Nasalization in Korebaju: An Evaluation of Reproducibility in Laboratory Phonology and Phonetic Fieldwork

Nasalization in Korebaju: An Evaluation of Reproducibility in Laboratory Phonology and Phonetic Fieldwork Jenifer Vega Rodriguez1,2, Thiago Costa Chacon1,Nathalie Vallée2, Christophe Savariaux2 1, Brasilia University, Brasilia, Brazil. 2,Univ. Grenoble Alpes, CNRS, Grenoble INP, GIPSA-lab, 38000 Grenoble, France Abstract This study investigates nasalization in the Korebaju language (also known as Koreguahe, a Tukanoan language spoken in Colombia) through an approach that explores reproducibility in Laboratory Phonology and Phonetic Fieldwork. The major goal is a study of inferential replicability (“Type E Reproducibility”) as it applies laboratory and statistical methods aiming to test the phonological analysis of nasal harmony in Korebaju – including aspects such as directionality, target segments, blockers, and transparent segments – as proposed by Cook and Criswell (1993) using qualitative and impressionistic methods more than 30 years ago. This general goal is supported by two experiments on reproducibility. In the first, we explore the inferential replicability of our findings (“Type B Reproducibility”) by an experiment where a single speaker produced two datasets with comparable stimuli using two different phonetic instruments, each with a different method for capturing nasal airflow data: the EVA2, which employs nasal tubes, and the EGG-D800 uses a silicone mask. The study compared how these distinct instruments enable similar conclusions to be reached despite the differing natures of the data. In a second experiment, we tested for the repeatability of the results (“Type C Reproducibility”) by using the same stimuli and instrument EGG-D800 on two different subjects, with the same gender and age, but under different environmental conditions, one in a Phonetics Laboratory in France and another in a Korebaju village in the Colombian Amazon. In all experiments, the nasalization index from the original study was maintained, calculating the proportion of nasal airflow. Further statistical procedures were equally applied to each dataset individually, evaluating their comparability and how each of them corroborated the phonological hypotheses about phonetic differences between nasal and nasalized vowels, as well as nasal, nasalized, opaque and transparent consonants The findings provide evidence for the robustness of nasalization indices, as well as the relative equivalence from a phonological perspective of data generated by the use of different instruments and under distinct environmental conditions. We also make further contributions to the phonetic and phonological understanding of nasalization phenomena in Korebaju, in particular by observations concerning the nasalization of vowels, the distribution of nasal airflow in the language’s phonotactics and syllable structure, and the behavior of transparent segments in the language. 1. Introduction 1.1. Production and Aerodynamics of Nasal Sounds 1.1.1. Velopharyngeal Function and Nasalization Vega et al. (2024) state that nasal sounds occur when the velum lowers, opening the velopharyngeal port and allowing airflow from the lungs to pass through the larynx, pharynx, and nasal cavity. Both nasal consonants and vowels involve oral articulation, with varying degrees of oral closure and vocal fold vibration as the sound source. The opening of the velopharyngeal port while the oral cavity remains partially closed enables resonance through the nasal passages (Ladefoged & Maddieson, 1996; Stevens, 1998). Vega et al (2024) also note that studies using the pressure-flow method (Warren & DuBois, 1964; Shosted, 2011) and other aerodynamic measurements (Benguerel et al., 1975; Lubker & Moll, 1965) indicate that different oral articulations influence nasal airflow, with velar sounds producing greater nasal airflow than bilabials. Subsequent research confirms that nasal airflow depends on both velum movement and oral articulation (Thompson & Hixon, 1979; Texeira et al., 2000; Basset et al., 2001; Delvaux et al., 2008; Chi et al., 2015), while lower nasal vowels are produced with reduced velum displacement (Shosted, 2011; Hajek, 1997). Velum movement typically occurs at approximately 5 Hz (less than 200–300 ms), facilitating the propagation of nasal sounds during production (Stevens, 1998). Nasal coarticulation—the influence of nasal segments on adjacent sounds—has been widely studied and varies across languages (Basset et al., 2001; Delvaux et al., 2008; Passavant, 1863; Sole, 1995, 2007; Ohala, 1971, 