A coding scheme for request for confirmation sequences across languages

3.2.1 B1: Syntactic complexity

Which syntactic design does the confirmable exhibit in terms of its complexity?

The variable syntactic complexity [B1] captures basic aspects of the syntactic design, i.e. differences in the syntactic complexity that the confirmable of the RfC exhibits (Section 2.3).

– Clause

The confirmable is realised in a full-blown clausal/sentential form that contains a verb. Note that this category captures declaratives but also insubordinate clauses, as well as other complex sentence formats.

– Phrase

The confirmable is realised as a phrase that consists of one or more lexical items without a verb.

– Other

The confirmable consists of more than a single phrase, but it does not exhibit a syntactically full-blown structure (such as ellipses or anacolutha).

3.2.2 B2: Syntactic formats

Which syntactic formats are used for formulating RfCs?

We added another open-class variable without a pre-set list of options in order to account for language-specific ways of formulating RfCs. For the variable syntactic formats [B2], coders were asked to indicate different syntactic structures that proved to be relevant in their specific language (such as verb-first vs verb-second sentences in English (Küttner and Szczepek Reed 2024) or Low German (Weber 2024)). Due to the large variety of language-specific coding options, this variable is not examined in the cross-linguistic paper comparing RfCs in ten different languages (Pfeiffer et al. forthcoming) and is not included in the ICR check (Section 4). Some articles in the special issue nevertheless address this variable for a more detailed account of the RfC formatting in their specific languages.

3.2.3 B3: Polarity

In which polarity is the RfC cast?

Like the variables syntactic complexity [B1] and syntactic formats [B2], this category is also concerned with the morphological or syntactic design of the confirmable. It asks whether an RfC is formatted with positive or negative polarity, as this may have an impact on the design of confirming or disconfirming responses (Heritage and Raymond 2021, Sadock and Zwicky 1985). The syntactic marking of negative polarity can vary from language to language. It can be done with forms such as negation articles (such as kein (‘no’) in keine Kohle (‘no money’) in German, Deppermann et al. 2024), negation particles (such as miš (‘not’) in Egyptian Arabic, Marmorstein 2024), or it can be marked by a bound morpheme such as post-verbal affixes in Korean (Kim 2024b). We did not include markings that operate on a lexical level (such as English satisfied vs dissatisfied). All RfCs with no explicit marker for negative polarity (including phrasal RfCs such as Arabic ʔāxir šahrĭ ʔuktobar, ‘at the end of October’, Marmorstein 2024) were coded as positive. In sum, this variable is primarily form-oriented in its design. We did not include action-based interpretations of polarity, like counter-oriented argumentative pragmatic implicatures, for example.

– Negative polarity

This code is used if the RfC contains a negative polarity item.

– Positive polarity

This code is used if the RfC does not contain a negative polarity item.

3.2.4 B4: Type of negative polarity

Which type of negative polarity marking is used?

Similar to the coding of syntactic complexity [B1], the coding of polarity [B3] provides a restricted set of coding options, which is unspecific regarding the respective language-specific options. Thus, if the variable polarity [B3] was coded as ‘negative polarity’, coders were asked to list the negative polarity items used in the RfC in this variable [B4]. As resources differ between languages, there was no pre-set list of options. If the polarity of the confirmable was positive, the cells were coded as ‘NA’ (not available). Like the variable syntactic formats [B2], this variable is not included in the quantitative cross-linguistic comparison (Pfeiffer et al. forthcoming), but was relevant for some of the language-specific studies (see e.g., Kim 2024a).

3.2.5 B5: Modulation

Is the confirmable modulated concerning the requester’s stance towards the confirmable’s validity?

