ILLUSTRATIONS OF THE IPA

Kejom [k $ ›dÉZç@m], the preferred autonym for the language more commonly known as Babanki, is a Central Ring Grassfields Bantu language (ISO 693-3: [bbk]) spoken in the Northwest Region of Cameroon (Hyman 1980, Hammarström et al. 2017, Simons & Fennig 2017). The language is spoken mainly in two settlements shown in Figure 1, Kejom Ketinguh [k $ ›dÉZç@m $k›@tçg $ u/] and Kejom Keku [k $ ›dÉZç@m k›@k $ u], also known as Babanki Tungoh and Big Babanki, respectively, but also to some extent in diaspora communities outside of Cameroon. Simons & Fennig (2017) state that the number of speakers is increasing; however, the figure of 39,000 speakers they provide likely overestimates the number of fluent speakers in diaspora communities. The two main settlements’ dialects exhibit slight phonetic, phonological, and lexical differences but are mutually intelligible. The variety of Kejom

Stem-initial onsets do not undergo a great deal of allophonic variation. Stem-onset velar stops /k g/ are palatalized before the front vowels /i/ and /e/, e.g.
[kJe@ ] 'allow', [gJÚ$ t›$ ] 'add'. Analogously, the labial-velar approximant /w/ is realized as rounded palatal [Á] when it occurs before the high front vowels /i/ and /e/, as in [ÁÚ$ /] 'person', [Áe@ ] 'plant (v.)'. This process could be described as palatalization applying to a natural class of velars, excepting only /N/ and /Â/, which do not occur before front vowels. The sequence /b¨/ is sometimes produced as [bB]; this could well be transcribed simply as a syllabic bilabial trill [B], given that bilabial trills are typically prestopped (Ladefoged & Maddieson 1996). The trill is debatably an allophone of the vowel rather than the consonant, a point we discuss in more detail below in connection with the several allophones of the vowel /¨/. The stem-final syllable coda is associated with more (and more salient) allophony, and only six consonants /f s k m n N/ may occur in the syllable coda, as exemplified in (1) above. In coda position, and in fact any non-stem-initial position, /k/ is realized as a glottal stop [/]. For the present speaker, in utterance-final position, coda [/] consistently results in full glottal closure, which is occasionally strongly released, i.e. [/H], in the accompanying recordings. 1 Word-medially, glottal stop is variably reduced to a period of creaky phonation. The vowel /i/ is produced with non-contrastive nasalization when it follows a word-internal glottal stop, as in [vç $ /Ú) @ ] 'people', which we take to be an instance of 'rhinoglottophilia', or nasalization triggered by a glottal syllable onset (Matisoff 1975). Coda /n/ conditions diphthongization of all non-high vowels that precede it, namely the set /a e o/, resulting in syllable rhymes [aI 9n EI9n çI9n]; other vowels followed by /n/ do not undergo diphthongization. Occasionally, the coda nasal stop is produced without closure, resulting in the free variants [aI 9 ) EI 9 ) çI 9 ) ]. This diphthongization pattern is shared in common with the neighboring language Kom and is attested mainly in the dialect of Kejom Ketinguh. All coda consonants except /m/ are frequently deleted intervocalically; this is discussed further below.

Velar approximant
The segment /Â/ is commonly transcribed as a voiced velar fricative [ƒ] in both Kejom (Hyman 1979b, Akumbu & Chibaka 2012 and other languages in the area (Hyman 1979a, Schaub 1985. A frictionless, approximant version of the phone is sometimes reported instead (Hyman 1981). The typical production of /Â/ in the second author's speech in Kejom shows little to no frication, motivating our description of it here as an approximant rather than a voiced fricative.
Two representative examples of stem-initial, intervocalic [Â] are given in Figure 2; no aperiodic energy in the low-frequency region can be seen, as would be expected for a velar fricative, although a general reduction in intensity and a weakening of formant structure can be seen, indicating the presence of an approximant constriction. Figure 2 and all subsequent spectrographic displays were produced using Praat (Boersma & Weenink 2001).