1975; Clumeck, 1976). Recent studies propose a continuum of nasality based on velopharyngeal opening, ranging from pauses to nasal vowels and consonants, with velum closure during oral segments (De Boer et al., 2023). Another notable phenomenon is nasal consonants with oral release at the end, observed in some African languages, which help limit the spread of nasality to the following vowel by reducing nasal airflow at the consonant’s release (Ladefoged & Maddieson, 1996; Anderson, 1976; Wetzels & Nevins, 2018; Demolin et al., 2006). A particularly common phenomenon among Amazonian languages is rhinoglotofilia, that is, the nasalization of vowels adjacent to [h], as in Baniwa, an Arawakan language from Northwest Amazonia, where a word such as /oho/ ‘yes’ is regularly produced as [ˈõhõ] (Matisoff, 1975; Chacon & Carvalho, 2020). Vega et al. (2024) emphasize that these aerodynamic and articulatory patterns are crucial for understanding nasality in Korebaju, highlighting both segmental and suprasegmental interactions in vowel and consonant production. 1.1.2. Aerodynamic Studies in Amerindian and Tukanoan Languages Few studies have examined nasality from an aerodynamic perspective. In Karitiana, an isolated indigenous language from Brazil, a post-stopped nasal allophone (e.g., [mb]) is characterized by increased intraoral pressure and nasal airflow, illustrating contrasts between nasal segments (Demolin, Haude, & Storto, 2006). Research on Kakataibo in Peru shows nasal coarticulation near nasal vowels, with effects modulated by process directionality, stress, syllable boundaries, and prosody (Avelino, Zariquiey, & Pérez-Silva, 2020). In Tukanoan languages, including Korebaju, phonetic studies are limited; in Siona, a Western Tukanoan language, velum movement during nasal and oral productions was measured using the earbuds method, revealing progressive and regressive nasal harmonies influenced by suffixes (Bruil & Stewart, 2022). Anticipatory and carry-over nasal coarticulation has also been documented at the phonological level across Western and Eastern Tukanoan languages (Sylak-Glassman, Farmer, & Michael, 2014; Stenzel, 2007, 2013; de L. Silva, 2012; Gómez-Imbert, 2003, 2004; Kaye, 1971).. 1.1.3. The Korebaju Language According to Vega et al. (2024), Korebaju (autonym [kòrèβàhɨ], also known as koreguaje, ISO 639-3: coe) is a language of the Western Tukanoan family, spoken in the Colombian Amazon along the Orteguaza, Peneya, Consaya, and Caquetá rivers, with approximately 2,000 native speakers (Korebaju Community, 2011; Moseley, 2010). The contemporary Korebaju-speaking community emerged from the historical union of four populations—Korebaju, Tama, Macaguaje, and Carijona—who adopted Korebaju as their main language, leading to the extinction of their original tongues, though some cultural practices have been preserved. Dialectal variation exists as communities settled in distinct territories and maintain clan-based distinctions that reinforce group identity (Gralow, 1985; Chacón, 2016). Previous descriptions have noted features such as tone, nasal harmony and glottalization (Dupont, 1988; Cook & Gralow, 2001; Vega, 2019; Vega & Vallée, 2021), but detailed phonetic and acoustic-spectrographic data were limited. Notably, since 2017, a consistent observation has been made that most nasal consonants in Korebaju exhibit a burst at the end of production, initially hypothesized to result from glottic cycle modifications but later refuted (Vega, 2019; Vega & Vallée, 2021). The Korebaju vowel inventory includes six oral vowels /i, e, a, o, u, ɨ/, six nasal vowels /ĩ, ẽ, ã, õ, ũ, ɨ/, three glottal vowels /aˀ, eˀ, oˀ/, seventeen consonants /p, t, k, pʰ, tʰ, kʰ, β, ɸ, s, h, w, c, m, n, ɲ, ʰɲ, r/, and a system of segmental and suprasegmental glottalization. The Tama and Korebaju dialects differ prosodically and in the production and distribution of glottalization. Communities are located about one hour apart by canoe but share cultural events, educational institutions, and organizational activities, including a boarding school within the Korebaju reserve (Vega et al., 2024). The previous analyses are grounded in the nasal harmony hypothesis formulated by Dupont (1988), later expanded and refined in his 2021 publication. Within this theoretical framework, we propose to further develop Dupont’s description in the following paragraphs. In the phonological system of Koreguaje, nasality plays a central role as a suprasegmental phenomenon. Following the analysis proposed by Dupont Moreno (2021), this feature is not associated with isolated segments, but rather with the vocalic