In this category, we coded for the lexical (e.g., modal particles, modal adverbs, modal verbs in epistemic use, vagueness markers or approximators), morphological (e.g., subjunctive verb mode) and syntactic resources (e.g., conditional clauses) that requesters use to modulate the RfC and therefore position themselves with regard to the subjectively assumed validity of the confirmable (epistemic stance). These markers can mitigate or reinforce the requester’s commitment to the validity of the confirmable (such as the commitment marker =la in Yurakaré, Gipper 2024). Hedging devices, which serve to mark approximation or imprecise reference (e.g., in the Czech RfC to prostě- to je trošku jako krb; jako kamna; ‘that simply it’s a bit like a fireplace; like a stove;’ Oloff 2024), are also coded as modulations. RfCs formatted as negative interrogatives can be used to put up a positive assertion for confirmation (Heritage 2002, Hentschel 1996, such as in English: “Wasn’t your phone broken for a while?,” Küttner and Szczepek Reed 2024). Such instances of non-propositional negation were also coded as a type of modulation. If a confirmable contains modulation markers, coders were asked to list the markers. For the cross-linguistic analysis by Pfeiffer et al. (forthcoming), the variable [B5] was later transformed into a binary dummy variable consisting of a numerical ‘0–1’ code.

– Modulation marker

If a confirmable contains one or several modulation markers, the items are coded as characters.

– No modulation

This code is used if a confirmable is not modulated.

3.2.6 B6: Inference marking

Is the RfC marked as an inference from prior talk?

This variable refers to linguistic resources speakers deploy to mark the confirmable as inferred from prior talk (upshots, candidate understandings, formulations; see Heritage and Watson 1979, Zinken and Küttner 2022). While there may be RfCs in which speakers present an inference without explicitly framing it as such, here we were only interested in coding those forms which explicitly mark the confirmable as an assumption or upshot which is inferred from what the other speaker has said before (see König and Pfeiffer forthcoming a for a methodological reflection). These markers include, among others, expressions such as Hebrew ‘az (‘so/then’, Ben-Moshe and Maschler 2024), Mandarin jiu(shi) (‘just’, Li 2024), Spanish pues (‘so/because’, Ehmer 2025) or change-of-state tokens if they indicate that new information has been inferred from prior talk (as opposed to marking something as ‘just remembered’, see e.g., Betz and Golato 2008). Comparable to the coding of modulations [B5], coders were asked to list an inference marker, if an RfC consists of some form of overt inference marking. For the quantitative cross-linguistic analysis (Pfeiffer et al. forthcoming), this character variable was also transformed into a dummy variable by replacing the characters with a binary ‘0–1’ code.

– Inference marker

If an RfC contains forms of overt inference marking, the items are coded as characters.

– Without inference marking

This code is used if an RfC does not contain any overt inference marking.

3.2.7 B7: Connectives

Does the RfC contain connective devices?

In this category, we coded for connective devices, such as conjunctions and conjunctive adverbs (e.g., Hebrew ‘aval ‘but’, Ben-Moshe and Maschler 2024 or Czech tak ‘so/then’, Oloff 2024). In contrast to the established linguistic concept of connectives, the present study also subsumes discourse markers (such as English well, Küttner and Szczepek Reed 2024, or Arabic ṭayyib ‘okay’, Marmorstein 2024) and change-of-state tokens (e.g., German achso ‘I see’, Deppermann et al. 2024) under the category of connectives as they link the RfC to previous discourse. Inference markers in [B6] are also coded as connectives in this category. Similar to the categories modulation [B5] and inference marking [B6], if an RfC contains connective devices, coders were asked to list them. For the quantitative comparison (Pfeiffer et al. forthcoming), this character variable was transformed into a dummy variable by replacing the characters with a binary ‘0–1’ code.

– Connective item(s)

If an RfC contains connective devices, the items are coded as a character.

– Without connective(s)

This code is used if an RfC does not contain any connective devices.

3.2.8 B8: Tag

Is the RfC appended by a tag?

