2.1 Fricativization

Step I involves the fricativization of voiced stops. The articulation of voiced stops makes conflicting demands: On one hand, airflow must be maintained to produce voicing. On the other hand, the airflow must be obstructed to produce a stop. This competition means that voiced stops are often cross-linguistically avoided. Intervocalically, voiced stops are commonly lenited to fricatives (Kirchner 2001: 102–122; Kümmel 2007: 55–75; Kaplan 2010: 13; Beguš 2018: 101; Beguš 2019; 2020). Beguš (2020) counts about 47 cases of postvocalic fricativization of voiced stops based on approximately 294 languages surveyed in Kümmel (2007). We propose that in Kayan-Murik, Berawan dialects, and Kalabakan Murut, voiced stop fricativization was generalized to any postvocalic position, including word-finally. Evidence supporting this claim is presented in section 3.

2.2 Devoicing

Step II involves (at least partial) word-final devoicing of voiced fricatives. Voiced fricatives, like voiced stops, are articulatorily difficult. Voicing requires the pressure in the oral cavity to be as low as possible, while frication requires it to be as high as possible (Ohala 1983; 1997; Smith 1997; Ohala 2006). Due to these opposing demands, maintaining voicing in fricatives is strenuous. Additionally, voicing in the word-final position is particularly dispreferred, making final devoicing one of the most common phonological processes across many language families (Beguš 2020). As such, final devoicing is attested and articulatory motivated. Note that due to the fricativization in Step I, there are no word-final voiced stops left, so Step II can be stated simply as word-final devoicing, without restricting it to the devoicing of word-final fricatives to the exclusion of word-final stops. This being said, Davidson (2016) shows that English word-final fricatives tend to devoice more often than stops, which suggests that voiced fricatives are particularly difficult to produce word-finally.

2.3 Nasalization

Step III involves spontaneous nasalization of voiceless fricatives. While spontaneous nasalization (henceforth SN) is generally rare typologically, the vicinity of voiceless fricatives constitutes the one environment where this development is attested and phonetically grounded (other such environments include affricates and aspirated stops; Ohala & Ohala 1993). Instances of spontaneous nasalization can be observed e.g. in some developments from Indo-Aryan to Modern Hindi (2) and Early Modern Breton borrowings from French (3). These instances of spontaneous nasalization are recurrent, i.e. they affect large classes of words, but they are not regular.

Ohala & Ohala (1992; 1993) propose a phonetically grounded explanation for the relationship between high airflow segments, such as fricatives, and perceived nasalization—voiceless fricatives require a greater glottal opening which ends up creating an acoustic effect similar to nasalization. Ohala & Ohala’s (1993) explanation is reproduced verbatim in (4).

Ohala & Amador (1981) demonstrate that the spectral effects of fricatives are indeed similar to those of nasalization. Ohala & Amador (1981); Ohala & Ohala (1992) show that speakers of English, Spanish, and Hindi judge fricative-adjacent vowels as significantly more nasal than vowels not adjacent to fricatives. Khattab & Al-Tamimi & Alsiraih (2018) demonstrate that Iraqi Arabic pharyngeals exhibit various degrees of nasalization when combined with nasals and suggest that their variable realization may act as a precursor for sound change.

In particular, nasality shows a particularly strong connection to glottal articulation. The close affinity between the two classes of sounds has been dubbed rhinoglottophilia by Matisoff (1975). Ohala’s (1975) summary of its phonetic basis is given in (5).

The rhinoglottophilic connection between nasals and glottals is reflected in their frequent synchronic and diachronic connections. For example, Johnson et al. (2019) show that Thai non-high vowels spontaneously nasalize after the glottal consonants h and ʔ. Blevins & Garrett (1992; 1993) propose that Proto-Ponapeic geminates developed into Ponapean nasal-stop sequences through a stage of preaspiration (*Tː > *hT > NT), demonstrating, if correct, that spontaneous nasalization can give rise to a regular sound change (6).

While we propose that in Austronesian and Dakota, spontaneous nasalization was triggered by all voiceless fricatives (not only h), the phonetics of rhinoglottophilia is similar to that of other cases of spontaneous nasalization, lending indirect support to our account.

