Obstruents

Nafsan exhibits a single plosive series; voicing is not contrastive, and the plosives are generally phonetically voiceless, though some voicing may be present when they occur between voiced segments in rapid speech. They contrast at four places of articulation: bilabial /p/, alveolar /t/, velar /k/, and labial-velar /k͡p/. The labial-velar plosive /k͡p/ is a double articulation, produced with overlapping constrictions at the lips and velum. /t/ may sometimes be produced with a dental place of articulation. The plosives are unaspirated; /p t k/ typically have short positive voice onset times, sometimes with slightly longer release phases for /k/, while the labial-velar plosives generally have a small amount of pre-voicing as well as slightly longer durations than the other plosives, as can be seen in Figure 2d compared to Figures 2a–c for examples from the above list, and as shown in Table 1, based on a small sample of stops produced by three speakers. The plosives are frequently unreleased when utterance-final.

Table 1 Mean closure duration and VOT (in ms) for Nafsan stops, for 61 tokens produced by three female speakers (word-medially in disyllabic words, in an utterance-medial frame).

Figure 2 Spectrograms showing word-medial Nafsan plosives /p t k k͡p/: (a) /naːpas/ ‘coconut branch’; (b) /naːtap/ ‘statue’; (c) /kaːkas/ ‘to be sweet’; (d) /naːk͡pak/ ‘tree sp.’.

The timing of the gestures for /k͡p/ may depend on syllable position. Figure 3 shows first and second formant trajectories for 136 tokens of /aː/ adjacent to /k͡p k p/ in word-initial, word-medial and word-final positions, recorded with three female speakers in an utterance-medial frame. These indicate that as an onset, /k͡p/ is acoustically more like /k/ than /p/ as the articulators close, but more like /p/ than /k/ as the articulators open, as evidenced by the higher second formant frequency for /k͡p/ at the transition from /aː/ into the onset and the lower second formant frequency for /k͡p/ at the transition from the onset to /aː/. This suggests that the velar closure is typically initiated before the labial component when /k͡p/ occurs in syllable onsets, a pattern also supported by the typical realisation of /n/ as [Ŋ] when undergoing place assimilation preceding /k͡p/, as seen for ‘head’ in (4) below. In the transition from /aː/ to /k͡p/ as a syllable coda, /k͡p/ is acoustically more like /p/, suggesting the labial closure has greater overlap with, and perhaps even precedes, the velar closure.

Figure 3 Smoothed trajectories of first and second formant frequencies in tokens of /aː/ preceding and following /k͡p k p/ as syllable onsets and preceding /k͡p k p/ as syllable codas, for three female speakers.

However, /k͡p/ remains distinct from medial sequences of /k/ followed by /p/, as in the examples given in (1).Footnote 4

Figure 4 Spectrograms showing word-medial /k͡p/ compared to /kp/ sequence: (a) /nak͡pel/ ‘yam sp.’; (b) /nakpes/ ‘tree sp.’.

As can be seen in Figure 4, /kp/ sequences typically exhibit noticeably longer durations than /k͡p/, and sometimes a detectable burst at the segment transition. In a sample of 60 word-medial tokens of /k͡p/ and /kp/ in disyllabic words, recorded with three speakers in an utterance-medial frame, the mean duration of /k͡p/ is 155 ms (s.d. 23 ms) and the mean duration of /kp/ sequences is 200 ms (s.d. 24 ms).

There are two fricatives, labio-dental /f/ and alveolar /s/. A non-phonemic postalveolar fricative [ʃ] may also occasionally be heard, for example as in (2) (from French couchez), used to hush a dog (Thieberger Reference Thieberger2006: 101).

As is the case for the plosives, the fricatives are generally phonetically voiceless. A subset of verbs with /p/-initial stems undergo stem-initial consonant mutation, a process often observed for languages of Vanuatu (Crowley Reference Crowley1991). Under this process, /p/-initial forms marked for mood, reduplicated or nominalised are produced with stem-initial /f/ (as in (10) below, for example). This is strictly a morphological alternation, and /p/ is otherwise not lenited in this way (Thieberger Reference Thieberger2006: 162–170; Reference Thieberger2012). However, /k/ may optionally be lenited in various ways; sometimes, it is realised as a velar fricative [x] or [ɣ], approximant [ɰ], or glottal fricative [h], and sometimes it is completely elided (Thieberger Reference Thieberger2006: 47–49). Velar lenition can occur intervocalically between various vowels and also when /k/ is part of a medial consonant sequence. While velar lenition has not yet been comprehensively investigated for Nafsan, it appears to be present in the speech of some speakers more than others, though it is used by speakers of a range of ages and backgrounds. It is also more characteristic of rapid rather than careful speech.