This category documents the presence or absence of tags in RfCs, understood as more or less formulaic expressions appended to an RfC that put the confirmable up for debate. Tags may function as epistemic stance markers that mobilise response (Enfield et al. 2012, Stivers and Rossano 2010). However, they can also express affective stances (for Egyptian Arabic see Marmorstein 2024, for Korean see Kim 2024b), which is why the common expression ‘question tag’ seems not to be suitable for all languages. They differ from grammaticised final particles (as, for instance, the question particle -ma in Mandarin, Li 2024) in that they are optional. While the category as we operationalise it here is open to different forms (such as reversed polarity tags like isn’t it in English (Küttner and Szczepek Reed 2024) or particles such as Czech jo ‘yes’ or ne ‘no’, Oloff 2024) and different degrees of prosodic (dis)integration (in contrast to other definitions of final particles, e.g., Hancil et al. 2015), it is specified in terms of its position as an RfC-final element, i.e. we are interested in syntactically peripheral or post-positioned resources and their interactional import. Unlike turn-final conjunctions, tags do not refer to ‘unverbalised’ propositions, e.g., to ‘hanging implications’ (Mulder and Thompson 2008) or to expandable utterances (Koivisto 2012). Again, if an RfC includes a tag, coders were asked to list the tag as a character. The cross-linguistic quantitative analysis (Pfeiffer et al. forthcoming) turns this variable into a binary ‘0–1’ coding.

– Tag

If an RfC contains a tag, the tag is coded as a character.

– Without a tag

This code is used if an RfC does not contain a tag.

3.2.9 B9: Prosodic integration of tag

If there is a tag: Is the tag prosodically integrated into the RfC?

If the RfC contains a tag, this category coded if the tag is prosodically integrated into the RfC (i.e. confirmable and tag are part of one intonation contour) or if it is prosodically disintegrated (i.e. the tag is realised in a separate intonation contour). There may be pauses between confirmable and tag, but these were not criterial for a tag to be coded as disintegrated; it is also possible that latching occurs between the two components.

– Integrated

This code is used if the tag is prosodically integrated into the confirmable (if they are part of one intonation contour).

– Disintegrated

This code is used if the tag is not prosodically integrated into the confirmable (if they form separate intonation contours).

– NA

This code is used if the RfC does not contain a tag.

3.2.10 B10: Final intonation confirmable

What is the confirmable’s final pitch movement?

In this category, we coded for the final pitch movement of the confirmable (including those RfCs with a prosodically integrated tag, see variable prosodic integration of tag [B9]). If a tag is appended in a separate intonation contour, its prosodic design was coded separately (see variable final intonation tag [B11]). If a confirmable is realised in more than one intonation phrase, only the final pitch movement of the last intonation phrase was coded.

– Rise

This code is used if a confirmable is deployed with a rise-to-mid or a rise-to-high pitch contour.

– Level

This code is used if a confirmable is deployed with a level pitch contour.

– Fall

This code is used if a confirmable is deployed with a falling-to-mid and falling-to-low pitch contour.

3.2.11 B11: Final intonation tag

If there is a tag in a separate contour: What is the tag’s final pitch movement?

If the RfC contains a tag in a separate intonation contour, the final pitch movement of the tag was coded separately.

– Rise

This code includes rise-to-mid and rise-to-high pitch movements.

– Level

This code captures level pitch movements.

– Fall

This code includes fall-to-mid and fall-to-low pitch movements.

– 0

This code is used if an RfC contains a tag but without a separate contour.

– NA

This code is used if the RfC does not contain a tag.

3.3.1 C1: Response

Is there a response?

In order to analyse RfCs as an adjacency pair sequence, we coded whether an RfC receives a response or not. Responses in [C1] consist of verbal only, embodied only, or combinations of verbal and embodied features.

– Yes

This code is applied if a verbal, an embodied, or a verbal and embodied response follows the RfC.

– No

This code is applied if no verbal, embodied, or verbal and embodied response follows the RfC.

3.3.2 C2: Verbal response

Is there a verbal response?

This category documents the occurrence or non-occurrence of a verbal response (or several verbal responses by different participants) to the RfC. It includes all utterances by other participants of the conversation that can be heard as dealing with the RfC, even if they do not directly answer it (so-called non-answer responses, among them repair initiations, hesitation particles, claims of not knowing, Stivers 2022, Chapter 3). Since verbal responses (in contrast to embodied responses [C3]) can be compared across all languages that are part of the research network, the [C2] coding forms the basis for the following NA-codings.

– Yes

This code is used when the RfC receives a verbal response.

– No

This code is used when the RfC does not receive a verbal response (e.g., a purely embodied response ([C3])).

– NA

This code is used if the variable reponse [C1] is tagged as ‘no’.

3.3.3 C3: Embodied response

Is there an embodied response?