Prenasalized onsets
All oral onset consonants except /v/ may occur with prenasalization. The absence of / n v/ onsets is most likely an accidental gap due to the low frequency of onset /v/ in general. Postnasal neutralization of obstruent voicing to [+voi] does not occur at either the phonological or phonetic levels, as is common in other Bantu languages (Hyman 2003: 50), see, for instance, / n pÉ f¨$ / 'rope' versus / n bÉ v¨@ / 'chicken'; /k›$ n sjs/ '(act of) scattering' versus /k›$ n zjs/ 'key'. We are agnostic as to whether prenasalized consonants as presented here are best phonologically represented as single complex segments or sequences of a placeless nasal and the simple consonants presented above. Both analytical routes equally well represent the phonetic events observed in the language, and there is a dearth of evidence that definitively supports one analysis over the other. Nonetheless, we opt to transcribe prenasalization with the secondary articulation diacritic as a necessary abstraction: prenasalization has several realizations that are dependent on the manner of articulation of the modified segment (continuant or non-continuant) and its position within the stem (stem-initial or preceded by a vowel), and which cannot be reduced to such a relatively invariant target as a placeless nasal stop. If the prenasalized segment is a non-continuant (a stop or an affricate), prenasalization is generally realized as a homorganic nasal stop preceding the oral segment's closure, as seen in [k›$ 6tÉ Sç/] 'lid' or [h $ Nghm] 'next week' (Figure 3, left). On the other hand, if the prenasalized segment is a continuant (an approximant, lateral approximant, or fricative), prenasalization is typically realized without a full closure in the oral cavity. Representative examples of / n S/ and / n Â/ are given in Figure 3 (right). Vowels preceding prenasalized fricatives tend to be nasalized, with the nasalization of the consonant shifting completely onto the vowel, as in [f›) $ SÉ/] 'grass beetle' (Figure 3, upper right). If the prenasalized segment is not a fricative, the entire segment AND the preceding vowel are typically nasalized, as in [k›) $ Â) ç% m] '(act of) beating' (Figure 3, lower right). No abrupt reduction of formant intensity or rapid shift in formant frequencies, either of which might suggest an oral closure with continuing nasal airflow and resonance, is visible in the spectrograms for either / n S/ or / n Â/ in Figure 3 (bottom left, bottom right). This can be contrasted with the clear division between a vowel [›] or [a] and the following nasal stop in the other two examples in Figure 3 (top left, top right). We speculate that the lack of a closure in the oral cavity for prenasalized continuant segments could be a learned aspect of articulatory timing in Kejom that has the effect of avoiding excrescent oral stops in nasal-continuant sequences (Browman & Goldstein 1990, Ohala 1993. Such a scenario would be particularly damaging to the fricative-affricate contrasts of Kejom, which have a relatively high functional load and could easily be lost if gestural timing promoted excrescent oral stops in this context.
The examples of prenasalized consonants used above all have at least a vowel preceding the prenasalized consonant in the stem. When prenasalized consonants occur at the left edge of a stem, with no other prefixal material present, the nasal portion of the segment gives the percept of having syllabic prominence and being produced with low tone, similar to most prefixal material in Kejom. Previous transcriptional schemes based on auditory impression, used by the second author and others (Hyman 1980, Akumbu 2008a, generally reflect this intuition, marking the nasal portion of prenasalized segments in this position as bearing a low tone, akin to most other prefixal material in Kejom. Prenasalized noncontinuants in particular show this tendency, e.g. / n dç% N/ 'potato' typically being transcribed as [n$ dç% N].