nucleus of the first syllable of a morpheme, from which position it may extend to other segments within the morpheme. This nasal spreading process—governed by formal rules that also determine blocking environments—affects both vowels and certain adjacent consonants, resulting in fully nasalized morphological forms as well as allomorphic alternations (pp. 353–354). Nasality is therefore represented phonologically not as a property of a specific segment, but as a suprasegmental [+nasal] feature. Its spreading triggers a nasal realization of underlyingly oral segments. However, this propagation may be blocked by particular segments, especially voiceless stops and, most notably, the glottal stop /ʔ/, which prevents the nasal feature from spreading (p. 360). At the morphological level, nasal harmony also gives rise to allomorphic variation depending on whether the root is nasal or oral. The alternation of the verbal suffix illustrates the notion of virtuality in the morphology of Koreguaje (Dupont, 1988, p.114), appearing as {-ʧe} after oral roots and as {-ɲe} after nasal roots, when the root contains a nasalized vowel, demonstrating the direct influence of suprasegmental nasality on morphological derivation (pp. 360–361). In conclusion, the author presents an analysis in which nasality is treated as a lexically relevant feature of morphemes, surfacing according to structured rules of spreading and blocking. This framework enables an efficient description of the Koreguaje phonetic inventory, while coherently integrating its morphological and phonological processes with those attested in other languages of the same family (p. 365). 2. Objectives and Hypotheses 2.1. Objectives ● To evaluate nasalization in nasal and nasalized vowels in Korebaju, using the original indicator that calculates the proportion of nasal airflow. ● To compare data obtained with different instruments (EVA2, EGG-D800, and microphone) in order to analyze how different data collection methods influence results and whether they allow for equivalent conclusions. ● To test inferential replicability by examining whether datasets collected from the same speaker on different occasions, using different instruments, corroborate the hypothesis of phonetic differences between nasal and nasalized vowels. ● To assess results replicability by comparing data collected from different speakers in different environments (laboratory vs. indigenous village) using the same instruments. ● To investigate conceptual replicability by confronting the results obtained in the previous stages with phonological analyses of nasal harmony in Korebaju, as proposed by Cook and Criswell (1993), particularly regarding directionality, target segments, blockers, and transparent elements within the nasal harmony system. 2.2. Hypotheses ● Oral, nasal and nasalized vowels in Korebaju exhibit measurable phonetic differences, reflected in the proportion of nasal airflow. ● Despite differences in data collection methods (EVA2 intranasal tubes versus EGG-D800 silicone mask), results obtained with different instruments will consistently support conclusions regarding the phonetic distinction between nasal and nasalized vowels. ● Data collected from different speakers and in different contexts will maintain statistical consistency in identifying phonetic differences between nasal and nasalized vowels. ● Statistical analyses applied to the various datasets will conceptually support phonological descriptions of nasal harmony in Korebaju, including patterns of directionality, target segments, blockers, and transparent elements. 3. Data Generation Here we present the generation of our corpora, who were the participants and what were the recording procedures in each experiment, as summarized in Table 1 and described in the following subsections. Table 1. Corpora Instrument Stimuli Location Date Participant EVA Wordlist A France 2023 Participant-A EGG-D800 Wordlist B Agua Negra 2021 Participant-A +11 participants Wordlist C Agua Negra 2023 Participant-A +5 participants (subset of 11 above) 3.1 Corpus Three different corpora were used for this study. Wordlist A consists of 54 words produced in isolation and repeated three times in random order. This corpus covers all possible occurrences of nasal consonants within roots and morphemes, although the complete tonal inventory is not included due to ongoing research on the tonal system. Each word contains at least one nasal consonant, with target syllables including CV and NV structures in both word-initial and medial positions, as well as in certain affix morphemes. Words consist minimally of