A glottal fricative may also be optionally produced word-initially for some words which are otherwise vowel-initial. As is the case for velar lenition, initial glottal insertion appears to be more frequently used by some speakers than others, in both rapid and careful speech. Similar variation between vowel-initial and glottal-initial forms is observed for Bislama, which as noted is spoken by all Nafsan speakers who have contributed to the present study (Crowley Reference Crowley2004: 14–15). A non-contrastive glottal stop also sometimes occurs in Nafsan (Thieberger Reference Thieberger2006: 53–54). It may optionally be produced after a vowel which is word-final and pre-pausal, and also occurs medially in the high-frequency feedback particles shown in (3).

Nasals

Nafsan has nasal consonants at four places of articulation, corresponding to the plosives: bilabial /m/, alveolar /n/, velar /Ŋ/, and labial-velar /Ŋ͡m/. As is the case for the labial-velar plosive /k͡p/, the nasal /Ŋ͡m/ is produced with overlapping labial and velar closure. However, as also noted for /k͡p/, formant trajectories indicate that in syllable onsets the velar closure for /Ŋ͡m/ is initiated before the labial closure, and the consonant release is similar to that of /m/, for example as seen in the low F1 and F2 following initial /Ŋ͡m/ in Figure 5c below.

Figure 5 Spectrograms showing word-medial prenasalised alveolar trill /n^dr/ compared to alveolar nasal /n/ and alveolar trill /r/: (a) /naːn^dra/ ‘banana sp.’; (b) /nanan/ ‘goat’; (c) /Ŋ͡maːrak/ ‘to be clever’.

Though a range of heterorganic clusters are permitted in Nafsan, as discussed further below, alveolar nasals typically undergo a process of assimilation to the place of articulation of a following plosive. Many examples can be seen for nouns with a word-initial alveolar nasal which originates from an article /na/, also found in many other Oceanic languages (Crowley Reference Crowley1985). In Nafsan this article is frequently applied to common nouns (Thieberger Reference Thieberger2006: 123–124) and the vowel is often deleted, which then allows anticipatory assimilation to take place, as shown in the examples in (4).

The vowel deletion suggests some level of accretion, as also noted for this article in various other languages of Vanuatu (Lynch Reference Lynch2001), and in some cases the use of this article may be lexicalised, and forms with the assimilated nasal could be considered the phonemic representation. However, in Nafsan the article is also still used productively, for example in nominalisation processes (Clark Reference Clark1985: 31; Lynch Reference Lynch2000: 322; Thieberger Reference Thieberger2006: 132–137), as seen later in (10).

As in many languages (e.g. Gordon Reference Gordon2016: 131), the place assimilation does not appear to apply to non-coronal nasals; for example, as shown in (5) below, verb stems which include the historical (though no longer productive) de-transitivising prefix /ma/, which also has the form /m/, are all produced with a bilabial nasal regardless of the place of the following consonant (Thieberger Reference Thieberger2006: 70, 214–216).

This lack of assimilation in some nasal-initial onset clusters is perhaps also not surprising given that Nafsan freely permits various heterorganic clusters.

When the consonant following the alveolar nasal is labio-dental fricative /f/, assimilation generally does not occur, and /n/ of course already shares place features with fricative /s/ and liquids /l/ and /r/. When the segment following /n/ is also a nasal, assimilation may not occur in careful speech (e.g. as in the last two examples in (14) below, especially if the nasal originates from a productive use of the nominal article), but assimilation often occurs in rapid speech.

