The Phonological System of TumɁi

As part of a linguistic research team I recorded a Khoisan language currently spoken by three people in the Northern Cape province of South Africa. Since the variety of language spoken in this location is close to varieties of both the Khoekhoe and Tuu language families, the question of genetic affiliation and classification within the Khoisan language cluster becomes significant. Although reported to have significant lexical similarities due to intensive language contact (Güldemann 2006), extensive research provides evidence of numerous linguistic differences which distinguish between the varieties within the Khoisan families mentioned above (Beach 1938, Bleek 1930, Ladefoged & Traill 1994, Miller, Brugman, Sands, Namaseb, Exter & Collins 2007). Overall, this project attempts to answer the question: How unique is this undocumented language TumɁi in comparison to varieties of geographically neighbouring Khoisan language clusters? This comparative analysis is comprised of a detailed description of the vowel and consonant systems, as well as evidence of phonetic and phonological contrasts. The clear focus on the analysis of sound contrasts is a consequence of limited data due to speaker competence. As a result of intense incomplete acquisition and linguistic attrition, the consultants produce utterances using Khoisan content words within an Afrikaans framework (Killian 2009). Specific research questions include:


Introduction
The goal of this research is the documentation and phonological codification of the Khoisan variety Tumʔi, within the fieldwork and constraints of 'salvage linguistics' (Grinevald 2003).
Tumʔi is a variety discovered in the small town of Prieska in the Northern Cape province of South Africa. The linguistic community consists of a group of three siblings between the ages of 64 and 80. The variety of Khoisan spoken by these individuals has been consensually named Tumʔi, as the speakers originally referred to their language as hottentotstaal, which is considered a derogatory term in South Africa. Hottentot language is also referred to by Beach (1938) as a click language originally spoken by people living in South-West Africa. This language was found situated geographically close to varieties of both the Khoekhoe and Tuu language families, therefore the question of genetic affiliation and typological similarity within the Khoisan lineages becomes significant. This is addressed through the analysis of phonetic, phonological and lexical similarities and oppositions between Tumʔi and representative Tuu and Khoekhoe languages (Beach 1938, Bleek 1930, Ladefoged & Traill 1994, Miller 2007.
The results presented in this paper are the sound inventories which make up the phonological system. This includes all phonemic contrasts as well as typologically significant phonetic features.

Sociolinguistic context: Endangered languages and speaker competence
Based on our investigation, there are three known speakers of this language; Elsie George, age 80; Francina George, age 70; and Robert George, 64. As siblings raised in the same home in Prieska, they were all exposed to the same linguistic variety. However, their levels of exposure and resulting knowledge vary. This variation is mainly attributed to their substantial differences in age, with Elsie noticeably being the authority on the language and Robert being the least knowledgeable speaker. Aside from the age differences there is also the gender distinction -as a non-white male in South Africa, Robert was required to leave home to work on the nearby farms and contribute financially to the household. Hence, Robert's exposure to the language ended when he became a teenager, unlike his sisters who attended school for a longer period and spent most of their time helping in and around the home. Hence, within this linguistic community, which is made up of only three speakers, the speaker knowledge constitutes a continuum of proficiency, though not extending from full fluency (Dorian 1977).
The importance of speakers as a source for endangered language research and fieldwork cannot be overstated. Unlike speakers of non-endangered languages, the complete population of an endangered language may consist entirely of marginal speaker types, speakers with limited linguistic competency as a result of several interrelated factors. At the time these speakers were born, Khoe and San people had already been subjected to decades of oppression and coercion by the South African government to reidentify themselves as "Afrikaans-speaking Christian 'coloured'" communities (Killian 2009:12). This process of reclassification involved not only abandoning their indigenous identities but also their languages. Hence, the speakers of Tumʔi acquired an already endangered language--a language used only in the home and rarely directed to them as children, due to the fact that the general attitude toward the use of this language was one of trepidation.
Though the 'semi-speaker' has been found to be emblematic of the endangered language situation, the consultants involved in the documentation of Tumʔi are better described as 'terminal speakers'. The negative connotation associated with this profile has caused some debate across the literature on speaker typology, however it most accurately encompasses the distinction between the linguistic abilities of the 'semi-speaker' and that of the consultants involved in this project. The speakers of Tumʔi retain a passive knowledge of the language, even the eldest and most knowledgeable speaker has very limited productive skills. Conversation between speakers is conducted primarily in their dominant language Afrikaans, with the insertion of fixed phrases and expressions from their endangered language (Grinevald 2003). The reality of the situation in the case of endangered language study is that the reduced use of the language generally leads to a reduced form of the language (Dorian 1977:24).