a bimoraic root morpheme, which takes the form of CVCV, CVV, or CVGV, where C can be oral or nasal, G a glottal sound /h/ or /ʔ/, and V can be a modal or a laryngealized vowel. The second and third corpora consist of Wordlist B and C, containing 118 and 145 items and recorded between 2021 and 2023, respectively. Different from Wordlist A, in B and C, words were embedded within a carrier sentence. The first list of 118 words was collected from 12 speakers and was designed to elicit productions of minimal and quasi-minimal pairs across as many lexical contexts as possible. The second list, consisting of 145 words, was recorded from six speakers, three from each variety (say again their names), balanced for gender and generation, who had already participated in the initial 118-word survey. This second list aimed to complete the identification of minimal and quasi-minimal pairs in all possible contexts for each variety and to verify the presence of nasal harmony and tonal patterns. The carrier sentence was constructed as follows: /cɨkɨnà ìkámè ___ kóˀrèbàhɨ cɨò pí/ {cɨkɨńà ìká-mè ___ kóˀrèbàhɨ cɨò pí} we say-PL ___ Korebaju language ‘We say___in Korebaju’ 3.2 Participants The 54-word corpus was recorded with a single participant, a 67-year-old native speaker of Korebaju. Before this study, she had never been away from her community for more than two weeks, and her stay in France marked her first prolonged absence, lasting 88 days. Spanish, acquired during her schooling, is her second language. For the 118and 145-word lists, twelve native speakers of Korebaju participated, evenly divided by sex with six males and six females, and representing two generations: G1, aged 18 to 31, and G2, aged 42 to 70. All participants were of Korebaju descent and used Spanish as a second language, both with local settlers and at the regional boarding secondary school. None had previously been away from the community for more than two weeks at the time of recording. Importantly, the 67-year-old participant from the 54-word corpus is included in this larger group and is the only individual who had spent an extended period outside the community, totaling 66 days. All twelve participants also took part in the recordings of the other two word lists. 3.3. Recording Procedure The 54-word corpus and the 118-word list were recorded in France by the 67-year-old native speaker using the EVA system. The remaining two lists (118and 145-word lists) were recorded in the Korebaju community (Agua Negra) with the eleven and six participants using electroglottography (EGG). It is important to note that, for this study, only 18 of the 54 items in Corpus A will be analyzed. This selection was made based on their correspondence with items in Corpus B, ensuring comparability across datasets. By focusing on lexical items that are shared between both corpora, we aim to obtain reliable results derived from equivalent linguistic material recorded in both datasets. 3.3.1 Recording Instruments The present study utilized advanced instrumentation to capture detailed acoustic, aerodynamic, and physiological data during speech production. Two primary systems were employed: the Assisted Vocal Evaluation (EVA2™) workstation and the Rothenberg OroNasal mask with electroglottography (EGG). These tools allow for precise measurement of airflow, subglottal pressure, glottal activity, and acoustic properties, providing a comprehensive view of the vocal production mechanisms that cannot be captured through auditory analysis alone. The EVA2™ system integrates multiple data streams synchronously, including acoustic signals, oral and nasal airflow, and physiological parameters, enabling multiparametric analyses of voice production in both laboratory and field conditions. In parallel, the Rothenberg mask, in its original and OroNasal versions, has historically enabled reliable aerodynamic measurements while accounting for potential acoustic effects induced by the mask design. Together, these instruments form a robust methodological framework for investigating the phonetic and phonological properties of Korebaju speech, including nasalization, glottalization, and other segmental and suprasegmental features. 