Previous work on polar questions in sign languages

Zeshan [14,15] provides a typological overview of interrogatives in sign languages, focusing mainly on canonical question forms. Based on data from 37 sign languages (though in some cases the amount of available data for a given language is minimal), Zeshan concludes that canonical polar questions are most commonly marked by non-manual markers alone (without specific manual signs or a characteristic constituent order). Raised eyebrows, wide-open eyelids, eye contact with the addressee, and a forward head/body position are cross-linguistically the most common markers. Zeshan further notes that, depending on various factors discussed in no further depth, signers may use different non-manuals than these, such as “lowering instead of raising of the eyebrows in pragmatically marked questions” ([14], p. 20), like when expressing doubt (cf. [22]). About one fourth of the sign languages in Zeshan’s sample have optional manual question particles, with East Asian sign languages being particularly rich in them. Subject pronoun doubling and verb doubling are also common in canonical polar questions across sign languages.

Not much is known about non-canonical polar question forms in sign languages. Question tags have been mentioned briefly for Italian Sign Language [23], for Chinese Sign Language [24], and for British Sign Language [25]. Negative polar questions have been described for New Zealand Sign Language, Japanese Sign Language, American Sign Language (ASL), Croatian Sign Language (HZJ), Turkish Sign Language (TİD), Flemish Sign Language, and Indo-Pakistani Sign Language (for an overview, see [26]), Swedish Sign Language [27], Catalan Sign Language (LSC; [28]), and Hong Kong Sign Language (HKSL; [29]). Most of these studies report that negative polar questions combine (non-manual) markers of negation and interrogativity, for instance a headshake or backward head movement to mark negation, and raised eyebrows to signal that the utterance is a polar question. Additionally, brow furrowing and forward or backward head thrusts have been observed to occur in polar questions (e.g., in HKSL and LSC) and are said to mark surprise or doubt. For ASL, both Gökgöz and Wilbur [26] and Gonzalez and colleagues [30] report that, when negation is expressed manually, negative polar questions sometimes lack non-manual marking of negation (a headshake), while in negative assertions non-manual marking of negation is obligatory.

Very few studies have investigated the contexts of use of canonical and non-canonical polar question forms in sign languages. Gökgöz and Wilbur [26] and Sze and Lee [29] investigated the form and contexts of use of non-canonical polar questions involving negation in HZJ, ASL, and TİD. In terms of form, their main finding is that in HZJ and TİD negative polar questions exhibit a particular, rather fixed combination of non-manual markers, while in ASL there is not one particular pattern that stands out as there is quite some variation. A representative example from TİD is given in (5) (example from [26], p. 27; adapted to match our glossing conventions). In this example, backward head movement (‘h.bw’) and upward chin movement (‘ch.u’) signal negation, while the forward head movement (‘h.fw’), downward chin movement (‘ch.d’) and eyebrow raise (‘br.r’) are typical non-manual markers of polar questions. The translation suggests that the authors interpret the construction as a high negation question. Indeed, Gökgöz and Wilbur argue that “a positive epistemic implicature arises in TİD when a yes/no question bears negation” (p. 27), whereas negation is claimed to be interpreted low in negative polar questions in HZJ and ASL.

As for the interpretation of negative polar questions and the contexts in which they can be used, the main finding in [26] is that negative polar questions in TİD, like the one in (5), signal positive SB, but negative polar questions in HZJ and ASL do not, i.e., they are neutral with respect to the signer’s prior belief. Gökgöz and Wilbur do not consider CE as a contextual factor.

Sze and Lee [29] investigate the form and contexts of use of negative polar questions in HKSL. They consider four types of contexts, of which we discuss the three most relevant ones here. In the first type of context, there is (implicit) positive SB and neutral CE. In this case, negative polar questions, and high negation questions specifically, do not seem to be licensed in HKSL. In the second and third type of context, there is positive SB and negative CE. The difference is that in the second type of context, the positive SB is explicit, while it is implicit in the third type. The second type of context is reported to trigger high negation questions, while the third triggers low negation questions. These different question types are claimed to be distinguished through different non-manual markings, although the specific non-manual marker patterns used are also reported to be inconsistent between signers.

Finally, Cañas Peña [28] investigates the contexts of use of canonical and various non-canonical polar question forms in Catalan Sign Language (LSC). More specifically, Cañas Peña elicited polar question forms in different contexts from two native LSC signers. In these contexts, epistemic bias (comparable to SB in our study) and evidential bias (comparable to our CE) were manipulated. In a follow-up step, the same signers were shown different combinations of contexts and question forms to determine which forms could be used in which contexts. Cañas Peña found that different combinations of epistemic and evidential bias were associated with different combinations of brow configuration (raised vs. furrowed) and head and body position (backward vs. forward). In total, Cañas Peña describes five different non-manual marker combinations. Example (6) ([28,28], p. 12; ‘b.fw’ = body forward) illustrates one such combination (brow raise, head and body forward), felicitous when epistemic bias is positive and evidential bias is not negative (e.g., in (6-a)), and infelicitous otherwise (e.g., in (6-b)). The English translation and context descriptions have been slightly adapted for clarity.

Although the study by Cañas Peña considers various polar question forms and various contexts differing in terms of SB and CE, it does not cover the entire spectrum: in total, five polar question forms are considered and each is associated with a particular combination of SB and CE. For two possible combinations of SB and CE (neutral SB with positive CE, and neutral SB with negative CE), Cañas Peña leaves open which polar question form(s) they are compatible with. The range of polar question forms that are investigated is most likely not exhaustive either, because question forms involving negation are not taken into consideration.

Previous work on polar questions in NGT

The reference grammar for NGT by Klomp [10] characterizes canonical polar questions as being marked by raised eyebrows and a forward position of the head. The grammar mainly draws on the work of Coerts [12] and inspection of the Corpus NGT [31], and its findings align with [13], who find exactly the same non-manuals in canonical polar questions in NGT. Klomp adds that the forward position of the head is often combined with a head turn or tilt, and that polar questions also optionally involve a manual element, namely the sign palm-up, which usually appears sentence-finally (see also [32]). Finally, Klomp notes that polar questions and declarative clauses do not differ from each other in terms of constituent order (again, see also [13]).

It is worth summarizing the work by Coerts [12], the main source for of information for Klomp, in a bit more detail. Coerts elicited data from 19 informants, all L1 signers of NGT of different ages and from different regions in the Netherlands. During the elicitation sessions, informants worked in pairs and were given four tasks. Three of these involved storytelling by just one of the informants. In these tasks, polar questions directed toward the addressee only occurred twice. The fourth task was a guessing game: one informant was given a drawing of a certain object and the other informant had to find out what kind of object this was by asking questions. This yielded a total of 75 polar question instances.

Across these 77 instances in total, the marker ‘head forward’ occurred 78% of the time, while ‘raised eyebrows’ occurred 57% of the time. They occurred together in 51% of the cases, so 27% of the cases featured only ‘head forward’ without ‘raised eyebrows’ and 6% of the cases featured only ‘raised eyebrows’ without ‘head forward’. Almost all of the polar questions that did not involve either of these two markers did involve one or more of the following markers: ‘eyes wide open’, ‘head tilt’, ‘body/shoulder up/forward’. Coerts remarks that “one could argue that these features replace the characteristic features ‘eyebrows up’ and ‘head forward’ in certain contexts” (p. 107). Only in one case, the markers ‘head and body backward’ and ‘brows down’ were observed.

It is important to note that the task in Coerts’s study was such that the informants asking questions generally did not have any particular prior belief with respect to the answer (they had no idea what the object in the drawing was), and the context generally did not provide any evidence leading them to expect a particular answer either. Further, they presumably did not have a preference for any particular answer, and likely asked their questions in a polite and emotionally neutral way.

Besides Coerts, the only other previous experimental study of polar questions in NGT is that of De Vos and colleagues [33]. In this study, two L1 signers of NGT were shown videos of ten polar questions and were asked to pose the same ten questions in a neutral, angry, and surprised affective state, resulting in 30 target items per participant. The questions did not involve negation or question tags.

The analysis focused on the position of the eyebrows only. Video recordings were annotated using the Facial Action Coding Scheme (FACS, [34]). This scheme includes an annotation value for raised eyebrows (AU 1 + 2) and one for lowered eyebrows (AU 4), as well as a third value for a configuration of the face which results from engaging both the brow lowering muscles and the brow raising muscles simultaneously (AU 1 + 2 + 4). In terms of eyebrow position, this facial configuration is much more similar to a configuration with lowered eyebrows than to one with raised eyebrows.

When participants posed the ten questions in a neutral affective state, raised eyebrows (AU 1 + 2) occurred 67% of the time, while lowered eyebrows with forehead wrinkles (AU 1 + 2 + 4) were attested in the remaining 33%. In an angry state, lowered eyebrows (AU 4) occurred in 78% of the cases, while the remaining 22% involved either sequential or simultaneous combinations of AU 1 + 2 and AU 4. Finally, in a surprised affective state, AU 1 + 2 was observed in 58% of the cases, AU 1 + 2 + 4 in 17% of the cases, and sequential or simultaneous combinations of AU 1 + 2 and AU 4 in the remaining 25%. The authors further report that the intensity of the brow raises in a surprised affective state was higher on average than in a neutral affective state, though they also note that annotations of the intensity of facial action units may not always be reliable/reproducible.

Overall, these results indicate that (i) there is considerable variation in eyebrow configuration in polar questions even in a neutral affective state, and (ii) the signer’s affective state can have a considerable impact on eyebrow configuration. In particular, in an angry affective state, polar questions frequently involved ‘pure’ lowered eyebrows (AU 4) while in a neutral or surprised affective state, lowered eyebrows were attested less frequently and only in combination with AU 1 + 2, resulting in forehead wrinkles. We hypothesize that contextual parameters like SB and CE similarly affect eyebrow configuration and other aspects of the form of polar questions. The production experiment detailed in the following section serves to test this hypothesis.

Participants

Participants were recruited between 26 April and 12 July 2022 via a video-recorded call for participation in a study on questions in NGT, placed on the Facebook page of the university’s sign language group. Eight deaf signers of NGT responded to the call. One of the signers participated in a trial session to evaluate and optimize the experimental procedure and stimuli; the data resulting from this session were not analyzed and are not publicly available. After annotating the data of the remaining seven signers, we decided to exclude the data from one participant from the study. The signer in question had indicated in the participant survey that they began learning Sign Supported Dutch from the age of four (primary school) and only started learning NGT in high school. Because the participant indicated that they used NGT as their primary means of communication in daily life, the signer was allowed to participate in the study. However, the influence from Sign Supported Dutch was obvious in their signing, and their productions clearly diverged from those by the other signers. Since our main focus lies on the expression of different kinds of questions in NGT, we decided to exclude this participant from the study. Thus, we ended up including the data of six of the participants in our analysis.

Prior to the start of the study, participants answered a number of questions about their (language) background via the online survey platform Qualtrics. The questions, most of them multiple-choice, were presented in Dutch only. Of the six signers whose data were analyzed – all of them women –, two were between 18 and 29; one between 30 and 39, and three between 50 and 59 years of age. All signers self-identified as deaf, and one of the participants reported having a cochlear implant. Five of the signers reported to be right-handed and one ambidextrous. All participants reported that they use NGT on a daily basis, and all indicated that they acquired NGT either from birth or within the first four years of life. In the Netherlands, there are five main variants of NGT connected to the locations of the (former) five deaf schools in Groningen (North East), Amsterdam (North West), (Sint-Michiels)gestel (South), Rotterdam, and Voorburg (both West) [35,36]. All variants except one (Gestel) are used by the participating signers. Rotterdam (N = 2) and Voorburg (N = 2) are both represented by two signers, while the Groningen and Amsterdam variants are represented by one signer each. All participants indicated that they use Dutch either on a daily basis or most days. Other languages used by participants include English (‘most days’, N = 1; ‘occasionally’, N = 2) and American Sign Language (‘occasionally’, N = 1). Three of the signers also reported that they occasionally use International Sign. All except one signer, whose parents are both deaf, have hearing parents. Approximately half of the hearing parents were reported to know little to no NGT; the reported signing skills of the remaining hearing parents ranged from ‘limited’ to ‘good’.

Participants were financially compensated for their time and travel expenses.

Stimuli

Questions were elicited from participants in a role-play setting in which they interacted in NGT with two members of our research team, who we refer to as confederates A and B. Both are deaf and early acquirers of NGT. During the experimental sessions, the confederates were referred to as Ria (confederate A) and Tom (confederate B) to contribute toward the sense of a play-acting setting. The confederates signed pre-scripted utterances in response to participant productions that were prompted by stimulus materials projected on a laptop screen. The confederates’ utterances were intended to introduce SB (confederate A) and CE (confederate B). The only language used during the experimental sessions was NGT.