Finally, we speculate that preaspiration may be an intermediate step linking voiceless fricatives to nasalization. Gordeeva & Scobbie (2010: 169–170) show that female speakers of Scottish Standard English (SSE) often produce word-final voiceless fricatives with a considerable preaspiration. For example, grass may be realized as [gɹahs]. SSE preaspiration affects only fricatives; stops are not preaspirated. Thus, we suggest that final nasalization in the cases at hand may have arisen via the intermediate step of preaspiration. Word-final voiceless fricatives first preaspirated, and then preaspiration gave rise to rhinoglottophilic nasalization, i.e. VS > VhS > ṼS̬̃ / _ #. Positing this intermediate stage has two advantages. First, it connects spontaneous nasalization to preaspiration, which has been argued to give rise to nasalization as a regular sound change (Blevins & Garrett 1993). Second, since fricative preaspiration has been reported word-finally (Gordeeva & Scobbie 2010: 170), it helps explain why voiced stops nasalized only in the word-final position.

2.4 Occlusion

Step IV involves the occlusion of nasalized fricatives. Frication requires high oral air pressure while nasalization consists of lowering the velum and redirecting a part of the pulmonic airflow through the nasal cavity. Ohala (1975); Ohala & Ohala (1993) suggest that the remaining oral air pressure might be too low to produce a nasal fricative. This makes nasal fricatives very articulatorily difficult (and typologically rare). As such, a change away from nasalized fricatives is to be expected.

The development from spontaneously nasalized fricatives to nasal stops might have proceeded along several different paths. First, nasalized fricatives could have developed into nasal stops as a single sound change, i.e. ṼS̬̃ > VN. While we are not aware of previously reported instances of this change, we speculate it might be due to the overall typological rarity of nasal fricatives.

Second, the occlusion of nasalized fricatives might have itself involved a combination of two changes: fricative approximantization and approximant occlusion, i.e. ṼS̬̃ > ṼR̃ > VN. The change of a fricative into an approximant is very well attested, with many reports of unconditioned β, v > w; ð > l, ɾ (> r); ʝ, ʓ > j; and ɣ > w (Kümmel 2007: 78–88).⁵ In the scenario at hand, the fricatives undergoing the change are at least partially nasalized. The aerodynamic difficulty involved in the production of nasalized fricatives (Ohala 1975; Ohala & Ohala 1993) makes their approximantization all the more likely. Indeed, one of the possible implementations of Scottish Gaelic phonologically nasal fricatives is as nasal approximants (Warner et al. 2015).⁶

In this scenario, the approximantization of nasal fricatives is then followed by the occlusion of nasal approximants. This again is a typologically attested change. For example, the Medieval Polish nasal vowel ã (< Proto-Slavic *ę and *ǫ) developed into the Standard Polish ɔw̃ (and ɛw̃,⁷ following a conditioned split). The diphthongized ɔw̃ further developed into ɔm in Greater Poland (Stieber 1973: 129; Baranowska & Kaźmierski 2020; Kaźmierski & Szlandrowicz 2020). In A’ingae, nasal stops in the coda position nasalized the preceding vowel and underwent deletion. Additionally, most instances of ɲ and some instances of m go back to sequences of *nj and *nʋ (Sanker & AnderBois t.a.). Regardless of the relative ordering of the two changes, A’ingae developments testify to the occlusion of nasalized approximants: *Vnj (> *Ṽj) > Ṽɲ; *Vnʋ (> *Ṽʋ) > Ṽm. In Samoyed and in Lappish, word-initial *j- became ɲ- if a nasal consonant is present later in the word, i.e. *j- > ɲ- / # _ VN (Collinder 1960: 63; Mikola 1988: 228f; Kümmel 2007: 159). In Kamassian, *w- changed into m- in the same environment, i.e. *w- > m- / # _ VN (Collinder 1960: 66; Mikola 1988: 228f; Kümmel 2007: 159). Both changes presumably resulted from nasal spreading with nasalized approximants as the intermediate stage, i.e. *j- > * j̃- > ɲ- and * w- > * w̃- > m-.

Finally, the step of occlusion might have proceeded as follows: (i) the spontaneously nasalized vowel unpacks into an oral vowel followed by a nasal stop, (ii) the postnasal fricative hardens to a stop, (iii) the resulting stop is ultimately deleted. This sequence of changes can be schematized as ṼS̬̃ > VNS̬ > VND > VN. The unpacking is attested in the development of Portuguese nasal vowels into sequences of oral vowels followed by a nasal stop in Portuguese-based creoles. E.g., in Casamancese Creole (Biagui & Quint 2013), kõˈtar > konta ‘tell’ (p. 42), ũ > uŋ ‘one’ (p. 43), ˈsĩku > siŋku ‘five’ (p. 43). In Papamientu (Birmingham, Jr. 1970), ʒɐ̃ˈtar > žaːnta ‘eat’ (p. 6), tɐ̃ˈbɐ̃j̃ > taːmbɛ ‘also, too’ (p. 6), ˈbõ > bõŋ ‘good’ (p. 34). (Yet another example is the previously discussed development of the Medieval Polish ã into the Greater Poland Polish ɔm.) The latter sequence of fricative hardening followed by stop deletion is, among others, a familiar development from German, e.g. Germanic *tanθ > Old High German zand > German Zahn ‘tooth.’

