1 Introduction

This paper has two main goals: one descriptive and one analytical. The first is to describe the formation of polar questions (PolQs) and content/wh-questions (WhQs) in Sm’algyax (Maritime Tsimshianic, ISO 639-3: tsi; VSO), based on novel data from elicitation with fluent first language speakers. I show that in main clauses, both PolQs and WhQs in Sm’algyax bear two morphological reflexes associated with question formation, while embedded clauses in each case only exhibit one. This bipartite structure for main-clause questions is illustrated in the examples below. In (1a) we see a baseline declarative sentence, and in (1b) we see its PolQ counterpart, marked by the left-edge element ał and the polar-interrogative clitic =ii.

    1. (1)
    1. a.
    1. Nam dawłin.
    1. nah=m
    2. pfv =2sg.i
    1. dawł-n
    2. leave-2sg.iii
    1. ‘You left.’ Baseline
    1.  
    1. b.
    1. Ał nah dawłinii?
    1. ał
    2. q
    1. nah
    2. pfv
    1. dawł-n=ii
    2. leave-2sg.ii=int
    1. ‘Did you leave?’ PolQ

In the following examples, we see a baseline declarative (2a) and a WhQ (2b). The WhQ is characterized by a clause-initial wh-word and the wh-interrogative clitic =u.

    1. (2)
    1. a.
    1. Dawła aadit.
    1. dawł
    2. leave
    1. =a
    2. =cn
    1. aadit
    2. fisherman
    1. ‘The fisherman left.’ Baseline
    1.  
    1. b.
    1. Naayu dawłit?
    1. naa=u
    2. who=int
    1. dawł-ət___
    2. leave-sx
    1. ‘Who left?’ WhQ

Embedded questions, as we will see, are marked by the obligatory presence of a complementizer dzi in embedded PolQs in the same position as ał, and by wh-initiality in embedded WhQs. The interrogative clitics =ii and =u appear only in main-clause questions and are obligatorily absent in embedded questions, certain non-canonical questions, and other non-interrogative contexts. This basic distribution is given in Table 1.

Table 1: Reflexes of question formation.

Main Embedded
WhQs wh-initial + =u wh-initial
PolQs ał + =ii dzi

The analytical goal, then, is to situate these reflexes within a broader typology of interrogative particles. I argue for a two-layer syntax associated with question formation. Both main- and embedded-clause questions feature a clause-typing particle C[+Q] (Cheng 1991), which heads an interrogative CP. C[+Q] is instantiated by ał/dzi in PolQs, and a silent wh-operator in WhQs that is associated with clause-initial wh-words. Main clauses additionally feature a Speech Act Phrase (SAP) that is headed by an illocutionary operator SAASK associated with the conversational function of asking (see Speas & Tenny 2003; Farkas & Bruce 2010; Krifka 2014; Woods 2016; Sauerland & Yatsushiro 2017; Miyagawa 2022; Dayal 2023 and others). In Sm’algyax, SAASK is instantiated by the interrogative clitics =ii and =u; throughout, I will refer to these elements as markers of interrogative illocutionary mood. This is sketched below in (3) and (4) for embedded and main-clause PolQs, respectively. The main-clause restriction of =ii and =u falls out if question embedding predicates such as ask or know select for interrogative CPs, as in (3), and never SAPs.

    1. (3)
    1. (4)

This analysis, if correct, bears on a recent debate in the literature concerning whether speech acts can occur in embedded contexts. Drawing on data from embedded root/main-clause phenomena or quasi-subordination—such as embedded inversion in English questions—authors such as Krifka (2014; 2023) and Woods (2016) argue that speech acts (that is, peripheral projections headed by a speech act or illocutionary operator) can be embedded in a restricted way: namely, under rogative predicates such as ask or wonder, but not under responsive predicates such as know or tell.1

    1. (5)
    1. I wondered [was he illiterate]. Embedded inverted question
    2. (Roddy Doyle, The Woman Who Walked into Doors, cited in McCloskey 2006: 1)

Dayal (2023), on the other hand, provides extensive argumentation, based in part on the behaviour of several classes of interrogative particle in different embedding contexts, that quasi-subordination (such as the embedded inverted question in (5)) targets an intermediate projection between CP and SAP (called Persp(ective)P), and that SAP is in fact not embeddable. The clearest evidence for the unembeddability of SAP comes from the behaviour of so-called meta question particles (MQPs), that are taken to modify speech acts, and are strictly confined to main clauses. The unembeddability of the English MQP quick(ly) is shown below.

    1. (6)
    1. a.
    1.   Quick(ly), where did you hide the matza?
    1.  
    1. b.
    1. *Mary asked Sue [quick(ly) where she hid the matza].
    1.  
    1. c.
    1. *Mary asked Sue [quick(ly) where did she hide the matza]. (Dayal 2023: 8)
    1. (7)
    1. [SAP … Not embeddable
    2.             [PerspP … Embeddable under rogative predicates
    3.                         [CP … Embeddable under all question-embedding predicates

While the Sm’algyax facts presented here do not provide evidence for a three-layer interrogative left periphery (that is, I do not provide evidence for PerspP in Sm’algyax), they do provide compelling evidence for an unembeddable SAP layer: not from particles that modify speech acts, but from particles associated with canonical question formation.

The remainder of this section introduces Sm’algyax, methodology, and basic morphosyntactic background. Sections 2 and 3 introduce the reflexes of polar and wh-question formation. Section 4 provides a formal analysis of question-formation reflexes in Sm’algyax, arguing for a bipartite interrogative syntax that minimally contains a clause-typing projection and a speech act projection. Section 5 considers additional evidence for this analysis with reference to non-canonical questions, and Section 6 concludes.

Finally, a note on terminology: this paper deals with three distinct notions of clause type: (i) the Tsimshianic-internal distinction between independent and dependent clauses; (ii) the more general crosslinguistic distinction between main (or root/matrix) and embedded (or subordinate) clauses; and (iii) the grammatically defined classes of sentences which closely correspond to speech acts: declarative, interrogative, imperative, and exclamative (see Portner (2018)); of which this paper only deals with declarative and interrogative clauses. Whenever I refer to the first of these, I will explicitly call it the independent/dependent clause distinction. Likewise, I will explicitly refer to the second as the main/embedded clause distinction. Any bare references to “clause typing” should be understood as referring to the third notion, i.e., the distinction between declarative and interrogative clauses.

1.1 Data and methodology

All uncited examples are drawn from novel fieldwork conducted in Prince Rupert, British Columbia, between 2018 and 2025, in collaboration with fluent Sm’algyax speakers: Velna Nelson, Ellen Mason (Txałgiiw/Hartley Bay), and Beatrice Robinson (Gitxaała/Kitkatla). Additional sessions with Velna Nelson took place via Zoom from 2020 to 2025. In person sessions typically involved group sessions with two or more language consultants, whereas Zoom sessions were conducted individually.

All language data was collected following standard semantic fieldwork methodology (Matthewson 2004): the consultant was given a context and an English sentence and asked to translate it into Sm’algyax. Acceptability judgments were obtained by presenting the consultant with a Sm’algyax sentence and asking for feedback on its acceptability in a specific context.

The four-line glossing convention used here is structured as follows: the first line includes the community orthography used across British Columbia (Dunn 1978). The second line uses the same orthography but includes morpheme boundaries. The third line provides grammatical category labels following the Leipzig glossing rules, and the fourth line gives an English translation. I have also presented examples drawn from traditional Ts’msyen narratives (or adaawx) (Boas 1912) in the community orthography.

1.2 Language background

Sm’algyax, also known as Coast Tsimshian or the Ts’msyen language, is a Maritime Tsimshianic language spoken in Northwestern British Columbia, and Southeastern Alaska. The Interior branch of the family is made up of Gitksan and Nisga’a while the Maritime branch is made up of Sm’algyax and Southern Tsimshian, or Sgüüxs. See Forbes (2023) for an overview of the family, and see Hill & Matthewson (2025) and Matthewson (to appear) for recent discussion and analysis of PolQ formation in Gitksan. This section outlines morphosyntactic facts about Sm’algyax relevant to the discussion that follows.

Unmarked word order in Sm’algyax is predicate initial, with transitive subjects preceding direct objects:

    1. (8)
    1. ’Makdis Bill misoo.
    1. ’mak-t-ə
    2. catch-t-tr [-3.ii]
    1. =s
    2. =pn
    1. Bill
    2. Bill
    1. =a
    2. =cn
    1. misoo
    2. sockeye.salmon
    1. ‘Bill caught a sockeye salmon.’2 Predicate-initial word order

Sm’algyax, like all Tsimshianic languages, exhibits a distinction between independent and dependent clauses, which are characterized by distinct person-marking configurations and morphology on the predicate. Independent clauses are typically main clauses, but also appear in Ā-movement constructions (including relative clauses, wh-questions, and focus fronting) that target the direct object argument, or O-argument. Embedded clauses (except for those that are characterized by Ā-movement of the O-argument), imperatives, and main clauses introduced by a dependent marker such as aka/ałga ‘negation’, or yagwa ‘progressive aspect’ are all characterized by dependent clause morphology.

The distinct person marking configurations of independent and dependent clauses are shown in the following examples. The independent clause in (9) is characterized by the presence of suffixal (Series II) agreement marking the ergative argument. The dependent clause in (10), triggered by the negation marker aka, features a prepredicative clitic (Series I) marking the ergative argument.

    1. (9)
    1. Nah gabu naasü.
    1. nah
    2. pfv
    1. gap-ə-u
    2. eat-tr-1sg.ii
    1. =a
    2. =cn
    1. naasü
    2. raspberry
    1. ‘I ate raspberries.’Independent clause
    1. (10)
    1. Akandi gaba hoon.
    1. aka=n=di
    2. neg =1sg.i =foc
    1. gap
    2. eat[-3.ii]
    1. =a
    2. =cn
    1. hoon
    2. fish
    1. ‘I didn’t eat fish.’ Dependent clause

The distribution of the transitive suffix -ə (glossed as “tr”) is likewise sensitive to the independent/dependent clause distinction.3 The following pair of examples shows the presence of the transitive suffix in independent clauses and its absence in dependent clauses:

    1. (11)
    1. a.
    1. Gabit.
    1. gap-ə-t
    2. eat-tr-3.ii
    1. S/he ate it. Independent clause
    1.  
    1. b.
    1. Yagwat gapt.
    1. yagwa=t
    2. prog =3.i
    1. gap-t
    2. eat-3.ii
    1. ‘S/he is eating it.’ Dependent clause

I note here, and reiterate this point below in footnotes 5 and 16 that the independent/dependent clause distinction does not affect the distribution of the interrogative clitics =ii and =u. That is, either clitic may appear in both independent and dependent clauses, so long as they grammatically function as main clauses.

All Ā-extraction constructions in Sm’algyax are marked by extraction morphology that indicates the grammatical role of the extracted element. This is illustrated below with relative clauses (but as we will see in Section 3, this same verbal morphology appears in wh-questions). In (12) we see an intransitive subject (S) relative clause marked by the suffix -ət (glossed as “sx” for “subject extraction”), in (13) an object (O) relative clause marked by the transitive suffix -ə, and in (14) a transitive subject (A) relative clause marked by the prepredicative morpheme in (“ax”) and a third person agreement marker =t. In each case, a common-noun connective cliticizes to the right edge of the extracted element.

    1. (12)
    1. a.
    1. Sis’aaxsa gyet.
    1. sis’aaxs
    2. laugh
    1. =a
    2. =cn
    1. gyet
    2. person
    1. ‘A person laughed.’ Baseline
    1.  
    1. b.
    1. Wilaayu gyeda sis’aaxsit.
    1. wilaay-u
    2. know-1sg.ii
    1. =a
    2. =cn
    1. gyet
    2. person
    1. =a
    2. =cn
    1. sis’aaxs-ət ___
    2. laugh-sx
    1. ‘I know the person who laughed.’ S-relative
    1. (13)
    1. a.
    1. Gaba gyeda ts’ik’aaws.
    1. gap-ə
    2. eat-tr [-3.ii]
    1. =a
    2. =cn
    1. gyet
    2. person
    1. =a
    2. =cn
    1. ts’ik’aaws
    2. split.salmon
    1. ‘The people eat split dried salmon.’ Baseline
    1.  
    1. b.
    1. Niidzu ts’ik’aawsa gabit.
    1. niist-ə-u
    2. see-tr-1sg.ii
    1. =a
    2. =cn
    1. ts’ik’aaws
    2. split.salmon
    1. =a
    2. =cn
    1. gap-ə-t ___
    2. eat-tr-3.ii
    1. ‘I saw the split dried salmon they ate.’ O-relative
    1. (14)
    1. Wilaayu gyeda int gapt.
    1. wilaay=u
    2. know-1sg.ii
    1. =a
    2. =cn
    1. gyet
    2. person
    1. =a
    2. =cn
    1. in=t
    2. ax =3.i
    1. gap-t ___
    2. eat-3.ii
    1. ‘I know the people who ate it.’ A-relative

The differences in person marking and verbal morphology in O-argument and A-argument extraction aligns with the independent/dependent clause distinction. O-extraction features independent clause morphology, including the transitive -ə and suffixal agreement with the A-argument. A-extraction features dependent clause morphology, including preverbal clitic agreement with the A-argument, and the absence of transitive -ə.

Having introduced the basic word order, the independent/dependent clause distinction as well as a background to Ā-movement, I now turn to the empirical core of the paper: the morphosyntactic reflexes associated with question formation.

2 Polar question reflexes

In this section, I outline the morphosyntactic reflexes of polar question (PolQ) formation, focusing on a structural asymmetry between main-clause and embedded-clause questions. In either environment, PolQs are introduced by an interrogative complementizer element (ał in main clauses, dzi in embedded clauses). Main-clause PolQs, however, display an additional reflex: the polar-interrogative clitic =ii, which is restricted to this environment.

One aspect of interrogative clitics that I will not address in detail in this paper is their linear positioning. Both =ii (discussed in this section) and =u (introduced in Section 3) exhibit the same linearization properties: they can appear in a range of positions in the clause, as documented in previous work (Brown & Davis 2024a; b; Davis & Brown 2024; Brown 2024a).4 Importantly, this positional variability does not affect the interpretation of the question. I set aside issues of linearization here, focusing instead on the syntactic distribution of these clitics and their restriction to main-clause interrogative sentences.

2.1 Main-clause polar questions

PolQ formation typically involves the left-edge element ał and the polar interrogative clitic =ii (15b). The word order of a declarative sentence (15a) is retained.

    1. (15)
    1. a.
    1. ’Waays Judy naletayu.
    1. ’waa-ə
    2. find-tr [-3.ii]
    1. =s
    2. =pn
    1. Judy
    2. Judy
    1. =a
    2. =cn
    1. na-leta-u
    2. poss-letter-1sg.ii
    1. ‘Judy received my letter.’ Baseline
    1.  
    1. b.
    1. Ałt waadit Judiyii naletayu?
    1. Ał=t
    2. q=3.i
    1. waa-t
    2. find-3.ii
    1. =t
    2. =pn
    1. Judi=ii
    2. Judy=int
    1. =a
    2. =cn
    1. na-leta=u?
    2. poss-letter-1sg.ii
    1. ‘Did Judy receive my letter?’ PolQ ([+ał, +ii])

The element ał is likely diachronically related to the negation markers ałga/aka. The Interior Tsimshianic languages Gitksan and Nisga’a likewise form neutral/unbiased PolQs with morphemes associated with negation (Rigsby 1986; Tarpent 1987; Hill & Matthewson 2025; Matthewson to appear). The interrogative clitic =ii is cognate with Sgüüxs =ii, which occurs in both PolQs and WhQs, and with Gitksan and Nisga’a =aa, which is restricted to PolQs.