We created six situations (one used as practice trial) designed to elicit polar questions from participants in different contexts. The situations were loosely based on selected scenarios from a study on biased polar questions in spoken German and English [19], which was not a production experiment but a forced-choice task using written stimuli in which participants had to pick the best option from a number of sentence forms given a particular context description. For each situation, there were seven experimental conditions with different combinations of SB and CE (Table 1), which we refer to as variations. Only double positive and double negative evidence combinations were not included in the study, as it is unnatural to ask a question in such cases. Note that Domaneschi et al. [19] tested only six experimental conditions in their study, because they also excluded the combination negative SB – neutral CE. While it can indeed be somewhat unnatural to ask a question in such a setting, it is not impossible, so we decided to include this condition in our task. All situations in the experiment are described in English in S1 File.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Experimental conditions.

https://doi.org/10.1371/journal.pone.0354015.t001

Each trial, i.e., each variation within a situation, consisted of three short interactions between the participant and confederate A (first interaction) and confederate B (second and third interaction). The structure of each trial is visualized in Fig 1, which was also printed on a poster and hung in the studio where the experimental sessions took place.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Visualization of the structure of each experimental trial.

Jij = ‘you’ in Dutch. Ria and Tom are the confederates. Reprinted from [37] under a CC BY license, with permission from University of Amsterdam / Amsterdam University of Applied Sciences, original copyright 2022.

https://doi.org/10.1371/journal.pone.0354015.g001

Participants first saw a context video on a laptop screen, prompting them to ask a question to confederate A, who through their answer introduced positive, neutral, or negative SB with regard to the issue at stake (e.g., whether or not Kim is a vegetarian). In the second interaction, participants watched another context video in which they were instructed to ask another question to confederate B, who then gave a response which provided positive, neutral, or negative CE for the question at issue. Participants were subsequently shown a picture prompt on the laptop, which was intended to trigger participants to ask another question, the target question, to confederate B in the final interaction. After the participant asked the target question, confederate B gave a brief unscripted response to conclude the role play.

Participants were instructed to minimally include the concepts that were represented in the picture prompts. For instance, in situation 1 (see S1 File), the concept ‘vegetarian’ is represented by the V label (logo for vegetarianism) with two small icons of an apple and corn, and ‘Kim’ is represented by a silhouette of a woman’s head and the word ‘Kim’ below it. The question mark on the right was intended to remind the participant that their response should come in the form of a question, not a statement. Any images that were unclear to the participant were explained during the practice round of each first variation of a situation (see the section Experimental procedure for more details on the experimental procedure). We chose to use picture prompts for triggering the target question in order to avoid having to use video recordings of signs, which could influence participants in terms of lexical choices but also sign order. We also purposely aligned the pictorial representations of concepts vertically rather than horizontally. However, we realize that this vertical alignment could still have an effect on constituent order. We therefore created two sets of picture prompts, where the top and bottom images on the left of the prompts were reversed (see S1 File for all picture prompts). Two participants saw version 1 and four participants saw version 2 of the picture prompt set. A seventh participant saw version 1 but was excluded from the study after collection of the data; hence the imbalance in the number of participants who saw each version.

All stimulus materials are available online. The context videos, signed in NGT, can be found at https://doi.org/10.21942/uva.21695150, and recordings of the scripted confederate utterances can be found at https://doi.org/10.21942/uva.26870170. All target utterances from our participants (the final utterance in each trial) are available at https://doi.org/10.21942/uva.21666203.

Recording studio

The experimental sessions took place at a recording studio at the University of Amsterdam, equipped with studio lighting and a green screen. Fig 2 presents an overhead view of the studio set-up during the sessions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Overhead view of the recording studio setup.

Reprinted from [37] under a CC BY license, with permission from University of Amsterdam / Amsterdam University of Applied Sciences, original copyright 2022.

https://doi.org/10.1371/journal.pone.0354015.g002

Participants were stood in front of a green screen. Confederates, when in dialogue with the participant, stood slightly toward the signer’s right, positioned exactly between two cameras on tripods. They would stand and wait at the back of the room when not interacting with the participant. We used two cameras to record the participant: a Sony FDR-AX53 (Full HD) positioned slightly toward the signer’s left (confederate’s right) and an iPhone with depth camera for 3D recording, toward the signer’s right (confederate’s left). A second Sony FDR-AX53 camera (top left corner in the figure) recorded the confederates, to make it possible later on to do a double-check when needed, e.g., in case of an unexpected utterance from the participant in reaction to a confederate. The confederate recordings made during the experimental sessions are not publicly available, but we recorded the scripted confederate utterances separately after the sessions (available at https://doi.org/10.21942/uva.27043453).

In addition to the participant and confederates, there were two experimenters in the room. The first stood behind a high table (3 in Fig 2) with a keyboard and mouse connected to a laptop facing the signer (also on table 3), and a monitor visible to both the confederates and the experimenter on a low table (2). This experimenter guided each session and was responsible for providing instructions, projecting the stimuli (see the section Stimuli) on the laptop and monitor, as well as handling the participant camera. A second experimenter was positioned between the confederate camera and the studio lamp on the left in Fig 2, and was responsible for handling the confederate camera and depth camera. This experimenter also kept track of the trial number on a poster on the wall as well as on a clapperboard, which was displayed in front of the cameras at the start of each trial.

Experimental procedure

Participants were given a brief introduction by the lead experimenter and were then shown pre-recorded instruction videos in NGT on the laptop on Table 3. In the video, participants were told that they were going to participate in short role plays together with the confederates, who were introduced as ‘Ria’ and ‘Tom’ and referred to with initialized sign names, and that they had to ask these confederates a number of questions during the role plays. The sign name for ‘Kim’, which was used in two of the situations, was also introduced in the instruction video. The participants then witnessed two example role plays in real time, where the lead experimenter took the role of the participant and played two variations within the practice situation with the confederates (see S1 File). The lead experimenter is a hearing L2 signer of NGT with nine years of signing experience at the time the data collection took place. She uses NGT mostly in interaction with signing colleagues and participants.

Then, the structure of a role play was explicated using the visualization in Fig 1. Participants were instructed that they always had to ask questions to confederates, but that there were no restrictions on sign order or use of facial expressions. They were instructed to keep their productions brief, preferably restricted to a single sentence, and to sign them as naturally as possible. It was explained that there were five different situations with seven different variations within each situation, as well as a practice situation. Participants were told that, within each situation, the context videos and picture prompts were always identical, but that the responses of the two confederates would differ from variation to variation. Participants were instructed to adapt their questions depending on the confederates’ utterances, meaning that they might want to ask the same question slightly differently for each variation within a situation. To help participants get a sense of what we were after, they were shown three examples of variations of the same base question (e.g., “Is Kim home?”), but with different use of facial expressions and other non-manual markers (e.g., headshake for negation).

Then the practice session commenced, which included several variations within the practice situation (see S1 File). This allowed the participants to get used to the structure of each role play and to experience how the variations within a given situation would slightly differ from each other and as a result may lead them to ask the final question in each variation in a different way. In our experience, the practice phase usually took up quite some time, but once participants got the hang of the experimental procedure, the actual experimental phase would typically go rather smoothly. The main phase of the experiment began after the practice session and after participants had no more questions about the procedure. In total, all sessions lasted between 3–4 hours including breaks. Usually, the instructions and the practice session lasted about 1.5 hours, after which the tempo sped up considerably.

Remember from the section Stimuli that, in each trial, participants were presented with context videos preceding their first two interactions with the confederates. In practice, these context videos were not replayed over and over again. Rather, when the first variation within a new situation was introduced, participants would take some time to absorb the information in the context videos and to remember the questions they had to ask to the confederates. They would then do a practice run, only after which we would start recording. The context videos would not be shown anymore after this, unless the signer requested it. Each trial, then, consisted solely of interactions between the participant and the confederates, and the laptop where videos and stimuli were presented would display just the picture prompt for the target question. This way of going through the trials worked well and generally made the participant’s interactions with the confederates feel more natural, both to the participant and to the observing experimenters.

Dataset and data processing

The data of all six participants whose productions were analyzed are available at https://doi.org/10.21942/uva.21666203. Each file includes a compilation of five target questions involving the same participant and the same experimental condition (combination of SB and CE), one for each of the five different situations represented in S1 File. Video file names refer to the participant number, stimuli version (picture prompt set; see the section Stimuli), and experimental condition, where, e.g., ‘PosNeut’ stands for positive SB and neutral CE. For all participants, there is also a video with ‘baseline’ questions; these include for each situation one of the questions asked by the participant to confederate A at the beginning of each variation. Since these questions were asked without any bias yet involved, we considered these baseline, or ‘out of the blue’, polar questions. For participants with participant numbers 03, 05, 06, and 07, we also elicited declarative versions of the target questions for each situation (e.g., ‘Kim is a vegetarian’); these recordings are also included in the online dataset. No such recordings were made with participants 01 and 02.

Thus, the dataset comprises 52 files with five target sentences each, amounting to 260 constructions in total. This total includes 30 baseline questions (five from each signer) and 20 declarative sentences (five each from four signers). These 50 sentences were all annotated but will not be discussed in detail in the results sections below. We simply remark here that the declarative sentences all consisted of a single clause and did not involve overt polarity marking except in two cases in which the signer used subtle head nodding. The baseline questions involved somewhat more variation, albeit considerably less so than the question forms elicited in the various experimental conditions. Six baseline questions involved embedding (“Do you know if...”; cf. section Sentences with embedding), 14 consisted of a single clause without polarity marking (cf. section Sentence radical only), and ten were made up of a sentence radical (SR) followed by an or... tag or a hesitation tag (cf. section Sentence radical and tag(s)). Of those ten questions, two involved positive polarity marking, while the others did not show marking of polarity.

Table 2 indicates the number of target questions included in the analysis per condition. A total of 210 target questions were elicited across the six participants and seven experimental conditions. A total of thirteen of these question sentences – generally one or two per condition – had to be excluded from the analysis, leaving 197 questions for analysis.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Number of included target questions per condition.

https://doi.org/10.1371/journal.pone.0354015.t002

In most cases, the reason for exclusion was that the construction was unanimously assessed by two of the authors (two hearing L2 signers of NGT) and a deaf L1 language consultant as (i) not representing a question sentence (N = 4), or (ii) a clear mismatch with the experimental condition (N = 5). In two cases, the signer used a predicate sign antonymic to the target predicate (closed instead of open), inversing the polarity of the sentence which inadvertently complicated the analysis of the results. In one case, the signer inquired into the interlocutor’s knowledge about the issue at stake (“Did you know that...”) rather than the issue itself. One construction had a structure that was unclear to both the authors and the language consultant.

It can be observed that in the condition with neutral SB and negative CE, a total of five constructions were excluded; in four of those cases the reason for exclusion was that the elicited question form was judged not to be a good match with the preceding context. It is possible that this particular combination of SB and CE renders the asking of a follow-up question somewhat unnatural, making it harder for participants to provide a fitting response. Nonetheless, we obtained 25 question sentences that were suitable for analysis in this condition.

The remaining 210 target questions, plus the baseline and declarative sentences, were subsequently glossed and annotated for manual and non-manual markers in ELAN Linguistic Annotator across a set of 21 newly created tiers. All steps of the annotation procedure are described in the subsections below. The annotation files are publicly available at https://doi.org/10.21942/uva.22737074.

Delimiting and glossing target questions

We first determined the scope of each target question by creating time-aligned annotations on the main tier labeled ‘Question’, and subsequently gave it a unique identifier code indicating participant number, condition (e.g., PosNeg for positive SB and negative CE), and situation number.

On two child tiers, we added ID-glosses for manual signs, both in English (‘Glosses_EN’) and in Dutch (‘Glosses_NL’). The glosses correspond to the ID-glosses in the NGT database on Global Signbank (https://signbank.cls.ru.nl/). Signs that were not included in Signbank at the time data annotation took place, such as vegetarian, were prefixed with % in the annotations on these tiers. In Global Signbank, lexical variants of the same concept are distinguished from one another with letter suffixes in the ID-glosses (e.g., free-of-charge-a and free-of-charge-b); these are also included in the annotations, as well as in the glossed examples that we include in this paper. A comment tier (‘Comments’) was also included.

The annotations for this first triptych of tiers were made by the first author of this research paper, a hearing L2 signer. Cases of uncertainty were discussed with the NGT consultant.