Labialized and palatalized onsets
Syllable onsets in Kejom may also be accompanied by palatalization or labialization, as shown in Table 1. Complex onsets that have both prenasalization and either palatalization or labialization (i.e. n CJ, n CW) are attested fairly frequently and are referenced throughout this section; this results in strikingly complex onset consonants (or lengthy clusters, depending on one's analysis). As with prenasalization above, there is some ambiguity as to whether labialized and palatalized consonants are single complex segments with secondary articulations or sequences of the plain segments and approximants /j w/. Also as before, there is little evidence at present that supports one analysis to the exclusion of the other. In this sketch, we opt to treat palatalization and labialization as attributes of complex segments, primarily because it allows for a simpler description of velar palatalization. This analytical decision greatly increases the number of distinct consonants, since the secondary articulations are relatively freely combined with onsets. Palatalization can primarily be observed on bilabial and labiodental segments, and it is particularly frequent before the front and central non-low vowels /i eˆ¨›/. Minimal or nearminimal pairs hinging on the contrast between palatalized and non-palatalized versions of the same segment are relatively easy to find for the bilabials and labiodentals, e.g. conditioned by a following front vowel /i/ or /e/ and occur nowhere else; no minimal pair contrast between palatalized and non-palatalized velars occurs in any Kejom lexeme. The [Á] allophone of /w/ before /i/ and /e/ could also be thought of as a non-contrastively palatalized /w/, in a sense palatalized [wJ].
Unlike palatalization, labialization is never an allophonic feature of the consonant, and co-occurs with a somewhat broader range of consonants compared to palatalization. Except for the alveolar plosives /t, d/, all consonants that can be palatalized can be labialized, and several that cannot be palatalized (the postalveolars /tÉ S dÉ Z S/) or contrastively palatalized (the velars /k g N/) may be contrastively labialized. Labialization is typically produced with a labial-velar constriction, as in [SWiIn] 'shuck corn', [›@ NW›$ /] 'INF-be bright, clean'. Akin to the allophony described for /w/ above, labialized consonants are realized with a rounded palatal secondary articulation when followed by the high front vowel /i/, e.g.
Bilabial and labiodental consonants with this secondary articulation, however, are still produced with a labial-velar secondary articulation, e.g.

Vowels
Kejom has eight vowel phonemes contrasting in height, backness, and rounding. Lengthened vowels frequently occur in running speech owing to a process of vowel coalescence, but length is only marginally phonemic (see the following section). Two of these lengthened vowels, the low-mid vowels /E˘/ and /ç˘/, are marginally contrastive but frequently observed due to vowel coalescence. Kejom has several typologically unusual vowel contrasts: there is a contrast between mid and high central unrounded vowels /ˆ/ and /›/, and a contrast between a high central unrounded vowel /ˆ/ and a high central rounded vowel /¨/ is typically described. Of additional interest are the various allophones of /i/ and /¨/, which exhibit carryover coarticulation of the supralaryngeal constrictions of certain immediately preceding onset consonants.
The mid vowels /e/ and /o/ exhibit higher or lower allophones depending on syllable shape and palatalization or labialization of the onset. In open syllables, mid-high allophones are consistently observed (i.e.  , we do not call the contrast between the PHONEMES /ˆ/ and /¨/ into question, given the high functional load of the contrast. The contrast between /ˆ/ and /¨/ is particularly robust once the assimilated allophones of /¨/ are taken into account: both /ˆ/ and /¨/ appear after postalveolars and contrast robustly in that position, although the nature of the contrast in this environment is different owing to /¨/'s adoption of a postalveolar tongue shape. The surface allophone [¨] which appears in Figure 4 has a restricted distribution, occurring only after onset /l/, /6/, and the velar consonants; /ˆ/ does not appear following initial /6/. For the present speaker, the contrast between [ˆ] and [¨] in this environment is likely reinforced visually by the visibly different activity of the lips in the two phones.

Vowel coalescence
Phonetic long vowels are quite common in Kejom, but these can generally be analyzed as a reduction of /VCV/ sequences. All coda consonants except /m/ may be deleted when followed by a /›/ whether or not this additional vowel is in the same morpheme or word as the coda. The resulting /VV/ sequence coalesces to a single phonetically long vowel [V˘], which often does not match the quality of either the first or second vowel. Examples of this process are given in (2). (2) Some attested mappings between stem VC and coalesced vowels resulting from the addition of a following /9/ Sequences of /VCV/ that lead to coalescence generally arise when a morpheme of the form /›/ (of which there are several in Kejom) follows a stem ending in a /VC/ sequence, either through suffixation (as in the imperative or progressive) or placement of a separate morpheme (as in the associative marker). In addition, any sequence of a vowel immediately followed by /›/ with no intervening consonant coalesces to a lengthened version of the first vowel, e.g. /tí-›@ / → [tí˘] 'be strong-PROG'. When the deleted stem-final consonant is /N/, coalescence results in a different, generally higher set of long vowels compared to deletion of /f s k n/. This sketch of the coalescence process given here skims over a number of complications, in particular morphophonological factors. Deletion and coalescence do affect /N/ and /n/ and their flanking vowels in a wide variety of morphological contexts, but affect the set /f s k/ only when they occur in certain high-frequency lexemes, e.g.