The particle ał is not obligatory in PolQs; many PolQs are marked solely by the clitic =ii.5

    1. (16)
    1. ’Waays Judiyii naletayu?
    1. ’waa-ə
    2. find-tr [-3.ii]
    1. =s
    2. =pn
    1. Judi=ii
    2. Judy=int
    1. =a
    2. =cn
    1. na-leta-u
    2. poss-letter-1sg.ii
    1. ‘Did Judy receive my letter?’ [–ał, +ii]

A less common pattern features ał without =ii, as in (17). Further work is needed to determine whether this configuration which features ał and lacks =ii has any systematic semantic or pragmatic effects.6

    1. (17)
    1. Ał yagwał sikopis Henili?
    1. Ał
    2. q
    1. yagwa
    2. prog
    1. =ł
    2. =irr.cn
    1. si-kopi
    2. make-coffee[-3.ii]
    1. =s
    2. =pn
    1. Henili?
    2. Henry
    1. ‘Is Henry making coffee?’ [+ał, –ii]

Preliminary investigation in Brown (2021) reveals no difference between declarative and interrogative sentences with respect to intonation.7 Attempts to form a question marked solely by rising intonation (therefore lacking both ał and =ii) are rejected:

    1. (18)
    1. #Yaa waas? (rising intonation)
    1.    yaa
    2.    walk
    1. =a
    2. =cn
    1. waas
    2. rain
    1.    Intended: ‘Is it raining?’ # [–ał, –ii]

Though ał is likely derived from a morpheme associated with negation, its presence or absence in a question doesn’t seem to signal a biased interpretation as one might expect. We see below that both options are possible in a context where the asker of the question does not appear to have any evidence or bias for or against the prejacent proposition (that is, whether there is tea) being true:

    1. (19)
    1. [You are at a friend’s house; you have no idea whether they keep tea in the house, but would like to have some tea]
    1.  
    1. a.
    1. Sgüüyiił dii?
    1. sgüü=ii
    2. exist=int
    1. =ł
    2. =irr.cn
    1. dii
    2. tea
    1. ‘Is there tea?’ Neutral PolQ ([–ał])
    1.  
    1. b.
    1. Ał sgüüdiił dii?
    1. ał
    2. q
    1. sgüü-t=ii
    2. exist-3.ii =int
    1. =ł
    2. =irr.cn
    1. dii
    2. tea
    1. ‘Is there tea?’ Neutral PolQ ([+ał])

PolQs both with and without ał are also freely volunteered and accepted in contexts where there is contextual evidence for p:

    1. (20)
    1. [Lucy is in her windowless office. Alan has just arrived holding a wet umbrella and raincoat. (adapted from Gunlogson 2008)]
    1.  
    1. a.
    1. Du! Yaayiił waas?
    1. du
    2. oh
    1. yaa=ii
    2. walk=int
    1. =ł
    2. irr.cn
    1. waas
    2. rain
    1. Oh, it’s raining? Biased PolQ ([–ał])
    1.  
    1. b.
    1. Du! Ał waasii gyelx?
    1. du
    2. oh
    1. ał
    2. q
    1. waas=ii
    2. rain=int
    1. gyelx
    2. outside
    1. Oh, it’s raining? Biased PolQ ([+ał])
    1. (21)
    1. [Lucy mentioned a few days ago that her mother is visiting her. You run into her at a coffee shop with an older woman who bears a close resemblance to Lucy. You assume it’s her mom, but want to confirm it. (adapted from Poschmann 2008)]
    1.  
    1. a.
    1. ’Niit Lucy, ’niis noonii gwa’a?
    1. ’niit
    2. 3.iii
    1. Lucy
    2. Lucy
    1. ’nii
    2. dem
    1. =s
    2. =pn
    1. noo-n=ii
    2. mother-2sg.ii =int
    1. =a
    2. =cn
    1. gwa’a
    2. dem
    1. Hi Lucy! This is your mother? Biased PolQ ([–ał])
    1.  
    1. b.
    1. ’Niit Lucy, ał ’niis noonii gwa’a?
    1. ’niit
    2. 3.iii
    1. Lucy
    2. Lucy
    1. ał
    2. q
    1. ’nii
    2. dem
    1. =s
    2. =pn
    1. noo-n=ii
    2. mother-2sg.ii =int
    1. =a
    2. =cn
    1. gwa’a
    2. dem
    1. Hi Lucy! This is your mother? Biased PolQ ([+ał])

The acceptability of both PolQ variants in the contexts above highlights a contrast between Sm’algyax and Gitksan: in Gitksan, the presence of the negation morpheme nee in PolQs carries a pragmatic effect, favouring contexts with neutral or negative contextual evidence for p, whereas PolQs without nee are preferred in contexts with contextual evidence for p (Hill & Matthewson 2025; Matthewson to appear). That is, in a neutral context like the one in (19), a PolQ featuring negative nee is preferred, while in a context where there is evidence for p, such as those in (20) and (21), a PolQ without negative nee is preferred. In Sm’algyax, however, PolQs with and without ał appear to occur in free variation.

Sm’algyax does however have a grammaticalized biased PolQ construction, characterized by the negative response particle ayn in initial position, as well as =ii. These ayn-questions are biased toward the truth of the prejacent proposition.

    1. (22)
    1. [We’re looking for a restaurant in the neighbourhood. I remember there being a decent Chinese restaurant close by. I say to you: (adapted from Ladd (1981))]
    2. Aynł t’aadii waptxooxgm dzeena a gwa’a?
    1. ayn
    2. no
    1. =ł
    2. =irr.cn
    1. t’aa-t=ii
    2. stand-3.ii=int
    1. =a
    2. =cn
    1. wap=txooxk-m
    2. house=eat-attr
    1. dzeena
    2. Chinese
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. gwa’a
    2. here
    1. ‘Isn’t there a Chinese restaurant around here?’ Biased PolQ ([+ayn])
    1. (23)
    1. [There’s a feast in early May; Lucy tells you she’s planning to attend it. This surprises you because she previously mentioned she would be in Vancouver from mid-April to mid-May. You ask her:]
    2. Aynł dm waanii a Ts’a’mas?
    1. ayn
    2. no
    1. =ł
    2. =irr.cn
    1. dm
    2. prosp
    1. waal-n=ii
    2. lv-2sg.ii =int
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. Ts’a’mas
    2. Vancouver
    1. ‘Won’t you be in Vancouver?’ Biased PolQ ([+ayn])

These data point to further variation between Sm’algyax and Gitksan question formation. In Gitksan, both neutral PolQs and biased PolQs such as (22) are formed using the same negation element nee (Hill & Matthewson 2025; Matthewson to appear).

Finally, we also find that both ał and =ii appear in alternative questions—those questions offering a disjunctive choice of answers, and expecting one of those options as a response.

    1. (24)
    1. a.
    1. Ał waasii dzi ligi ayn?
    1. ał
    2. q
    1. waas=ii
    2. rain=int
    1. dzi
    2. irr
    1. ligi
    2. dwid
    1. ayn
    2. no
    1. ‘Is it raining or not?’
    1.  
    1. b.
    1. Waasii dzi ligi ayn?
    1. waas=ii
    2. rain=int
    1. dzi
    2. irr
    1. ligi
    2. dwid
    1. ayn
    2. no
    1. ‘Is it raining or not?’8(Polar) alternative question
    1. (25)
    1. [You’re at a diner where breakfast includes a choice of coffee or tea. You order breakfast, and the waiter asks you:]
    2. Ał dm hasaganii kopii dzi ligi dii?
    1. ał
    2. q
    1. dm
    2. prosp
    1. hasax-n=ii
    2. want-2sg.ii =int
    1. =a
    2. =cn
    1. kopii
    2. coffee
    1. dzi
    2. irr
    1. ligi
    2. dwid
    1. dii
    2. tea
    1. ‘Do you want coffee, or tea?’ Alternative question
    1. (26)
    1. [You are responding to (25)]
    2. Dm xkopii’nu.
    1. dm
    2. prosp
    1. x-kopii-'nu
    2. ingest-coffee-1sg.iii
    1. ‘I’ll have coffee.’

Wrapping up this section, we have seen that two particles (ał and =ii) are used to form polar questions in Sm’algyax. Ał, which is likely derived from negation, is often omitted; contra Gitksan, the absence of the initial negative element in PolQs does not appear to carry a pragmatic effect. In the following sections, we investigate the syntactic distribution of these elements in more detail.

2.2 Embedded polar questions

So far, we have only considered main-clause PolQs. In embedded contexts, neither element that characterizes main-clause PolQs, that is, ał or =ii, can appear. Embedded PolQs are instead uniformly introduced by the complementizer dzi, glossed as ‘irrealis’ in the Tsimshianic literature.9 This generalization holds for PolQs embedded under responsive predicates, which select for declarative and interrogative complements, and rogative predicates, which select only for interrogative complements (Lahiri 2002).10

The contrast in (27)–(28) illustrates the pattern with the responsive predicate wilaay ‘know’: the embedded clause appears with initial dzi—where ał would occur in a main-clause PolQ—and lacks both ał and =ii.

    1. (27)
    1. Ał hasaxł haasii wineeya?
    1. ał
    2. q
    1. hasax
    2. want[-3.ii]
    1. =ł
    2. =cn
    1. haas=ii
    2. dog=int
    1. =a
    2. =cn
    1. wineeya
    2. food
    1. ‘Does the dog want food?’ Baseline PolQ
    1. (28)
    1. Wilaays Dzon dzida hasaga haasa wineeya.
    1. wilaay-ə
    2. know-ə[-3.ii]
    1. =s
    2. =pn
    1. Dzon
    2. John
    1. [dzi=da
    2. irr=spt
    1. hasax
    2. want[-3.ii]
    1. =a
    2. =cn
    1. haas
    2. dog
    1. =a
    2. =cn
    1. wineeya]
    2. food
    1. ‘John knows whether the dog wants food. Responsive embedding

The PolQ embedded under the rogative predicate güüdax ‘ask’ patterns the same way: dzi introduces the embedded interrogative clause, and neither ał nor =ii appears.

    1. (29)
    1. Ał hasaganii dm yeltgn?
    1. ał
    2. q
    1. hasax-n=ii
    2. want-2sg.ii =int
    1. dm
    2. prosp
    1. yeltk-n?
    2. return-2sg.ii
    1. ‘Do you want to go back?’ Baseline PolQ
    1. (30)
    1. Yagwan güüdagan dza hasagan ła dm yeltgn.
    1. yagwa=n
    2. prog =1sg.i
    1. güüdax-n
    2. ask-2sg.ii
    1. [dza
    2. irr
    1. hasax-n
    2. want-2sg.ii
    1. ła
    2. incep
    1. dm
    2. prosp
    1. yeltk-n]
    2. return-2sg.ii
    1. ‘I’m asking you if you want to go back.’ Rogative embedding

Attempts to insert =ii into an embedded PolQ, with or without an initial dzi, are rejected:

    1. (31)
    1. *Wilaays Dzon (dzida) hasaxł haasii wineeya.
    1.   wilaay-ə
    2.   know-tr [-3.ii]
    1. =s
    2. =pn
    1. Dzon
    2. John
    1. [dzi=da
    2. irr =spt
    1. hasax
    2. want
    1. =ł
    2. =cn
    1. haas=ii
    2. dog=int
    1. =a
    2. =cn
    1. wineeya]
    2. food
    1.   Intended: ‘John knows whether the dog wants food. *Embedded =ii
    1. (32)
    1. *Yagwan güüdagan (dza) hasaganii dm yeltgin.
    1.   yagwa=n
    2.   prog =1sg.i
    1. güüdax-n
    2. ask-2sg.ii
    1. [dza
    2. irr
    1. hasax-n=ii
    2. want-2sg.ii
    1. dm
    2. prosp
    1. yeltk-n]
    2. return-2sg.ii
    1.   Intended: ‘I’m asking you if you want to go back.’ *Embedded =ii

Likewise, attempts to insert ał into an embedded question are rejected, as shown in (33b). The example in (34) shows that, even when the embedded question itself appears as the complement of a main-clause question, ał is still prohibited from embedded PolQs.

    1. (33)
    1. a.
    1.   Akandi wilaay dzi dmt liiłgidit Meeli haas.
    1.   aka=n=di
    2.   neg =1sg.ii =foc
    1. wilaay
    2. know[-3.ii]
    1. [dzi
    2. irr
    1. dm=t
    2. prosp =3.i
    1. liiłk-t
    2. care.for-3.ii
    1. =t
    2. =pn
    1. Meeli
    2. Mary
    1. =a
    2. =cn
    1. haas]
    2. dog
    1.   ‘I don’t know if Mary will look after the dog.’ Baseline
    1.  
    1. b.
    1. *Akandi wilaay (dzi) ał (dzi) dmt liiłgidit Meeli haas.
    1.   aka=n=di
    2.   neg =1sg.ii =foc
    1. wilaay
    2. know[-3.ii]
    1. [dzi
    2. irr
    1. ał
    2. q
    1. dzi
    2. irr
    1. dm=t
    2. prosp =3.i
    1. liiłk-t
    2. care.for-3.ii
    1. =t
    2. =pn
    1. Meeli
    2. Mary
    1. =a
    2. =cn
    1. haas]
    2. dog
    1.   Intended: ‘I don’t know if Mary will look after the dog.’ *Embedded ał
    1. (34)
    1. *Ał wilaaynii ał dm yaał waas?
    1.   ał
    2.   q
    1. wilaay-n=ii
    2. know-2sg.ii=int
    1. [ał
    2. q
    1. dm
    2. prosp
    1. yaa
    2. walk
    1. =ł
    2. =irr.cn
    1. waas]
    2. rain
    1.   Intended: ‘Do you know if it will rain?’ *Embedded ał

This section has shown that, in contrast to main-clause PolQs, which receive bipartite marking, embedded PolQs bear only one reflex: the clause-initial element dzi.

2.3 Assumption: ał and dzi occupy the same position

This section considers three shared properties of ał and dzi: (i) both elements are markers of PolQs, (ii) they occupy the far left edge of the clause, and (iii) they never co-occur. In addition, I show that ał does not share a syntactic position with the propositional negation marker aka, despite being historically derived from a negative element. Taken together, these observations suggest that ał synchronically functions as a main-clause interrogative C element, and that ał and dzi are best analyzed as allomorphs of one another.

Supporting evidence for this claim comes from their co-occurrence with, and consistent ordering relative to, aka: both ał and dzi may appear with aka, and in all cases they precede it.11 This ordering is notable because propositional negation itself occupies a high clausal position, as shown by the fact that it precedes ergative Series I person agreement n, the aspectual marker dm, and the low focus element di. (The notation a > b stands for “a linearly precedes b”.)

    1. (35)
    1. Ha’ligoodu akanmdi goo Txałgiiw.
    1. ha’ligoot-u
    2. think-1sg.ii
    1. [aka=n=dim=di
    2. neg =1sg.i =prosp =foc
    1. goo
    2. go.to
    1. =a
    2. =cn
    1. Txałgiiw]
    2. Hartley Bay
    1. ‘I think I won’t go to Hartley Bay.’ aka > n > dim > di

Functional elements that precede negation include complementizers (e.g. wil in (36)), clausal coordinators (e.g. da’al ‘but’ in (37)), and discourse particles (e.g. ap ‘verum’ in (37)).