In sum, we propose that final nasalization in Austronesian and Siouan arose from a sequence of sound changes which are natural, i.e. motivated by facts of speech production and perception. First, voiced stops underwent fricativization. Second, the fricatives devoiced. Third, voiced fricatives triggered spontaneous nasalization. Finally, fricatives in the vicinity of nasalization occluded to nasal stops. Nasal occlusion might have involved an intermediate approximantization, proceeded via post-nasal hardening, or consisted of a single sound change. Our accounts are compatible with any of the three scenarios. As such, we remain agnostic about the possible incremental changes involved in the final step.

3.1 Kayan-Murik

Let us first take a look at the Kayan-Murik (KM) data. Kayan-Murik (or Kayanic) is a branch of Austronesian languages spoken in Borneo, Indonesia, and Malaysia. Blust (2016) reports (at least) two dialects to have undergone final nasalization: Uma Bawang (UB) and Long Atip (LgA) dialects of Kayan (7). Proto-Kayan-Murik (PKM) final *-b and *-d (7a) correspond to Uma Bawang and Long Atip -m and -n (7b-c). In Uma Juman (UJ), the PKM word-final *-b fricativizes to -v and *-d lenites to -r (7d). While the data is limited, Blust (2016) observes that Murik seems to have devoiced PKM *-b, but nasalized PKM *-d (7e).

Word-initially, the voiced stops *d- and *b- do not undergo any changes in Uma Juman, Uma Bawang, or Long Atip. Word-internally, *-d- lenites to -r- in all four languages. Crucially, Uma Juman intervocalic *-b- fricativizes to -v-. In Murik, word-initial *d- lenites to l-; Murik non-final *b does not change. These developments are summarized in (8).

Additionally, Murik also shows phonetically unnatural post-nasal devoicing (henceforth PND), whereby Proto-Kayan-Murik voiced stops devoice after nasal consonants (8).

The development of unnatural post-nasal devoicing in Murik and twelve other languages receives extensive treatment in Beguš (2019), who demonstrates that post-nasal devoicing always arises from a sequence of (at least) three changes, in the course of which (i) “[a] set of segments enters complementary distribution,” (ii) “[a] sound change occurs that operates on the changed/unchanged subset of those segments,” and (iii) “[a]nother sound change occurs that blurs the original complementary distribution” (p. 735). Beguš (2019) refers to a combination of sound changes that fit that description as the blurring process. Changes involved in the post-nasal devoicing in Murik are summarized in (10). For more on the mechanics and subtypes of the blurring process, see e.g. Beguš (2018; 2019; 2020; 2022); Beguš et al. (2022).

First (10i), voiced stops *b and *d fricativize to *β and *ð if not directly preceded by a nasal. Second (10ii), all the remaining (i.e. post-nasal) voiced stops devoice. Finally (10iii), word-initial and intervocalic *ð lenites to l- and -r- respectively,⁸ and *β occludes to b. This combination of sound changes in Murik gives rise to the appearance of unnatural post-nasal devoicing. However, due to the preceding fricativization (*b, *d > *β, *ð) in other positions, the devoicing of stops was unconditioned (and hence phonetically natural) at the time of its actuation.

Evidence for the proposed trajectory of development (10) comes from the reflexes of Pre-PKM *d. The alveolar stop lenites word-initially and intervocalically but devoices post-nasally, suggesting a stage of complementary distribution. The lenited reflexes point to a stage of fricativization since *d is likely to develop into l and r through an interstage with *ð (Beguš 2019).

We extend Beguš’s (2019) argument to propose that Uma Bawang and Long Atip underwent a similar fricativization. Unlike in Murik however, the UB and LgA voiced stops develop into voiced fricatives in all post-vocalic positions, including word-finally. This reconstruction is supported by evidence from the closely related dialect of Uma Juman. Uma Juman lacks final nasalization, but voiced stops develop into voiced fricatives in all post-vocalic positions, including word-finally (7d): PKM *-b develops into *-β > -v and PKM *-d develops into -r, likely through an interstage with *-ð. In Uma Bawang and Long Atip, the word-final fricatives devoice, creating a precondition for spontaneous nasalization. Finally, nasalized fricative occlude to nasal stops. The developments of PKM word-final *-b and *-d are exemplified in (11).