Annotating sentence structure

Annotations for sentence structure were made across three tiers. On the tier ‘Structure’, a child tier of the main ‘Question’ tier, we indicated the general sentence structure of each question. There were five possible annotation values for this tier, listed in Table 3.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Annotation values on the tier ‘Structure’.

https://doi.org/10.1371/journal.pone.0354015.t003

Firstly, constructions that were excluded from the analysis, of which there were thirteen (see the section Dataset and data processing), were labeled ‘EXCL’. Secondly, constructions involving embedding (N = 24) were labeled ‘EMB’. For questions without embedding, there were three possible annotation values. Preliminary analysis of the data revealed that each question sentence consisted of either just the SR (‘SR’), or a SR followed by one (‘SR-T1’) or two (‘SR-T1-T2’) sentence-final phrases, which we refer to as tags. We define tags as particular combinations of manual and non-manual markers following the SR, fulfilling a specific semantic function (see the section Sentence radical and tag(s) for examples and discussion).

On the tier ‘Structure-components’, a dependent of the ‘Structure’ tier, we specified further details about the properties of SRs and tags, if present. The annotations on this tier concerning the SR took the form of combination labels indicating (i) whether the SR was overtly marked for polarity, and (ii) whether it included a prosodically integrated palm-up sign. palm-up (see Fig 4 and 7 for examples) is a frequently used sign/gesture in both sign languages and (co-speech) gesture expressing a wide range of epistemic meanings [38]. Table 4 gives an overview of the annotation labels and frequencies.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Annotation values and frequencies for SRs on the tier ‘Structure-components’ (N = 179).

https://doi.org/10.1371/journal.pone.0354015.t004

Polarity was often marked non-manually by signers by means of head nodding or headshaking. Concerning head nodding, both single and repeated head nods were attested in the data. Single head nods were not annotated as markers of positive polarity on the sentence structure tiers, because all instances appeared to fulfill functions other than polarity marking. It is known, for instance, that a single head nod in NGT may be used to signal imperatives [39], to mark an intonation phrase boundary and/or to signal emphasis [10]. In contrast, instances of repeated head nodding, typically spanning multiple signs, were analyzed as positive polarity markers. Headshake is a known marker of negation in NGT [12,40], but – as we shall see – it may also fulfill other types of (pragmatic) functions, in particular when it occurs on certain types of tags (see the section Hesitation tag for discussion). When a headshake accompanies the SR, we found that it usually served a negation function, thus marking negative polarity. Some SRs (also) included manual polarity markers, such as the basic clause negator not-a or the sign wel, which appears to function similarly to the word ‘wel’ in spoken Dutch, a multifunctional particle signaling positive polarity [41]. Given these commonalities, and given that English does not possess a suitable translational equivalent, we use the Dutch Signbank ID-gloss wel rather than the English one (yes-a). SRs that were marked for positive and negative polarity, manually or non-manually, were annotated as ‘SR.pos’ and ‘SR.neg’, respectively. SRs that were not overtly marked for polarity received the annotation ‘SR.no-pol’. The annotation labels for SRs that included a prosodically integrated palm-up sign – that is, a palm-up not functioning as (part of) a tag – received the suffix ‘.PU’.

For the tag parts of SR-T1(-T2) constructions, we annotated the tag type. The options are listed in Table 5; see the section Sentence radical and tag(s) for descriptions and examples of all tag types.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Annotation values for tags on the tier ‘Structure-components’.

https://doi.org/10.1371/journal.pone.0354015.t005

Finally, the tier ‘Structure-elements’, also a child tier of the ‘Structure’ tier, was used for free annotation of additional potentially relevant information. For instance, several items included a contrastive discourse particle in the SR, of which note was made on this tier.

The annotation protocol for the structure tiers was developed and evaluated by the first and last author of this paper. Annotations on these tiers were subsequently added by the first author. Cases of uncertainty were discussed with one of the confederates in the production experiment.

Annotating non-manual markers

All question forms were further annotated for non-manual markers (NMMs) across a set of 12 tiers. For this, we developed, evaluated, and then applied a new annotation guideline that went through several stages of development; we report extensively on this process in [42]. The version of the guideline that was used during the main annotation phase is available at https://doi.org/10.21942/uva.24080868.

For every question form, we made continuous NMM annotations on each of the 12 NMM tiers. We used Controlled Vocabularies containing between three to nine labels, i.e., different non-manual feature specifications, per tier. Table 6 lists the annotation labels available for each tier. The Controlled Vocabularies of all NMM tiers included at least the labels neutral and other (except ‘NMM.eye-gaze’, which does not include a neutral label). Neutral was used in cases where the relevant non-manual feature was not actively engaged, e.g., when the eyebrows were in neutral position. Other could be selected when none of the annotation labels were considered to fit with the observed NMM. In such cases, the marking had to be specified on the ‘Comments’ tier.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. NMM tiers and annotation label inventories.

https://doi.org/10.1371/journal.pone.0354015.t006

20% of the question forms in the dataset were randomly selected for independent annotation by two annotators, a hearing L2 signer of NGT (the first author of this paper) and a deaf L1 signer of NGT. The resulting two sets of annotations were evaluated for inter-rater agreement. The outcomes of this evaluation are reported in detail for all tiers and all annotation labels in a technical report (https://doi.org/10.21942/uva.25563540). Here, we just concentrate on the NMMs which we identified as playing a role of interest in the question forms in our dataset, and which will thus feature prominently in our discussion of the results in the sections Results: Polar question structures and their contexts of use (annotations on the ‘NMM.head-move’ tier) and Results: Nonmanuals (annotations on other NMM tiers).

The Cohen’s Kappa () scores for the relevant labels are listed in Table 7. Before we further discuss these scores, a couple of notes are in order on (i) the method of calculation of these scores, and (ii) the label selection presented in Table 7.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Cohen’s scores for various NMM labels, under a frame-based approach.

https://doi.org/10.1371/journal.pone.0354015.t007

With regard to the first issue, in the technical report, κ scores are reported based on two different approaches: a frame-based approach, where inter-annotator agreement is evaluated on a frame-by-frame basis, and an event-based approach, where inter-annotator agreement is evaluated at the level of the event, i.e., the annotation labels assigned to time intervals spanning an entire event rather than individual frames. The event-based approach tends to be somewhat more conservative. While we conclude in the report that the event-based approach – enriched with an automatic error analysis – is more suitable for the purposes of understanding the sources of coder disagreements and further improving the guideline, here, we report scores under a frame-based analysis. The reason for this is that in certain situations, the event-based approach is arguably too punitive, leading to sometimes much deflated agreement scores that do not seem to be an accurate reflection of the situation. To give an example that is also discussed in the extended report, imagine that one annotator coded an entire sentence as involving a downward head position (down) on the ‘NMM.head-y’ tier, while a second annotator broke up the same unit into three separate long down segments, with two intervening short neutral segments. Under the event-based method, all events coded on this tier would be analyzed as ‘unmatched’, thus heavily impacting the inter-annotator agreement score. Under the frame-based method, only the two short segments labeled neutral by the second annotator would be labeled as disagreements. Moreover, we observe that labels that are not so frequently used (such as down on the ‘NMM.lip-corners’ tier) generally yield much lower agreement scores under an event-based than under a frame-based method. For the purposes of the current paper, we therefore consider reporting the -scores under the frame-based method the most appropriate.

With regard to the label selection, firstly, for the analysis of sentence-structure patterns (section Results: Polar question structures and their contexts of use), it was important to consider non-manual polarity marking in our dataset. The primary NMMs for marking polarity are nodding (positive polarity) and shaking (negative polarity) on the ‘NMM.head-move’ tier; we therefore report the inter-annotator agreement scores for these labels in the table. However, we do not report the -scores for the labels nod and shake, which were used for single rather than repeated head nods and headshakes. While single head nods can be used as polarity markers (see (9-b) for an example), they can also have other linguistic functions. We therefore decided to exclude single head nods from the analysis altogether. The label shake was hardly ever used, as headshakes typically tend to involve multiple side-to-side head movements in NGT. Indeed, at least in a negating function, headshakes in NGT are known to be prone to spread over larger parts of the sentence [12,40].

Secondly, unlike the other tiers, the ‘NMM.eye-shape’ tier included ‘degree’ labels: squint-full vs. squint-half, and wide-full and wide-half. It turned out that the ‘half’ labels were more difficult to reach agreement on between the two annotators, as is also reflected by the lower inter-annotator agreement scores (0.45 for squint-half and 0.32 for wide-half). In cases of disagreement, when one of the coders had selected the ‘full’ label, the other coder had typically selected the corresponding ‘half’ label. In other words, coders agreed on the NMM itself but not on the degree to which it was engaged. But the pattern does not hold in the opposite direction: in cases where one coder had selected the ‘half’ label and the other coder disagreed, the nature of the disagreement varied: sometimes the other coder had selected the ‘full’ label and sometimes the neutral label. Given this, the analysis of NMM patterns in our dataset presented in the section Results: Nonmanuals is based on the annotation labels squint-full and wide-full only, which is the more conservative approach. Similar reasons also partially motivate our decision to exclude the labels raised-low and inner-raised on the ‘NMM.eyebrows’ tier from the analysis in the section Results: Nonmanuals. Both these labels received pretty low agreement scores (0.33 and 0.18, respectively), and the label inner.raise in particular was also applied relatively infrequently. A more theoretically-motivated reason not to include these labels in our analysis is that the literature on question forms in sign languages typically identifies specifically brow lowering and (supposedly full) brow raising as non-manual interrogative markers.

Returning to Table 7, overall, inter-annotator agreement for all but one of the NMMs that play a role in the discussions in this paper is quite decent, reaching moderate (0.41–0.60) or substantial (0.61–0.80) and in one case even almost perfect (0.81+) agreement. The exception is the label forward on the ‘NMM.head-z’ tier, which got a -score of 0.38, demonstrating only fair (0.21–0.40) agreement. This relatively low score can in large part be explained by mismatches due to annotators not agreeing on whether or not the label forward should apply (i.e., cases where one of the coders selected forward and the other neutral). Additionally, in the cases where both coders had identified a forward event, they sometimes disagreed on the onset and offset, further impacting the overall reliability score. We believe that the fact that annotations had to be made based on a single 2D camera view – and lacking a side-view camera perspective – made it difficult for annotators to reliably code head position. Indeed, not just the agreement scores for the ‘NMM.head-z’ tier were low but also for the ‘NMM.head-y’ tier with the labels up and down. As such, we expect that the addition of a side-view camera would increase reliability scores. For the present study, this issue cannot be overcome; hence, the discussion on the inquisitive function of forward head (and body) position in the section What is the main non-manual marker of polar questions / inquisitiveness should be appreciated with this relatively low inter-annotator agreement score in mind.

After we evaluated inter-rater reliability of the 20% of the dataset that was annotated by the two coders, 100% of the data was annotated for NMMs by one master coder, the first author of the paper.

Annotating additional elements of interest

Finally, every annotation file included three additional tiers that could be used for annotating further elements of interest. Instances of palm-up, a multifunctional marker which can also function as a question particle in NGT [12,32], were annotated on the tier ‘MM.PU’. Other potentially interesting manual markers we wanted to keep track of were annotated on the tier ‘MM.sign’, while possibly relevant mouthings of Dutch spoken words accompanying manual signs were annotated on the tier ‘NMM.mouthing’.

We should note that we did not systematically analyze these additional elements of interest, which can be considered a limitation of the present study. Nevertheless, mouthings will be shown in the section Or... tag to play an important role in one particular type of question form.

Sentence radical only

A total of 114 question forms in our dataset consist of a SR only. We indicated for such constructions whether or not they ended with the sign palm-up, as this marker has been observed to occur frequently in question contexts in various sign languages (e.g., [32,43,44]). Thirteen constructions (11.4%) were found to include a prosodically integrated palm-up in sentence-final position. From this we can conclude that palm-up does not systematically take on a question-marking function in this construction type (but see the section Sentence radical and tag(s) on the use of palm-up in tags). In the remainder of this section, we will discuss question forms with and without palm-up together.

It is interesting to consider polarity marking in questions made up of just a SR. Of the 114 constructions, 38 did not show any overt polarity marking, 43 displayed positive polarity marking, and 33 showed negative polarity marking. Representative examples are given in (7): (7-a) does not include any manual or non-manual marking of polarity, while (7-b) and (7-c) are non-manually marked for positive and negative polarity by means of (repeated) head nodding and headshaking. (7-c) is also manually marked for negative polarity by means of the sign not-a.

Table 8 shows the distributions of these different subtypes of question forms across experimental conditions. Darker shaded cells indicate that a particular type of polarity marking occurred in the majority of question forms within a condition.