    1. (36)
    1. Lu aam goodu wil akadmdi yaa waas.
    1. lu
    2. in
    1. aam
    2. good
    1. goot-u
    2. heart-1sg.ii
    1. [wil
    2. comp
    1. aka=dm=di
    2. neg =prosp =foc
    1. yaa
    2. walk
    1. =a
    2. =cn
    1. waas]
    2. rain
    1. ‘I am happy that it won’t rain.’ wil > aka
    1. (37)
    1. Da’al ap akadi haws dip gwa’a.
    1. da’al
    2. but
    1. ap
    2. ver
    1. aka=di
    2. neg =foc
    1. haw
    2. say[-3.ii]
    1. =s
    2. =pn
    1. dip
    2. pl
    1. gwa’a
    2. dem
    1. ‘But these people didn’t say.’ (SLLTD 2017) da’al > ap > aka

Turning to ał, we see that it may co-occur with aka, obligatorily preceding it. This positional evidence suggests that the two occupy distinct syntactic positions. Furthermore, the absence of a double negation reading in such contexts indicates that ał, despite being derived from a negation marker, does not contribute negative semantics.12

    1. (38)
    1. [Lucy previously said she would be in Vancouver in May, but now she’s saying she’ll attend a local feast during that time. You ask:]
    1.  
    1. a.
    1.   Ał aka dm waanii a Ts’a’mas?
    1.   ał
    2.   q
    1. aka
    2. neg
    1. dm
    2. prosp
    1. waal-n=ii
    2. lv-2sg.ii =int
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. Ts’a’mas
    2. Vancouver
    1.   ‘Will you not be in Vancouver?’ ał > aka
    1.  
    1. b.
    1. *Aka ał dm waanii a Ts’a’mas?
    1.   aka
    2.   neg
    1. ał
    2. q
    1. dm
    2. prosp
    1. waal-n=ii
    2. lv-2sg.ii =int
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. Ts’a’mas
    2. Vancouver
    1.   Intended: ‘Will you not be in Vancouver?’ *aka > ał

Likewise, in embedded PolQs, dzi may co-occur with, and precedes, aka:

    1. (39)
    1. a.
    1.   Güüdagu dzi aka nm dii goo Txałgiiw.
    1.   güüdax-ə-u
    2.   ask-tr-1sg.ii
    1. [dzi
    2. irr
    1. aka
    2. neg
    1. n=dim
    2. 1sg.i =prosp
    1. dii
    2. foc
    1. goo
    2. go[-3.ii]
    1. =a
    2. =cn
    1. Txałgiiw]
    2. Hartley.Bay
    1.   ‘I asked if I will not go to Hartley Bay.’ dzi > aka
    1.  
    1. b.
    1. *Güüdagu aka dzi nm dii goo Txałgiiw.
    1.   güüdax-ə-u
    2.   ask-tr-1sg.ii
    1. [aka
    2. neg
    1. dzi
    2. irr
    1. n=dim
    2. 1sg.i =prosp
    1. dii
    2. foc
    1. goo
    2. go[-3.ii]
    1. =a
    2. =cn
    1. Txałgiiw]
    2. Hartley.Bay
    1.   Intended: ‘I asked if I will not go to Hartley Bay.’ *aka > dzi

As established in Section 2.2, ał and dzi cannot co-occur in embedded PolQs. This restriction also holds in main-clause PolQs:

    1. (40)
    1. a.
    1. *Ał dzi yaayii waas?
    1.   ał
    2.   q
    1. dzi
    2. irr
    1. yaa=ii
    2. walk=int
    1. waas
    2. rain
    1.   Intended: ‘Is it raining?’ *[+ał, +dzi]
    1.  
    1. b.
    1. *Dzi ał yaayii waas?
    1.   dzi
    2.   irr
    1. ał
    2. q
    1. yaa=ii
    2. walk=int
    1. waas
    2. rain
    1.   Intended: ‘Is it raining?’ *[+dzi, +ał]

The ability of ał to co-occur with aka provides evidence that it is distinct from propositional negation. Meanwhile, the shared linear positioning of ał and dzi before negation, along with their complementary distribution in main vs. embedded PolQs, suggests a case of contextual allomorphy: ał (and its null alternant) functions as the main-clause realization of dzi. I adopt this assumption in Section 4, treating ał and dzi as allomorphs of a single interrogative complementizer.13 Let us now turn to WhQs.

3 Wh-question reflexes

WhQs, like PolQs, display two morphosyntactic reflexes in main-clause questions and a single reflex in embedded questions. We see in (41) a WhQ characterized by (i) a clause-initial wh-word and (ii) the wh-particle =u:14

    1. (41)
    1. a.
    1. Nah baat Dzeen.
    1. nah
    2. pfv
    1. baa
    2. run
    1. =t
    2. =pn
    1. Dzeen
    2. Jane
    1. ‘Jane ran.’ Baseline
    1.  
    1. b.
    1. Naayu nah baat?
    1. naa=u
    2. who=int
    1. =a
    2. =cn
    1. nah
    2. pfv
    1. baa-ət ___
    2. run-sx
    1. ‘Who ran?’ WhQ

I outline below the distribution of wh-words and the wh-clitic =u. I show that =u, but not the clause-initial wh-word, is restricted to main-clause WhQs: =u is prohibited from occurring in embedded WhQs, while both main and embedded clauses are introduced by a wh-word. I further show that =u is not associated directly with the wh-word itself. This section lays the groundwork for Section 4, where I propose that the trigger for the clause-initial wh-word is an interrogative C element, while =u (alongside polar interrogative =ii) is the overt instantiation of a speech act operator.15

3.1 Main-clause wh-questions

Main-clause WhQs bear the same extraction morphology introduced in Section 1.2, but are further marked by the presence of the enclitic =u, which appears in every wh-question configuration, with all wh-words. Below we see S, O, and A questions marked with their respective extraction morphology as well as the wh-particle =u.16,17 The position of =u is variable and it has several alternants, which are outlined in Brown (2024a). These factors do not affect my core claim here that =u is a marker of illocutionary mood.

    1. (42)
    1. Naayu sis'aaxsit?
    1. naa=u
    2. who=int
    1. =a
    2. =cn
    1. sis'aaxs-ət ___
    2. laugh-sx
    1. ‘Who laughed?’ S wh-question
    1. (43)
    1. Goyu gabin?
    1. goo=u
    2. what=int
    1. =a
    2. =cn
    1. gap-ə-n ___
    2. eat-tr-2sg.ii
    1. ‘What did you eat?’ O wh-question
    1. (44)
    1. Goyu int łak'an?
    1. goo=u
    2. what=int
    1. =a
    2. =cn
    1. in=t
    2. ax=3
    1. łak'-n ___
    2. bite-2sg.ii
    1. ‘What bit you?’ A wh-question

In WhQs, the wh-word must appear in clause-initial position: wh-words cannot appear in their in-situ argument positions, either as a canonical question or an echo/surprise question (Brown 2024b):

    1. (45)
    1. a.
    1.   Tgi k'apaaytga ‘yuuta.
    1.   tgi
    2.   down
    1. k'apaaytk=a
    2. fall=cn
    1. ’yuuta
    2. man
    1.   ‘The man fell down.’
    1.  
    1. b.
    1. Naayu tgi k'apaaytgit?
    1.   naa=u
    2.   who=q
    1. tgi
    2. down
    1. k'apaaytk-ət ___
    2. fall-sx
    1.   ‘Who fell down?’
    1.  
    1. c.
    1. *Tgi k'apaaytgit naa(yu)?
    1.   tgi
    2.   down
    1. k'apaaytk-ət
    2. fall-sx
    1. naa(=u)
    2. who=q
    1.   Intended: ‘Who fell down?’ *wh-in-situ
    1. (46)
    1. [Mary is talking about her young child, Pat, and mentioned that he ate sea lion. You are surprised by this:]
    1.  
    1. a.
    1. *Gabis Patł goo?!
    1.   gap-ə-t=s
    2.   eat-tr-3.ii =pn
    1. Pat=ł
    2. Pat=irr.cn
    1. goo
    2. what
    1.   Intended: ‘Pat ate what?!’
    1.  
    1. b.
    1.   [Correction offered by speaker]
    2.   Goł gabis Pat?!
    1.   goo=ł
    2.   what=irr.cn
    1. gap-ə-t=s
    2. eat-tr-3.ii =pn
    1. Pat ___
    2. Pat
    1.   ‘What did Pat eat?!’ *wh-in-situ

Though WhQs are almost always volunteered to me with =u, it may be dropped in colloquial or rapid speech.

    1. (47)
    1. a.
    1. Naayu łimooys Dzon?
    1. naa=u
    2. who=int
    1. =a
    2. =cn
    1. łimoo-ə
    2. help-tr[-3.ii]
    1. =s
    2. =pn
    1. Dzon ___
    2. John
    1. ‘Who did John help?’
    1.  
    1. b.
    1. Naał łimooys Dzon?
    1. naa
    2. who
    1. =ł
    2. =irr.cn
    1. łimoo-ə
    2. help-tr[-3.ii]
    1. =s
    2. =pn
    1. Dzon ___
    2. John
    1. ‘Who did John help?’ =u may be omitted

Therefore, like the polar interrogative clitic =ii, the appearance of =u is optional (but preferred; my consultants describe wh-questions without =u as a “shortcut” and it appears to be judged as a difference in register). I note here, echoing footnote 6, that in both question types, when an interrogative clitic is absent, there is some other morphosyntactic cue of an interrogative clause: in polar questions there is the clause-initial particle ał, and in wh-questions there is a clause-initial wh-word.

Finally, though the examples above have shown the wh-particle occurring with the question words naa ‘who’ and goo ‘what’ in core-argument (S, O, A) questions, we see below that it also occurs in locative (48), temporal (49), discourse-linked (50), manner (51), and reason (52) questions, and with all other wh-words such as the underspecified ndaa/ndeh and all the wh-words derived from it (48)–(51), as well as the quantificational wh-predicates t’masool ‘how many (people)’ (53), and t’maays ‘how many (things)’ (54).18

    1. (48)
    1. Ndeyu nam wil niidzu?
    1. ndeh=u
    2. where=int
    1. =a
    2. =cn
    1. nah=m
    2. pfv=2sg.ii
    1. wil
    2. comp
    1. niist-u ___
    2. see-1sg.ii
    1. ‘Where did you see me?’ Locative WhQ
    1. (49)
    1. Dzindał dm ap yaltgidut Norman?
    1. dzi=ndaa
    2. irr =when
    1. =ł
    2. =cn
    1. dm
    2. prosp
    1. ap
    2. ver
    1. yaltk-t=u
    2. return-3.ii =int
    1. =t
    2. =pn
    1. Norman ___
    2. Norman
    1. ‘When is Norman really coming back?’ (Sasama 2001: 64) Temporal WhQ
    1. (50)
    1. Ksindeyu gan diduulsit?
    1. ksi=ndeh=u
    2. out=which=int
    1. =a
    2. =cn
    1. gan
    2. tree
    1. =a
    2. =cn
    1. diduuls-ət ___
    2. alive-sx
    1. ‘Which tree is alive?’ Discourse-linked WhQ
    1. (51)
    1. Ndeyu wila waan?
    1. ndeh=u
    2. how=int
    1. wila
    2. manr
    1. waal-n ___
    2. do-2sg.ii
    1. ‘How are you doing?’ Manner WhQ
    1. (52)
    1. Goł gan sis’aaxsdut Kayla?
    1. goo
    2. what
    1. =ł
    2. =cn
    1. gan
    2. reas
    1. sis’aaxs-t=u
    2. laugh-3.ii=int
    1. =t
    2. =pn
    1. Kayla ___
    2. Kayla
    1. ‘Why is Kayla laughing?’ (SLLTD 2017) Reason WhQ
    1. (53)
    1. T’masooldu gyet?
    1. t’masool-t=u
    2. how.many.people-3.ii=int
    1. =a
    2. =cn
    1. gyet
    2. people
    1. ‘How many people are there?’ ‘How many’ WhQ
    1. (54)
    1. T’maaysdu dooni?
    1. t’maays-t=u
    2. how.many.things-3.ii=int
    1. =a
    2. =cn
    1. dooni
    2. dem
    1. ‘How many are way over there?’ (SLLTD 2017) ‘How many’ WhQ

In this section, we have seen that main-clause WhQs are obligatorily wh-initial and feature the interrogative clitic =u.

3.2 Embedded wh-questions

Embedded WhQs, like their main-clause counterparts, are strictly wh-initial. They appear as complements of standard question-embedding predicates, including responsive wilaay ‘know’ and rogative güüdax ‘ask’. The wh-clitic =u, which characterizes main-clause WhQs, cannot appear in embedded questions.

    1. (55)
    1. Wilaayu naa(*yu) łimoom sm’ooygit.
    1. wilaay-ə-u
    2. know-tr-1sg.ii
    1. [naa(*=u)
    2. who(*=int)
    1. =a
    2. =cn
    1. łimoom-ə
    2. help-tr [-3.ii]
    1. =a
    2. =cn
    1. sm’ooygit _____]
    2. chief
    1. ‘I know who the chief helped.’ Responsive embedding; *embedded =u
    1. (56)
    1. Güüdagu naa(*yu) łimoom sm’ooygit.
    1. güüdax-ə-u
    2. ask-tr-1sg.ii
    1. [naa(*=u)
    2. who(*=int)
    1. =a
    2. =cn
    1. łimoom-ə
    2. help-tr [-3.ii]
    1. =a
    2. =cn
    1. sm’ooygit ____]
    2. chief
    1. ‘I asked who the chief helped.’ Rogative embedding; *embedded =u

Embedded WhQs therefore parallel embedded PolQs in that both constructions exhibit a proper subset of the reflexes that appear in main clause questions.

3.3 Wh-interrogative =u is not directly associated with wh-words

A class of particles that appear both in WhQs and in a variety of other contexts has received considerable attention in previous literature. These particles are referred to as Q-particles (Kuroda 1965; Hagstrom 1998; Cable 2007; 2010a; b; Kotek 2014; 2019; Uegaki 2018: a.o.). While their distribution varies across languages, Q-particles exhibit a number of shared syntactic behaviours. I focus here on the distribution of Q-particles in Tlingit (or Lingít, a Na-Dene language of Northern British Columbia, Yukon, and Alaska), because Tlingit is a northern neighbour of Sm’algyax and has a relatively well-studied Q-particle sá (Cable 2007; 2010a; b).

First, Q-particles often appear in WhQs, main and embedded. We see this in the Tlingit examples below. The Q-particle sá obligatorily surfaces in a main-clause question in (57) and an embedded question in (58).

    1. (57)
    1. Tlingit (Cable 2007: 26)
    1. Aadóo
    2. who
    1. yaagú
    2. boat
    1. sá
    2. q
    1. ysiteen?
    2. you.saw
    1. ‘Whose boat did you see?’
    1. (58)
    1. Tlingit (Nyman & Leer 1993: 200; cited in Cable 2007: 111)
    1. Tlél
    2. not
    1. xwasakú
    2. I.know
    1. [daa
    2. what
    1. sá wé
    2. q.foc-part
    1. a káx
    2. its surface.about
    1. xat x’aywóos’].
    2. you.ask.me
    1. ‘I don’t know what you are asking me about.’