3.3.2 Assisted Vocal Evaluation (EVA2™) The Assisted Vocal Evaluation system (EVA2™) (Figure 1) is an advanced workstation for measuring and analyzing the mechanisms of human voice production. Developed Ghio et al. (2022), it is designed to perform multiparametric voice analyses by integrating acoustic, aerodynamic, and physiological data in a synchronized manner (Ghio et al., 2022). Technical Features EVA2 allows simultaneous acquisition of multiple types of data: ● Acoustic: Intensity, fundamental frequency, and formant spectra, using external microphones and electroglottography (sampling at 25 kHz). ● Aerodynamic: Oral airflow (up to 10 L/s) and nasal airflow (up to 3 L/s), as well as intraoral and subglottal pressure (up to 100 hPa), sampled at 6.25 kHz. ● Physiological: Glottal activity recorded via electroglottography, and detection of velar and tongue movements. All sensors are connected to a single analog-to-digital converter, allowing synchronized and multiplexed data acquisition, ensuring stable calibration and precise measurement. Innovations and Advantages EVA2 enables the capture of information that cannot be observed through acoustic sound alone, such as nasal airflow, subglottal pressure, and glottal movements. Its ability to integrate these data synchronously provides a comprehensive view of voice production mechanisms, overcoming the limitations of previous instruments and complementing traditional auditory assessments (Ghio et al., 2022). Figure 1: EVA2 Station’ Applications of EVA2 in Phonetic Research Ghio et al. (2023) note that EVA2 has been used in numerous phonetic studies, for example: ● For laboratory speech production analysis: Meynadier, Ghio, Cesari-Lietard, & Robert (2019) developed a protocol for aerodynamic analysis of nasalization in French, addressing both regional varieties and clinical variations. ● For field speech production analysis, including linguistic research within the GDR “Grand Rift Africain”: Carbone, Bouchet, Ghio, Legou, André, et al. (2022) created the Speed-Vel Project, a corpus of acoustic and aerodynamic data to measure droplet emission during speech interaction. ● For neurological analysis of Parkinsonian dysarthria: Duez, Ghio, & Viallet (2020) studied the effect of linguistic context on the perception of consonants in Parkinsonian Read French speech. 3.3.3 Rothenberg Mask: Evolution and Acoustic Considerations The Rothenberg mask, developed by Rothenberg (1973, 1977), is a central instrument in experimental phonetics for measuring airflow during vocal production. Its rigid design and geometry produce measurable acoustic effects, such as formant reduction. This is particularly noticeable in open vowels, like /a/, where the first two formants can decrease by 50 to 100 Hz due to the effective elongation of the vocal tract. However, these variations are relatively small (around 5%) and do not significantly compromise basic aerodynamic assessment. Later versions, such as the OroNasal Mask (Figure 2), enabled important advances by allowing simultaneous measurement of oral and nasal airflow, as well as nasality (nasalance), defined as the proportion of nasal acoustic energy relative to the total (oral plus nasal) within approximately the first formant range of vowels (350–750 Hz). This measure has become a clinical reference for assessing oral-nasal resonator balance (Fletcher, 1978; Dalston et al., 1991; Hardin et al., 1992). Recent research highlights the phenomenon of transpalatal nasality, i.e., the transfer of acoustic energy from the oral cavity to the nasal cavity via palatal structures, even when there is complete velopharyngeal closure. In speakers without cleft palate, up to 80% of nasality in voiced stop consonants (/b, d, g/) can be attributed to transpalatal effects, while about 20% results from acoustic crossover between microphones (Gildersleeve-Neumann & Dalston, 2001). Additionally, a positive, although non-significant, correlation has been observed between speaker fundamental frequency (F0) and nasality derived from both the first formant (F1) and F0, suggesting that the vibration of palatal structures partially depends on the acoustic energy generated by the voice. From a clinical perspective, these findings are relevant: patients with repaired cleft palate may achieve complete velopharyngeal closure and still exhibit persistent oral-nasal imbalance due to the acoustic transmission characteristics of the soft palate (Gildersleeve-Neumann & Dalston, 2001; Rothenberg, 2006). This underscores the need to consider transpalatal nasality when interpreting airflow and resonance measurements in voice assessments and speech therapy. ● Gralow, F. (1985). The Coreguaje suprasegmental system: Tone, stress, and intonation. In R. Brend (Ed.), Phonology to discuss: Studies in six Colombian languages (pp. 3–11). Dallas: SIL. ● Hajek, J. (1997). Universals of sound change in nasalization. Oxford: Blackwell. ● Hardin, J., Dalston, R. M., & Watterson, T. (1992). Nasalance measures in speech assessment. Cleft Palate