Trills and lateral

In Nafsan there is an alveolar lateral /l/ and both an alveolar trill /r/ and prenasalised alveolar trill /n^dr/. The alveolar trill may occasionally be produced as a tap inter-vocalically. The prenasalised alveolar trill is a complex segment, involving a sequence of gestures beginning with an alveolar closure and open velar port, followed by some pre-stopping in the transition from the nasal to the oral articulation, then releasing into the voiced alveolar trill. While there are many possible consonant clusters in Nafsan, discussed further in relation to the examples in (12)–(16) below, distributional evidence suggests the prenasalised trill operates as a unit; for example, if it were a cluster it would be the only consonant cluster to occur in coda position, and would also occur with disproportionately high frequency as an onset cluster (Thieberger Reference Thieberger2006: 51–52, 58–62). A contrastive prenasalised trill may be present in other closely related varieties spoken elsewhere on Efate and the small islands to the north (Clark Reference Clark1985: 13; Sperlich Reference Sperlich1991: 59), and is also found in various other languages of Vanuatu, as well as in Oceanic languages such as Fijian and a number of languages of Manus Island (Tryon Reference Tryon1976, Blust Reference Blust, Jeff, John and Diana2007). The presence of the prenasalised trill in these languages is suggested to be reflective of a larger set of coronal contrasts in Proto-Oceanic, compared to that retained in most other contemporary Oceanic languages (Ross Reference Ross1988: 31; Lynch, Ross & Crowley Reference Lynch, Ross and Crowley2002: 64; Clark Reference Clark2009: 11). Both the alveolar trill and the prenasalised alveolar trill may be partially or fully devoiced when word-final and pre-pausal (e.g. as can be heard for ‘mouth’ in (4) above), or when preceding a voiceless consonant, and devoicing is also sometimes observed for the alveolar lateral in these contexts. For some speakers, the alveolar lateral may be produced with some pre-stopping when it follows the alveolar nasal in an onset cluster, as [n^dl] (Thieberger Reference Thieberger2006: 52). This is similar to the pre-stopping noted for the prenasalised trill, but optional. Similarly, some speakers may produce onset clusters of bilabial nasals followed by a trill or lateral with pre-stopping, as [m^br] or [m^bl].

Glides

Nafsan has two glides: palatal /j/ and labial-velar /w/. Phonemic /j/ occurs very infrequently; it occurs initially in only a very small number of words, including borrowed words and personal names, and as a word-medial phoneme it is also uncommon. However, a palatal glide is frequently produced in the transition between a close front vowel and a more open vowel in vowel sequences, or as an allophone of the close vowel in such sequences, for example as in (6), which comprises a stem ending in /i/ followed by a suffix /en/.

Phonemic instances of /w/ are more common than those for /j/, but similarly, it may also occur phonetically, in the transition between a close back vowel and a more open vowel in a vowel sequence, or as an allophone of /u/, forming a cluster with the preceding consonant, as seen in (7), with a stem ending in /u/ followed by a suffix /om/.

These instances of glide formation and insertion are frequent but optional, and the syllabification of words with vowel sequences beginning with /i/ and /u/ depends to some extent on speech rate, and likely interacts with other prosodic and phonotactic considerations. Both /j/ and /w/ also occur phonemically in onset clusters, as seen in (12) below, for example.

Monophthongs

Figure 6 First and second formant frequencies (in Hz) at midpoints of Nafsan monophthongs in CVC words, for three speakers ( $95\%$ confidence intervals).

Figure 7 Duration values (in ms) for Nafsan monophthongs in CVC words, for three speakers.

The phoneme inventory of Nafsan includes ten monophthongs, /i iː e eː a aː o oː u uː/.Footnote 5 Some examples are shown above. Contrasts can be observed between five distinct vowel qualities, and each of these may occur phonemically short or long. A schematic representation of the ten monophthongs is shown above, and Figure 6 shows the acoustic vowel space of Nafsan, based on measurements of the first and second formant frequencies at midpoints of 325 vowel tokens in CVC monosyllables (with stop onsets and codas), recorded with three female speakers of Nafsan in an utterance-medial frame. Figure 7 shows the duration values for the same set of vowels, and clearly illustrates the length difference; the long vowels (mean 151 ms, s.d. 23 ms) are overall 1.91 times longer than the short vowels (mean 79 ms, s.d. 15 ms), and the duration differences are quite consistent across the five vowel qualities. Length is reportedly also used contrastively to at least some extent in the vowel systems of closely related language varieties (Schütz Reference Schütz1969, Sperlich Reference Sperlich1991, Lacrampe Reference Lacrampe2014). While the phonemically short vowels can be produced with a more centralised quality compared to the phonemically long vowels, in general the quality difference is very small in monosyllabic words or accented syllables. However, in syllables preceding an accented syllable, short vowels can show substantial centralisation (Billington et al. Reference Billington, Fletcher, Thieberger, Volchok, Julien, Joe, John and Caroline2018, Reference Billington, Fletcher, Thieberger and Volchok2020). A recent study of Nafsan monophthongs examines the acoustic and durational correlates of quality and quantity distinctions in more detail (Billington, Thieberger & Fletcher Reference Billington, Thieberger and Fletcher2021).

Diphthongs

The five contrastive vowel qualities in Nafsan are also drawn on in the formation of various diphthongs. These include the closing diphthongs shown in (8).