Already, this behaviour contrasts with the distribution of Sm’algyax =u, which is restricted to main-clause questions (see the previous section).

Another hallmark of Q-particles is their more general association with wh-expressions, even in non-interrogative contexts. This is again exemplified by Tlingit sá, which appears in the examples below alongside wh-indefinite pronouns formed from daa ‘what’ and aadóo ‘who’:

    1. (59)
    1. Tlingit (Nyman & Leer 1993: 14; cited in Cable 2007: 75)
    1. Daa
    2. what
    1. sá wé
    2. q.foc
    1. yóo
    2. yonder
    1. dikéenax.á.
    2. far.out.across.one
    1. ‘There was something up there.
    1. (60)
    1. Tlingit (Dauenhauer & Dauenhauer 1990: 310; cited in Cable 2007: 111)
    1. Aadóo
    2. who
    1. sá
    2. q
    1. du
    2. his
    1. éet
    2. to
    1. shukawdudlixúxu
    2. song’s.words.are.for.sbjv
    1. áwé.
    2. foc
    1. ‘Whoever the words of a song are for.’

These core behaviours are shared by Japanese ka and Sinhala də, which also appear in WhQs as well as wh-indefinites (Kuroda 1965; Hagstrom 1998).19

Turning back to Sm’algyax, we find that wh-expressions appear not only in WhQs, but also in a range of non-interrogative contexts, including existential and universal expressions (61)–(63), free relative clauses (64), and exclamatives (65). The wh-clitic =u is strictly prohibited from occurring in any of these non-interrogative wh-constructions.

    1. (61)
    1. Ła’a ligi goo(*yu) haasgu.
    1. ła’a-ə
    2. bite-tr [-3.ii]
    1. =a
    2. =cn
    1. ligi
    2. dwid
    1. goo(*=u)
    2. what(*=int)
    1. =a
    2. =cn
    1. haas-k-u
    2. dog-pass-1sg.ii
    1. ‘Something bit my dog.’ Wh-existential
    1. (62)
    1. Nah niidzu ligit naa(*yu).
    1. nah
    2. pfv
    1. niits-ə-u
    2. see-tr-1sg.ii
    1. =a
    2. =cn
    1. ligi
    2. dwid
    1. =t
    2. =pn
    1. naa(*=u)
    2. who(*=int)
    1. ‘I saw someone.’ Wh-existential
    1. (63)
    1. Dm małdu txa’nii goo(*yu) da k’wan.
    1. dm
    2. prosp
    1. mał-t-ə-u
    2. tell-t-tr-1sg.ii
    1. =a
    2. =cn
    1. txa’nii
    2. all
    1. goo(*=u)
    2. what(*=int)
    1. da
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. k’wan
    2. 2sg.obl
    1. ‘I will tell you everything.’ Wh-universal
    1. (64)
    1. Waayu naa(*yu) dmt in dzaba ts’ikts’igu.
    1. Waa-ə-u
    2. find-tr-1sg.ii
    1. =a
    2. =cn
    1. [naa(*=u)
    2. who(*=int)
    1. dm=t
    2. prosp =3.i
    1. in
    2. ax
    1. dzap
    2. do[-3.ii]
    1. =a ___
    2. =cn
    1. ts’ikts’ik-u]
    2. car-1sg.ii
    1. ‘I found someone who will fix my car.’
    2. Lit. ‘I found who will fix my car.’ Wh-free relative
    1. (65)
    1. Goł/*(goyu) waalt!
    1. goo(*=u)
    2. what(*=int)
    1. =ł
    2. =irr.cn
    1. waal-t ___
    2. be-3sg.ii
    1. ‘What a thing!’20Wh-exclamative

These examples show that the wh-clitic is not associated with wh-expressions themselves and should not be analyzed as a Q-particle, unlike Japanese ka or Tlingit sá (Kratzer & Shimoyama 2002; Beck 2006; Cable 2007; 2010a; Kotek 2014; 2019; Uegaki 2018). Brown (2024a) further argues, based on the linearization of =u, that it does not form a constituent with the wh-expression, in contrast to the behaviour of Q-particles in wh-movement languages (e.g. Cable 2007; 2010a).

4 Bipartite question formation: the CP/SAP distinction

Sections 2 and 3 showed that the formation of PolQs and WhQs in Sm’algyax involves two distinct reflexes. PolQs feature a left-edge particle, while WhQs are obligatorily wh-initial. Both main-clause and embedded-clause questions show these reflexes. In addition, both question types feature a clitic: polar =ii and wh =u, respectively. These clitics occur only in main-clause questions, are insensitive to the Tsimshianic independent/dependent clause distinction, and are not directly associated with wh-expressions. The distribution of these elements across main and embedded questions is summarized in Table 2.

Table 2: Reflexes of question formation.

Main Embedded
WhQs wh-initial + =u wh-initial
PolQs ał + =ii dzi

This section develops a two-layer syntax associated with the formation of questions in Sm’algyax, which consists of a CP layer and a higher SAP (Speech Act Phrase) layer. All interrogative clauses (main and embedded) contain a C[+Q] element marking clause type. In Sm’algyax, this is instantiated by ał or its null variant in main-clause PolQs, and by dzi in embedded PolQs. In both main and embedded WhQs, C[+Q] is realized as a null element that triggers movement of argument wh-words to clause-initial position. I treat =ii and =u as interrogative illocutionary mood operators: morphemes linked to the conversational function of asking. Syntactically, these particles occupy a projection high in the syntactic superstructure and take an interrogative CP as their complement. Evidence for this structure comes from embedding and coordination facts. Semantically, the operator selects a sentence radical (a set of propositions) and sets it as the question under discussion (QUD). The proposed structure is given in (66).

    1. (66)

This analysis contributes to recent debates on the division between propositional content (typically associated with TP/CP) and illocutionary content, often thought to be restricted to root clauses. Sm’algyax provides evidence for the existence of overt, syntactically projected illocutionary operators in the form of interrogative clitics, and also supports the view that such operators are strictly non-embeddable (see Dayal 2023), contra approaches that permit embedded speech acts (e.g., Krifka 2014; 2023; Woods 2016).

4.1 Dayal’s (2023) three-way distinction between interrogative particles

To situate the Sm’algyax system within a broader typology of interrogative particles, I draw on the distinctions proposed in Dayal (2023).21 Dayal proposes a three-layered left periphery for interrogatives, motivated partly by the crosslinguistic distribution of interrogative particles. This architecture consists of: (i) a CP layer responsible for clause typing, (ii) a PerspP (Perspective Phrase) layer associated with introducing centered questions, and (iii) a higher SAP layer hosting speech act marking.22 These layers differ in their embeddability: CPs embed under all question-embedding predicates, PerspPs embed only under rogatives, and SAP cannot be embedded (setting aside direct quotation).23

    1. (67)
    1. The interrogative left periphery (Dayal 2023)
    2. [SAP … Not embeddable
    3.             [PerspP … Embeddable under rogative predicates
    4.                         [CP … Embeddable under all question-embedding predicates

Evidence for a CP projection comes from Japanese-type particles such as ka. Setting aside its non-interrogative uses (see Section 3.3), two defining properties of ka are: (i) embedded questions must contain it, and (ii) its presence alone is sufficient to mark the clause as interrogative. (In matrix contexts, ka/no mark questions but may be omitted; see footnote 13.)

    1. (68)
    1. Japanese (Miyagawa 2022: 174)
    1. Taroo-wa
    2. Taro-top
    1. [dare-ga
    2. who-nom
    1. kuru
    2. come
    1. *(ka)]
    2. q
    1. sitteiru.
    2. know
    1. ‘Taro knows who will come.’

Similar facts hold for Turkish and Finnish. In both languages, interrogative particles obligatorily occur in main and embedded PolQs, and their presence is sufficient to signal an interrogative interpretation (Gonzalez 2021; 2023, and references within). This is illustrated below with the Finnish PolQ marker -kO:

    1. (69)
    1. Finnish (Gonzalez 2023: 9)
    1. Jenna
    2. Jenna
    1. tietää
    2. knows
    1. [on-ko
    2. is-PolQP
    1. huomenna
    2. tomorrow
    1. hyvä
    2. good
    1. sää].
    2. weather
    1. ‘Jenna knows whether tomorrow’s weather will be nice.’

Dayal analyzes ka-type markers (along with English whether and wh-movement) as indicators of interrogative clause type in the sense of Cheng (1991).24

A second class is exemplified by Hindi-Urdu kya: (Bhatt & Dayal 2020; Dayal 2023, see also Biezma et al. 2025). Like ka, it can appear in matrix PolQs. Unlike ka, it embeds only under rogatives (e.g. ask, wonder, want to know), not under responsives (e.g., know). This restricted distribution, known as quasi-subordination (McCloskey 2006; Dayal & Grimshaw 2009; Woods 2016; Dayal 2023; Farkas & Roelofsen 2024), is illustrated below.

    1. (70)
    1. Hindi-Urdu (Dayal 2023: 9)
    1. Ti:char-ne
    2. teacher-erg
    1. anu-se
    2. Anu-ins
    1. pu:cha:
    2. asked
    1. [ki
    2. sub
    1. kya:
    2. pqp
    1. vo
    2. she
    1. ca:i
    2. tea
    1. piyegi:].
    2. will.drink
    1. ‘The teacher asked Anu if she will drink tea.’
    1. (71)
    1. Hindi-Urdu (Dayal 2023: 9)
    1. *
    2.  
    1. ravi
    2. Ravi
    1. ja:nta:
    2. knows
    1. hai
    2. aux
    1. [ki
    2. sub
    1. kya:
    2. pqp
    1. anu
    2. Anu
    1. ja:egi:].
    2. will.go
    1. Intended: ‘Ravi knows whether Anu will go.’

Similar restrictions hold in Mandarin, where neutral PolQs are marked with the sentence-final particle ma (Li & Thompson 1981; Cheng 1991; Liu & Luo To appear). Like kya:, ma-questions embed under rogatives and shifty responsives such as xiang zhidao ‘want to know,’ but not under responsives such as zhidao ‘know’ (Bhatt & Dayal 2020; Liu & Luo To appear).

    1. (72)
    1. Mandarin (Liu & Luo To appear: 3)
    1. Xiaoming
    2. Xiaoming
    1. xiang
    2. want
    1. zhidao
    2. know
    1. [ta
    2. 3sg
    1. de
    2. poss
    1. fenshu
    2. score
    1. bi
    2. than
    1. wo
    2. 1sg
    1. gao
    2. high
    1. ma].
    2. ma
    1. ‘Xiaoming wants to know whether his score is higher than mine.’
    1. (73)
    1. Mandarin (Liu & Luo To appear: 4)
    1. #
    2.  
    1. Xiaoming
    2. Xiaoming
    1. zhidao
    2. know
    1. [ta
    2. 3sg
    1. de
    2. poss
    1. fenshu
    2. score
    1. bi
    2. than
    1. wo
    2. 1sg
    1. gao
    2. high
    1. ma].
    2. ma
    1. Intended: ‘Xiaoming knows whether his score is higher than mine.’

Bhatt & Dayal (2020) and Dayal (2023) argue that these restricted embedding facts preclude a clause-typing account for kya:-type particles. Instead, kya:-type particles are placed in the intermediate PerspP projection.

The highest layer, SAP, hosts illocutionary operators associated with speech acts in addition to speech act modifiers, referred to as meta-question particles (MQPs). Japanese -kke (a “remind-me” question particle; Sauerland & Yatsushiro 2017) and English MQP quick(ly) (Dayal 2016) are placed in SAP. These particles cannot embed, as shown below:

    1. (74)
    1. Japanese
    1.  
    1. a.
    1. Namae-wa
    2. name-top
    1. nan-da-kke?
    2. what-cop-kke
    1. ‘What is your name again?’ (Sauerland & Yatsushiro 2017: 4)
    1.  
    1. b.
    1. *
    2.  
    1. Boku-wa
    2. I-top
    1. [(kimi-no)
    2. you-gen
    1. namae-ga
    2. name-nom
    1. nan-da-kke
    2. what-cop-kke
    1. (ka)]
    2. q
    1. siri-tai.
    2. know-want
    1. Intended: ‘[I used to know but now I’ve forgotten, so] I want to know your name, please remind me.’ (Dayal 2023: 8)
    1. (75)
    1. a.
    1.   Quick(ly), where did you hide the matza?
    1.  
    1. b.
    1. *Mary asked Sue [quick(ly) where she hid the matza].
    1.  
    1. c.
    1. *Mary asked Sue [quick(ly) where did she hide the matza]. (Dayal 2023: 8)

This three-way distinction is summarized in Table 3: ka-type particles occupy CP, kya:/ma-type particles occupy PerspP, and MQPs like -kke/quick(ly) occupy SAP.

Table 3: Interrogative particles three ways.

Main-clause? Quasi-embedded? Embedded? Domain
Japanese ka yes yes yes CP
Finnish -kO yes yes yes CP
Hindi-Urdu kya: yes yes no PerspP
Mandarin ma yes yes no PerspP
Japanese -kke yes no no SAP
English MQP quick(ly) yes no no SAP

Turning back to Sm’algyax, we find a clear parallel between ka-type particles and both ał/dzi in PolQs and the wh-initial requirement in WhQs. As shown in the preceding sections, all interrogatives—main and embedded—feature one of these reflexes (with ał optional in main-clause PolQs, similar to the optionality of Japanese ka/no in main clauses). I treat the obligatory wh-initial position in WhQs as the reflex of a null operator CWhQ heading interrogative CP, and propose that PolQs likewise involve an operator CPolQ in the same position, spelled out variably as ał, dzi, or ∅.

The interrogative clitics =ii and =u, by contrast, clearly fall outside this class: their inability to appear in embedded contexts rules out a clause-typing analysis.25 Likewise, they should not be treated as PerspP level particles, as they do not exhibit quasi-subordination, as shown in Sections 2.2 and 3.2. The example in (76), repeated from (32) shows that =ii cannot appear as the complement of a rogative predicate:26

    1. (76)
    1. *Yagwan güüdagan (dza) hasaganii dm yeltgin.
    1.   yagwa=n
    2.   prog =1sg.i
    1. güüdax-n
    2. ask-2sg.ii
    1. [dza
    2. irr
    1. hasax-n=ii
    2. want-2sg.ii
    1. dm
    2. prosp
    1. yeltk-n]
    2. return-2sg.ii
    1.   Intended: ‘I’m asking you if you want to go back.’ *Embedded =ii

Given these embedding facts, the interrogative clitics are best analyzed as occupying the highest layer, SAP. They are different from the other particles used to argue for SAP in Dayal, however, in that they are associated with canonical information seeking questions, rather than functioning as modifiers of illocutionary acts such as MQPs -kke and quick(ly). I therefore do not analyze Sm’algyax’s interrogative clitics as MQPs, but rather as instantiations of an illocutionary operator. In this way, while Sm’algyax doesn’t provide direct evidence for a three-layer interrogative syntax, it provides a different kind of evidence for a separate, unembeddable SAP layer.27

The remainder of this section presents an analysis of Sm’algyax’s interrogative particles within a two-layer interrogative syntax.