Data collection and elicitation
The elicitation sessions were conducted across two data collection trips, the first only three days long, and the second spanning six days. The data accumulated on the initial trip is comprised of a total of three hours and five minutes of recorded data. A total of ten hours and six minutes of elicitation was conducted on the second trip, significantly more than the initial trip as the second trip was initiated solely to collect data of this specific variety. The method of data collection consisted mainly of single word elicitations using word lists adapted from previous fieldwork expeditions. The word list consists of basic verbs and actions as well as body parts and a few possible nouns relevant to the lifestyle and surroundings of the indigenous community (Kilian 2020, appendix G). Picture prompting was also used as a method of elicitation. During this process participants were shown pictures of indigenous plants and animals taken from Branch et al. 2001, Picker et al. 2002, Iwu 2014, and Van Wyk 2013, for which they were asked to provide a name, generally resulting in the elicitation of a different but semantically related lexical item.
The methodological process of endangered language documentation is pervaded with obstacles related to competence, insecurity, and multiple layers of both linguistic and social deprivation. Due to this reality, the research team which consisted of Dr William Bennett, Dr Levi Namaseb, and myself, was unable to collect any traditional stories or productive interaction solely conducted in Tumʔi. The data therefore consists mainly of individual lexical items which do not appear to display any derivable morphological information. Further conditioned by the context of endangered language research there appears to be extensive phonetic, phonological and semantic variation across the lexical items. Within this research framework, variation of this nature is attributed to the lack of any documented norm as many endangered languages such as Khoisan are historically spoken and not written. Ultimately, the limitations on elicitation and the collected data have directly constrained the type of linguistic analysis realizable for the documentation of this language variety.

Click inventory
The click inventory of Tumʔi as spoken today is limited to four distinct click types and three accompaniments. Two of the three accompaniments are attested across most click languages, which provides little evidence of exclusivity or typological distinctiveness (Ladefoged & Traill 1984, Ladefoged et al. 1999). However, the third accompaniment discovered in the inventory of Tumʔi is the audible uvular stop accompaniment which is rarely reported in the documentation of click languages (Miller 2007). This accompaniment is particularly unusual in that the resulting click is composed of two noise bursts. Reports of this click accompaniment are limited to a select number of languages most of which are classified as belonging to the Tuu family, hence the presence of this accompaniment in Tumʔi is a typologically remarkable find. The plain clicks presented above are the most under-utilized of the click segments transcribed across this data set. Generally, the speakers produce clicks with some form of accompaniment The most common is the pulmonic uvular accompaniment which constitutes a second distinct noise burst. A similar distribution of this lingual pulmonic segment has been reported for the languages ǃXóõ and |Gui, in which the uvular position appears to be exploited more frequently than predicted (Güldemann 2001). This unexpected distribution may reflect the speaker's intention to retain the unique features of the language by extending the use of the infrequently attested uvular accompaniment. Together the nasal and uvular stop accompaniments are the most frequently occurring posterior release types. However, the audible uvular accompaniment is what sets the click inventory of Tumʔi apart as typologically complex. The velar fricative accompaniment occurs infrequently throughout the data set, however, this accompaniment is clearly audible and phonetically distinct in particular lexical items such as |xei 'give birth' and ǃxara 'female genitals-type'.

