*5.1. Data*

The current study is based on the Kata Kolok Corpus, a naturalistic data set of deaf Kata Kolok signers of generations II through VI (de Vos 2016). The corpus is stored and archived in *The Language Archive* at the Max Planck Institute for Psycholinguistics, The Netherlands (König 2011). For the purpose of this study, three dialogues between close friends with a relatively high level of transcription detail were selected. The most important selection criterion was to cover signers from different generations. Given that there are very few recordings of a single generation II-signer available and that generation VI consists of infants and small children, the final data set comprised generations III through V. Details of the sample, including their length in minutes, are provided in Table 1. The variance in the length of recordings will be taken into account by reporting values of negation per minute rather than absolute frequency.

**Table 1.** Detailed information on the sample used in the present study.


With the exception of Palfreyman (2019), who reports a correlation between gender and syntactic position of the negator in two urban signing varieties of Indonesia, gender has never been reported to affect the grammatical realization of negation. For Kata Kolok, there is some indication that gender may affect lexical variation, and we therefore account for individual variation in the statistical analyses by adding signer ID as a random intercept (cf. Mudd et al. 2020).

#### *5.2. Coding and Procedure*

Although the selected data included detailed transcriptions, all files were enriched by manual coding, using the annotation software ELAN (Wittenburg et al. 2006; *ELAN [Computer Software]* (version 5.9) 2020).<sup>5</sup> Moreover, while negative forms such as negative interjections, negative existentials, etc. were included in the initial coding, our report here focuses on standard negation. In addition to coding the manual and non-manual activity, separate tiers were dedicated to selected functional and analytic information (for the coding scheme, see Appendix A). All coding was done by the first author and proceeded in three rounds, initially targeting all instances of negation, then completing the information in the remaining tiers, before reviewing the coding in a final round. Data were systematically checked for missed tokens by searching for NEG, headshakes, and tongue protrusion in the

prior transcriptions. Some instances had to be excluded due to reasons such as (i) absence of a felicitous translation of the utterance, (ii) the camera being out of focus temporarily during the recording, and/or (iii) bad lighting conditions in the video. Coding presented several challenges, of which the most frequent ones are addressed briefly.

First, just as in Spanish and other spoken languages, the means of negative interjection and the clause negator are formally identical in Kata Kolok. Second, Kata Kolok relies heavily on shared knowledge and context, which makes the omission of sentential constituents a very common pragmatic strategy; as a result, elliptic standard negation and negative interjections are not always distinguishable. We coded conservatively by excluding instances with subtle articulatory breaks or changes within the accompanying non-manuals, as these features indicate separate prosodic domains as would be expected in the case of a negative interjection (Sandler 1999). Third, together with the frequent omission of constituents in spontaneous discourse, it was not always straightforward whether the negative element operated on a declarative clause or a negative existential. Fourth, every instance that did not clearly involve a negative existential meaning was coded as standard negation. Fifth, and finally, instances of negator doubling are noted as such in the comment tier, but counted as a single negative sentence. Similarly, immediate repetitions of the same negative utterance were counted as a single instance of negation.

We double-coded 10% of the data (11 min) to provide an intra-coder reliability measure, ensuring the validity of the findings of the present study. Cohen's Kappa was calculated using the irr package (Gamer et al. 2012) in R (R Core Team 2019) and yielded substantial intra-coder agreemen<sup>t</sup> between both rounds of coding (κ = 0.951; z = 15.6; *p* < 0.05) (Fleiss et al. 2003).