4.2 What happens at CP

This section breaks down the interrogative clause-type marker C[+Q] into two separate elements: CPolQ and CWhQ. Starting with PolQs, the form of a polar interrogative sentence radical is given in (77). (I address allomorph selection in Section 4.3.)

    1. (77)
    1. Polar question radical

I assume a standard Hamblin (1973) semantics for PolQs, where a PolQ denotes the set containing the positive and negative answers, {p, ¬p}. The role of CPolQ is to convert a proposition p into this bipolar set, as illustrated below for the English question ‘Did Betty leave?’.28

    1. (78)
    1. Did Betty leave?
    1. (79)
    1. a.
    1. [CPCPolQ [TP Betty left] ]
    1.  
    1. b.
    1. ⟦TP⟧ = λw.leftw(b)
    1.  
    1. c.
    1. ⟦CPolQ⟧ = λq⟨s,t⟩.λp⟨s,t⟩. [p = q˅p = ¬q]
    1.  
    1. d.
    1. ⟦CP⟧ = λp⟨s,t⟩.[p = λw.leftw(b)˅p = λw.¬leftw(b)] ⇒ {λw.leftw(b), λw.¬leftw(b)}

Wh-question radicals also feature a clause-typing element, which I label CWhQ:

    1. (80)
    1. Wh-question radical

I adopt a Hamblin-Karttunen approach to wh-questions: a wh-question denotes the set of possible answers (Hamblin 1973) and CWhQ shifts the clause from a proposition to a set of propositions; the wh-expression, an indefinite generalized quantifier, binds the variable denoted by the trace (Karttunen 1977). This is shown below for the English question ‘Who left?’.

    1. (81)
    1. Who left?
    1. (82)
    1. a.
    1. [CP whoi [C′ C[+Q] [TP ti left ] ] ]
    1.  
    1. b.
    1. ⟦TP⟧ = λw.leftw(xi)
    1.  
    1. c.
    1. ⟦CWhQ⟧ = λq⟨s,t⟩.λp⟨s,t⟩.p = q
    1.  
    1. d.
    1. ⟦CWhQ(⟦TP⟧)⟧ = λp.[p = λw.leftw(xi)]
    1.  
    1. e.
    1. ⟦CP⟧ = λp.∃x.[human(x) ᴧ p = λw.leftw(xi)
    2. ⇒ {Betty left, Mary left, Henry left, …}

Assuming that questions denote the set of true propositions (Karttunen 1977), I adopt the answerhood operator Ans-D from Dayal (1996), which applies to a set of propositions Q in a given world, and picks out the unique proposition p in Q that is true in that world.

    1. (83)
    1. Ans-D: λQ.ιp.[Q(p)∧p(w)]

The denotations in (79c) and (82c), which convert their propositional arguments p into sets of propositions Q, explain the embedding facts: rogative verbs like ‘ask’ exclusively select for complements of type Q, responsive verbs like ‘know’ variably do so, and anti-rogative verbs like ‘believe’ never do so. As pointed out in Dayal (2023), this is equivalent to treating these elements as markers of interrogative clause type. Embedded questions are schematized below. Embedded polar questions, under my analysis, proceed as follows: a question-embedding verb such as ‘ask’ or ‘know’ selects an interrogative CP headed by CPolQ (which is realized in this context as dzi).

    1. (84)

Embedded wh-questions involve a question-embedding predicate taking a wh-question radical headed by CWhQ as in (80) as its complement.

    1. (85)

I now turn to the interrogative clitics =ii and =u.

4.3 What happens at SAP

As sketched in (66), I locate the interrogative clitics in the clausal super-structure, heading a SAP projection. Like question-embedding predicates, they select for a CP complement that features a C[+Q] element (see Section 4.2). A PolQ is shown in (86) and a WhQ is shown in (87).

    1. (86)
    1. (87)

I propose that both =ii and =u are instantiations of an illocutionary mood marker SAASK. In terms of its semantics, SAASK combines with a question radical Q (a set of propositions) and returns Q only if its not-at-issue condition is satisfied—namely, that Q is set as the question under discussion (QUD; see, e.g., Roberts 2012).29 The denotation is given in (88).

    1. (88)
    1. SAASK operator
    2. ⟦SAASK⟧(Q) is defined only if Q is set as the question under discussion.
    3. If defined, ⟦SAASK⟧(Q) = Q

Two types of evidence support the structures in (86) and (87). First, as already established, interrogative clitics are strictly unembeddable. As argued for in Section 4.2, question embedding predicates select for interrogative CP complements, and not SAP complements, which explains the absence of the interrogative clitic in embedded questions. Second, in coordinated questions, a single clitic scopes over both conjuncts, as in (89) and (90). The clitic cannot c-command the second conjunct from its surface position, nor can it move there without violating the Coordinate Structure Constraint (Ross 1967). The simplest analysis is that interrogative clitics are base-generated above both conjuncts, as in (91), and are positioned inside the first conjunct post-syntactically (as argued for in Brown & Davis (2024a; b); Davis & Brown (2024); Brown (2024a)).

    1. (89)
    1. Ał kwdiinii ada su’naałan?
    1. ał
    2. q
    1. kwdii-n=ii
    2. hungry-2sg.ii =int
    1. ada
    2. and
    1. su’naał-n
    2. tired-2sg.ii
    1. ‘Are you hungry and (are you) tired?’
    1. (90)
    1. Goyu nah gabn ada naał nah int dzapt?
    1. [goo=u
    2. what=int
    1. =a
    2. =cn
    1. nah
    2. pfv
    1. gap-ə-n ____]
    2. eat-tr-2sg.ii
    1. [ada
    2. and
    1. naa
    2. who
    1. =ł
    2. =irr.cn
    1. nah
    2. pfv
    1. in=t
    2. ax =3.i
    1. dzap-t ____]
    2. make-3.ii
    1. ‘What did you eat and who made it?’
    1. (91)

The syntactic structure in (91) dovetails with the semantics of coordination: as an anonymous reviewer helpfully points out, coordinated questions freely embed, as in (92), and are straightforwardly captured under a Hamblin-Karttunen approach, as in (93), which creates a set of possible joint answers to Q1 and Q2.

    1. (92)
    1. I’m not sure [who arrived and who left].
    1. (93)
    1. λp⟨s,t⟩.∃p ∈ Q1. ∃p′ ∈ Q2. [q = p ∩ p′]

The questions in (89) and (90), are therefore best analyzed as involving coordinated question radicals, that are then selected by an illocutionary operator =ii/=u, as in (91).30

Let us turn to the question of allomorphy, starting with the CPolQ operator. Recall that embedded PolQs (such as (84)) are always introduced by the element dzi, while main-clause polar questions are never introduced by dzi, and instead feature either a clause-initial particle ał, or no additional marking ∅ (besides =ii). I treat the variation in surface forms as a case of outward contextual allomorphy (see , e.g., Bobaljik 2000). In main-clause polar questions, where SAASK selects an interrogative clausal complement, CPolQ is sensitive to the structurally adjacent SAASK element. In this configuration, CPolQ spells out alternately as ał or ∅. Elsewhere—specifically in embedded PolQs where no such adjacency obtains—CPolQ surfaces as dzi. The lexical entry for CPolQ is given in (94).

    1. (94)

Turning to the alternation between =ii and =u, both of which are analyzed as allomorphs of a single illocutionary mood head SAASK. The distribution of =ii and =u can be treated as a case of inward contextual allomorphy. Specifically, =u is conditioned by the presence of a CWhQ element heading its complement, whereas =ii appears elsewhere (i.e., in polar and alternative questions). The lexical entry for SAASK is given in (95).

    1. (95)

This captures the complementary distribution of the two interrogative clitics: =u surfaces only in the context of a WhQ, while =ii appears elsewhere (including PolQs and alternative questions). Further support for treating both interrogative clitics as allomorphs of a single head comes from Sm’algyax’s Maritime Tsimshianic sister, Sgüüxs, where both PolQs and WhQs may be marked by the same interrogative clitic =ii.31

    1. (96)
    1. Sgüüxs (Tarpent 1994: 1)
    1. Ał
    2. q
    1. mi
    2. 2sg.i
    1. wilaax
    2. know[-3.ii]
    1. =hl
    2. =cn
    1. waa-t=ii
    2. name-3.ii=int
    1. ‘Do you know her name?’
    1. (97)
    1. Sgüüxs (Tarpent 1994: 2)
    1. Nda
    2. how
    1. =ł
    2. =cn
    1. gasgoo
    2. amount[-3.ii ]
    1. =hl
    2. =cn
    1. hasaga-n=ii
    2. want-2sg=int
    1. ‘How much do you want?’

To conclude, interrogative clitics =ii and =u appear only in main-clause questions; they are absent from embedded questions and incompatible with non-interrogative uses of wh-expressions. I argued, based on this distribution, and further evidence from coordinated questions, that these elements head a SAP projection and select for an interrogative CP complement. Semantically, they set a question radical Q as the QUD. In Section 5, I present additional data from marked non-canonical question constructions that further support these conclusions.

5 Non-canonical questions

If =u and =ii are best analyzed as operators marking the speech act of asking, as opposed to functioning as clause-typing particles, we would expect them to be absent in certain interrogative clauses that exhibit a mismatch between clause-type and illocutionary mood. We already saw data that pointed to this conclusion in embedded questions (which I have argued lack illocutionary mood entirely in Sm’algyax). That is, in an embedded question such as ‘I know who left’, who left is interrogative in terms of its semantics (i.e. it denotes a set of propositions), but it does not have the conversational function associated with canonical, information-seeking questions. It follows that the Sm’algyax equivalent lacks the interrogative clitic =u:

    1. (98)
    1. a.
    1. Wilaayu naa ksi dawłit.
    1. wilaay-u
    2. know-1sg.ii
    1. =a
    2. =cn
    1. [naa
    2. who
    1. =a
    2. =cn
    1. ksi=
    2. out=
    1. dawł-ət]
    2. leave-sx
    1. ‘I know who left.’
    1.  
    1. b.
    1. *Wilaayu naayu ksi dawłit.
    1.   wilaay-u
    2.   know-1sg.ii
    1. =a
    2. =cn
    1. [naa=u
    2. who=int
    1. =a
    2. =cn
    1. ksi=
    2. out=
    1. dawł-ət]
    2. leave-sx
    1.   Intended: ‘I know who left.’

We also predict that certain flavours of non-canonical question should prohibit the appearance of the interrogative clitics. This prediction is borne out. After providing a quick background to non-canonical questions, I show that Sm’algyax possesses a number of grammaticalized non-canonical question types, and that those that do not expect or require a response from an addressee also lack an interrogative particle.

5.1 Default assumptions accompanying question acts

I outline the empirical facts below with reference to Farkas’s (2022) default assumptions accompanying question acts, which are inspired by traditional speech act theory (Searle 1969):

    1. (99)
    1. Default assumptions accompanying question acts (Farkas 2022)
    1.  
    1. a.
    1. Speaker ignorance: The speaker’s epistemic state is neutral relative to the possible resolutions of the issue she raises.
    1.  
    1. b.
    1. Addressee competence: The speaker assumes that the addressee knows the information that settles the issue she raises.
    1.  
    1. c.
    1. Addressee compliance: The speaker assumes that the addressee will provide this information in the immediate future of the conversation as a result of the speaker’s speech act.
    1.  
    1. d.
    1. Issue resolution goal: It is assumed that the main aim the speaker pursues when raising an issue is to have it resolved in the immediate future of the conversation.

According to Farkas, a canonical, information-seeking question is one in which all the assumptions in (99) hold. However, these assumptions may be weakened or suspended, resulting in different types of non-canonical question (e.g. rhetorical questions, biased questions, etc.).

Sm’algyax exhibits three types of grammaticalized non-canonical question: rhetorical questions, conjectural questions, and biased questions.32 In the subsections that follow, I discuss these flavours of non-canonical question, with reference to the distribution of interrogative clitics.

5.2 Rhetorical questions

Rhetorical questions (RQs) can be informally characterized as interrogative sentences that have the feeling of an assertion; that is, the speaker does not expect an answer, though the interlocutor may still provide one (Sadock 1974; Han 2002; Caponigro & Sprouse 2007; Biezma & Rawlins 2017: a.o.). Although English RQs are not formally distinct from information-seeking questions (Caponigro & Sprouse 2007), a diagnostic for RQs is their ability to license “strong” negative polarity items (NPIs), such as minimizers (‘budge an inch’, ‘lift a finger’), which are degraded in information-seeking questions. The examples in (100) and (101) illustrate this contrast. In (100), the rhetorical question licenses the strong NPI ‘lift a finger’ and can be paraphrased as a negative existential: ‘(After all,) nobody helped me’ (Ladusaw 1980). (101) shows that the same NPI is not licensed in information-seeking questions.

    1. (100)
    1.   (After all,) who lifted a finger to help me? Rhetorical question
    1. (101)
    1. #(I’m curious,) who lifted a finger to help you? Information-seeking question

In terms of the default assumptions given in (99) above, a RQ is one in which Speaker ignorance and the Issue resolution goal are suspended—that is, the Speaker and Addressee are both expected to know the answer (by Addressee competence together with the suspension of Speaker ignorance), and the communicative function of uttering a rhetorical question is not one of asking (due to the suspension of the Issue resolution goal).

Sm’algyax possesses a dedicated RQ construction, as discussed in Brown (2023). Like information-seeking questions, this construction features a clause-initial wh-expression, but it is further distinguished by the presence of the C element dzi—also found in embedded PolQs—and by the obligatory absence of the interrogative clitic =u (103).33

    1. (102)
    1. [Allie is talking to Ben about a problem that cannot be fixed. She says:]
    2. Goo dzi wila waali. Nah baaltu txa’nii goo…
    1. goo
    2. what
    1. dzi
    2. irr
    1. wila
    2. manr
    1. waal-i
    2. lv-irr.1sg.ii
    1. nah
    2. pfv
    1. baal-t-ə-u
    2. try-t-tr-1sg.ii
    1. =a
    2. =cn
    1. txa’nii
    2. every
    1. goo
    2. what
    1. ‘What can I (even) do. I’ve tried everything.’ dzi-RQ [– u]
    1. (103)
    1. *Goyu dzi wila waali?
    1.   goo=u
    2.   what=int
    1. dzi
    2. irr
    1. wila
    2. manr
    1. waal-i
    2. lv-irr.1sg.ii
    1.   Intended: ‘What can I (even) do.’ dzi-RQ [– u]

This RQ construction occurs frequently in narrative contexts, and is variably translated either as a wh-question or as a declarative sentence with a negative existential element such as ‘nowhere’, or ‘nothing’. This construction is always associated with what I refer to as a “negative implication”: the implication that the speaker believes that the answer corresponds to a negative existential.34 In these narrative contexts, they are never answered.

Below, we see examples of marked rhetorical questions occurring in texts, with the narrative context indicating that the negative implication is present. In (104a), the narrator utters the marked rhetorical question “Therefore what then could he use now?” after making it clear that Asdiwaal has nothing to use to get himself out of the situation he is in, while in (104b) we see the marked rhetorical question “Where could he go?” preceding expository material asserting that there is in fact nowhere to go.

