade these cells in Figure 3 with diagonal lines. We can, however, be certain that, in cases with no source of bias for p or else only sources of bias for $\neg p$, there cannot be projected bias for p, and hence, à questions will always be infelicitous in these four bottom right cells.

The behavior of Vietnamese à questions thus attests to the possibility of a question’s use conditions making conventionalized reference to B&K’s notion of projected bias, in addition to specifications of original and contextual bias. Therefore, we advocate for the reification of this notion for the description of question bias.8

5. Bias in Embedded Questions

Next, we turn to the behavior of embedded questions, which leads to an important consequence for the theoretical analysis of question bias in Vietnamese. We have shown that particle questions with á, à, and hả as well as phải không tag questions all exhibit some form of bias, whereas matrix standard polar questions require an unbiased situation. Recall as well that only standard polar questions can be embedded (Duffield, 2013; Phan, 2024; Phan & Starke, 2022, among others), as illustrated in (4a,b) above. Interestingly, embedded standard polar questions are not limited to unbiased contexts.

We first illustrate the bias-insensitivity of embedded questions with respect to original bias. The example in (15) includes an embedded standard polar question with prejacent p = ‘it’s raining’, and is felicitous in all three variant contexts below: (a) without original bias, (b) with original bias for p, and (c) with original bias for $\neg p$. The contexts here are based on situations S1–S3 above, but with both the speaker and addressee in a windowless room. (We thank an anonymous reviewer for suggesting this example.)

(15)	Embedded standard question is insensitive to original bias:
	You are sitting in a windowless room.
	You have no information about current weather conditions.  (no bias) A while ago, a friend told you on the phone that it was rainy outside.                         (original bias for p) A while ago, a friend told you on the phone that it was sunny outside.                         (original bias for $\neg p$)
	You ask your officemate, in the same room as you: Nếu tớ muốn biết [trời có đang mưa không] thì tớ phải  hỏi  ai? if   I   want   know  it     aff  prog  rain  neg    then I   must ask who ‘If I want to know whether it’s raining, who should I ask?’    a ✓ b ✓ c ✓

Next, we illustrate this bias-insensitivity with respect to contextual bias, with example (16) below. This utterance embeds a standard polar question with prejacent p = ‘Sam got the first prize’ and is felicitous in all three contexts: (a) without contextual bias, (b) with contextual bias for p, or (c) with contextual bias for $\neg p$.

(16)	Embedded standard question is insensitive to contextual bias:
	You know Sam wanted to get the first prize in the contest.
	You don’t know how Sam did.                      (no bias) You see Sam smiling afterwards, so you think he got it.      (contextual bias for p) You see Sam frowning afterwards, so you think he didn’t get it.                              (contextual bias for $\neg p$)
	Nếu tớ muốn biết [Sâm có đạt giải  nhất không] thì tớ phải  hỏi ai? if   I want  know  Sam aff  get  prize  first  neg    then I   must ask who ‘If I want to know whether Sam got the first prize, who should I ask?’   a ✓ b ✓ c ✓

The striking difference between the bias sensitivity of matrix and embedded standard polar questions lends itself to an analysis in terms of pragmatic competition between different question forms. To facilitate this discussion, we present a summary of the bias profiles of all question forms discussed here, in Figure 4 below. Note that we illustrate the bias profiles for matrix standard questions (a) and embedded standard questions (f) separately.

Suppose that the basic semantics of standard questions has no conventionalized bias sensitivity. Matrix standard polar questions must be neutral because in contexts with original or contextual bias, a corresponding matrix particle or tag question can be used instead.9 However, for embedded questions, such pragmatic blocking does not occur, as the particle and tag questions are unavailable.10 This derives the difference in bias profiles for matrix versus embedded standard questions, as in Figure 4a vs. Figure 4f. Such competition-based accounts have been proposed for explaining the bias profiles of English rising declaratives (Rudin, 2018, 2022) and high negation questions (Goodhue, 2022).

6. The Limits of Pragmatic Competition

Although we have proposed that pragmatic competition can explain the difference in bias-sensitivity between matrix and embedded standard polar questions, we also caution that this type of explanation is unlikely to provide further insight into the bias profiles of other polar question forms.

Concretely, consider Situation 6, repeated in (17) below, where there is both original bias for $\neg p$ and contextual bias for p = ‘it’s raining’. Such contexts where new, surprising evidence can lead to belief revision are the canonical contexts for á particle questions, as we reported above in (10). However, as we indicate here in (17), other question forms with the particles à or hả, as well as the phải không tag question, are all also acceptable here. In other words, it must not be the case that the availability of any of these particle or tag question forms blocks the use of another in the same context.

(17)	Surprise context supports various question forms:
	You are sitting in a windowless room. A while ago, a friend told you on the phone that it was sunny outside. You see your co-worker enter the office with a wet raincoat.
	✓ Trời đang mưa { á / à / hả / phải  không }? it    prog  rain    á   à   hả   correct neg ≈ ‘Is it raining?’

This point is of particular note for hả questions, which roughly pattern with English rising declaratives in their use conditions. (We thank two anonymous reviewers for noting this similarity between hả and English rising declaratives.) Rudin (2018, 2022) has proposed that we can derive the bias profile of English rising declaratives pragmatically, by considering rising declarative utterances as being in competition with corresponding falling declaratives and standard polar questions. The overlapping contexts of use between hả questions and other particle polar questions makes it difficult, if not impossible, to import Rudin’s account to the analysis of Vietnamese hả questions. Although this does not directly challenge Rudin’s account for English itself, this comparative analysis demonstrates that it must be possible for a language to arrive at a polar question construction with the same or similar bias profile as English rising declaratives, without the proposed form of pragmatic competition.

We conclude then that the observed bias profiles of Vietnamese polar question constructions reflect a combination of conventionalized (i.e., learned, construction-specific) requirements on the use of individual particle and tag question constructions, as well as dynamically induced restrictions on the use of standard questions when unembedded.

7. Summary

Like many languages, Vietnamese has multiple morphosyntactic strategies for the expression of polar questions, with different pragmatic restrictions on their use. In this short paper, we have investigated the contexts of use for a number of such constructions, including standard (negation-final) questions, phải không tag questions, and á, à, hả particle questions, in Northern Vietnamese.

Previous work on biased polar questions in other languages have highlighted the importance of distinguishing between so-called original bias, based on the speaker’s prior beliefs, and contextual bias, based on shared evidence available in the conversational context (see overviews in Bill & Koev, 2023; and Romero, 2024). We have shown that the contexts of use for a number of polar question constructions in Vietnamese can effectively be described in these terms. We additionally argue that the bias profile of à particle questions is best described by adopting Bill and Koev’s (2023) notion of projected bias, reflecting the speaker’s current expectation for how the question will be resolved. As B&K discussed, this notion of projected bias has played a role in the conceptualization of question bias in some previous theoretical proposals, but has not been clearly reified as a necessary notion for the description of question bias.

Our study also contributes to the question of whether, and to what extent, the bias profiles of different question types can be derived in terms of pragmatic competition, as has been pursued in recent work on English questions (see, e.g., Goodhue, 2022; Rudin, 2018, 2022). We observe that embedded standard questions have no bias requirements, whereas matrix standard questions must be unbiased, and suggest that this difference can be captured via pragmatic competition, where competing, more specific forms are available for matrix questions but not in embedded contexts. However, the different bias profiles of individual polar question particles cannot be derived in the same way, thereby highlighting the need for multiple types of mechanisms and explanations for observed bias effects.