Download:

• PNG
  larger image
• TIFF
  original image

Table 8. Distribution of question forms consisting of just a SR with different types of polarity marking.

https://doi.org/10.1371/journal.pone.0354015.t008

We can observe that the sort of polarity marking that is used tends to correspond with the polarity of the contextual evidence (CE) provided by the second confederate. A link with speaker bias (SB) as introduced by the first confederate is less clear. Still, the only question forms with negative polarity marking in the absence of negative CE were observed in the NegNeut condition, and ten out of the 17 question forms with positive polarity marking but without positive CE occurred in the two conditions with positive SB. For constructions without polarity marking, it is worth noting that a fair few of them (12/38) were uttered following positive CE. For comparison: none of the baseline questions consisting of a SR only (N = 14) involved overt polarity marking. Thus, polarity marking in the SR is clearly a hallmark of biased polar questions, where the type of polarity marking has the tendency to reflect the polarity of the provided CE.

Given that questions with negative polarity marking are rather common in our dataset, it is worth considering whether some of these question forms could be analyzed as high negation questions. Specifically, negative-polarity marked questions following positive SB (and neutral or negative CE) are interesting to consider here. As shown in Table 4, our data include a total of 11 question forms containing negative polarity marking that were uttered following positive SB and negative CE (and no cases following neutral CE). Inspection of these tokens reveals that about half of them cannot be high negation questions, as the non-manual marking patterns (specifically, eyebrow raising) and/or use of the mouthing ‘dus’ (‘so’) clearly show the signer prepared to accept the CE just provided by the second confederate. In other words, these questions clearly function to seek confirmation of the (negative) CE, making them incompatible with a high negation reading. The remaining cases have the potential to be high negation questions, although judgment data is needed in order to verify this. We provide an example in (8), where headshake is observed on the predicate. In this question form, the sentence structure itself does not provide any clues as to whether this might be a high negation question, but the non-manual marking pattern might (like in TİD and HKSL; see the section Previous work on polar questions in sign languages). The lowered eyebrows and squinted eyes could be interpreted as signaling that the signer has doubts about the CE provided by the second confederate (see the section Factors associated with brow lowering for more discussion), which is compatible with a high negation reading. However, it is also possible that these markers actually fulfill a different function – for instance, that the signer hopes the SR turns out to be false (again, see the section Factors associated with brow lowering) – which is compatible with a low negation reading of this question form. Further research is needed in this area.

Finally, some question forms (either consisting of just a SR or a SR followed by a tag) include specific manual particles within the SR that are interesting to briefly mention here. Eight question forms included within the SR a sign we gloss as toch-a or toch-b (after the spoken Dutch particle ‘toch’); three forms included a sign we gloss as wel (after the spoken Dutch particle ‘wel’), and three forms included the sign hesitate-a. Both toch and wel can be analyzed as manual positive polarity markers. We provide an example for each type of particle in (9) below, where the English translation represents the closest approximation of the resulting meaning of the sentence the particle appears in. Interestingly, as will be shown in the section Sentence radical and tag(s), all three particles illustrated in (9) can also occur as a tag following the SR.

Sentence radical and tag(s)

A total of 65 constructions consist of a SR and one (N = 63) or two (N = 2) tags. Tags are particular combinations of manual and non-manual markers that follow the SR, making up a separate prosodic unit. Constructions with tags typically display a marked shift in non-manuals between the SR and the tag, with usually multiple non-manual markers that are actively engaged in the SR changing their configuration in the tag. Sometimes, there is a small pause between the two sentence parts. We can distinguish five types of tags in our data, which we discuss below. Some of these tags were first described in [45].

Hesitation tag.

The most frequently occurring tag is what we will refer to as a hesitation tag (annotated as ‘HES’, N = 39). The precise realization of this tag may differ, but it always involves a combination of manual and non-manual markers, which together appear to signal hesitation or uncertainty about the truth of the SR on the part of the signer. Manual signs that may be used in a hesitation tag include palm-up, hesitate, and what-c (see https://signbank.cls.ru.nl/datasets/NGT for video illustrations of the signs corresponding to these glosses). In each of these signs, the palms face upward and the fingers of both hands are either fully extended or slightly curved, however the movement differs slightly. Non-manual markers that we observed frequently in hesitation tags include a forward body and/or head movement and a mouth configuration with the lip corners down (‘mouth shrug’); see the section Results: Nonmanuals for more in-depth discussion. Moreover, we often observed a headshake on the hesitation tag that clearly did not function to express negation. Headshakes are known to fulfill multiple functions beyond negation both in co-speech gesture [46,47] and sign languages [47–49]. In the cases we describe here, we argue that the headshake expresses uncertainty and possibly a request for a response from the addressee. We refer to such cases as inquisitive headshakes. We first described this function of headshake in NGT polar questions in [45], in which we analyzed a subset of the dataset discussed in the article; specifically, the 68 single-clause question forms that contain a headshake.

Fig 4 shows two video stills illustrating hesitation tags; the corresponding glossed examples are provided in (10). In both examples, there is a marked change in non-manuals between the SR and the tag. Example (10-b) also includes an inquisitive headshake on the tag. In fact, there are two headshakes in (10-b): one on the SR and one on the tag, separated by a brief interruption. The headshake on the SR appears to have the dual function of expressing negation and hesitation. The amplitude (the head’s degree of rotation from side to side) of the second, inquisitive, headshake is smaller than the first, making the headshake less visually salient.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Two examples of hesitation tags.

A: Hesitation tag as articulated in (10-a). B: Hesitation tag as articulated in (10-b).

https://doi.org/10.1371/journal.pone.0354015.g004

Table 9 displays the distribution of the hesitation tag across experimental conditions. The tag is quite common across all experimental conditions, but in particular when there is neutral CE. An inquisitive headshake occurs in about half of all the hesitation tags (N = 19/39) and is fairly evenly distributed across experimental conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. Distribution of question forms with a hesitation tag.

https://doi.org/10.1371/journal.pone.0354015.t009

Hesitation tags may follow SRs with any type of polarity marking. There seems to be an interaction between the presence (or rather, absence) of an inquisitive headshake and the type of polarity marking. Of the 19 constructions with inquisitive headshake, positive polarity marking, negative polarity marking, and absence of polarity marking on the SR occur in approximately equally many cases (six, six, and seven cases, respectively). In contrast, of the 20 constructions with a hesitation tag but without inquisitive headshake, 15 involve a SR without overt polarity marking. Four cases involve positive polarity marking, and just one case involves negative polarity marking on the SR.

toch tag.

The second type of tag we discuss is what we refer to as the toch tag. In eleven constructions, the SR was followed by the sign toch-a (English gloss in Signbank: sure-a; two-handed sign) or toch-b (sure-b; one-handed); in two of these cases, the tag was followed by a second tag. The English ID-gloss sure stems from the fact that the sign toch can (also) be used to mean ‘certain(ly)’.

To understand the meaning of this particle, it is useful to discuss its apparent equivalent in Dutch, which has previously been investigated in various studies. Dutch ‘toch’ is a modal particle, used in biased polar questions, that may occur within the sentence or sentence-finally [50,51]. In sentence-final position, ‘toch’ roughly translates as ‘right’, as in, e.g., “Kim is a vegetarian, right?”. In sentence-medial position, ‘toch’ can have different functions, where there is also an interaction with word stress [52]. When stressed, ‘toch’ may function as a contrastive marker of concession, or to indicate inconsistency with the common ground, in which case it can be translated as ‘still’ or ‘after all’. When stress lies elsewhere, ‘toch’ typically functions as a reminder of the common ground, in which case its meaning approximates ‘surely’ [52]. On the face of it, the two variants of the NGT sign seem to function equivalently to Dutch ‘toch’, although their precise semantic and distributional properties require further investigation. The sign toch-a/b occurs both within the SR (see (9-a) and (13) for two examples of sentence-medial toch with different meanings) as well as sentence-finally, as a tag, in our NGT data. Here, we focus on the instances in which toch-a/b functions as a tag.

Our data show that constructions with a toch tag typically involve eyebrow lowering and/or eye squint on both the SR and the tag. A head forward position is also very common, while eyebrow raise and mouth shrug are typically not observed. As such, sentences with a toch tag altogether involve a less pronounced shift in non-manuals than the other tag sentence types described in this section. Two representative examples from our data are illustrated in Fig 5 and glossed in (11).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Two examples of toch tags.

A: TOCH tag as articulated in (11-a). B: TOCH tag as articulated in (11-b).

https://doi.org/10.1371/journal.pone.0354015.g005

Distributionally, toch tags occur in conditions with non-neutral SB (Table 10). In all but two cases, polarity marking in the SR conforms to the polarity of the SB; that is, in the case of negative SB, there is negative polarity marking in the SR (as in (11-a)), and in the case of positive SB, there is positive polarity marking in the SR (as in (11-b)). The two exceptions both involve positive SB but no polarity marking on the sentence radical.

Download:

• PNG
  larger image
• TIFF
  original image

Table 10. Distribution of question forms with a toch tag.

https://doi.org/10.1371/journal.pone.0354015.t010

Or... tag.

The third tag type we discuss is the or... tag, which occurred eleven times in our dataset. Examples of this tag type are illustrated in Fig 6 and are shown in context in the glossed examples in (12). In two of the eleven cases, the tag directly followed a toch tag; see (12-c) for an example. Or... tags look quite similar to hesitation tags, with the crucial difference that they involve a mouthing, which is either of (‘or’) or of niet (‘or not’). Like hesitation tags, or... tags typically involve manual signs that carry little lexical content but fulfill a more pragmatic or other (e.g., prosodic) function, such as hesitate-a (12-a), sentence-final pronoun copies (12-b), or palm-up (12-c). Regarding non-manual marking, the or... tag generally involves forward body and/or head movements, but headshakes and head nods also frequently occur. Headshakes may be used to express negation (‘or not’), uncertainty about the truth of the SR, inquisitiveness (a request for a response), or a combination. In example (12-a), for instance, we interpret the headshake on the tag as signaling uncertainty about the truth of the SR and/or inquisitiveness, while the headshake that is present in the preceding SR functions to mark negation. If the headshake (or head nod) on the or... tag functions to mark polarity, we find contrastive polarity marking on the SR. This is exemplified by the constructions in (12-b) and (12-c), where we can observe head nodding in the SR but headshaking on the tag, expressing negation (or not).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Three examples of or... tags.

A: Or... tag as articulated in (12-a). B: Or... tag as articulated in (12-b). C: Or... tag as articulated in (12-c).

https://doi.org/10.1371/journal.pone.0354015.g006

Distributionally, it is striking that all question forms with or... tags were uttered in contexts in which CE is neutral (Table 11). In cases where the tag was positive and followed a SR with negative polarity marking (12-a), we can furthermore observe that SB was either neutral or negative but not positive (Table 11, left), although note that the number of attested examples is very small here. In contrast, seven out of the eight sentences with a negative or... tag and positive polarity marking on the SR were attested in contexts with either positive or neutral SB (and neutral CE; Table 11, right). In other words, the polarity of the SR in constructions of this type tends to match SB.

Download:

• PNG
  larger image
• TIFF
  original image

Table 11. Distribution of constructions including an or... tag.

https://doi.org/10.1371/journal.pone.0354015.t011

Other tags.

Finally, we found two other tag types in the data, which are exemplified by only a handful of tokens. Firstly, four elicited questions, produced by two different signers, included a tag type we gloss as CONF (for ‘confirmation’). The tag consists of a combination of palm-up or, in one instance, the sign wel, and a particular set of non-manuals. These include head nodding and raised (inner) eyebrows – both sustained from the SR – as well as pressed lips. The tag type occurred in three different conditions – NeutNeut (twice), NeutPos, and PosNeut – none of which involved negative SB or CE. We provide one example of the tag in Fig 7, with the corresponding example glossed in (13); also note the use of the sign toch-a in the SR.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. CONF tag as articulated in (13).

https://doi.org/10.1371/journal.pone.0354015.g007

Interestingly, all four examples were uttered in the context of situation 5, in which participants had to ask a ticket seller at the train station about the availability of a 9am train to Paris the next day. Although we tried to control for differences in authority between the confederate roles as much as possible, this is the one scenario where the second confederate has a relatively high degree of authority. This tag type, and the non-manuals in particular, thus appears to function as a politeness strategy when asking a confirmation-seeking question.