    1. (104)
    1. [Asdiwaal carries several magical tools that have helped him out of difficult situations. However, this time he is stranded on a mountain in a storm without his magical tools:]
    1.  
    1. a.
    1. Gan goo dzi gik hoyt gya’wn?
    1. gan
    2. reas
    1. goo
    2. what
    1. dzi
    2. irr
    1. gik
    2. again
    1. hoy-t
    2. use-3.ii
    1. gya’wn
    2. now
    1. ‘Therefore what then could he use now?’
    1.  
    1. b.
    1. Ndaa dzi yaakit? Man duulxgit, ada tgi duulxgit…
    1. ndaa
    2. where
    1. dzi
    2. irr
    1. yaak-t
    2. go-3.ii
    1. man
    2. up
    1. duulxk-t,
    2. stuck-3.ii
    1. ada
    2. and
    1. tgi
    2. down
    1. duulxk-t
    2. stuck-3.ii
    1. ‘Where could he go? He could not go up, he could not go down…”
    2. (Boas 1912: The Story of Asdiwaal; 144–145)

Shortly after the narrator poses these rhetorical questions, Asdiwaal dies on the mountain.

These data show that Sm’algyax possesses a dedicated RQ construction, and that the interrogative clitic never appears in it. According to Farkas (2022), RQs are characterized by the suspension of Speaker ignorance and the Issue resolution goal: perhaps the presence of =u is associated with one or both of these pragmatic assumptions. In the next section I provide evidence from conjectural questions that suggests that the presence of =u is associated with the Issue resolution goal.

5.3 Conjectural questions

Conjectural Questions (CQs), like RQs, are a flavour of non-canonical question that has the form of an interrogative but the feel of an assertion (See e.g. Faller 2002; Littell et al. 2010; Matthewson 2010; Eckardt & Beltrama 2019; Eckardt 2020; Farkas 2022). Like RQs, they do not require or expect an answer. They are set apart from RQs, however, in that they express the Speaker’s curiosity/ignorance about the interrogative prejacent Q. They are often translated as ‘I wonder Q’ or ‘I don’t know Q’. We see an example of a CQ in German below. The CQ in (105) is characterized by the presence of the particle wohl, appearing in an interrogative sentence with verb-final syntax.

    1. (105)
    1. German (Eckardt 2020: 2)
    1. Wo
    2. where
    1. wohl
    2. wohl
    1. der
    2. the
    1. Schlüssel
    2. key
    1. ist?
    2. is
    1. ‘Where might the key be, I wonder.’

For Farkas (2022), a conjectural question is one in which Addressee competence and the Issue resolution goal are suspended—the Speaker and the Addressee are thus both not expected to know the information that settles the issue, and the goal of uttering a CQ is not one of immediately resolving the issue.

Sm’algyax possesses a dedicated CQ construction that is characterized by the presence of the epistemic particle =sn/=si’in (a second-position clitic) in a sentence with interrogative syntax (i.e. a fronted wh-expression and extraction morphology). In (106), we see a declarative sentence marked with si’in translated to English as might. In (107), we see a CQ that is uttered in a context where the speaker (Mary) does not assume that the Addressee (Paul) is able to resolve the question under discussion.35

    1. (106)
    1. [There are some dark clouds in the sky]
    2. Dm yaasi’inł waas.
    1. dm
    2. prosp
    1. yaa=si’in
    2. walk=epis
    1. =ł
    2. =irr.cn
    1. waas
    2. rain
    1. ‘It might rain.’ Conjectural/epistemic =sn/=si’n
    1. (107)
    1. [There is a knock on the door in the middle of the night. Mary has no idea who it could be, and doesn’t expect Paul to know either. Mary says to Paul: (adapted from Farkas 2022)]
    2. Naasi’in gwii a taaym gwa’a?
    1. naa=si’in
    2. who=epis
    1. =a
    2. =cn
    1. gwii
    2. dem
    1. a
    2. prep[-3.ii]
    1. =a
    2. =cn
    1. taaym
    2. time
    1. gwa’a?
    2. dem
    1. ‘Who could it be at this time, I wonder.’ CQ [– u]

Like the RQ construction in the previous section, the interrogative clitics do not appear in CQs.36

    1. (108)
    1. *Naasnu t’aadit a gwii?
    1. naa=sn=u
    2. who=epis =int
    1. =a
    2. =cn
    1. t’aat-ət
    2. sit-sx
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. gwii
    2. dem
    1. Intended: ‘Who sat here, I wonder.’ CQ [– u]

Further evidence for the dissociation between CQs and the notion of asking come from the following examples. In (109) we see that a CQ may be used as a reply to an ordinary question to indicate that the Speaker of the CQ does not know the answer. Here it is obvious that the Speaker of the information-seeking question (109a) doesn’t know the answer either, and is not expected to respond.

    1. (109)
    1. a.
    1. Naał int gapdu txa’nii maay?
    1. naa
    2. who
    1. =ł
    2. =cn
    1. in=t
    2. ax =3
    1. gap-t=u
    2. eat-3.ii=int
    1. =a
    2. =cn
    1. txa’nii
    2. all
    1. maay?
    2. berries
    1. ‘Who ate all the berries?’
    1.  
    1. b.
    1. Naasi’in (int gapt)…
    1. naa=si’in
    2. who=epis
    1. =a
    2. =cn
    1. in=t
    2. ax =3.i
    1. gap-t
    2. eat-3.ii
    1. ‘I wonder who (ate them)…’ / ‘I don’t know who.’ CQ as response [– u]

CQs may also function as self-addressed questions:

    1. (110)
    1. [You’re home alone and you can’t find your keys. You say to yourself:]
    2. Ndesn nahak'a’ayu.
    1. ndeh=sn
    2. where=epis
    1. =a
    2. =cn
    1. na-hak'a’a-u
    2. poss-key-1sg.ii
    1. ‘I wonder where my keys are.’ Self-addressed CQ [– u]
    1. (111)
    1. [You have a missed call from a mysterious phone number. You ask yourself:]
    2. Naasn naht in si’is huutgu.
    1. naa=sn
    2. who=epis
    1. =a
    2. =cn
    1. nah=t
    2. pfv =3.i
    1. in
    2. ax
    1. si’is
    2. try
    1. huutk-u
    2. call-1sg.ii
    1. ‘I wonder who called me.’ Self-addressed CQ [– u]

I conclude that Sm’algyax has a dedicated CQ construction, that, according to the characterization in Farkas (2022), arises due to the suspending of the assumptions of Addressee competence and the Issue resolution goal. Given that RQs and CQs share in common the suspension of the Issue resolution goal, as well as the absence of an interrogative clitic, I suggest that interrogative clitics are associated with the notion of Issue resolution. This is compatible with the analysis proposed for SAASK in Section 4.3.

5.4 Biased questions

Sm’algyax has two dedicated constructions associated with biased polar questions. The first type features the negative response particle ayn, and the second has the form of a tag question, formed with a pronoun ’niit.

    1. (112)
    1. [There’s a feast in early May; Lucy tells you she’s planning to attend it. This surprises you because she previously mentioned she would be in Vancouver from mid-April to mid-May. You ask her:]
    2. Aynł dm waanii a Ts’a’mas?
    1. ayn
    2. no
    1. =ł
    2. =irr.cn
    1. dm
    2. prosp
    1. waan-n=ii
    2. lv-2sg.ii =int
    1. a
    2. prep [-3.ii]
    1. =a
    2. =cn
    1. Ts’a’mas
    2. Vancouver
    1. ‘Won’t you be in Vancouver?’ Ayn question [+ ii]
    1. (113)
    1. [Michael told Lucy about a surprise party for a friend and asked her to keep it secret. Knowing she struggles with secrets, he asks her the next day]
    2. Akadiim małit as ’niit, ’niidii?
    1. aka=dii=m
    2. neg =foc =2sg.i
    1. mał-ə-t
    2. tell-t-3.ii
    1. a
    2. prep [-3.ii]
    1. =s
    2. =pn
    1. ’niit
    2. 3.iii
    1. ’niit=ii
    2. 3.iii =int
    1. You didn’t tell him, did you? ’Niidii-tag question [+ ii]

Note that both of these non-canonical question types feature an interrogative clitic, contra the RQ and CQ constructions described above. This is not especially concerning, if the analytical goal is to link the presence of the interrogative clitics to information-seeking questions. For Farkas, a biased question arises via the weakening of Speaker ignorance: the Speaker’s epistemic state is not neutral relative to the possible resolutions of the issue. Unlike RQs and CQs, the Issue resolution goal is still assumed, and an answer is therefore expected.

6 Conclusion

This paper examined the morphosyntactic reflexes of polar and wh-question formation in Sm’algyax. I argued that main-clause questions exhibit a bipartite structure marked by both a clause-typing element (ał/wh-initiality) and an interrogative clitic (=ii/=u), whereas embedded questions only feature the clause-typing element (dzi/wh-initiality). This distribution motivates a two-layer interrogative syntax: all questions feature a CP headed by C[+Q], while canonical main-clause questions additionally feature a higher speech act layer (SAP), the locus of interrogative clitics. This account explains both the main-clause restriction of the interrogative clitics and their association with the speech act of asking. Evidence from embedding, coordination, and non-canonical questions further supports this analysis. Taken together, the Sm’algyax facts provide clear evidence for an unembeddable speech act layer and contribute to ongoing debates about the architecture of the interrogative left periphery and the typology of interrogative particles.

Future work should investigate the extent to which Sm’algyax exhibits effects comparable to quasi-subordination and identify the formal properties of such structures. One potential avenue concerns the grammar of quotation. As illustrated in (114), while English SAP-level elements such as the MQP quick(ly) cannot straightforwardly appear in embedded questions, they may appear within quoted speech embedded under predicates such as ask, which, in addition to being a rogative predicate, can also select interrogative quotative complements (Grimshaw 2012; Dayal 2023).37

    1. (114)
    1. a.
    1. She asked (*quick(ly),) who left.
    1.  
    1.   b.
    1.   She asked, “quick(ly), who left?”

Preliminary investigation reveals comparable effects in Sm’algyax: interrogative clitics are permitted in embedded contexts only when the complement is interpreted as a quotation. In (115b), a prosodic break following güüdaga’nut Klalens ‘Clarence is asking me,’ and the shift in person marking (first person -u in the non-quoted complement in (115a) vs. second person -n in the quotation in (115b)), support this analysis.

    1. (115)
    1. [Clarence phones you from the cafe, you report what he’s saying to your friend who is there with you]
    1.  
    1. a.
    1. Güüdaga’nut Klalens goo(*yu) dm hasagu.
    1. güüdax-’nu
    2. ask-1sg.iii
    1. =t
    2. =pn
    1. Klalens
    2. Clarence
    1. [goo(*=u)
    2. what
    1. =a
    2. =cn
    1. dm
    2. prosp
    1. hasax-u]
    2. want-1sg.ii
    1. ‘Clarence is asking me what I want.’
    1.  
    1. b.
    1. Güüdaga’nut Klalens, “gooyu dm hasagn?”
    1. güüdax-’nu
    2. ask-1sg.iii
    1. =t
    2. =pn
    1. Klalens
    2. Clarence
    1. [goo=u
    2. what=int
    1. =a
    2. =cn
    1. dm
    2. prosp
    1. hasax-n]
    2. want-2sg.ii
    1. ‘Clarence is asking me, “what do you want?”’

A related question concerns the Sm’algyax quotative particle daya. Data from texts and preliminary elicitation show that it can occur with a range of complements, including exclamations such as hobiyee ‘hurrah’ in (116) and jussive clauses featuring the prohibitive particle giloo in (117). Whether daya can likewise select interrogative quotative complements remains to be determined.

    1. (116)
    1. “Hobiyee!” daya gyet wil g̱a̱lmiilga na ts’apt.
    1. hobiyee
    2. hurrah
    1. daya
    2. quot
    1. gyet
    2. people
    1. wil
    2. comp
    1. galmiilk
    2. play[-3.ii ]
    1. =a
    2. =cn
    1. na
    2. poss
    1. ts’ap-t
    2. team-3.ii
    1. ‘The people said “hurrah!” where their team was playing.’ (SLLTD 2017)
    1. (117)
    1. “Giloom gapt” daya.
    1. giloo=m
    2. prohib =2pl.ii
    1. gap-t
    2. eat-3.ii
    1. daya
    2. quot
    1. ‘He said “don’t eat it.”’

Finally, the embeddability of other sentential mood markers and speaker-oriented elements—such as evidential enclitics (reportative gat, conjectural/epistemic sn/si’in, prior evidence gn/gi’in) and jussive mood markers, including the hortative laan, prohibitive giloo, and the zero-marked imperative—merits closer examination, particularly for what their distribution may reveal about the architecture of the Sm’algyax clausal periphery.

Abbreviations

Glosses not following the Leipzig Glossing Conventions are as follows:

ax – “Agent extraction morpheme”; attr – “Attributive”; cn – “Common noun connective”; dwid – “Domain widener”; epis – “Epistemic modal”; exist – “Existential”; i – “Series I clitic”; ii – “Series II suffix”; iii – “Series III pronoun”; incep – “Inceptive aspect”; int – “Interrogative mood”; l.o.c. – “Lack of committment” lv – “Light verb”; manr – “Manner subordinator”; pn – “Proper noun connective”; prep – “Preposition”; prohib – “Prohibitive mood”; prosp – “Prospective aspect”; reas – “Reason subordinator”; spt - “Spaciotemporal particle” sx – “Subject extraction morpheme”; t – “T voice suffix”; ver – “Verum”.

Acknowledgements

First and foremost, I thank Velna Nelson, Ellen Mason, and Beatrice Robinson for sharing their language with me. T’oyaxsut nüüsm! I am grateful to Yael Sharvit, Harold Torrence, Ethan Poole, Lisa Matthewson, Henry Davis, Margaret Anderson, Hilda Koopman, and Jessica Rett for their guidance and support during the development of this project, and audiences at CUSP 2019, LSA 2020, WSCLA 2023, the UBC Q-Lab, and the UBC Gitksan Research Lab for feedback. Finally, I thank three anonymous reviewers for detailed comments that substantially improved the paper, and Associate Editor Shin Fukuda for overseeing the review process. This work was supported by SSHRC Doctoral and Postdoctoral Fellowships. All errors are my own.

Competing interests

The author has no competing interests to declare.