Clicks in Khoisan varieties
Click segments are a unique class of sounds defined by the rarefaction of air concealed by two articulatory closures (Ladefoged & Maddieson 1996:246). Numerous descriptions of clicks are presented throughout the literature on Bantu and Khoisan languages, and are generally concerned with the phonological oppositions between different clicks. In-depth phonetic analyses of the click systems of particular Khoisan languages have provided evidence for extending these phonological distinctions and identifying more fine-grained differences between the articulation of different clicks (Ladefoged & Traill 1984). The general acoustic analysis of clicks is concerned with the waveform and spectra which align with the release of the anterior closure. The analysis of this closure release provides a phonetic interpretation of the sound produced as a result of the articulators separating and the rapid change in the configuration of the vocal tract (Ladefoged & Maddieson 1996:257).
The release of the posterior closure is generally described as a voiceless velar accompaniment which is not reflected in the acoustics, or a velar nasal accompaniment depending on the position of the velum (Ladefoged & Traill 1994). However, the stance adopted in this paper is one proposed by Miller (2007) which states that the place of articulation of the posterior release is uvular, with possible contrasts in voicing and discrepancies relating to the position of the anterior closure. The proposition that the posterior release is positioned at the uvula and not the velum is based on the analyses of ultrasound data collected from speakers of N|uu. The importance of this claim is related to the discovery of the lingual-pulmonic click, which is acoustically remarkable in that it produces two noise bursts (Miller 2011:420). The latter burst displays the features of a uvular stop. Following the assumption that this is a voiced counterpart to the general posterior release, both accompaniments are described as uvular (Miller et al. 2009, Miller 2007. As click sounds are indigenous to Khoisan languages, it is challenging to distinguish between particular languages and dialects based solely on their click inventories. Comparative studies of representative languages belonging to different Khoisan language families discuss the differences and similarities between these click inventories, with the general diagnosis that languages of the Tuu family contain a larger variety of clicks and accompaniments (Childs 2003). However, results of areal typological analyses have reported the effects of language contact on click inventories to be extensive, with Khoekhoe languages exhibiting a similar reliance on click segments as observed in Tuu languages . Therefore, the distribution of click types within a linguistic inventory, and the functional load of click segments across collected data sets may not be sufficient to distinguish between the phonological typologies of Khoisan varieties classified within different lineages.

Consonant inventory
Consonant phonemes are generally described by the characteristics of place of articulation and manner of articulation. As indicated by the columns in table 2 below, the articulators involved in producing egressive consonants in Tumʔi include the lips, the tongue positioned at the alveolar ridge or velum, and the closure of the glottis. The consonant inventory of Tumʔi also consists of different manners of articulation. This includes seven plain plosives produced with complete closure of the vocal tract, followed by a release burst. The aspirated alveolar plosive is the only aspirated segment and is phonetically distinct from the plain voiceless plosive [t].
The voiceless alveolar plosive is the only segment proven to consistently exhibit a contrast in aspiration. The consonant inventory also includes three nasal stops produced with complete closure of the oral cavity at either the bilabial, alveolar or velar places of articulation, with a lowered velum allowing airflow through the nose. Sounds with close approximation of articulators but not complete closure include the six fricatives and single affricate, as well as the labio-velar and lateral approximant presented in the inventory below. Finally, this consonant inventory also includes an alveolar trill which is produced with continuous tapping of the tongue-tip against the alveolar ridge.
Lateral approximant l Trill r The /t/ phoneme is one of the most frequently used plosives, occurring word initially, word medially, and word finally (see table 3 below). This is also the only phoneme which may be produced with additional aspiration, or as distinctly unaspirated. Though the degree of aspiration has been exhibited to vary across tokens of the same lexical item, [tumʔi] 'speak', the aspirated segment [tʰ] is produced consistently in tokens of the lexical item [tʰi!qo] 'God', while the unaspirated [t] is consistently produced in tokens of the lexical item [tɔrəŋtɔrəŋ] 'crazy'. /k/ exhibits the most frequent distribution across the non-click consonants, occurring both in the word initial and word medial position. The final voiceless stop /p/ occurs both word initially and in word final position, indicating that the voiceless stop consonants are more widely distributed than the voiced stops, /d/, /b/, and /g/, which occur in either the word initial or word medial position and are each transcribed in no more than three lexical items.
The fricatives /s/, /f/, and /v/, are limited to either word initial position, or word medial position. The phoneme /h/ also occurs both word initially and medially, however it appears to only precede the low vowel /a/. /x/ appears to be the most frequently occurring and unrestricted fricative, extending to the accompaniment of clicks in variations of particular lexical items. The bracketed phoneme /β/ is transcribed in medial position of tokens of a single lexical item [taβa] 'handiwork'. This bilabial fricative appears to alternate with the phonemes /v/ and /w/ across tokens of this particular lexical item. The nasal phoneme /n/ is frequently transcribed across the data set, however it occurs mainly as an accompaniment to one of the four click types. The actual phoneme /n/ as well as the phoneme in the form of a click accompaniment appear in both the onset and coda position of the syllable, however in the word initial position the phoneme /n/ is only present as a click accompaniment. The other pulmonic egressive nasal phonemes /m/ and /ŋ/ share similar wide-spread distributions, however the phoneme /ŋ/ is clearly more restricted in its distribution than /m/, occurring only in coda position of words.
Particular phonemes recorded infrequently and inconsistently include /l/, /tʃ/ and /t j /. Consequently, these phonemes are bracketed and not definitively proposed as phonemes of the consonant inventory, due to alternating production across tokens of a single lexical item, and their absence in any of the other lexical data.