Finally, in two instances, both of which involved the same signer, the SR was followed by palm-up combined with non-manuals that seem to express a mixture of confusion and slight exasperation at the confederate’s previous response. As can be observed in Fig 8, these non-manuals include a backward head/body position, eye squint, brow lowering, and mouth shrug. We refer to this tag type as HUH. In contrast to all other tag types, the HUH tag involves backward rather than forward head and/or body movement. Both instances occurred in reaction to conflicting pieces of information offered by the two confederates (conditions PosNeg and NegPos). One example is given in (14). Various relevant non-manuals extend across the entire question form.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. HUH tag as articulated in (14).

https://doi.org/10.1371/journal.pone.0354015.g008

Sentences with embedding

A total of 18 sentences in the dataset involve embedding, which we qualitatively describe here. The sentences can be divided into two categories based on the matrix clause predicate. In about two thirds of the examples (N = 11), the matrix clause includes the predicate know and the interlocutor is directly addressed (‘Do you know...’). In one of these cases, the predicate used is actually call-a, inflected for a first-person subject and a second-person object. The main clause can be roughly translated as ‘I want to ask you...’. Because we consider this a rough paraphrase of ‘do you know...’, we categorize it with the other tokens discussed here. In the remaining examples (N = 7), the matrix clause serves to introduce the signer’s own expectations about the matter of discussion (‘I thought...’; ‘I heard that...’). We discuss both construction types in greater detail below. The section will be concluded with a brief discussion about the distribution of the two types of embedded question forms across the five situations in the experiment.

Do you know....

The 11 question forms with the embedding predicate know come from four different signers. We observed two variations of this general construction type, illustrated by two representative examples in (15). A few examples also include a hesitation tag or an or... tag at the end of the sentence.

The two construction variations illustrated in (15) differ only in that the matrix clause in (15-a) additionally includes the sign glossed as sure. With the exception of the accompanying mouthing, this sign is formally identical to the sign used in toch tag constructions (see the section toch tag), but – as the mouthing (‘ze(ker)’; ‘sure’) indicates – the meaning is clearly different: the matrix clause roughly translates as ‘Are you certain that...”. This construction type was used twice, by the same signer. The utterance followed positive CE in one case, and negative CE in the other; in both instances, SB was neutral. In the case with negative CE, the polarity of the embedded SR mirrored the polarity of the CE; in the other case, polarity marking was neutral.

The other nine embedded clause constructions with know are of the type illustrated in (15-b), with a main clause that translates as ‘Do you know...’. In four cases, the main clause was reiterated at the end of the construction, which is a fairly common phenomenon in NGT. All nine tokens were uttered following neutral CE expressed by the second confederate. SB was positive in two cases, negative in three cases, and neutral in four cases. Thus, neutral CE appears to be a precondition for use of this construction type. This contrasts with ‘Do you know...’ constructions that also include the sign sure, which rather appear to require non-neutral CE (although the number of examples is, of course, low).

I thought....

Seven embedded clause constructions, uttered by three different signers, include a matrix clause that makes reference to the signer’s own expectations regarding the answer to the question at hand. Two representative examples, one with the predicate think and one with the predicate listen, are shown in (16).

At first glance, the sentences in (16) may not have the appearance of interrogative constructions. However, the non-manuals that accompany the manual signs in these sentences, as well as the use of a hesitation tag in (16-b), indicate clearly that the signer is soliciting a response from the interlocutor. As such, we consider them to be questions both in function as well as in (primarily non-manual) form. Indeed, six of the seven constructions of the type discussed here include a hesitation tag. Of these seven constructions, four were uttered in one of the conditions in which the participant was presented with conflicting evidence from the two confederates (PosNeg or NegPos). Three of these instances involved the predicate hear/listen; the fourth case included the predicate think. The other three cases all involved think and were attested among the responses in the NegNeut (N = 1) and NeutNeg (N = 2) conditions.

SR-only questions

The first generalization that can be drawn concerns the relationship between polarity marking on the SR of a polar question, on the one hand, and the type of contextual evidence that is available for the SR, on the other.

1. (17)  Generalization: In polar questions consisting only of an SR (i.e., without any tags) polarity marking correlates with the type of CE. More specifically:
   1. A. Questions with negative polarity marking (i.e., negation) mostly occurred in contexts with negative CE (29 out of 33 cases). The remaining four cases occurred in contexts with neutral CE and negative SB. CE was never positive.
   2. B. Questions with non-negative polarity marking (i.e., either positive polarity marking or no explicit polarity marking) mostly occurred in contexts where CE was not negative either (73 out of 81 cases).
   3. C. Questions with explicit positive polarity marking mostly occurred in contexts with positive CE (27 out of 43 cases) or neutral CE (12 out of 43). Of the remaining five cases, four occurred in contexts with negative CE but positive SB.

The first part of the generalization (A), concerning negative polarity marking, aligns with a generalization proposed by Büring and Gunlogson [16] for English: according to them, a polar question in English with low negation, i.e., a question of the form , is felicitous only if there is compelling contextual evidence against ¬p?, i.e., in our terminology, if there is negative CE. Note that Büring and Gunlogson [16], like many other authors, make a distinction between high negation polar questions (e.g., Isn’t there a vegetarian restaurant around here?) and low negation polar questions (e.g., Is there no vegetarian restaurant around here?). Only the latter are relevant for our discussion here.

However, Van Rooij and Safárová [20] have argued that Büring and Gunlogson’s generalization for English is too strong. They point out that questions of the form can sometimes be felicitous without any particular contextual evidence for or against p. One of their examples is given in (18).

1. (18) [A woman goes to the doctor with her son. The doctor asks the mother:]
   1. Has he not been eating properly?

Experimental evidence in line with this observation, i.e., evidence demonstrating that low negation polar questions in English can be felicitous in contexts without any particular contextual evidence for or against the SR, has been given in [18].

Turning back to NGT, our intuition is that a polar question with negative polarity marking on the SR would be felicitous in the equivalent of Van Rooij and Safárová’s example (18) in NGT as well. And indeed, our dataset also includes four instances of questions with negative polarity marking in contexts with neutral CE. So Büring and Gunlogson’s generalization seems to be too strong for NGT as well as for English.

For English, a weaker version of Büring and Gunlogson’s generalization has been proposed by Roelofsen and colleagues [53], who propose that a polar question with low negation of the form in English is felicitous only if there is no compelling contextual evidence for p, i.e., in our terminology, if there is no positive CE. This weaker generalization for English is compatible with Van Rooij and Safárová’s observation and the experimental evidence in [18], and our data suggests that it also holds for NGT.

Before turning to a possible explanation of this pattern, let us consider part B of Generalization (17): our results indicate that in NGT, SR-only polar questions without negative polarity marking (i.e., either without any polarity marking at all or with positive polarity marking) mostly occur in contexts without negative CE. This pattern also aligns with what has been observed for English (see [16–18,53], among many others). For instance, the polar question in (19-b) is clearly infelicitous, while (19-a) is completely natural (example adapted from [16]).

1. (19) Scenario: A enters B’s windowless office wearing a dripping wet raincoat (so there is contextual evidence for it being rainy outside and against it being sunny). B asks:
   1. a. Is it raining outside?
   2. b. #Is it sunny outside?

Roelofsen et al. [53] formulate a generalization for polar questions in English that captures both cases with low negation and cases without negation in one go: namely, polar questions are felicitous only if there is no compelling contextual evidence that the SR is false. Our results suggest that this generalization also applies to NGT.

Finally, part C of Generalization (17) concerns polar questions in NGT with explicit positive polarity marking (in most cases head nodding). In this case, there is no clear correspondence with a polar question form in English (at least not one that has been investigated in depth so far). In particular, polar questions in English with an emphatic positive marker such as really (see, e.g., [19]) have rather different felicity conditions than polar questions with positive polarity marking in NGT. We leave open here whether there are any polar question forms in other languages, signed or spoken, which have the same felicity conditions as polar questions with explicit positive polarity marking in NGT. We are not aware that any such forms have been discussed in the literature, but of course this does not at all exclude the possibility that such forms do exist.

A possible explanation of the generalization that polar questions in English are felicitous only if there is no compelling contextual evidence that the SR is false has been proposed by Roelofsen et al. [53] in terms of a pragmatic principle which they term Avoid Reverse. This principle says that, other things being equal, when formulating a polar question, cooperative speakers formulate the question in such a way as to minimize the chance of eliciting a ‘reverse’ response from their interlocutor, where a ‘reverse’ response is one that rejects the SR of the question. This explains the generalization, because it disfavors the asking of a polar question if there is compelling contextual evidence that the SR is false. After all, in this case the question is very likely to elicit a ‘reverse’ response. Note that this explanation relies on a general pragmatic principle, not on anything specifically related to English. So, it plausibly applies beyond English as well, in particular to NGT.

Questions with tags

We now turn to questions with tags. We will discuss hesitation tags, toch tags, and or... tags in turn.

Head/body forward

In simple polar questions without a tag, ‘head forward’ is the most frequently used non-manual marker. The label forward on the ‘NMM.head-z’ tier was used in over 80% of SR-only polar questions in all experimental conditions except in NeutPos. In this condition, the proportion of question forms with head forward is slightly lower (0.71), but it is still the most commonly used non-manual marker. A comparable pattern can be observed in constructions ending with a hesitation tag, where it can additionally be noted that ‘head forward’ occurs more frequently on the tag than on the SR. Across all conditions and both question types, ‘body forward’ tends to occur less frequently, and in almost all questions in which a body forward position was observed, a head forward position occurred as well (but not necessarily vice versa).

Eyebrow configuration and eye shape

Both eyebrow raising / eyes wide (green bars) and eyebrow lowering / eye squint (orange bars) are attested in all experimental conditions and for both question types under consideration (SR-only and SR + HES). However, there is variation between conditions with respect to the relative distribution of these markers. Brow raising and lowering sometimes occur in approximately equal amounts (e.g., PosNeut; NeutNeut), while in some conditions there is a preference for brow raising (NeutPos) or – more commonly – brow lowering (e.g., NegNeut; PosNeg). The patterns for eye shape (squint-full vs. wide-full) by-and-large mirror these patterns for brow configuration, with a general trend toward slightly lower proportions for these markers. This may be the result of the fact that the annotation labels wide-half and squint-half were not included in the analysis, because of the poor inter-annotator agreement scores for these labels. In other words, it is likely that we are looking at an underrepresentation of eyes wide/eye squint events compared to brow raising/lowering events.

Mouth shrug

For SR-only questions, in all experimental conditions we find at least some uses of what we may refer to as a ‘mouth shrug’ (annotated as down on the ‘NMM.lip-corners’ tier). Proportions range between 14% and 37% across conditions. Strikingly, the use of mouth shrug is much more common in constructions made up of a SR and a hesitation tag, where this non-manual marker almost always accompanies the tag.

What is the main non-manual marker of polar questions / inquisitiveness?

We have seen that a head and/or body forward position is common across all experimental conditions. This leads us to argue in this section that a head/body forward position is the main marker of polar questions in NGT. Of course, this is not to say that a head/body forward position is not plurifunctional, as we argue is the case for eyebrow raising (and lowering), but we believe it is so to a lesser extent. Moreover, in contrast to what we observe for eyebrow raising vs. lowering, we do not find a competition between a forward vs. backward head or body position in NGT polar question forms.

First, let us comment on the relationship between ‘head forward’ and ‘body forward’. Figs 11 (SR-only constructions) and 12 (SR + HES constructions) present the same information about the proportions of head/body forward per experimental condition as Figs 9-10 previously, but additionally include a third bar in each subplot which indicates the proportion of question forms that include a forward label on either the ‘NMM.head-z’ tier, the ‘NMM.body-position’ tier, or both. From this additional information, we can infer that when a particular question form has a forward annotation on the ‘NMM.body-position’ tier, it almost always has a forward annotation on the ‘NMM.head-z’ tier, too (but not necessarily vice versa), since in almost all experimental conditions, the third bar in the subplot is equal in length to that of the first bar (head forward). This suggests that head and body forward are likely expressions of the same underlying non-manual feature, which we will refer to as ‘forward’. The fact that we observe a forward head position more frequently than a forward body position is probably best explained by articulatory considerations: it is easier to only move the head forward than to move the body forward without also moving the head.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 11. Proportions of head forward (‘h.fw’) and body forward (‘b.fw’) non-manuals in question forms per experimental condition, for questions consisting of just a SR.

The third bar in each subplot indicates the proportion of question forms with either head forward or body forward, or both (indicated by the disjunction symbol ‘∨’). The number of tokens per condition based on which the proportions are calculated are reported in the upper right corner of each bar chart.

https://doi.org/10.1371/journal.pone.0354015.g011

The proportion of ‘forward’ we observe in our dataset is in line with Coerts [12], who reported that 77.9% of the 95 polar questions in her NGT dataset include the feature ‘head forward’, making it the most common non-manual marker of polar questions in her data, too. In our dataset, 77.8% of SR + HES constructions involve a ‘forward’ feature on the tag (and 61.1% on the SR; see further discussion below). In SR-only polar questions, the proportion of instances involving a head and/or body forward position is even higher: 86.8%. We therefore suggest that a core function of ‘forward’ in NGT is to mark polar questions – or possibly inquisitiveness (i.e., a request for information from the interlocutor) more generally.