Vowels with postalveolar and labial constrictions
The phonetic realization of the phonemes /i/ and /¨/ is highly dependent on the preceding consonant, particularly for /¨/. This allophony can be broadly described as a process of assimilation to the syllable-initial consonant's place of articulation; all four allophones are presented in (3).
(3) Allophones of /i/ and /0/ by the preceding consonant's place of articulation This is most obviously the case for two allophones of /¨/, which exhibit labial constrictions other than the outrounding observed elsewhere in Kejom. After labiodental fricatives and affricates, /¨/ is produced with a labiodental constriction continuous with that of the onset that persists through the vowel, which is denoted here as [¨v]. Slight labiodental frication is produced through most of the duration of the vowel as a consequence of the labiodental constriction, as seen in Figure 5 (left). A central vowel with significant lip compression, denoted as [¨B] here, is an allophone of /¨/ after non-labialized, non-palatalized /b/; bilabial trilling frequently occurs after the release of the [b] and may persist for the duration of the vowel, as seen in Figure 5 (right). Both /i/ and /¨/ acquire a postalveolar, [Z]-like constriction following postalveolar fricatives and affricates. The vowel /i/ has a postalveolar-constricted allophone transcribed here as [i6 ] that occurs after postalveolars; 2 the vowel /¨/ has a similarly constricted allophone [¨6 ] occurring in the same environment. Both allophones are transcribed using the laminal diacritic to indicate the change in active articulator. Phonetic [i] occurs as an allophone of /i/ following bilabial, labiodental, alveolar, and velar consonants; phonetic [¨], the only vowel quality associated with /¨/ that cannot be attributed to assimilation to a preceding consonant, occurs only after velar consonants, /l/, and /6/. There is strong frication present during some productions of the postalveolar allophones, particularly when they are preceded by a voiced postalveolar /dÉ Z/ or /Z/, as shown in Figure 6.
In this context it can be difficult to segment the onset from the vowel, given the continuous voiced frication over the entire onset-vowel sequence. This issue is encountered in Figure 6 (left), where the speaker is essentially producing a syllabic fricative for the sequence /Zi/. An analogous rounded segment, not shown in this figure, is also attested for the sequence /Z¨/, e.g. [´@ Z% W] 'INF-be cold'. Nonetheless, the postalveolar allophones of /i/ and /¨/ cannot simply be described as syllabic fricatives, given that the intensity of frication varies substantially depending on aerodynamic conditions. contrasts are predictable from the preceding consonant, it is not unexpected that acoustic differences among allophones would be slight, since the functional load that this acoustic difference would assume is low to nonexistent.

Articulation of vowels with postalveolar and labial constrictions
In this section, we use ultrasound and video records to characterize the vowels with postalveolar and labial constrictions in greater detail, and in particular to demonstrate the existence of the extra supralaryngeal constrictions described above. This data confirms the initial impression that /i/ and /¨/ assimilate in place of articulation to preceding postalveolar consonants. We provide evidence that the fricativized allophones of /i/ and /¨/ exhibit lingual or labial constrictions not normally expected in vowel sounds, which canonically have unobstructed central channels and laminar airflow.

Method
All articulatory data in this section was collected in the UC Berkeley PhonLab in April and May of 2016. All records are of the second author. Video of the speaker's face was collected using a Canon XF100 HD camcorder. The lingual ultrasound records shown in this section are still frames from videos taken from the approximate acoustic midpoint of the segments examined. Ultrasound video was recorded at a frame rate of 107 Hz using an Ultrasonix SonixTablet and a C9-5/10 microconvex transducer. The transducer was held in place under the chin and approximately perpendicular to the speaker's occlusal plane by an Articulate Instruments stabilization headset (Scobbie, Wrench & van der Linden 2008). Tongue surface contours were extracted from the ultrasound data using EdgeTrak (Li, Kambhamettu & Stone 2005); comparisons among segments are carried out below using smoothing-spline analysis of variance (SSANOVA) calculated using polar coordinates (Davidson 2006, Mielke 2015, which models the typical position of the tongue surface contour based on multiple input contours (here, in excess of 20 tokens per vowel or consonant type). The approximate location of the speaker's hard palate is superimposed on the SSANOVA plots below as a landmark; this information was collected by asking the speaker to drink water while scanning the swallowing action with the ultrasound.