We observe that Proto-Malayo-Polynesian (PMP) did not have phonemic non-strident fricatives (except for *h) (Blust 2013: 585–6, 591–3), and we have no reasons to believe that the immediate ancestors of Kayan-Murik, Berawan (§3.2), Kalabakan Murut (§3.3), or Karo Batak (§3.4) had the non-strident phonemic fricatives either. As such, the stages of fricativization and devoicing did not result in any mergers with *β, *ð, *ɣ, or *ɸ, *θ, *x.

Proto-Malayo-Polynesian had two fricatives *s and *h (Blust 2013: 591–3). The alveolar fricative *s does not trigger spontaneous nasalization. However, *s is a sibilant. Sibilants have a higher amplitude and pitch than non-sibilant fricatives. As a consequence, sibilants and non-sibilants are often found to pattern differently. Our discussion of Austronesian fricatives is restricted to the non-sibilant ones, to the exclusion of the *s.

Finally, PMP had the non-strident glottal fricative *h (Blust 2013: 591–3). In most Malayo-Polynesian languages, word-final *-h has been lost (Blust & Trussel & Smith 2023). The PMP word-final *-h does not have nasal reflexes, suggesting that it had been lost in the daughter languages before the sequence of changes in (1) took place.

In sum, we propose that final nasalization in Uma Bawang and Long Atip proceeded through the interstage of fricativization. Fricativization is independently motivated by the facts of Murik post-nasal devoicing and synchronically attested in the closely related Uma Juman.

3.2 Berawan

The second language with reported final nasalization that shows independent evidence of fricativization of voiced stops is Berawan. Berawan is a group of closely related dialects spoken in the Malaysian state of Sarawak. Proto-Malayo-Polynesian (PMP) word-final voiced stops (12a) correspond to nasal stops in the Berawan dialects of Long Terawan (12b), Batu Belah (12c), and Long Jegan (12d).

Intervocalically, the Berawan voiced stops undergo other unusual developments. The Proto-Malayo-Polynesian *-g- devoices to -k-; *-b- also devoices and velarizes to -k-. PMP *-d- does not devoice, but rather lenites to -r-. Word-initially, the three voiced stops do not undergo any changes. Some of these developments are exemplified in (13). The developments of voiced stops in all positions are summarized in (14).

The phonetically unnatural intervocalic devoicing receives treatment in Beguš & Dąbkowski (2023), who propose that like the post-nasal devoicing in Murik, Berawan’s intervocalic devoicing is also the outcome of a sequence of sound changes (the blurring process) (15).

First (15i), voiced stops fricativize intervocalically. Second (15ii), *ð lenites to *r,⁹ while the other voiced fricatives devoice unconditionally. Third (15iii), the labial fricative *ɸ velarizes to *x. The velarization of labial fricatives is typologically attested (Kümmel 2007: 222) and phonetically motivated (because non-strident fricatives are perceptually confusable; Redford & Diehl 1999). Finally (15iv), the remaining *x occludes to k. The unconditioned change of *x > k has been reported e.g. from Proto-Tacanan to Cavineña (Key 1968). A similar change of *x > kʰ has been reported from Proto-Tai to several Tai languages (Li 1977: 207–214).

Several pieces of evidence support Beguš & Dąbkowski’s (2023) reconstruction. First, the lenition of *-d- to -r- likely went through an intermediate step of *-ð-, pointing to a stage of fricativization. Second, the velarization of labial fricatives is more common than velarization of labial stops (Kümmel 2007: 221–2). Third, unless an interstage of intervocalic fricativization is postulated, the changes of devoicing and velarization cannot be ordered. Ordering devoicing before velarization incorrectly predicts the velarization of Pre-Berawan *p > ^×k. Ordering velarization before devoicing incorrectly predicts the velarization of Pre-Berawan word-initial *b- > ^×g-. The only way out of the ordering paradox is to say that the velarization of stops (an already unusual change) occurred only intervocalically. This is highly implausible since the place of articulation of stops is best articulatorily cued specifically between vowels. For a full exposition of the argument, see Beguš & Dąbkowski (2023).

We extend Beguš & Dąbkowski’s (2023) analysis in proposing that—similarly to Kayan-Murik—Berawan fricativization takes place in all post-vocalic positions, including word-finally. Next, we reconstruct that the fricatives devoice. Note that the unconditioned devoicing of fricatives in Berawan is independently motivated in the account of Berawan’s intervocalic devoicing (15ii). Finally, word-final fricatives nasalize and occlude, yielding apparent word-final nasalization. This development in Long Terawan is exemplified in (16).¹⁰

In Berawan, only the word-final voiceless fricatives trigger spontaneous nasalization. There are at least three ways in which one may try to account for this fact. First, Berawan devoicing might have taken place in two steps, separated by spontaneous nasalization. At first, word-final fricatives devoice. Next, voiceless fricatives nasalize. Lastly, devoicing affects all other fricatives.