In order to be able to determine the relevance of bodily resources in responses to RfCs, coders were asked to identify them in variable [C3], if video recordings were available. Note that we do not include this category in the cross-linguistic quantitative analysis, as some researchers studied telephone conversations or only had access to audio recordings of the conversations investigated. Given that recent studies include head nods or head shakes as type-conforming interjections in their (comparative) quantifications (Enfield et al. 2019, Stivers 2019, 2022), we nevertheless asked network members working with video data to code them to be able to determine their role in responses to RfCs – in particular those without verbal uptake. As various collections show that bodily resources usually co-occur with verbal responses (Oloff 2024) and rarely form a full response on their own (Deppermann et al. 2024), we decided not to include head nods and head shakes as response tokens (see [C6]).^[5]

– Yes

This code is used for (standalone or co-occurring) bodily resources that potentially deal with the conditional relevance established by the RfC (such as head nods, head shakes, shrugs, hand or facial gestures).

– No

This code is used if the RfC does not receive an embodied response.

– NA

This code is used if the variable response [C1] is tagged as ‘no’. Moreover, NA is used for audio-only data or if resources are not clearly identifiable.

3.3.4 C4: Multiple responses

Are there responses from different co-participants in a multi-party conversation?

This variable captures instances in multi-party interactions in which more than one speaker responds or starts to respond to a previous RfC. Here, the coding scheme mainly works as a heuristic instrument. That is, rather than analysing all responding turns and their relationship, the scheme makes ‘multiple responses’ identifiable for future research on speaker selection, response timing, or response design in multi-party interaction. In addition, marking dyadic conversations with a particular code (NA) enables researchers to filter for differences in the participation framework.

If multiple responses occur, we decided to code only one of them. The decision not to study all responses raised the question of which of them should form the basis for further coding. In the case of (slightly) time-delayed responses, coders were asked to code the response that starts first (Stivers and Enfield 2010, 2625). In rare instances in which responding turns start simultaneously, we coded the response of the selected speaker (in particular through gaze, but also address terms, etc.) or, if the selected speaker could not be clearly identified, the speaker with higher epistemic rights to respond. If none of these criteria could be applied, which virtually never happened, the more elaborate response was coded (e.g., if a minimal and a non-minimal response occurred, see [C9], the decision was made in favour of the non-minimal response).

Since some of the languages in our sample were investigated based exclusively on dyadic conversations, we did not include this category in the cross-linguistic analysis (Pfeiffer et al. forthcoming). Therefore, it was also excluded from the ICR test (see Section 4).

– Yes

This code is used if more than one speaker responds or starts to respond to the RfC.

– No

This code is used if only one speaker responds or starts to respond to the RfC.

– NA

This code is used if the variable response [C1] is tagged as ‘no’ or a conversation is dyadic.

3.3.5 C5: Prefacing elements

Does the response contain prefacing elements?

This category captures if the response is introduced by particular prefacing elements such as discourse markers (e.g., Spanish bueno ‘well’, Ehmer 2025). Laughter, audible breathing, or other liminal signs (Dingemanse 2020) were not included in this category, as the recording quality of the different data sets differs. Moreover, response tokens (see [C6]) are not counted as prefacing elements, but are part of a response ‘proper’.

– Yes

This code is used if a response is introduced by (a) prefacing element(s).

– No

This code is used if a response is not introduced by (a) prefacing element(s).

– NA

This code is used if the variables response [C1] and verbal response [C2] are tagged as ‘no’.

3.3.6 C6: Response token

Is the response done with a response token?

A response token is defined as the smallest or minimal conventionalised communicative unit of a language that can provide confirmation or disconfirmation. It is a “dedicated answer form” (Enfield et al. 2019, 289) of a given language that “[does] not assert a proposition in and of [itself] but [does] confirm one” (Enfield et al. 2019, 288). In contrast to Enfield et al. (2019), embodied resources such as nodding or head shakes are not included in the coding of a response token (see [C3]). In this study, a response token is always a verbal (or combined verbal and embodied) entity and has a certain stability of form which does not mirror elements of the RfC (so repeats are excluded, see [C10]). The term ‘token’ recognises that, in some languages, there seems to be a gradual rather than a categorical boundary between non-inflecting monomorphemic expressions (often referred to as particles or interjections) and formats that might still exhibit some traces of inflection or syntactic integration (Gipper 2024, Kim 2024b). Sometimes, several response tokens occur within a single response. If a response contain at least one response token (irrespective of its position in the turn), coders were asked to list them. For the quantitative overview, this was later transformed into a numerical ‘0–1’ code (Section 4).