Notes

  1. Embedded inversion is possible under responsives in certain conditions. For example, while know doesn’t license embedded inversion, want to know and not know can. These predicates are dubbed “shifty responsives” in Dayal (2023). [^]
  2. Differences between the surface and underlying forms in (8) reflect two pervasive morphophonological processes. The first involves the determiner-like connectives (glossed as ‘cn’ for common-noun connective or ‘pn’ for proper-noun connective in the Tsimshianic literature), which are syntactically associated with the nominal to their right but encliticize to the left. The common-noun connective =a deletes after vocalic or sonorant segments (Anderson & Ignace 2008). The second process involves deletion of the underlying third-person agreement suffix -t when immediately followed by the DP it agrees with (Davis 2018). Throughout this paper, t -deletion environments are indicated in the third line of glossed examples by square brackets around the third-person suffix: [- 3.ii]. [^]
  3. The underlying forms in examples in (9) and (10) show that the transitive suffix -ə, appears in independent clauses but is absent in dependent clauses. However, this distinction is not immediately obvious in the surface forms (compare with (11a) and (11b) where this distinction is obvious). This surface opacity results from a morphophonological process: when the underlying transitive suffix -ə is followed by a vocalic or sonorant segment, such as the first-person suffixal agreement marker -u, the transitive suffix is deleted. For a detailed account of the morphophonology of the transitive suffix, see Brown et al. (2020). [^]
  4. Both =ii and =u, along with their allomorphs, surface in clause-internal and clause-final positions, encliticizing either to the predicate or an argument DP, or alternatively to left-peripheral focused XPs and wh-phrases. With respect to linearization, Sm’algyax interrogative clitics differ from those in the rest of the Tsimshianic family: in Gitksan (Rigsby 1986; Hill & Matthewson 2025; Matthewson to appear), Nisga’a (Tarpent 1987), and Sgüüxs (Tarpent 1994), interrogative clitics always surface in clause-final position. [^]
  5. As shown in (15a), (15b), and (16), ał functions as a dependent marker. That is, ał consistently triggers dependent clause morphology, diagnosed by (i) the use of Series I ergative clitics (e.g., =t in (15a)) and (ii) the obligatory absence of the transitive suffix -ə. These examples also show that =ii freely appears in both independent and dependent clauses; I conclude from this that the Tsimshianic-internal clause-type distinction is orthogonal to the distribution of the PolQ clitic =ii. [^]
  6. It is noteworthy that either ał or =ii (but not both) may be omitted from a main-clause PolQ. I suggest that this optionality arises because the presence of either element is sufficient to signal a question interpretation. A similar pattern occurs in main-clause WhQs, where the wh-clitic =u can be dropped, since the clause-initial wh-word alone is sufficient to mark the clause as a question (see Section 3.1). [^]
  7. This claim is consistent with Tarpent (1987: 149) on the Interior Tsimshianic language Nisga’a. Shamei (2019) shows that the Gitksan polar interrogative clitic =aa is lexically encoded with a rising pitch. [^]
  8. These examples are better classified as polar alternative questions, in that the alternatives are a positive and negative counterpart. [^]
  9. Matthewson (to appear) and Aonuki (to appear) show that embedded PolQs in Gitksan likewise must be introduced by the cognate element ji. [^]
  10. Lahiri’s rogative/responsive distinction subsumes the classification for question embedding verbs in Karttunen (1977) and Groenendijk & Stokhof (1984) (what the latter refers to as intensional and extensional complement embedding verbs). [^]
  11. An anonymous reviewer asks whether ał and dzi must immediately precede aka. In my elicited data, they always do. Future elicitation should test whether other high clausal elements that also precede aka—such as the verum marker ap—may intervene between ał/dzi and aka. I additionally note here that dzi never co-occurs with the complementizer wil. [^]
  12. I thank an anonymous reviewer for highlighting the relevance of the absence of a double negation reading here. [^]
  13. This treatment is similar in spirit to Krifka (2017: 382), who suggests that the English question operator whether spells out as an overt complementizer in embedded clauses but as a null form in main clauses. A comparable pattern is found in Japanese: in embedded clauses, questions must be marked by the interrogative particle ka, while in main clauses, where questions are additionally marked by main-clause intonation, the interrogative particle ka/no may be omitted (see, e.g., Miyagawa 2022: 173–176 and references within). Sm’algyax differs from these languages in that the main-clause form is realized as a distinct allomorph ał, which is not droppable alongside question intonation, but alongside the presence of the interrogative clitic =ii. [^]
  14. An allomorph of the particle =u is =du:
      1. (i)
      1. Naadu baat?
      1. naa=u
      2. who=int
      1. =a
      2. =cn
      1. baa-ət ___
      2. run-sx
      1. ‘Who ran?’
      2. Consultant’s comment: “Same as Naayu baat.”
    The conditions affecting allomorph selection are described in Brown & Davis (2024b). [^]
  15. Sm’algyax exhibits two wh-question formation strategies, both of which can be characterized as rigidly wh-initial (Brown & Davis 2024b, see also Davis & J. Brown 2011; Davis & Nederveen 2021 on Gitksan). The first involves movement of a wh-argument, while the second employs a clause-initial predicative wh-word that selects a nominal argument and, by hypothesis, does not involve movement. Both constructions feature the wh-clitic =u. In this paper, I set aside the predicative wh-cases, but note that Richards (2021) accounts for similar facts in Tagalog by appealing to a linear-adjacency requirement between wh-phrases and interrogative C; linear-adjacency can be satisfied either through wh-movement of wh-arguments or base generation of wh-predicates. [^]
  16. O-argument WhQs differ from A-argument WhQs with respect to the Tsimshianic-internal independent/dependent-clause distinction. O-argument WhQs exhibit independent clause morphosyntax, while A-argument WhQs exhibit dependent clause morphosyntax (see Section 1.2). As (43) and (44) show, both configurations allow the wh-clitic =u; that is, like =ii (described in footnote 5), =u is insensitive to the independent/dependent-clause distinction. [^]
  17. In each wh-configuration, the left-peripheral connective =a/=ł typically associated with extraction does not co-occur with =u. The presence of this expected connective is also variably influenced by other mood and interrogative clitics, such as =ii (polar question), =sn (conjectural evidential), and the Series I person-marking clitics. While I set this issue aside here, I posit that there is a morphologically conditioned deletion process affecting the distribution of connectives in some clitic sequences. [^]
  18. Examples (53) and (54) feature predicative wh-words and are presented here without a proposed gap. I address the distinction between wh-movement cases and wh-predication cases in footnote 15. [^]
  19. A prominent family of analyses for Q-particles is couched within a two-tier alternative semantics (Rooth 1985; 1992; Beck 2006), under which wh-words have a focus-semantic value but lack an ordinary semantic value. The role of the Q-particle in this kind of analysis is to convert its argument’s focus-semantic value into an ordinary semantic value (Kotek 2014; 2019; Uegaki 2018). However, Q-particles such as ka have also been analyzed using Hamblin–Karttunen semantics, without adopting a two-tier semantics (e.g., Von Stechow 1996; Dayal 2023). [^]
  20. The exclamative sentence in (65) has a distinct syntax from WhQs. Attempts to elicit a WhQ equivalent by adding the wh-clitic =u are corrected to a form with the manner complementizer wila, often truncated to la:
      1. (i)
      1. Goł la waalda?
      1. goo
      2. what
      1. =ł
      2. =irr.cn
      1. wila
      2. manr
      1. waal-t=a
      2. be-3.ii =int
      1. ‘What happened?’
    Note that this WhQ also exhibits the allomorph =a, which surfaces whenever =u appears in clause-final position. [^]
  21. See also Gonzalez (2021; 2023), who builds upon this typology. [^]
  22. Centering “impl[ies] the presence of an individual who is potentially interested in obtaining the information conveyed by the question nucleus, the core proposition in the interrogative” (Dayal 2023: 4). [^]
  23. See Footnote 1 on so-called shifty responsives (Dayal 2023). [^]
  24. Gonzalez (2021; 2023) distinguish Finnish/Turkish particles from ka-type markers due to additional semantic effects, including focus effects, and an existential presupposition. For present purposes, I set these extra semantic effects aside and treat Finnish/Turkish particles as C-domain reflexes comparable to Japanese ka. [^]
  25. Further evidence that the interrogative clitics are not simply complementizers that are restricted to main clauses comes from their absence in (main clause) non-canonical questions, described in Section 5. [^]
  26. Other Tsimshianic languages show the same restriction, as illustrated in the following Gitksan examples:
      1. (i)
      1. Gitksan (Matthewson to appear: ex. 16)
      1.  
      1. a.
      1.   Gidax–a–t
      2.   ask–tr –3.ii
      1. ’nii’y
      2. 1sg.iii
      1. ji
      2. [irr
      1. wis(#=aa).
      2. rain(#=l.o.c.)]
      1.   ‘She asked me if it’s raining.’
      1.  
      1. b.
      1. #Gidax–a–t
      2.   ask–tr–3.ii
      1. ’nii’y
      2. 1sg.iii
      1. nee=hl
      2. [neg =cn
      1. wis(=aa).
      2. rain(=l.o.c.)]
      1.   ‘She asked me if it’s raining.’ Consultant’s comment on (ib) with aa: “In quotes.”
    Note that Matthewson glosses Gitksan’s polar interrogative clitic as “l.o.c.”, for Lack of Commitment. [^]
  27. Dayal (2023) provides additional evidence for SAP from (embedded) questions with disjunction, as well as from various intonational effects associated with questions and biased questions that are located at the level of both PerspP and SAP. [^]
  28. I steer clear of the ongoing debate in the literature about whether PolQs denote bipolar sets, monopolar sets, or allow both options. Monopolar approaches (e.g., Bolinger 1978; Biezma & Rawlins 2012; Roberts 2012; Roelofsen & Farkas 2015; Dayal 2023) treat PolQs as denoting singleton sets, in contrast to the bipolar Hamblin semantics adopted here for simplicity. Mixed approaches (e.g., Krifka 2015) argue that some PolQs are bipolar and some are monopolar. This latter option is adopted by Matthewson (to appear) for Gitksan, based on structural distinctions between neutral (analyzed as bipolar) and biased (analyzed as monopolar) PolQs. [^]
  29. This approach is comparable to the question operator proposed by Farkas & Bruce (2010: 10), which takes a sentence radical as its argument and adds the corresponding set of propositions to the stack of questions/issues under discussion. A related idea is found in Büring (2016), where the polar interrogative operator INT—realized as a high boundary tone in English—takes a proposition-denoting sentence radical, converts it into a bipolar set, and sets it as the current QUD. In this sense, Büring’s INT functions simultaneously as a clause-typing particle and an illocutionary operator, whereas the present analysis maintains a bifurcation between these roles. Finally, this proposal is also comparable to Matthewson’s (to appear) analysis of Gitksan’s PolQ clitic =aa, which is argued to encode a presupposition that the speaker lacks commitment to some proposition (by default the prejacent) that is relevant to the QUD. [^]
  30. This analysis, if correct, has implications for analyses of coordinated questions as involving coordinated illocutionary acts, as argued in Krifka (2001) and in particular Hirsch (2017). Hirsch (2017) proposes an inflexible semantics for coordinating connectives such as and that requires conjoined questions to be analyzed as separate illocutionary acts. The ability for a single instance of =ii/=u to scope over coordinated questions suggests, however, that this coordination is happening at the CP level. [^]
  31. As can be seen in these examples, and echoing the point made in footnote 4, Sgüüxs’ interrogative clitic =ii is a final-position clitic: it always surfaces in a right-peripheral position in the clause. [^]
  32. By grammaticalized, I mean that there are dedicated syntactic constructions that correspond to these non-canonical question interpretations. [^]
  33. A full explanation of the role of dzi in wh-rhetorical questions remains beyond the scope of this paper. By way of an initial hypothesis, the semantics for dzi proposed in Section 4.2 introduces a bipolar set denotation containing p and its complement. It is possible that the effect of such an element heading what would normally be the presupposed clause of a canonical information-seeking WhQ contributes to the non-canonical meaning observed in dzi RQs. [^]
  34. Rohde (2006) and Caponigro & Sprouse (2007) show that not all RQs possess such a negative implication (though the one in (100) does). [^]
  35. As with the dedicated RQ construction discussed in the previous section, there is no dedicated CQ construction for polar questions. Instead, a periphrastic wh-question construction is used:
      1. (i)
      1. a.
      1. Waasisn.
      1. waas=sn
      2. rain=epis
      1. ‘It might/must be raining’
      1.  
      1. b.
      1. Waasii?
      1. waas=ii
      2. rain=int
      1. ‘Is it raining?’
      1.  
      1. c.
      1. *waas=sn=ii / *waas=ii=sn
      1. (ii)
      1. Ndesi’inł waal dzi waas.
      1. nde=si’in
      2. wh=epis
      1. =ł
      2. =cn
      1. waal
      2. do/be
      1. dzi
      2. irr
      1. waas.
      2. rain
      1. ‘I wonder if it’s raining.’ Lit: How might it be if it rains?
    [^]
  36. One Ts’msyen elder I work with does however allow epistemic =sn to appear alongside =u, specifically in cases that appear to be information-seeking (corresponding to wh- do you think/assume).
      1. (i)
      1. [My brother and I are hiking through the woods in bear country and see animal tracks. Fortunately, my brother recently completed a wilderness class and is in a better position to judge. I ask him: (adapted from Korotkova 2016)]
      2. Goosnu galksa dawłit a gwa’a?
      1. goo=sn=u
      2. what=epis =int
      1. =a
      2. =cn
      1. galksa
      2. through
      1. dawł-ət
      2. pass-sx
      1. a
      2. prep [-3.ii]
      1. =a
      2. =cn
      1. gwa’a?
      2. dem
      1. ‘What passed through here, do you assume?’
    [^]
  37. See Dayal (2023: 6) for similar examples with English declarative questions, which likewise are only embeddable as quotations. [^]

References

Anderson, Margaret Seguin & Ignace, Marianne. 2008. Visible grammar: Twenty user friendly grammar modules for Sm’algyax. Prince Rupert, BC: Wap Sigatgyet, School District 52.

Aonuki, Yurika. to appear. ‘Irrealis’ ji in declarative and interrogative complements in Gitksan. In Proceedings of the 43rd West Coast Conference on Formal Linguistics.

Beck, Sigrid. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14(1). 1–56. DOI:  http://doi.org/10.1007/s11050-005-4532-y

Bhatt, Rajesh & Dayal, Veneeta. 2020. Polar question particles: Hindi-Urdu kya. Natural Language & Linguistic Theory 38(4). 1115–1144. DOI:  http://doi.org/10.1007/s11049-020-09464-0

Biezma, María & Butt, Miriam & Jabeen, Farhat & Mumtaz, Benazir. 2025. Urdu/Hindi polar kya as an expression of uncertainty. Natural Language & Linguistic Theory 43(3). 1417–1468. DOI:  http://doi.org/10.1007/s11049-024-09642-4

Biezma, María & Rawlins, Kyle. 2012. Responding to alternative and polar questions. Linguistics and Philosophy 35(5). 361–406. DOI:  http://doi.org/10.1007/s10988-012-9123-z

Biezma, Maria & Rawlins, Kyle. 2017. Rhetorical questions: Severing asking from questioning. Semantics and Linguistic Theory 27. 302. DOI:  http://doi.org/10.3765/salt.v27i0.4155

Boas, Franz. 1912. Tsimshian texts (new series). In Publications of the American Ethnological Society, vol. 3, 5–244. Leyden: Late E.J. Brill.

Bobaljik, Jonathan David. 2000. The ins and outs of contextual allomorphy. University of Maryland Working Papers in Linguistics 10. 35–71.

Bolinger, Dwight. 1978. Yes-no questions are not alternative questions. In Hiż, Henry (ed.), Questions, 87–105. Dordrecht: Springer Netherlands. DOI:  http://doi.org/10.1007/978-94-009-9509-3_3

Brown, Colin. 2021. A preliminary investigation of Sm’algyax intonation.

Brown, Colin. 2023. Irrealis-marked interrogatives as rhetorical questions. Open Linguistics 9(1). 20220239. DOI:  http://doi.org/10.1515/opli-2022-0239

Brown, Colin. 2024a. The form and function of interrogatives in Sm’algyax. Los Angeles, CA: University of California dissertation.

Brown, Colin. 2024b. Questions and their relatives in Sm’algyax. International Journal of American Linguistics 90(3). 277–326. DOI:  http://doi.org/10.1086/730303

Brown, Colin & Davis, Henry. 2024a. A second-last position clitic in Sm’algyax: A puzzle. In Bochnak, M. Ryan & Csipak, Eva & Matthewson, Lisa & Morzycki, Marcin & Reisinger, Daniel K. E. (eds.), The title of this volume is shorter than its contributions are allowed to be: Papers in honour of Hotze Rullmann, 79–94. Vancouver, BC: UBC Occasional Papers in Linguistics.