Labial vowels
Using video stills, substantial differences in labial activity can be confirmed between outrounded

Postalveolar vowels
Since labial activity does not determine the major characteristics of [i6 ] and [¨6 ], we turn to lingual ultrasound data for articulatory details. Figure 10 shows that [i] and [¨] have a higher tongue dorsum position compared to their respective allophones [i6 ] and [¨6 ] after postalveolars. The tongue blade is closer to the hard palate in the post-alveolar region; this configuration in fact closely resembles the blade position during articulation of the preceding fricative consonant [S]. It can thus be said that /i/ and /¨/ both exhibit a [Z]-like tongue posture when they follow postalveolar fricatives; acoustic consequences described above are likely subsequent to this change in constriction location relative to the [i] and [¨] allophones. This similarity to the preceding fricative segment is reminiscent of the so-called APICAL VOWELS found in Standard Chinese, which also adapt the lingual posture of their fricative or affricate onset and maintain it through production of the entire syllable (Lee-Kim 2014).
These data may inadvertently provide an explanation for the fact that [¨] appears to have a comparable or even slightly higher F2 than [ˆ] when these vowels are typically described as differing only in lip rounding. For the present speaker, the tongue position for [¨] is roughly as front as [i] but slightly lowered; this arrangement is frequently encountered in individuals' articulatory implementations of the contrast between front unrounded and front rounded vowels (Wood 1986). Although [¨] has typically been described as central, these data suggest a more anterior articulation for the present speaker, which may counteract the F2-lowering effect of lip rounding. We cannot speculate beyond the present speaker's idiolect on this point, and further research with multiple speakers is needed to evaluate the typical articulation of [¨].

Prosody
The most salient prosodic feature of Kejom is its extensive use of tone for lexical and grammatical purposes. At the phonological level, Kejom's tonal system has been described as a simple opposition between /H/ and /L/ tonemes in various combinations and associations with stem segments (Hyman 1979b). However, a relatively large number of surface tonal sequences are attested; lexical tone of stems is highly 'mobile' and spreads progressively to inflectional material; and tone is used for verb inflection, e.g. the imperative and hortative moods, both a suffixed high tone with no associated segmental material (Akumbu & Chibaka 2012: 167-170). A full discussion of Kejom's tonal phonology and morphology would necessarily be very complex and is not included here for the sake of space. Tonal contrasts are not evenly distributed in Kejom, with stems exhibiting several contrastive melodies while prefixal material (i.e. noun class inflectional prefixes) is generally only low-toned. Most suffixal material is toneless and receives tone from stems via spreading. There are exceptions to these generalizations, namely the infinitive verbal prefix /›@ $/ and the inflectional suffix /s›@ / for noun class 10, which both carry a high tone. We limit further discussion here to the phonetic realization of the contrastive tonal contours observed on short (i.e. non-coalesced) noun and verb stem vowels, in addition to the phonetic and phonological properties of two frequently occurring derived tones, the downstepped high tone and the mid tone.

Stem tone melodies
The tonal contours attested on single stem vowels are illustrated in (4) and visualized in Figure 11.
(4) Tone melodies on monosyllabic stems  Figure 11 f0 tracks for the stem vowels of the words in Table 4 (loess regression, 3-6 tokens per word). f0 estimates taken at 20 sampling points evenly distributed over the vowel's duration using an inverse filter control method (Ueda et al. 2007). Clear octave errors were manually removed. DH is downstepped high tone ( $ H); LF is low-falling tone.
Kejom contrasts simple high (H) and low (L) tone melodies on monosyllables. In addition, a downstepped high tone ( $ H) occurs when an unassociated L tone is stranded between two H tones (as with the infinitive prefix /›@ $/ in (4)). The L tone contrasts with a low falling tone (LF), which we transcribe here as super-low. 3 The latter starts at a pitch level typical of the low tone, but falls more rapidly to a lower level than the low tone and ends near the bottom of the speaker's pitch range. The L tone, by contrast, is relatively level and does not reach the extremely low pitch of the LF tone. The contrast between this pair of tones only occurs before pause; in all other positions, the distinction is neutralized to the level low tone. Tonal melodies involving multiple clear pitch levels are rare on monosyllabic stems, but do occur. High-low falling (HL) and low-high rising (LH) contours are attested, mainly in loanwords, as in [bJ›fl ] 'avocado', from Cameroonian Pidgin English [pia] '(avocado) pear'.
Phonetic contours occur in rapid speech with reasonable frequency, mainly due to dislocation of a following H or L tone onto a preceding syllable due to reduction or deletion of its host vowel, as occurs in the attached North Wind text in the phrase /mô Â›$ mo$ k/ → [mufl Â mç@ /] 'while he was wearing'. The contour tones that do consistently occur on monosyllabic stems regardless of speech rate (HM, HL, LH) are all rare outside of derived environments. In particular, a high-mid (HM) falling contour is only attested on the word [mu -@ ] 'water'.