– Response token

If a (verbal) response contains one or several response tokens, they are coded as a character.

– No (response token)

This code is used if a (verbal) response does not contain a response token.

– NA

This code is used if the variable verbal response [C2] is tagged as ‘no’.

3.3.7 C7: Cluster of response tokens

Is there a cluster of response tokens?

If a response token is identified in [C6], [C7] captures instances in which two or more response tokens are uttered in one prosodic unit. This includes multiple sayings of the same token (Stivers 2004), as long as they are not lexicalised in the given language (e.g., jaja, a double saying of ‘yes’ in German, Golato and Fagyal 2008, as well as uses of different response tokens such as ʔāh ṭabʕan ‘yes of course’ in Arabic, Marmorstein 2024).

– Yes

This code is used if there is more than one response token in a prosodic unit following the RfC.

– No

This code is used if there is only one response token or if two or more tokens are uttered in separate prosodic units.

– 0

This code is used if there is a verbal response [C2] without a response token (e.g., if the response is formatted as a repeat [C10]).

– NA

This code is used if the variable verbal response [C2] is tagged as ‘no’.

3.3.8 C8: Position of response token

In which position within the response turn do speakers use the (first) response token?

If there is a response token [C6], coders were furthermore asked to indicate where in the response turn the token is located (Raymond 2013). If there is more than one response token in the turn, the position of the first token is coded.

– Initial position

This code is used for turn-initial response tokens (including instances with prefacing elements [C5] and responses which consist of a response token only, see [C9]).

– Mid position

This code is used for response tokens in turn-mid position.

– Final position

This code is used for response tokens in turn-final position.

– 0

This code is used if there is a verbal response [C2] without a response token.

– NA

This code is used if the variable verbal response [C2] is tagged as ‘no’.

3.3.9 C9: (Non)-minimal response

Is the response minimal or non-minimal, i.e. does the response consist of more than a response token or a cluster of tokens?

This category differentiates between minimal and non-minimal responses that include a response token (Gubina et al. 2024, Keevallik 2010, Steensig and Heinemann 2013, Seuren and Huiskes 2017) and other response formats. Minimal responses are type-conforming responses (Raymond 2003) that only consist of a confirming or disconfirming response token (see [C6]) or a cluster of response tokens (see [C7]). Non-minimal responses are type-conforming responses that consist of an initial confirming or disconfirming response token or a cluster of response tokens and a turn expansion by the respondent – regardless of whether there is a pause in between the response token (or the cluster of response tokens) and the expansion. The code ‘other’ refers to cases in which a response does not contain a response token (e.g., if the response is formatted as a repeat [C10]).

– Minimal response

This code is used if the response contains a response token only or a cluster of response tokens.

– Non-minimal response

This code is used if the response contains more than one response token or a cluster of response tokens.

– Other

This code applies to responses without a response token, e.g., full repeats [C10]).

– NA

This code is used if the variable verbal response [C2] is tagged as ‘no’.

3.3.10 C10: Full repeat

Does the response include a full or full-expanded repeat?

This category codes for responses that repeat the RfC in full, i.e. reproduce the original proposition without replacing or deleting elements of the original utterance (like in the German RfC sequence A: ihr seid auto gefahren, ‘you drove by car?’ B: hm_hm? wir sin auto gefahren ‘uhu, we drove by car’, Deppermann et al. 2024). Moreover, we identified full-expanded responses that repeat the RfC in full and add lexical elements (such as noun phrases, adjectives, or focus particles [as for instance in the Arabic RfC sequence A: huwwa wi mrāt-u? ‘he and his wife?’ – B: huwwa wi mrāt-u wi l-bint wi kullu; ‘he and his wife and the daughter and everyone’, Marmorstein 2024, 23]).