Some support for this latter hypothesis comes from the temporal properties of this feature in our dataset. Based on qualitative inspection of the data, we observe that when ‘forward’ accompanies only a part of the question form, it is most likely to occur toward the end of it. Indeed, this observation is supported by Fig 12 for SR + HES constructions, where ‘forward’ accompanies the tag part at the end of SR + HES constructions at least as frequently as the SR part across all experimental conditions. This is significant, because this position – at the end of the signer’s turn – is expected under the assumption that ‘forward’ signals a request for information from the addressee.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 12. Proportions of head forward (‘h.fw’) and body forward (‘b.fw’) non-manuals in questions forms consisting of a SR and a hesitation tag, shown per experimental condition and separately for each part (SR or tag).

The third bar in each subplot indicates the proportion of question forms with either head forward or body forward, or both (indicated by the disjunction symbol ‘∨’). The number of tokens per condition based on which the proportions are calculated are reported in the upper right corner of each bar chart.

https://doi.org/10.1371/journal.pone.0354015.g012

However, if ‘forward’ is a general marker of inquisitiveness, it should not only occur frequently in polar questions but also in content questions. Coerts [12] reports that 70.6% of the content questions in her dataset (24 out of 34 in total) involve the feature ‘chin up’, of which seven are combined with ‘head forward’. She does not report an overall percentage for ‘head forward’ in content questions, which means that it must be less than the 50% that she takes as a lower limit for reporting features. Coerts’s work thus provides some indication that ‘forward’ is substantially less frequent in content questions than in polar questions. This would support the view that it is a marker of polar questions rather than of inquisitiveness more generally, though further data is required to verify this conclusion.

A possible alternative explanation is that the use of other markers of inquisitiveness in content questions (e.g., wh-signs) preempt the need for a forward position of the head and/or body. Indeed, in a recent corpus study on British Sign Language, Hodge et al. [55] show that wh-signs and wh-mouthings most strongly distinguish content questions from polar questions. Just 15% of polar questions were signaled manually by means of one or more question signs, as opposed to 76% of all content questions. However, the authors did not observe marked differences between polar and content questions with regard to how they are marked non-manually. Almost all question forms in their dataset were produced with at least one (and generally more) non-manual signals. Eyebrow movements and head movements occurred approximately equally often in both polar and content questions, although there were some distributional differences regarding the specific type of signals that were preferred. While almost all questions involved some form of eyebrow movement, polar questions involved proportionally more upward eyebrow movements, while content questions more often involved downward eyebrow movements. Still, raised and lowered eyebrows clearly occurred with both question types. With regard to head movements, 70% of polar questions and 58% of content questions included one or multiple head movements, where it is notable that polar questions involved proportionally many more head nods than content questions. Content questions, on the other hand, were more likely to include a downward head movement, and somewhat more likely to include a sideward head movement. Strikingly, and in contrast with the NGT data presented in this article, both question types involved a forward head movement relatively infrequently (15% for polar questions and 23% for content questions). Thus, it appears that ‘forward’ is not the main marker of polar questions in BSL, let alone of inquisitiveness more generally, while we have shown that it is a marker of polar questions in NGT (note that Hodge et al. [55] do not discuss body movements). At the same time, it is evident that – like in NGT – raised eyebrows are not a reliable marker of polar questions in BSL either.

It is further relevant to draw a connection with Göksel and Kelepir [56], who propose that ‘head forward’ is the main indicator of polar questions in Turkish Sign Language. In addition, they argue that ‘head tilt’, which for them encompasses both ‘head forward’ and ‘head backward’, functions to signal inquisitiveness (a “response seeking utterance”), while the direction of the tilt signals question type: a forward head tilt for polar questions and a backward head tilt for content questions. In NGT, one may similarly hypothesize that ‘forward’ and ‘chin up’ both articulate the feature inquisitiveness, the former being used when the issue that is raised concerns just two alternatives (yes/no) and the latter when there are multiple alternatives. Indeed, in a corpus-based study on wh-questions in NGT, Legeland [57] reports that ‘chin up’ was fairly common in her data; however, she did not annotate forward head/body positions. Thus, further study is needed here.

As an aside, Göksel and Kelepir also associate head nods and headshakes with polar and content questions, respectively. In NGT, we have seen that headshake can function as a marker of inquisitiveness and/or uncertainty about the truth of the SR, besides its function as marking negative polarity. Head nods do not seem to be involved in marking inquisitiveness in NGT.

A final connection relevant to draw here is with work on co-speech gesture which has found that head/body forward often occurs with inquisitive constructions in spoken languages as well, in particular in clarification questions [58–61]. It has been suggested that the relation between leaning forward and inquisitiveness potentially has a functional explanation: leaning forward reduces the space between interlocutors so it makes it easier for the speaker to pick up a response. For this reason, speakers may have a natural tendency to lean forward when asking questions. We should note that this explanation is more plausible for speech than for signing: in many contexts, moving closer to the interlocutor is likely to make it easier to hear what they say, but it will not necessarily make it easier to see their signed response.

Returning to the present study, if ‘forward’ indeed marks polar questions in NGT, then one may wonder why it is not attested in all polar question forms in our dataset. After all, the experimental design was such that participants were prompted to always request information from their interlocutors. We can think of at least three possible lines of explanation to account for this disparity.

Firstly, it is possible that a formally competing non-manual, like a backward head and/or body movement, blocks the use of ‘forward’ in some cases. One could imagine this happening, for instance, when a participant wishes to express strong surprise about or disagreement with the CE provided by the second confederate. However, our data do not offer much support for this explanation, as ‘forward’ is quite common in the contrastive conditions PosNeg and NegPos. Arguably, it is in these conditions that the likelihood that a signer will convey surprise or disagreement is the greatest. Possibly – and informal observation of the data supports this view – any competition between ‘forward’ and ‘backward’ may be resolved temporally, with the backward movement occurring earlier in the utterance, followed by a forward movement later on.

Secondly, it is possible that in some cases, participants may have felt a less pressing need to obtain information from the interlocutor because they already had some idea about the interlocutor’s answer. This could be the case in experimental conditions with positive or negative CE (some information already provided by the interlocutor), especially in combination with neutral SB (no conflicting prior information). The primary function of the target question may then subtly shift from a request for information to a request for confirmation, which could be less strongly associated with a forward head/body position. This explanation fits well with the observation that the proportion of SR-only constructions with ‘forward’ in the condition NeutPos is relatively low (0.71). However, it holds up less well against the fact that the proportion of SR-only question forms with ‘forward’ in NeutNeg is relatively high (0.84).

A third possible explanation is that the use of other markers related to inquisitiveness could make ‘forward’ superfluous. This may explain the difference between SR and SR + HES. A hesitation tag, as we will argue in the section Function of mouth shrug, signals uncertainty about the truth of the SR on the part of the signer, thus implicitly inviting a response from the addressee in order to resolve this uncertainty. This may obviate the need for using ‘forward’. Other potential markers that can make the use of ‘forward’ superfluous could be, for instance, prosodically integrated palm-up or inquisitive headshake on the SR.

Altogether, the data clearly show that ‘forward’ is the most consistently used non-manual marker in polar questions across experimental conditions. We have interpreted this as an indication that it is the main marker of polar questions, or possibly even inquisitiveness in general, rather than any particular eyebrow configuration. In the next section, we address the polyfunctionality of eyebrow raising and lowering.

Factors associated with raised and lowered eyebrows

A striking pattern emerging from our data is that eyebrow raise is not as common as one might have expected given both previous work on NGT [10,12] and the fact that brow raising is widely regarded as prototypical of polar question marking across sign languages. Cecchetto [62], for instance, remarks in an overview chapter that “the importance of the eyebrow raise feature should be stressed, since it also discriminates polar questions from content (wh) questions in the many sign languages in which [...] content questions are marked by eyebrow lowering” (p. 294). Our data show that, at least in NGT, eyebrow raise is not a general marker of polar questions, underscoring previous experimental findings by De Vos et al. [33]. In fact, we found remarkably many instances of eyebrow lowering in our data.

Our general proposal, therefore, is that brow raising and lowering are both associated with a range of pragma-semantic factors that may or may not come into play depending on the context in which a polar question is uttered and the particular intentions that the signer wants to convey beyond just requesting a response that either affirms or denies the SR. Our objective in this section is to identify the factors we think play a role in order to account for the distributional patterns of eyebrow raising vs. lowering in our data. In the discussion, we will also make connections with relevant studies on both sign languages and co-speech gesture which suggest similar associations between particular brow configurations and the factors we discuss.

Before we begin our discussion, a couple of notes are in order. Firstly, because our primary aim here is to generate rather than test hypotheses about the functions of eyebrow raise and lowering, this section is of a more exploratory nature. Further study, e.g., in the form of a perception or acceptability judgment task, will be needed to gather additional empirical support for our suggestions. Secondly, we shall focus primarily on the set of question forms consisting of a SR only. Where possible or useful, we will also consider or compare constructions involving a SR followed by a tag, or the set of 30 baseline question sentences that we annotated. Thirdly, similar to what we observed for head forward and body forward, the occurrence of eyebrow raising and eyes wide as well as eyebrow lowering and eye squint are strongly correlated with one another, where the proportions for eye shape are typically slightly lower than the proportions for brow configuration. As such, it does not seem possible to associate one (but not the other) of the two markers in each pair with an independent function. In what follows, we will therefore use ‘eyebrow raising’ as a container term for brow raise and/or eyes wide, and ‘eyebrow lowering’ as a container term for brow lowering and/or eye squint.

To support the discussion, Fig 13 displays the distribution of eyebrow raising and lowering, respectively, in each experimental condition in the set of sentences consisting of a SR only.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 13. Proportions of brow position and eye shape non-manuals in question forms per experimental condition, for questions consisting of only a SR.

Brow raising (‘br.r’) and eyes wide (‘e.w’) are displayed on the left in green, and brow lowering (‘br.l’) and eye squint (‘e.sq’) on the right in orange. The third bar in each subplot indicates the proportion of question forms with either brow raising/lowering or eye widening/squinting, or both (indicated by the disjunction symbol ‘∨’). The number of tokens per condition based on which the proportions are calculated are reported in the upper right corner of each bar chart.

https://doi.org/10.1371/journal.pone.0354015.g013

Fig 14 shows the distribution of brow raise vs. brow lowering across experimental conditions separately for each type of polarity marking (positive, none, negative) on the SR. This figure reveals the following interesting three-way interactions between polarity marking, CE, and brow configuration:

1. When both CE and polarity marking on the SR are positive, brow raise is quite common while brow lowering is infrequent. The opposite pattern (brow raise infrequent; brow lowering frequent) is observed when CE is positive but there is no polarity marking.
2. This contrast between the forms with positive vs. no polarity marking is not observed when CE is neutral. There, we see that both eyebrow raising and lowering are quite frequent across conditions.
3. Finally, when there is negative polarity marking on the SR, brow raising is much less frequent than when there is positive or no polarity marking. On the other hand, brow lowering is still quite frequent when there is negative polarity marking.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 14. Proportions of brow raising vs. lowering in SR constructions with different polarity marking, for each experimental condition.

A: Constructions with positive polarity marking. B: Constructions without polarity marking. C: Constructions with negative polarity marking.

https://doi.org/10.1371/journal.pone.0354015.g014

Below we discuss a number of factors which we believe may be responsible for these three patterns.

Bouletic bias.

First of all, one function that raised brows seem to fulfill is the marking of positive bouletic bias (cf. [20]), i.e., signaling that the signer hopes that the SR is true. A representative example is given in (22). In the given situation, the signer wants there to be a train at 9, so she hopes that the SR is true. The eyebrows are raised across the entire sentence.

If eyebrow raise is indeed associated with positive bouletic bias, then we would expect to find differences in the occurrence of eyebrow raise across the five different situations in the production task. Situations 2–5 were such that the signer would naturally have a positive bouletic bias, hoping for an affirmative answer, in case they asked a question with non-negative polarity marking (that there be a train at 9, that entrance be free of charge, etc., see again S1 File for a description of all situations). Situation 1 is different: here, the signer cannot be assumed to have a positive (or negative) bouletic bias toward the SR (that Kim is a vegetarian). In setting up our experiment, we did not purposely control for bouletic bias. Unfortunately, we did not include any situation in which the signer would have a negative bouletic bias toward the SR when uttering a question with non-negative polarity marking (e.g., “So I need a[n expensive] ticket?”).

Now, of the 11 target questions consisting of only a SR with non-negative polarity marking that were elicited for situation 1, only one (9.1%) includes eyebrow raising. This compares to 51.4% eyebrow raises among the 70 SR-only questions with non-negative marking in situations 2–5 (for the plots, see Fig. A and Fig. B in S2 File). Of course, the number of question forms that these statistics are based on, especially for situation 1, is small, so any conclusions need to be drawn with caution at this stage. Nevertheless, these patterns are in line with what we would expect when positive bouletic bias is one of the factors associated with raised eyebrows.