Non-click consonants in Khoisan varieties
Generally, a consonant inventory of this size is not commonly observed across Khoisan dialects, with most exhibiting a limited number of egressive consonants. However, there are segments which are commonly observed across the phonological systems of Khoisan languages but are not attested in Tumʔi, including the affricates /ts/ and /kx/ (Beach 1938:65-67, Güldemann 2006:11, Killian 2009. Based on previous typological analyses the consonant systems of Khoekhoe and |Xam are similar in overall organization. The major difference between these systems concerns the size of the inventory, which in turn is a typological consequence of an extended variety of distinctive features attested in Tuu languages .
Khoekhoe languages lack laryngeally marked stops including both aspirated egressives and ejectives; observations across the inventories indicate that these segments have been replaced with aspirated and ejective fricatives. The same laryngeally marked stops are however also unattested in the consonant inventory of |Xam which is considered typologically unusual (Vossen 2013:211). Previous analyses attribute this shared lack of complex stops to the phonological process of affricate lenition (Beach 1938, Vossen 2013. Also lacking from the phonological inventories of Khoekhoe varieties is a voicing distinction. Egressive segments with the feature [+voice] are absent from Khoekhoe varieties with the exception of !Ora and |Gui , Vossen 2013. Therefore, any evidence of the laryngeally marked stops, such as those discussed above, or a phonological voicing distinction would naturally align with the phonological typology of Tuu languages.

Vowel system
Distinctions in vowel quality are generally determined by three aspects including vowel height, vowel backness, and lip rounding. These aspects are reflected in the acoustic properties of vowels, the most prominent of which are the formants (Ladefoged & Maddieson 1996:104). Vowel height is proportional to the inverse of the frequency of the first formant (F1), while vowel backness is proportional to the frequency of the second formant (F2), or the difference between the frequencies of the first and second formants. Finally, the degree of lip rounding is generally indicated by the lowering of the second and third formants (

Monophthongs
The phonemes /i/ and /u/ constitute the high vowels of Tumʔi. /i/ is produced as a close front unrounded oral vowel with little variation across tokens. The phoneme /u/ is articulated as a close back rounded vowel and exhibits a wider distribution than its front counter-part /i/. /u/ occurs mainly as the first vowel in the initial syllable in a word, indicating that it may be a vowel phoneme designated to particular roots. This distribution may be explained by the fact that lexical roots ending in /u/ and /i/ are commonly observed in Khoisan languages (Beach 1938:42). The mid-vowels identified in the inventory include the phonemes /e/, /ɔ/ and /ə/. The phoneme /e/ is articulated as a half-closed front unrounded vowel and appears to occur mainly in word-final position, or as the nucleus of the final syllable. The back mid-vowel counterpart to /e/ is articulated as the open rounded vowel [ɔ], which appears frequently as the nucleus of the first and second syllable. Based on the vowel chart in figure 1 above, the mid-back vowel is positioned both lower and more centralised within the vowel space than the mid-front vowel

Diphthongs
The vowel system of Tumʔi includes a total of five diphthongs, all of which end with a high vowel, preceded by a low, mid or back vowel. This is a significantly large collection of 'unlike vowel sequences' (Güldemann 2006:384), which appear to be unrestricted in terms of position within the word as well as in terms of co-occurrence with other vowel phonemes. The phonemes /ei/, /oi/ and /ui/ all reflect a shift to the high vowel /i/. The vowel chart in figure 2 above displays three diphthongs which glide to /i/ and end at the same general height and front position. The diphthong /ae/ which could be perceived as the higher ending sequence [ai] appears to glide to a distinctly lower front position. The phoneme /i/ in the diphthong /ai/ is generally expected to be articulated as more open and flatter than when produced as a monophthong (Beach 1938:36), hence the end point of the low diphthong in the chart above is most accurately transcribed as [ae] and not [ai].