Brown, Colin & Davis, Henry. 2024b. A second-last position clitic in Sm’algyax (Coast Tsimshian). In Phadnis, Shaunak & Spellerberg, Carla & Wilkinson, Brynne (eds.), Proceedings of the fifty-forth annual meeting of the North East Linguistic Society, vol. 1, 89–102. Amherst, MA: GLSA.

Brown, Colin & Forbes, Clarissa & Schwan, Michael David. 2020. Clause-type, transitivity, and the transitive vowel in Tsimshianic. In Reisinger, Daniel K. E. & Green, Hannah & Huijsmans, Marianne & Mellesmoen, Gloria & Trotter, Bailey (eds.), Papers for the 55th international conference on Salish and neighbouring languages, 12–40. UBC Working Papers in Linguistics.

Büring, Daniel. 2016. Intonation and meaning (Oxford Surveys in Semantics and Pragmatics 3). Oxford: Oxford University Press.

Cable, Seth. 2007. The grammar of Q: Q-particles and the nature of wh-fronting, as revealed by the wh-questions of Tlingit. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Cable, Seth. 2010a. Against the existence of pied-piping: Evidence from Tlingit. Linguistic Inquiry 41(4). 563–594. DOI:  http://doi.org/10.1162/LING_a_00013

Cable, Seth. 2010b. The grammar of Q: Q-Particles, wh-movement, and pied-piping. Oxford University Press.

Caponigro, Ivano & Sprouse, Jon. 2007. Rhetorical questions as questions. In Puig-Waldmüller, Estela (ed.), Proceedings of Sinn und Bedeutung 11, 121–133.

Cheng, Lisa. 1991. On the typology of wh-questions. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Dauenhauer, Nora & Dauenhauer, Richard. (eds.) 1990. Haa tuwunáagu yís, for healing our spirit: Tlingit oratory (Classics of Tlingit Oral Literature v. 2). Seattle; Juneau: University of Washington Press; Sealaska Heritage Foundation.

Davis, Henry. 2018. Only connect! Determiners, case, and agreement in Tsimshianic. International Journal of American Linguistics 84(4). 471–511. DOI:  http://doi.org/10.1086/698854

Davis, Henry & Brown, Colin. 2024. A second-last position clitic in Sm’algyax: A solution. In Bochnak, M. Ryan & Csipak, Eva & Matthewson, Lisa & Morzycki, Marcin & Reisinger, Daniel K. E. (eds.), The title of this volume is shorter than its contributions are allowed to be: Papers in honour of Hotze Rullmann, 119–130. Vancouver, BC: UBC Occasional Papers in Linguistics.

Davis, Henry & Brown, Jason. 2011. On A-bar dependencies in Gitksan. In Lyon, John & Dunham, Joel (eds.), Papers for the 46th international conference on Salish and neighbouring languages, 43–80. UBC Working Papers in Linguistics.

Davis, Henry & Nederveen, Sander. 2021. Hli, focus and relativization in Gitksan. In Reisinger, Daniel K. E. & Green, Hannah & Griffin, Laura & Huijsmans, Marianne & Mellesmoen, Gloria & Trotter, Bailey (eds.), Papers for the 56th international conference on Salish and neighbouring languages, 1–32. UBC Working Papers in Linguistics.

Dayal, Veneeta. 1996. Locality in WH quantification, vol. 62 (Studies in Linguistics and Philosophy). Dordrecht: Springer Netherlands. DOI:  http://doi.org/10.1007/978-94-011-4808-5

Dayal, Veneeta. 2016. Questions (Oxford Surveys in Semantics and Pragmatics 4). Oxford, United Kingdom: Oxford University Press 1st edn.

Dayal, Veneeta. 2023. The interrogative left periphery: How a clause becomes a question. Linguistic Inquiry, 1–50. DOI:  http://doi.org/10.1162/ling_a_00507

Dayal, Veneeta & Grimshaw, Jane. 2009. Subordination at the interface: The quasi-subordination hypothesis.

Dunn, John Asher. 1978. A practical dictionary of the Coast Tsimshian language. Ottawa, ON: University of Ottawa Press/Les Presses de l’Universite d’Ottawa. DOI:  http://doi.org/10.1353/book65512

Eckardt, Regine. 2020. Conjectural questions: The case of German verb-final wohl questions. Semantics and Pragmatics 13(9). 1–54. DOI:  http://doi.org/10.3765/sp.13.9

Eckardt, Regine & Beltrama, Andrea. 2019. Evidentials and questions. In Empirical issues in syntax and semantics 12, 1–35. CNRS.

Faller, Martina. 2002. The semantics and pragmatics of evidentials in Cuzco Quechua. Stanford, CA: Stanford University dissertation.

Farkas, Donka. 2022. Non-intrusive questions as a special type of non-canonical questions. Journal of Semantics 39(2). 295–337. DOI:  http://doi.org/10.1093/jos/ffac001

Farkas, Donka & Bruce, Kim. 2010. On reacting to assertions and polar questions. Journal of Semantics 27(1). 81–118. DOI:  http://doi.org/10.1093/jos/ffp010

Farkas, Donka & Roelofsen, Floris. 2024. Quasi-subordinate questions.

Forbes, Clarissa. 2023. Tsimshianic. In Dagostino, Carmen & Mithun, Marianne & Rice, Keren (eds.), The languages and linguistics of Indigenous North America, 985–1012. De Gruyter. DOI:  http://doi.org/10.1515/9783110712742-042

Gonzalez, Aurore. 2021. Polar questions and interrogative particles: A crosslinguistic investigation. Cambridge, MA: Harvard University Graduate School of Arts and Sciences dissertation.

Gonzalez, Aurore. 2023. Interrogative particles in polar questions: The view from Finnish and Turkish. Glossa: A Journal of General Linguistics 8(1). DOI:  http://doi.org/10.16995/glossa.6487

Grimshaw, Jane. 2012. Quotes, subordination and parentheticals.

Groenendijk, Jeroen & Stokhof, Martin. 1984. On the semantics of questions and the pragmatics of answers. In Landman, Fred & Veltman, Frank (eds.), Varieties of formal semantics: Proceedings of the fourth Amsterdam Colloquium, 143–170.

Gunlogson, Christine. 2008. A question of commitment. Belgian Journal of Linguistics 22. 101–136. DOI:  http://doi.org/10.1075/bjl.22.06gun

Hagstrom, Paul. 1998. Decomposing questions. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Hamblin, Charles L. 1973. Questions in Montague English. Foundations of Language 10(1). 41–53.

Han, Chung-hye. 2002. Interpreting interrogatives as rhetorical questions. Lingua. International review of general linguistics. Revue internationale de linguistique générale 112(3). 201–229. DOI:  http://doi.org/10.1016/S0024-3841(01)00044-4

Hill, Hector & Matthewson, Lisa. 2025. Gidaxan aa? Yagayt halaayin You asked a question? You already know it. In Hannon, Ella & Diep, Brian & Griffin, Laura & Loginova, Mila & Oliver, Bruce & Schneider, Lauren & Steiner, Reed & Trotter, Bailey (eds.), Papers for the 60th international conference on Salish and neighbouring languages, 207–234. University of the Fraser Valley (Chilliwack) and The Stó:lō Shxwelí Halq’eméylem Language Program in Chilliwack: UBC Working Papers in Linguistics.

Hirsch, Aron. 2017. An inflexible semantics for cross-categorial operators. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Karttunen, Lauri. 1977. Syntax and semantics of questions. Linguistics and Philosophy 1(1). 3–44. DOI:  http://doi.org/10.1007/BF00351935

Korotkova, Natasha. 2016. Heterogeneity and universality in the evidential domain. Los Angeles, CA: University of California dissertation.

Kotek, Hadas. 2014. Composing questions. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Kotek, Hadas. 2019. Composing Questions. The MIT Press. DOI:  http://doi.org/10.7551/mitpress/10774.001.0001

Kratzer, Angelika & Shimoyama, Junko. 2002. Indeterminate pronouns: The view from Japanese. In Otsu, Yukio (ed.), The 3rd Tokyo Conference on Psycholinguistics, 1–25. Tokyo: Hituzi Syobo.

Krifka, Manfred. 2001. Quantifying into question acts. Natural Language Semantics 9(1). 1–40. DOI:  http://doi.org/10.1023/A:1017903702063

Krifka, Manfred. 2014. Embedding illocutionary acts. In Roeper, Tom & Speas, Margaret (eds.), Recursion: Complexity in cognition, vol. 43, 59–87. Cham: Springer International Publishing. DOI:  http://doi.org/10.1007/978-3-319-05086-7_4

Krifka, Manfred. 2015. Bias in commitment space semantics: Declarative questions, negated quetions, and question tags. Semantics and Linguistic Theory 25. 328. DOI:  http://doi.org/10.3765/salt.v25i0.3078

Krifka, Manfred. 2017. Negated polarity questions as denegations of assertions. In Lee, Chungmin & Kiefer, Ferenc & Krifka, Manfred (eds.), Contrastiveness in information structure, alternatives and scalar implicatures, vol. 91, 359–398. Cham: Springer International Publishing. DOI:  http://doi.org/10.1007/978-3-319-10106-4_18

Krifka, Manfred. 2023. Layers of assertive clauses: Propositions, judgements, commitments, acts. In Hartmann, Jutta M. & Wöllstein, Angelika (eds.), Propositionale Argumente im Sprachvergleich / Propositional arguments in cross-linguistic research (Studien Zur Deutschen Sprache 84), 116–183. Tübingen: Narr.

Kuroda, Shige-Yuki. 1965. Generative grammatical studies in the Japanese language. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Ladd, D. Robert. 1981. A first look at the semantics and pragmatics of negative questions and tag questions. In Papers from the 17th regional meeting of the Chicago Linguistic Society, 164–171.

Ladusaw, William. 1980. Polarity sensitivity as inherent scope relations. Austin, TX: University of Texas dissertation.

Lahiri, Utpal. 2002. Questions and answers in embedded contexts (Oxford Studies in Theoretical Linguistics 2). Oxford; New York: Oxford University Press.

Li, Charles & Thompson, Sandra. 1981. Mandarin Chinese: A functional reference grammar. Los Angeles, CA: California University Press. DOI:  http://doi.org/10.1525/9780520352858

Littell, Patrick & Matthewson, Lisa & Peterson, Tyler. 2010. On the semantics of conjectural questions. In Peterson, Tyler & Sauerland, Uli (eds.), Evidence from evidentials, vol. 28, 89–104. UBC Working Papers in Linguistics.

Liu, Yuyang & Luo, Yitang. To appear. Dissecting the illocutionary force phrase: New evidence from SFP embeddability. In Proceedings of Generative Linguistics in the Old World (GLOW) 47.

Matthewson, Lisa. 2004. On the methodology of semantic fieldwork. International Journal of American Linguistics 70(4). 369–415. DOI:  http://doi.org/10.1086/429207

Matthewson, Lisa. 2010. Cross-linguistic variation in modality systems: The role of mood. Semantics and Pragmatics 3. DOI:  http://doi.org/10.3765/sp.3.9

Matthewson, Lisa. to appear. Aren’t Gitksan polar questions biased? International Journal of American Linguistics.

McCloskey, James. 2006. Questions and questioning in a local English. Crosslinguistic research in syntax and semantics: Negation, tense, and clausal architecture, 87–126.

Miyagawa, Shigeru. 2022. Syntax in the treetops. The MIT Press. DOI:  http://doi.org/10.7551/mitpress/14421.001.0001

Nyman, Elizabeth & Leer, Jeff. 1993. Gágiwduł.àt: brought forth to reconfirm: The legacy of a Taku River Tlingit clan. Whitehorse, YT; Fairbanks, AK: Yukon Native Language Centre; Alaska Native Language Center.

Portner, Paul. 2018. Mood (Oxford Surveys in Semantics and Pragmatics 5). Oxford, United Kingdom: Oxford University Press 1st edn. DOI:  http://doi.org/10.1093/oso/9780199547524.001.0001

Poschmann, Claudia. 2008. All declarative questions are attributive? Belgian Journal of Linguistics 22. 247–269. DOI:  http://doi.org/10.1075/bjl.22.12pos

Richards, Norvin. 2021. Immobile wh-phrases in Tagalog. Glossa: A Journal of General Linguistics 6(1). 1–17. DOI:  http://doi.org/10.16995/glossa.5762

Rigsby, Bruce. 1986. Gitxsan grammar.

Roberts, Craige. 2012. Information structure in discourse: Towards an integrated formal theory of pragmatics. Semantics and Pragmatics 5. DOI:  http://doi.org/10.3765/sp.5.6

Roelofsen, Floris & Farkas, Donka. 2015. Polarity particle responses as a window onto the interpretation of questions and assertions. Language 91(2). 359–414. DOI:  http://doi.org/10.1353/lan.2015.0017

Rohde, Hannah. 2006. Rhetorical questions as redundant interrogatives. San Diego Linguistics Papers 2. 134–168.

Rooth, Mats. 1985. Association with focus. Amherst, MA: University of Massachusetts dissertation.

Rooth, Mats. 1992. A theory of focus interpretation. Natural Language Semantics 1(1). 75–116. DOI:  http://doi.org/10.1007/BF02342617

Ross, John Robert. 1967. Constraints on variables in syntax. Cambridge, MA: Massachusetts Institute of Technology dissertation.

Sadock, Jerold. 1974. Towards a linguistic theory of speech acts. New York: Academic Press.

Sasama, Fumiko. 2001. A descriptive study of the Coast Tsimshian morphology. Kyoto, Japan: Kyoto University dissertation.

Sauerland, Uli & Yatsushiro, Kazuko. 2017. Remind-me presuppositions and speech-act decomposition: Evidence from particles in questions. Linguistic Inquiry 48(4). 651–678. DOI:  http://doi.org/10.1162/LING_a_00257

Searle, John R. 1969. Speech acts. Cambridge: Cambridge University Press.

Shamei, Arien. 2019. An acoustic analysis of Gitxsan prosody.

SLLTD. 2017. Sm’algyax Living Legacy Talking Dictionary.

Speas, Peggy & Tenny, Carole. 2003. Configurational properties of point of view roles. In Sciullo, A. M. Di (ed.), Asymmetry in grammar, 315–344. Amsterdam: John Benjamins. DOI:  http://doi.org/10.1075/la.57.15spe

Tarpent, Marie-Lucie. 1987. A grammar of the Nisgha language. Victoria, BC: University of Victoria dissertation.

Tarpent, Marie-Lucie. 1994. Between realis and irrealis, Tsimshianic negation and the status of the morpheme =Ł. AAA/CAIL meeting, Atlanta, Ga.

Uegaki, Wataru. 2018. A unified semantics for the Japanese Q-particle ka in indefinites, questions and disjunctions. Glossa: A Journal of General Linguistics 3(1). DOI:  http://doi.org/10.5334/gjgl.238

Von Stechow, Arnim. 1996. Against LF pied-piping. Natural Language Semantics 4(1). 57–110. DOI:  http://doi.org/10.1007/BF00263537

Woods, Rebecca. 2016. Investigating the syntax of speech acts: Embedding illocutionary force. York, United Kingdom: University of York dissertation.