Returning to the three main patterns observed based on Fig 14, the hypothesized association between positive bouletic bias (hoping that the SR is true) and brow raising can account in particular for pattern 3, i.e., the fact that brow raising is much less frequent when there is negative polarity marking on the SR than when there is positive or no polarity marking on the SR. This is because in case of positive or no polarity marking on the SR, we can assume that there is (in situations 2–5) a positive bouletic bias, and so we expect more brow raising. In contrast, if there is negative polarity marking on the SR, then we can assume that there is (again in situations 2–5) a negative bouletic bias (the signer hopes that the SR is false), and so we expect less brow raising. This is exactly what we see.

We are not aware of any previous studies on sign languages and/or co-speech gesture that make a similar connection between bouletic bias and eyebrow raise. However, the link between eyebrow raise and positive emotions such as hope are well-reported (e.g., [63]; see [64] for a recent multimodal study on participants’ use of eyebrow raises when talking about future hopes).

Bouletic bias cannot account for patterns 1 and 2, where we believe other factors are at play.

Seeking confirmation.

A second function that raised brows seem to fulfill in some cases, especially when combined with positive polarity marking (head nodding), is to signal a request for confirmation of the SR. This function is exemplified in (23), repeated from (7-b) with the addition of non-manual markers. The entire utterance is marked by raised eyebrows (plus widening of the eyes) and head nodding accompanies part of the sentence. The sense one gets when inspecting the utterance – as is also reflected in the English translation (‘So...’) – is that the signer is seeking confirmation of the proposition that the train leaves at 9. In this sense, the utterance differs from more neutral information-seeking polar questions.

A number of previous studies on sign languages as well as co-speech gesture have also argued that eyebrow raising can be associated with requests for confirmation. Analyzing a set of dyadic conversations in spoken English between university students of different European nationalities, de Souza [65] reports a relatively high proportion of brow raises in confirmation-seeking polar questions as compared to information-seeking ones. Several other studies focus specifically on the use of eyebrow raising in other-initiated repair questions. For instance, in a qualitative study on Argentinian Sign Language, Manrique [66] reports that brow raise is used in ‘offer type’ other-initiated repairs, in which the signer offers their interlocutor a candidate understanding to request for confirmation. Similarly, in a corpus-based study on native Dutch speakers, Homke et al. [67] show that what they call ‘restricted offer’ other-initiated repair questions co-occur more often with eyebrow raises than eyebrow lowerings. And as mentioned before, Cañas Peña [28] claims for LSC that a combination of eyebrow raise with body/head forward in question forms may be used in contexts where epistemic bias (equivalent to SB) is positive and evidential bias (equivalent to CE) is not negative. Although implicit, it seems that she attributes a confirmation-seeking function to this non-manual pattern, as she notes that “[i]f the signer is expected to be answered with a negation, this combination of non-manual markers is not licensed” (p. 12).

Returning to the patterns observable in Fig 14, the association between seeking confirmation and brow raising in combination with positive polarity marking partly accounts for pattern 1, namely that the combination of raised brows and positive polarity marking is relatively frequent when CE is positive. The fact that there is positive CE in this case (which moreover has been provided by the interlocutor) makes it very likely that the answer to the question will be affirmative, therefore we expect more confirmation-seeking questions.

An interesting open issue is whether, when CE is negative rather than positive, raised brows can also be used in combination with negative polarity marking to signal that the question is confirmation-seeking. In our dataset, we hardly ever observe eyebrow raising in such cases. One conclusion we could draw from this is that brow raise is not associated with confirmation-seeking questions in general but only operates in tandem with positive polarity marking to signal confirmation-seeking in the presence of positive CE. There is, however, an alternative explanation. It could be that the broader association between brow raising and confirmation seeking does hold, but there are factors associated with brow lowering that weigh so heavily that brow lowering typically wins out over brow raising in questions with negative polarity marking – even if the question is confirmation-seeking (see the section Factors associated with brow lowering for further discussion).

It is important to highlight here that we are not suggesting that, when there is positive CE, a polar question should always be regarded as a confirmation-seeking one. Rather, we believe that this only happens if the person asking the question is willing to accept the positive CE. On the other hand, if she is not immediately willing to accept the positive CE, either because she has reason to doubt it (epistemic bias) or to disprefer it (bouletic bias) she would not seek confirmation but rather express her uncertainty/doubt or dispreference. In this case, we suggest, the signer does not use positive polarity marking on the SR. This would explain the contrast in Fig 14 between questions with positive polarity marking and questions without polarity marking in the presence of positive CE. In the former, raised brows are frequent and signal confirmation-seeking, while in the latter, lowered eyebrows are more common here. We will propose in the section Factors associated with brow lowering that lowered eyebrows, among other things, can signal uncertainty/doubt and negative bouletic bias.

Introducing a new conversational topic.

A third function that we suggest raised brows can fulfill is to signal that the signer introduces a new conversational topic. This would account for two of our findings. Firstly, it would partially account for the relatively high proportions of eyebrow raise in questions following neutral CE, independent of polarity marking (pattern 2). Recall that in our experiment, the information provided by the second confederate always followed a question by the participant which inquired about a different matter than the target question (e.g., ‘Do you know where the entrance is?’; target question: ‘Is entrance free of charge?’). In case of positive or negative CE, the confederate’s response indirectly offers evidence for (e.g., ‘It’s there by the white flag. You don’t need a ticket’) or against (e.g., ‘It’s there by the white flag, but you need to get a ticket first’) the SR. In contrast, in the neutral CE conditions, there is simply an absence of evidence concerning the SR (‘The entrance is there by the white flag’). This means that in the case of neutral CE, the signer’s target question opens up a new topic in the exchange between the participant and the second confederate. As such, these target questions are similar to out-of-the-blue polar questions, which we know from the literature are typically associated with raised eyebrows (e.g., [15], among many others). A representative example, uttered by a participant in situation 3 with both neutral SB and neutral CE, is shown in (24). Note that we analyze this example as a question form without polarity marking. The single head nod accompanying the lexical sign free-of-charge-a is analyzed as a prosodic boundary marker, a known function of single head nod in NGT [10].

Thus, the suggested association between eyebrow raise and marking a new conversational topic contributes to the relatively high proportions of eyebrow raise in conditions with neutral CE. Secondly, it also predicts relatively high proportions of eyebrow raising in the baseline questions that participants were instructed to ask the first confederate in each role play. These questions also introduce a new conversational topic (rather than, e.g., asking for confirmation or clarification). Indeed, we find a similar proportion of eyebrow raising among these baseline questions (57%) as we observe among target questions without polarity marking following neutral CE.

A caveat here is that eyebrow lowering is very common in the set of baseline questions we annotated, too. This is surprising, because these questions, which participants were instructed to ask to the first confederate, introduce a new conversational topic. We suspect that the frequent use of eyebrow lowering here is an artifact of the experimental set-up: within the experimental context, the baseline questions were not uttered entirely out of the blue by participants, because each situation was repeated multiple times. It is possible that this unintentionally affected signers, leading to more frequent use of eyebrow lowering because they felt less of a need to signal that they were initiating a new conversational topic with each iteration.

Several previous studies on co-speech gesture have similarly identified an association between eyebrow raising and the introduction of a new conversational topic (e.g., [68–70]). We do not know of any such previous findings concerning sign languages, but it has been observed by many authors that brow raising is often used to mark grammatical topics (rather than new conversational topics) in sign languages, including in NGT [71].

The three functions of eyebrow raising discussed so far – signalling a positive bouletic bias, confirmation-seeking, and the introduction of a new conversational topic – together explain the three patterns found in Fig 14 concerning eyebrow raising. There are, however, two more relevant functions which we think eyebrow raising can fulfill in NGT, and which explain some of our data though this is not directly reflected by Fig 14. We now turn to these two additional functions: marking politeness and surprise.

Politeness.

We suggest that there is an association between eyebrow raising and politeness in NGT. We think that this factor particularly played a role in one of the five situations considered in our experimental setup, namely situation 5, in which the participants addressed a ticket seller at a train station. It is likely that the participants felt more pressure in this situation than in others, which were more informal, to formulate their questions in a polite way. This may have affected the data elicited in this situation in various ways. Firstly, as discussed previously in the section Other tags, all four instances of the confirmation-seeking tag CONF – non-manually characterized by a combination of head nodding, pressed lips, and raised (inner) eyebrows sustained from the SR – were elicited in this particular situation. We already suggested there that the CONF tag in combination with the use of these particular non-manuals signal politeness. Secondly, looking at all questions elicited in situation 5 that consist of a SR only (for the plot, see Fig. C in S2 File), we observe a relatively high proportion of eyebrow raising across conditions, while eyebrow lowering occurs in only a few instances – most of which following negative CE. Thus, we suggests that eyebrow raising functions to express politeness here. For a representative example, see (23).

Conversely, there is one situation in which eyebrow raising is relatively infrequent and eyebrow lowering is much more common. This is situation 4, in which the participant asks whether their roommate Kim is home to let them in. We suspect that the infrequent occurrence of eyebrow raising in this situation is in part due to the sense of urgency behind the target question (the participant would not be able to enter their home if it turned out their roommate was not there). As a consequence, and combined with the fact that the second confederate takes the role of another roommate, politeness considerations become peripheral, resulting in markedly less frequent use of eyebrow raise in this situation. An illustrative example is provided in (25); also note the use of nose wrinkle in this example.

Some previous studies have also associated eyebrow raising with politeness marking. For instance, Mapson [72] reports for British Sign Language that eyebrow raise is the most consistently used non-manual feature by all participants in an experiment in which polite expressions (requests and apologies) were elicited. In a multimodal study, eyebrow raising was shown to positively influence participants’ perception of the politeness of verbal statements with speakers of American English, while eyebrow lowering had the opposite effect [73].

Surprise.

Finally, we propose that signaling surprise is also one of the functions that eyebrow raising can fulfill in NGT, though, importantly, only when we define surprise in a specific, rather narrow way. We will refer to this notion of surprise as accepting, hopeful surprise. An illustrative example is given in (26).

This example is an instance of ‘accepting’ surprise in the sense that the signer, being confronted with new, unexpected information, accepts this new information as true. It is also an instance of ‘hopeful’ surprise in the sense that the signer has a preference for the new, unexpected information to indeed be true. Another possible reaction to new, unexpected information, especially if the signer was previously told exactly the opposite, would be to enter into a state of uncertainty about the truth of this new information, to doubt it, or even to express their dismay. In that case, we may also say that the signer is surprised by the new information, but it is a case of ‘non-accepting’ surprise. This is more likely to be signaled by eyebrow lowering, as in example (27).

We have suggested before that, in case of positive CE, the way in which a signer marks her question depends on whether she is willing to accept this positive CE or not. If she is, she is likely to use eyebrow raise in combination with head nodding; if she is not, she is likely to use eyebrow lowering to signal uncertainty/doubt. Now, what we are suggesting is that in the first case, i.e., if the signer is willing to accept the positive CE – if she is furthermore surprised by this new positive CE because she had a different expectation (SB), and if she is hopeful that the new positive CE is indeed true, then she is even more likely to use eyebrow raising to signal ‘accepting, hopeful’ surprise. In Fig 14, this concerns the NegPos condition in the leftmost diagram (for question forms involving positive polarity marking on the SR). While the frequency of brow raising in this condition is not higher than in some other conditions, including NeutPos, we do believe that signaling surprise (of the relevant specific kind) is one of the factors promoting the use of brow raising in this condition, in addition to the other factors discussed above. Based on qualitative inspection of our data, it seems that if multiple factors including surprise-signaling independently promote the use of brow raising, this typically leads to brow raising with greater intensity. However, we did not annotate the intensity of non-manual markers in this study, and thus we cannot make any quantitative claims on a potential link between particular (combinations of) factors and the intensity of eyebrow raising or lowering. In future work, we think there is much potential for Computer Vision tools to be utilized for a more precise quantitative analysis than manual annotations can offer (see [74,75] for examples of such studies).

In previous work on sign languages, co-speech gesture, and the expression of emotions, eyebrow raising has often been associated with signaling surprise (see, e.g., [63]). The discussion above also resonates with previous experimental findings on NGT [33]. In this study, the authors show that polar questions in NGT uttered in a surprised affective state involved somewhat more frequent use of eyebrow raise compared to questions asked in a neutral affective state. Moreover, De Vos et al. [33] also report a higher intensity of brow raises for polar questions asked in a surprised affective state. Note here that De Vos et al. worked with a general, underspecified notion of surprise. They just asked their participants to imagine themselves in a ‘surprised’ affective state, not distinguishing between accepting/non-accepting or hopeful/fearful/neutral surprise. It is plausible, however, that participants imagined themselves in an accepting, hopeful surprise state, as this seems to be a relatively salient sub-category of surprise.