Vowel systems in Khoisan varieties
The vowel systems of Khoekhoe dialects are defined by three distinctive features: the limited number of vowel phonemes, the lack of diphthongs except in instances of very rapid speech, and the phonemic presence of both oral and nasal vowels (Beach 1938:35). The vowel systems of northern Khoekhoe languages as well as !Ora are reported to include the five basic vowels /i e a (ə) o u/, with the mid-central vowel generally considered neutral and non-phonemic. These five vowels are each reported to have a nasal counterpart, with the exception of the mid-front vowel /e/. Furthermore, laryngeal phonation and related vowel colourings are not reported as distinctive features in any of these varieties (Vossen 2013:150). Southern Khoisan varieties, though generally reported to exhibit the same basic five vowel system, additionally exhibit attested nasalized vowel contrasts, as well as distinctively laryngealised vowel contrasts (Garellek 2019, Miller et al. 2009, Vossen 2013

Non-phonemic features
The observations here are distilled from a larger unpublished MA thesis (Kilian 2020). Distinct differences in the phonation of particular vowel segments have been observed across the utterances of TumɁi speakers. However, unlike the phonemic distinctions presented in the sections above, the phonation contrasts observed in TumɁi are not the result of controlled variation. The majority of the previous phonation analyses have reported the resultant difference in voice quality as contrastive, serving as a distinguishing feature between particular vowel and nasal segments. However, there are instances such as in the language Hmong in which differences in phonation occur as idiosyncratic features exhibited only by particular speakers (Huffman 1987).
The phonetic characteristics of each phonation type are captured primarily using an acoustic analysis of the H1-H2 spectral slope parameter, and information extracted from the contrasted Harmonics to noise ratio (HNR) measures, as conducted across previous phonation analyses (Garellek 2019, Gordon & Ladefoged 2001, Miller 2007. Physiological characteristics, such as the medial thickness of the vocal folds and their aperture at the time of vibration can be interpreted by the H1-H2 spectral slope parameter (Garellek 2019:3, Huffman 1987. The HNR measure is reported to indicate the presence of any aspiration noise due to the glottis opening, as well as the distribution of irregular voicing (Garellek 2019:3) Figure 3: Spectra calculated across four glottal pulses for three low vowel segments. The initial vowel in tava 'handywork'; n|aski 'meat'; |qwans 'child'.
Individual spectral samples of tokens produced by Elsie and Francina displayed in figure 3 above provide evidence of three possible phonation types based on the H1-H2 slope, as well as the difference in spectral tilt across each sample. There is a clear discrepancy between the H1-H2 slopes across the tokens representing each phonation type. The steepest slope is visible in the breathy spectrum with a value of 14 dB. Both the modal and creaky tokens display a negative H1-H2 spectral tilt. The spectral tilt of the creaky segment is only slightly steeper than that of the modal segment with a value of -2.8. A negative spectral tilt is expected for segments articulated with creaky phonation as well as segments produced with modal phonation (Gordon & Ladefoged 2001). The most distinguishing factor is the spectral slope and the degree at which the intensity drops as the frequency of the harmonics increases. The token representing breathy phonation has a steeper negative slope than either of the other tokens. The modal token exhibits a less steep slope, while the creaky token exhibits only a slightly negative slope. characterized by less energy, as the glottal waveform is more sinusoidal due to smoother opening and closing phases. Alternatively, the sharp glottal closure and opening formed in the production of creaky voice corresponds to more energy in the higher frequency harmonics, with relatively less energy in the fundamental frequency, or first harmonic. The expectation is therefore that the H1-H2 spectral tilt should reflect a significantly steeper gradient for breathy vowels in comparison to creaky and modal vowels. The reverse effect is expected for creaky vowels, in which case a greater negative H1-H2 spectral tilt would best distinguish the creaky segment from the modal voice segment. Statistical results from previous phonation analyses provide evidence that the mean HNR for modal voice is expected to be significantly higher than other phonation types (Gordon & Ladefoged 2001, Miller 2007. However, in comparison to creaky voice, breathy segments are expected to have a higher HNR mean due to their limited glottal constriction and resultant noisy energy.  The results of the two-sample t-tests presented above provide evidence of a significant disparity between both the H1-H2 slope measures, as well as the HNR measures of the modal and nonmodal phonation groups. Overall, the results indicate a greater H1-H2 gradient for the nonmodal segments, but a smaller HNR measure. Translated, these results indicate that non-modal phonation is attested in Tumʔi, and the non-modal segments most likely consists of breathy segments rather than creaky. Furthermore, it must be reported that the eldest speaker who consistently produced particular words with non-modal phonation herself explicates the distinct form of articulation as "Die verskil kom van agter af, jy moet tril as jy dit sê" (the difference comes from the back of the throat, your articulators must trill when you say it). The speaker explains that the sound must be produced at the back of the throat with a consistent trill-like mechanism. Not only is there acoustic evidence of a non-modal phonation type, but this special phonation is recognised by the speakers.
The analyses above are most certainly constrained by the modicum of data available for the investigation of phonation contrasts. However, the statistical analyses of the spectral data provide evidence of a significant disparity in the phonation of low vowel segments produced in Tumʔi. The results of the t-tests show that the Harmonics and the HNR measures are successful in distinguishing between the modal and non-modal phonation groups. The mean values of these measures do not entirely align with the general expectations reported in previous phonation analyses, however the results are applicable when compared to the measures reported for particular phonation types identified in ǃXóõ. Comparison of the slope and HNR measures above with those reported for the pharyngealized breathy and modal segments in ǃXóõ indicate that the non-modal segments observed in Tumʔi may be similar to the breathy pharyngeal phonation type observed across multiple Southern Khoisan languages (Vossen 2013). Overall, the results of this analysis definitively tell us that speakers produce some form of non-modal phonation.