As noted earlier, there are also phonemic vowel sequences beginning with close vowels which may be produced phonetically with glide formation or insertion, and in addition, some other vowel sequences or possible diphthongs have been noted in a small number of lexical items (Thieberger Reference Thieberger2006: 64), but the phonemic status of these is not clear.Footnote 6 The closing diphthongs shown in (8) include contrasts between diphthongs which have similar trajectories but differ primarily in the closeness of the second target: /ai/ and /ae/, and /au/ and /ao/. This can be seen in Figure 8, which shows first and second formant trajectories for the 5–7 repetitions (total 37 tokens) produced for the items in (8). Further, there may be some durational differences; for these tokens produced in isolation, the mean duration for /ae/ is 244 ms and for /ao/ 245 ms, while means for /ai/ and /au/ are 182 ms and 172 ms, respectively (comparable to 196 ms for /ei/ and 180 ms for /oi/). Though /ae/ and /ao/ occur less frequently, various minimal pairs are attested in both open and closed syllables which show these diphthongs to be distinct from /ai/ and /au/.

Figure 8 Smoothed trajectories of first and second formant frequencies in tokens of /ai ae au ao ei oi/ in open syllables, for one speaker.

Vowel deletion

In Nafsan, short vowels may be deleted word-medially, as part of a process which appears to be both historical and ongoing (Clark Reference Clark1985, Lynch Reference Lynch2000, Thieberger Reference Thieberger2006). In comparing various recorded words with corresponding examples written down by early missionaries or reported for closely related language varieties spoken further to the north, it can be seen that many contemporary Nafsan words lack vowels which were produced at an earlier time in the language’s history, and which are retained in related languages. In many cases these forms with deleted vowels are the current phonemic form, consistently used by different speakers, though speakers may still be aware of older pronunciations. Many of the onset clusters given in (12)–(16) below have arisen as a result of this, and further discussion of historical vowel deletion can be found in Thieberger (Reference Thieberger2006: 2–3, 68–70) and Billington, Thieberger & Fletcher (Reference Billington, Thieberger and Fletcher2020: 127–128).

This process is also productive. While some words have stems which have lost historical vowels, others retain them but show regular patterns of vowel deletion in inflected and derived forms, for example in the monomorphemic nominal forms compared to forms marked for direct possession in (9).

Analyses of prosodic patterns and corpus materials indicate that this deletion process applies to penultimate short vowels of any quality preceding a final accented syllable; that is, vowel deletion is pre-tonic (Billington et al. Reference Billington, Fletcher, Thieberger, Volchok, Julien, Joe, John and Caroline2018, Reference Billington, Fletcher, Thieberger and Volchok2020; Billington, Thieberger et al. Reference Billington, Thieberger and Fletcher2020).

To some extent, the current productive deletion process appears to be mediated by phonotactic, grammatical, and lexical factors, described in detail in Billington, Thieberger et al. (Reference Billington, Thieberger and Fletcher2020: 136–140). For example, the words for both ‘language, story’ and ‘curse’ are derived from the stem /pes/ ‘talk’ via a regular nominalisation process (in which the stem form is /fes/ under the consonant mutation noted earlier), but, as shown in (10), the word for ‘language, story’ (also the autonym for the language) invariably shows vowel deletion (and the reduced form could be considered the phonemic form in this case).

The word for ‘curse’, which is not only semantically different but much less frequently used, is reportedly never produced with vowel deletion. Example spectrograms can be seen in Figure 9. Pre-tonic vowels in a grammatical morpheme (such as a verbal proclitic), or for which deletion would result in a dispreferred consonant sequence, are also typically retained though, as noted earlier, they may show substantial centralisation in this environment (Billington et al. Reference Billington, Fletcher, Thieberger, Volchok, Julien, Joe, John and Caroline2018, Reference Billington, Fletcher, Thieberger and Volchok2020).

Figure 9 Spectrograms showing presence compared to absence of vowel deletion: (a) [nafesaːn] ‘curse’; (b) [nafsaːn] ‘language, story’.

Historically, final short vowels have also been lost in Nafsan compared to other Oceanic languages, resulting in many closed syllables, and as noted earlier, /a/ in the common nominal prefix /na/ is also often deleted, allowing /n/ to form an onset cluster with other consonants (Clark Reference Clark1985, Lynch Reference Lynch2000, Thieberger Reference Thieberger2006).Footnote 7 These various deletion processes, combined with the possibilities for compounding as a word-formation strategy, lead to many possibilities for complex consonant sequences to arise, discussed in more detail below.