Mid and downstepped high tones
The downstepped high ( $ H) and mid (M) tones occur only in derived environments (which are themselves in complementary distribution) and have very different effects on the tonological properties of the phonological phrases in which they occur. Within a phonological phrase, downstep lowers the pitch ceiling at which all subsequent H tones are produced. This can be seen in Figure 12 (left), where after a H of about 160 Hz is produced, downstep occurs and constrains subsequent $ H tones to an f0 ceiling of about 140 Hz. The occurrence of a mid tone within a phonological phrase, on the other hand, does not have any observable effect on subsequently produced H tones, as seen in Figure 12 (right): M, which is produced with a pitch similar to the $ H that occurs later in the utterance, is followed by two successive H tones with an f0 about 20 Hz higher. In Kejom, the M and $ H tones appear to be phonetically indistinguishable, both from speaker intuition and acoustic evidence; however, they still exhibit the distinct phonological behaviors described above. For the speaker discussed here, both mid and downstepped high tones typically occupy an f0 range of 125-135 Hz (with the H tone occurring slightly higher at 130-150 Hz). This use of overlapping f0 ranges for two phonologically distinct tonal categories may be enabled by the fact that the two tones are derived in non-overlapping, complementary environments, and there is no possible need for speakers to distinguish the two tones from one another. A mid tone can only be derived in an environment immediately followed by a non-downstepped high tone (H), whereas downstepped high tones may occur with any other tone following ( $ H, L, LF, etc.) and must be preceded by a non-downstepped high tone in order to be derived.

Transcription of recorded passage 'The North Wind and the Sun'
We present a standard comparative text below at two levels of analysis. First, we provide a narrow phonetic transcription that illustrates some rapid-speech phenomena described above, particularly for tone. Second, for clarity, we provide a phonemic transcription with an interlinear gloss. Note that in this transcription, not all morphemes are assigned a phonemic tone. This is a deliberate representational choice, intended to reflect the analysis that much of Kejom's morphological material is underlyingly toneless and receives its phonetic tones from surrounding material via tone spread or reassignment.

pass-PROG who EMPH
The North Wind and the Sun were arguing about who was stronger than who, sɘ́s ͡t èn wù-d͡ ʒèk mú ɰɘ̀mòk dàlɘ́l mtɘ́vì. They agreed that the person who would first make the traveler take off his gown was stronger than the other. ɘ-fʷóf ɘ ɡɘ̀ɘ̀kʷì ɘ́mɘ̀zìtɘ̀sɘ̀t͡ ʃò-ɘ nókɘ̀nàntô.
C3-wind AM.C3 part DIR above SM.C3 then start PRS pass-PROG really much The North Wind then began to blow with great force. Then the Sun began to shine and make places hot, and the traveler quickly took off his gown.

ABBREVIATIONS
Interlinear morphemic glosses mainly adhere to the Leipzig Glossing Rules (Bickel, Comrie & Haspelmath 2015) except for the morphemes indicated below.
• Noun classes are glossed as C#, with the # indicating the class number. Noun class is most frequently expressed with agreement morphology (mostly prefixal) distributed throughout the noun phrase, but is also inherent to some nouns (e.g. 'sun', 'person'). • The associative marker is glossed as AM. This set of morphemes agrees with the noun class of the head of its noun phrase and expresses a variety of semantic relations between the nouns it conjoins (Akumbu & Chibaka 2012: 116-124).
• The subject marker is glossed as SM; this agrees with the noun class of the subject and immediately precedes the verb in some syntactic constructions (Akumbu & Chibaka 2012: 67). • A conjunction used specifically between serialized verbs is glossed as CONJ.