Conclusion
The various phonemic and non-phonemic contrasts presented above come together to form a complex sound inventory consisting of unique features generally unattested in Khoisan varieties with less typologically complex phonological systems. The vowel system consists of six monophthongs and five diphthongs, excluding indications of possible voice quality distinctions. The monophthong system is comprised of the five basic high, low, and mid vowels. However, the disparity in openness between the corresponding mid vowels causes an irregularity, which results in an asymmetric quadrilateral vowel space. This observation introduces the question of whether a phonemic contrast in openness had previously existed as part of the vowel system. That would suggest that the vowel inventory of Tumʔi consisted of seven vowels, which would align with many Bantu varieties and a select number of Khoisan varieties, generally not within the Khoekhoe family. A similar distinction is introduced with the identification of diphthongs, which appear frequently throughout the data set. Furthermore, though vowel phonation does not constitute a distinctive feature in Tumʔi at present, the statistical evidence of a non-modal voice quality distinction might be a remnant of a more complex phonological system.
The uniqueness of the sound system of Tumʔi is further indicated by the observation of phonemic laryngeal contrasts. While phonemic distinctions in aspiration have undergone fricative lenition in most varieties of Khoekhoe and even some of the Tuu family, the speakers of Tumʔi have retained or re-gained a controlled use of at least one aspirated segment. Hence, the sound system of Tumʔi exhibits a feature no longer attested in any of the Khoekhoe languages with the exception of !Ora. Another laryngeal distinction discovered in Tumʔi is the use of the glottalic egressive airstream mechanism. The results of the phonetic study suggest that ejectives are phonemically contrastive in the sound system of Tumʔi. As is the case with a few of the other distinctive features identified above, the ejective is only attested in Khoisan varieties with more complex phonological typologies. Overall, the egressive consonant inventory observed in Tumʔi may be described as extensive relative to other Khoisan varieties.
Finally, the click system observed in Tumʔi possibly contributes the major part of the complexity to the sound system, with the identification of the uvular click accompaniment. This particular click accompaniment is unique to Tuu varieties, including N|uu and !Xõó (Miller 2007). Hence, this unique click efflux provides the most obvious indication of a typological similarity to particular Khoisan varieties, including among others ǃXóõ, N|uu, ǂHoan, and |Auni, which all form part of the Tuu lineage. Interestingly, this accompaniment constitutes one of the two click accompaniments most frequently observed in Tumʔi, the other being the nasal accompaniment. Overall, the click inventory is comprised of a total of 15 clicks, including the four click types and accompaniments. Considering the size of click inventories in Khoisan languages such as !Xõó and ||Xegwi, an inventory of 15 clicks would appear relatively small. However, the distribution of these clicks across the data set is extensive, which indicate that clicks may have carried a heavy functional load in this Khoisan variety. There are few words in the lexical inventory of Tumʔi which do not include a click segment. Hence, based on observations across the data set, clicks in Tumʔi represent the most important consonant-like segments in the sound system. Setting aside consequences of language contact, this is a feature more commonly associated with the Tuu lineage as opposed to Khoekhoe.
While arguments have been made for the untenability of Khoisan language family distinctions and lineages, comparative research provides evidence of linguistic relations which cannot be ignored. The investigation of the phonological typology of this Khoisan variety was initiated based on evidence of strong lexical correspondences and cognates. The resultant phonological inventory and main findings are useful for filling the typological gaps within the areal isoglosses. Furthermore, similar to southern Khoisan varieties including those within the Tuu lineage, Tumʔi exhibits remnant features of what would be described as a typologically complex sound system.