In identifying full and full-expanded repeats, we allowed for pronominalisation and deictic or pronominal shifts (such as in the Yurakaré RfC sequence A: nij mim, ‘you didn’t catch any [fish]?’ B: nij mii, ‘I didn’t catch any’, Gipper 2024), the omission of tags (such as in the Low German RfC sequence A: de heff KLAssentreffen hat; =ne, ‘they had class reunion, right?’ B: de heff KLAssentreffen hat, ‘they had class reunion’, Weber 2024) and the omission of connective devices, inference marking or modulation (as in the British English RfC sequence A: oh Emma’s pregnant B: Emma’s pregnant, Küttner and Szczepek Reed 2024, 30, or the German sequence A: der hat aber auch ne eigene firma; =oder? ‘he has his own company though, hasn’t he?’ B: der hat ne eigene firma. ‘he has his own company’, Deppermann et al. 2024). Moreover, we included instances with inversion (‘Has he called X? – He has called X.’) or negation (‘She called X? – She did not call X.’).

Note that the operationalisation of full repeats presented here is stricter than in other studies as we exclude partial repeats in which propositional elements are deleted (e.g., ‘Is John coming?’ – ‘John is.’ or ‘He is.’ (Enfield et al. 2019, 288; see also Stivers and Enfield 2010) due to poor results in the reliability check (see Section 4). In line with Enfield et al. (2019), we did not include transformative responses (Stivers and Hayashi 2010).

– Full repeat

This code applies to responses that fully repeat the confirmable without adding any lexical elements except for response tokens.

– Full-expanded repeat

This code applies to responses which fully repeat the confirmable but also add lexical elements as outlined above.

– Other response formats

This code applies to responses which do not include a full or a full-expanded repeat (e.g., with a response token [C6]).

– NA

This code is used if the variable verbal response [C2] is tagged as ‘no’.

3.3.11 C11: Responsive action

Which action is implemented by the response?

This category codes for the types of responsive action that follow the RfC (confirmation or disconfirmation). It also registers instances in which speakers neither confirm or disconfirm. Note that this category does not relate to particular response formats, i.e. confirmations or disconfirmations could be done with token responses and full or full-expanded repeats as well as responses which are cast in other formats (such as in the Low German RfC sequence A: aber de kUmmt doch hier ut EMSland; =ne, ‘but he is from Emsland, right?’ B: ut WERLte; ‘from Werlte (a city in the Emsland region)’, Weber 2024), including embodied responses [C3].

– Confirmation

This code is used if the response clearly confirms the RfC, e.g., with a confirming ‘yes’-response token, a repeat, or a combination of verbal and embodied resources.

– Disconfirmation

This code is used if the response clearly disconfirms the RfC, e.g., with a disconfirming ‘no’ response token, a repeat in reversed polarity, or a combination of verbal and embodied resources.

– Neither

This code is used for a range of responsive actions that neither confirm nor disconfirm clearly. Responses in this category may push against epistemic assumptions like, for instance, non-answer turns that express lack of knowledge or that perform or initiate repair (Stivers 2022, chapter 3). They may be evasive or non-committed (Gipper 2024, Kim 2024b), offer conditional confirmation (Gipper and Groß 2024), transform the terms of agreement (Stivers and Hayashi 2010) or expressions of thankfulness, wish, or hope (such as German das hoff ich ‘I hope so’, Deppermann et al. 2024, or Arabic bi-zni llāh ‘with God’s permission’, Marmorstein 2024).

– NA

This code is used if the variable response [C1] is tagged as ‘no’.

As indicated above, the project’s main goal was to identify and compare the resources used in requesting confirmation and delivering confirming or disconfirming responses across languages. Accordingly, the scheme as we present it here primarily aims at capturing the linguistic design of the two turn types in a descriptive manner. So, instead of coding whether an RfC presents inferences from prior talk, for instance, we code the occurrence of inferential markers (category [B6]). Or, we code response tokens (C6) or full repeats (C10) as features of the response design without assuming particular interactional functions at the same time (König and Pfeiffer forthcoming a provide a methodological reflection). This procedure to code form and function separately allows us to untangle in subsequent case-by-case analyses how speakers of a language interpret certain resources functionally and how they are fitted to their context of use.