To conclude this section, we have put forth five factors which we believe promote the use of eyebrow raising in polar questions in NGT, and we have argued that each of these factors only comes into play under certain circumstances (e.g., depending on whether the signer wants the SR of the question to be true or not in that situation), in the presence of a certain type of CE, and/or in the presence of a certain type of polarity marking on the SR. Table 12 summarizes the discussion.

Download:

• PNG
  larger image
• TIFF
  original image

Table 12. Factors promoting the use of brow raising in polar questions in NGT.

https://doi.org/10.1371/journal.pone.0354015.t012

Negative bouletic bias

The other factor we believe weighs into the preferred use of brow lowering over brow raising in polar questions with a negative SR is negative bouletic bias. The question in (28) illustrates this.

In the given situation, the participant wants Kim to be home, but the SR is the negative clause ‘Kim isn’t home.’ So, the participant has a negative bouletic bias: they hope that the SR turns out to be false.

In general, we suggest that the polarity of the signer’s bouletic bias, i.e., whether or not the signer hopes the SR to be true, is associated with contrasting brow configurations: brow raising signals a positive bouletic bias, while brow lowering signals a negative bouletic bias. This makes sense, because where positive bouletic bias may be associated with positive emotions like hope and happiness, which in turn are typically associated with eyebrow raise, negative bouletic bias can be connected to negative emotions like fear, sadness, anger and contempt, which are typically associated with eyebrow lowering (e.g., [63]).

Unfortunately, our current dataset does not allow us to determine the extent to which the expression of negation and negative bouletic bias independently promote the use of brow lowering, since in almost all cases in which a polar question with a negated SR is asked, there is also a negative bouletic bias (the signer hopes that the negated SR is false), and vice versa. This means that our data do not provide evidence as to what happens, for instance, in polar questions with a negated SR but positive bouletic bias (e.g., ‘So I don’t need to pay to get in?’). Our hypothesis is that both brow raising and brow lowering are possible in such cases, but brow lowering would only be used to signal uncertainty/doubt (a factor discussed below). In other cases, in particular in confirmation-seeking questions with a negated SR but positive bouletic bias, we expect that brow raising will be used. If this is indeed the case, negative bouletic bias would be a better predictor of brow lowering in polar questions than the presence of negation. However, further research is needed to settle this issue.

In any case, while the hypothesized associations between eyebrow lowering and negation and negative bouletic bias account straightforwardly for the fact that eyebrow lowering frequently occurs in polar questions with negative polarity marking in our dataset, eyebrow lowering is also rather common in cases where there is no polarity marking, as well as, to a lesser extent, positive polarity marking on the SR. These occurrences of brow lowering are not accounted for by the associations discussed so far. We propose that in these cases, eyebrow lowering generally has a different function, namely to signal uncertainty/doubt. We distinguish three types of uncertainty/doubt here: uncertainty about the truth of the SR, uncertainty about the interlocutor’s ability to answer the question (in short, about ‘addressee competence’, cf. [80]), and uncertainty about the CE that has been provided by the interlocutor right before the question is asked.

Uncertainty/doubt about the SR.

Uncertainty about the SR can be expected to play a particularly important role in polar questions that do not involve polarity marking, as in such cases it appears that the signer does not have a particularly strong bias toward either answer. Indeed, eyebrow lowering is rather common in questions without polarity marking across all experimental conditions. A representative example from the NeutNeut condition, where one could expect participants to be especially uncertain about the SR given the lack of contextual cues, is given in (29).

Of course, questions with hesitation tags (see the section Hesitation tag) are also interesting to consider in this context. In the section Discussion: Polar question structures and their contexts of use, we proposed that this type of tag functions as an uncertainty marker. We thus expect eyebrow lowering to commonly accompany this tag type, and as Fig. 10 shows, this turns out to be the case. See (30) for an example.

However, brow raising is also fairly common in questions with a hesitation tag. In part, this can be explained by the fact that some questions of this type involve both eyebrow raising and lowering in different parts of the sentence. In example (10-a) in the section Hesitation tag, for instance, there is brow lowering on the SR but brow raising on the tag. This suggests that hesitation tags and eyebrow lowering are not inextricably linked – in fact, brow raising is slightly more common on the tag than on the SR in questions with a hesitation tag. Perhaps the use of a hesitation tag obviates the need for brow lowering for the purposes of uncertainty marking, such that brow raising may be used instead, e.g., to signal that the question is confirmation-seeking.

Similarly, we may expect to see common use of brow lowering in questions that include an or... tag, as the use of such a tag may signal uncertainty on the part of the signer, for whom both answer options are still viable. We observe that brow lowering is indeed common across these questions, often accompanying both the SR and the tag (see, e.g., (12-b) in the section Or... tag). Eyebrow raising, on the other hand, is rarely attested.

Uncertainty/doubt about the interlocutor’s ability to respond.

Uncertainty about the interlocutor’s ability to respond (in short, about addressee competence) can be expected to surface in cases where the second confederate provided neutral CE. This may lead participants to suspect that the confederate might know the answer to the target question. If the first confederate provided positive or negative SB, then the participant might have a bias for one of the possible answers (resulting in less uncertainty about the SR), but they may nevertheless be uncertain about whether the second confederate can provide confirmation.

We indeed observe that eyebrow lowering is fairly common across conditions with neutral CE; however, other factors (such as the above-mentioned uncertainty about the truth of the SR) may also partially account for this pattern. As such, it becomes interesting to take another look at the questions with embedding of the ‘Do you know...’ type, which we discussed previously in the section Do you know.... Inspection of these eleven complex question forms shows that eyebrow lowering is indeed attested in such forms, as in (31), which was uttered in a context with negative SB and neutral CE (however, so is eyebrow raising; see (15-b) for an example). Since SB is not neutral, it seems plausible that eyebrow lowering, which accompanies the sentence-initial matrix clause and extends somewhat into the embedded clause, marks uncertainty about addressee competence rather than uncertainty about the SR. Moreover, the construction does not involve negation or negative bouletic bias. We also observe mouth shrug on the matrix clause (in the beginning of the utterance) as well as the matrix clause double (at the end of the utterance), which also appears to contribute to signaling uncertainty about whether the interlocutor knows the answer to the question expressed by the embedded clause.

Uncertainty/doubt about the contextual evidence.

Recall from Fig 14 that eyebrow lowering is generally strongly preferred over eyebrow raising in the two contrastive conditions PosNeg and NegPos; an example is given in (32). Note that there is one notable exception, namely questions with positive polarity marking asked in a context with negative SB and positive CE. In such cases eyebrow raising is more common. We suggest that positive bouletic bias, confirmation-seeking, and (accepting, hopeful) surprise play a crucial role here, all promoting the use of eyebrow raising; see the section Factors associated with brow raising.

We suggest that uncertainty or doubt about the CE provided by the second confederate (which conflicts with the signer’s prior belief) is the most important trigger of eyebrow lowering in such cases. If this is true, then we should expect to see less brow lowering in contrastive conditions if the signer has relatively high confidence in the second confederate, i.e., brow lowering may be overruled by politeness/authority considerations. This is the case in situation 5, where the second confederate (in the role of a ticket seller at the train station) has a comparatively high degree of authority. Indeed, we observe relatively little use of brow lowering in the contrastive conditions here as compared to the other four situations (see Fig. C and Fig. D in S2 File for the plots); however, numbers are small so this issue warrants further study.

There are two additional question forms that are interesting to consider here. Firstly, we may expect questions with a toch tag, when uttered in contrastive conditions, to involve eyebrow lowering to signal doubt about the second confederate’s utterance. As previously discussed in the section toch tag, six of the eleven questions with a toch tag occur in one of the two contrastive conditions (the remaining five occurred after non-neutral SB and neutral CE), where the polarity of the SR aligns with SB. As such, this construction signals that the signer believes the information provided by the first confederate is true, thus leading the signer to doubt the CE offered by the second confederate. As expected, all of these instances involve either brow lowering and/or inner brow raising, but never brow raising (for two examples, see (11) earlier). In most of the cases, the brow position is sustained from the SR onto the tag. Secondly, four of the seven ‘I thought...’-type question forms with embedding, discussed in the section I thought..., occur in one of the contrastive conditions. Like toch-type questions, these forms foreground the signer’s own expectation with respect to the SR while signaling that it conflicts with the information provided by the interlocutor. Again, all of these instances involve eyebrow lowering; see (16) earlier for two examples.

The three factors just discussed all make a connection between eyebrow lowering and uncertainty/doubt. This can again be linked to previous work on other languages: several studies on nonverbal communicative cues have similarly claimed that brow lowering may signal uncertainty (e.g., [81,82]), while Haviland [83] has described the use of brow lowering (in combination with mouth shrug) as a signal of “doubt, uncertainty, lack of knowledge or understanding, typically directed toward an interlocutor” in Zinacantec Family Homesign (p. 545). Furthermore, in a set of multimodal studies on Catalan and Dutch listeners’ perception of incredulity questions, i.e., questions that convey disbelief about a situation which contradicts the speaker’s own expectations, it is shown that facial expressions play an important role in the interpretation of such questions [84,85]. Importantly, the facial expression associated with incredulity questions is described as “a furrowing of the brows and a squinting of the eyes, often accompanied by a head shake” (p.360). The contrastive NegPos and PosNeg conditions in our experiment may, too, trigger question forms that express a sense of disbelief, i.e., incredulity questions. As we have shown, such forms indeed frequently involve eyebrow lowering.

Table 13 lists all the factors we have argued to be associated with brow lowering, indicating in which situations, which contexts, and for which question forms they play a role.

Download:

• PNG
  larger image
• TIFF
  original image

Table 13. Factors promoting the use of brow lowering in polar questions in NGT.

https://doi.org/10.1371/journal.pone.0354015.t013

Function of mouth shrug

We found that a mouth shrug (lipcorners down) very frequently occurred on hesitation tags across all conditions (36 out of 39 cases). In SR-only questions, mouth shrugs are much less common, but still occur in all conditions (proportions range between 14% and 37% across conditions). We propose that a mouth shrug expresses uncertainty in NGT. More specifically, we propose that it can signal uncertainty about the SR, as well as uncertainty about addressee competence (the addressee’s ability to answer the question that is asked). This proposal is supported by two facts. First, in our own data, mouth shrugs very frequently co-occur with hesitation tags, and we have argued that such tags also signal uncertainty (either about the SR or about addressee competence). And second, expressing uncertainty, doubt, or ignorance is one of the functions that has been ascribed to mouth shrug in previous studies on several other sign languages (e.g., [89] on Italian Sign Language; [90] on Danish Sign Language) and (co-speech) gesture (e.g., [82,91]. It is interesting to note here that Debras [91], building on earlier work by [92] and [93], speculates that mouth shrugs may be used in so many spoken languages to express uncertainty or ignorance because they iconically involve a configuration of the mouth which literally makes it impossible to speak, indicating that the speaker ‘has nothing to say’ on the topic at hand. Of course, in sign languages, the mouth is not the only articulator, and not even the most essential one. So for sign languages, the idea that a mouth shrug would iconically convey that the signer has nothing to contribute to the topic at hand does not fly. It is possible, in principle, that the mouth shrug first conventionalized in many spoken languages, and then made its way into sign languages through contact. We must leave further investigation of this issue for another occasion.

To return to our proposal, for uses of mouth shrug in SR-only questions, the hypothesis that this marker functions to signal uncertainty is still viable, even if there is no co-occurring hesitation tag, but with an interesting twist. About two-thirds of the instances of mouth shrug in SR-only constructions are compatible with an uncertainty function. In most of these cases, the mouth shrug occurs at the end of the SR, typically accompanying a prosodically light sign such as a pointing sign or a prosodically integrated palm-up. Other instances of mouth shrug, most of which appear at the beginning of the SR, seem to express other types of speaker attitudes, such as disappointment, certainty(!), or surprise. Based on these observations, we suggest that sentence-final mouth shrugs express uncertainty (about the truth of the SR or about addressee competence), while mouth shrugs that do not appear sentence-finally fulfill other functions.

In our experimental setting there was always some degree of uncertainty on the part of the signer when asking a polar question. Why, then, did mouth shrug not always occur? In particular, why did it occur much less often in SR-only questions than in questions with a hesitation tag? We suggest that this can be explained by the fact that in SR-only questions the mouth was frequently used to articulate mouthings or mouth actions accompanying lexical signs. Since hesitation tags involve manual signs such as palm-up, which are not lexically associated with a particular mouth action or with mouthing, the mouth and lips were freed up in this case to articulate a mouth shrug.