Journal. ● Hertegård, S., & Gauffin, J. (1992). Effects of external microphones and vocal tract extensions on formant frequencies. Speech Communication, 11(4), 341–350. ● Hoole, P., & Kroos, C. (1998). Control of larynx height in vowel production. In Proceedings of the 5th International Conference on Spoken Language Processing (ICSLP 1998) (Paper 1097). https://doi.org/10.21437/ICSLP.1998-360 ● Hothorn, T., Bretz, F., & Westfall, P. (2008). Simultaneous inference in general parametric models. Biometrical Journal, 50(3), 346–363. https://doi.org/10.1002/bimj.200810425 ● Kaye, J. D. (1971). Nasal harmony in Desano. Linguistic Inquiry, 2(1), 37–56. ● Korebaju Community. (2011). Proposal of the Korebaju pedagogical model. Caquetá, Colombia. (Unpublished manuscript). ● Lubker, J., & Moll, K. (1965). Simultaneous oral-nasal airflow measurements and cinefluorographic observations during speech production. Cleft Palate Journal, 2, 257–272. ● Matisoff, J. A. (1975). Rhinoglottophilia: The mysterious connection between nasality and glottality. In C. A. Ferguson, L. M. Hyman, and J. J. Ohala (Eds.), Nasalfest: Papers from a Symposium on Nasals and Nasalization (pp. 265-287). Stanford, CA: Stanford University, Department of Linguistics. ● Meynadier, Y., Ghio, A., Cesari-Lietard, F., & Robert, J. (2019). Protocole d’analyse aérodynamique de la nasalité en français: Des variétés régionales aux variations cliniques. Journées de Phonétique Clinique, Mons, Belgium. ● Moseley, C. (Ed.). (2010). Atlas des langues en danger dans le monde (3rd ed.). Paris: Éditions UNESCO. ● Ohala, J. J. (1971). Monitoring soft palate movements in speech. In Project on Linguistic Analysis (Vol. 2, pp. 13–27). University of California, Berkeley. ● Ohala, J. J. (1975). Phonetic explanations for nasal sound patterns. In C. Ferguson, L. Hyman, & J. J. Ohala (Eds.), Nasalfest: Papers from a symposium on nasals and nasalization (pp. 289–316). Stanford, CA: Stanford University. ● Passavant, G. (1863). Ueber die Verschliessung des Schlundes beim Sprechen. Frankfurt a. M: J. D. Sauerländer. ● R Core Team. (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing. ● Riordan, C. J. (1980). Larynx height during English stop consonants. Journal of Phonetics, 8(3), 353–360. https://doi.org/10.1016/S0095-4470(19)31485-8 ● Rothenberg, M. (1973). A new device for measuring airflow in speech. Journal of the Acoustical Society of America, 53(2), 565–570. ● Rothenberg, M. (1977). Acoustic correlates of airflow in speech production. Phonetica, 34(4), 212–228. ● Rothenberg, M. (2006). Acoustic considerations in nasalance measurement. Journal of Voice. ● Shosted, R. K. (2011). An EMA-aerodynamic approach to the velic opening hypothesis: Evidence from Hindi vowel pairs. In Proceedings of the ICPhS (pp. 68–71). ● Sole, M. J. (1995). Spatio-temporal patterns of velopharyngeal action in phonetic and phonological nasalization. Language and Speech, 38(1), 1–23. ● Sole, M. J. (2007). The stability of phonological features within and across segments: The effect of nasalization on frication. In P. Prieto, J. Mascaró, & M. J. Solé (Eds.), Segmental and prosodic issues in Romance phonology (pp. 41–65). Amsterdam: John Benjamins. ● Stenzel, K. (2007). Glottalization and other suprasegmental features in Wanano. International Journal of American Linguistics, 73, 331–366. ● Stenzel, K. (2013). A reference grammar of Kotiria (Wanano). Lincoln: University of Nebraska Press. ● Sylak-Glassman, J., Farmer, S., & Michael, L. (2014). An agreement by correspondence analysis of Máíhɨki nasalization harmony. In ABC Conference, University of California, Berkeley. ● Texeira, A., Vaz, F., & Principe, J. C. (2000). Nasal vowels following a nasal consonant. In 5th Seminar on Speech Production (pp. 285–288). ● Thompson, A. E., & Hixon, T. J. (1979). Nasal air flow during normal speech production. Cleft Palate Journal, 16(4), 412–420. ● Vega, J. (2019). Glottalisation et aspiration en Korebaju [Master’s thesis, Université Sorbonne Nouvelle – Paris 3]. ● Vega, J., & Vallée, N. (2021). Glottal sounds in Korebaju. In INTERSPEECH 2021 (pp. 1011–1014). ISCA & Brno University of Technology. ● Vega Rodríguez, J., Vallée, N., Savariaux, C., & Gerber, S. (2024). Nasal air flow during speech production in Korebaju. In Interspeech 2024 – 25th Annual Conference of the International Speech Communication Association (pp. 3694–3698). https://doi.org/10.21437/Interspeech.2024-1674 ● Wetzels, W. L., & Nevins, A. (2018). Prenasalized and postoralized consonants: The diverse functions of enhancement. Language, 94(4), 834–866. https://doi.org/10.2307/26630402