Second, it may be the case that only word-final fricatives nasalized because articulatory cues in the final position are the weakest. i.e., word-final fricatives and adjacent vowels are most readily misperceived as nasalized (via the mechanism of Ohala & Amador 1981).

Third, Berawan final nasalization might have involved an intermediate stage of fricative preaspiration, as discussed in subsection 2.3. In Standard Scottish English, only word-final fricatives preaspirate (Gordeeva & Scobbie 2010: 170). If the same facts held of Berawan preaspiration, only word-final fricatives would be then affected by rhinoglottophilic nasalization. Due to a lack of additional evidence, we do not take a strong stance on why only word-final fricatives nasalize in Berawan.

In sum, we propose that final nasalization in Berawan arose through the intermediate stage of voiced stop fricativization. The interstage of fricativization is motivated independently by the facts of Berawan intervocalic devoicing.

3.3 Kalabakan Murut

In two out of four reported languages with final nasalization, a stage of voiced stop fricativization must be reconstructed for reasons completely independent of final nasalization. In Kayan-Murik, the interstage of fricativization accounts for phonetically unnatural post-nasal devoicing (Beguš 2019). In Berawan, fricativization explains the unusual devoicing of stops between vowels (Beguš & Dąbkowski 2023). In one additional language—Kalabakan Murut—there exists strong internal and dialectal evidence that final nasalization arises through an interstage with fricativization as well.

Kalabakan Murut (MKal) is a language of the Sabahan branch of Austronesian spoken in Malaysia. Blust (2016) reports that Kalabakan Murut underwent final nasalization (17). Proto-Murutic (PMur) final voiced stops *-b, *-d, and *-g (17a) develop to -m, -n, and -ŋ (17b). Notably, Proto-Murutic *-b-, *-d-, and *-g- yield -b-, -r-, and -h- intervocalically at morpheme boundaries when followed by a vowel-initial prefix (17c). Word-initially, voiced stops do not change. The Kalabakan Murut developments are summarized in (18).

To account for the Kalabakan Murut reflexes, we propose that—like in Kayan-Murik and Berawan—the Proto-Murutic voiced stops (19) fricativize word-medially and finally, i.e. in all post-vocalic positions (19i). The reflexes of Proto-Murutic *d and *g point directly to this interstage. Intervocalically, *d > *ð lenites to r. The velar fricative *g > *ɣ unconditionally devoices to *x. The other fricatives devoice word-finally (19ii). Word-final fricatives spontaneously nasalize (19iii). At last, fricatives occlude, yielding word-final nasal stops (19iv). In Karo Batak (discussed briefly in §3.4), the voiceless velar fricative x is word-finally in free variation with voiceless glottal fricative h (Woollams 1996: 17), pointing to an ongoing sound change x > h. We propose that this debuccalization takes place in Kalabakan Murut as well.

This proposed sequence of developments yields the synchronic alternation between final -m, -n, and -ŋ (17b) and medial -b-, -r-, and -h- (17c), which directly points to an interstage with frication, e.g. aŋ-ipaŋ ∼ ipah-in or an-akon ∼ tokor-on.

In Kalabakan Murut, only the word-final *-x spontaneously nasalized, even though the Proto-Murutic word medial *-g- had also fricativized to *-x-. It is possible that, like in Berawan, only the final fricatives triggered spontaneous nasalization because articulatory cues in the final position are least robust (i.e. word-final fricatives and adjacent vowels are most readily misperceived as nasalized). This scenario is particularly plausible if the velar fricative had debuccalized before nasalizing, i.e. *-x > *-h > *-h̃. Word-finally, the realization of h is very weak, which may have further precipitated reinterpreting the glottal fricative cues as nasal.

The reconstruction of the stages of fricativization (19i) and word-final devoicing (19ii) is further supported by data from closely related dialects (Lobel 2013). In Murut Paluan (MPal) and Murut Timugon (MTim), the Proto-Greater Murutic (PGMur) voiced velar stop *-g develops word-finally into the voiced velar fricative -ɣ (e.g. PGMur *liʔog ‘neck’ > MPal, MTim lijoɣ). In Murut Tagol (MTag), the fricative further debuccalizes to -h (> MTag lijoh). The Kalabakan Murut cognate shows word-final nasalization (> MKal lijoŋ). The developments of the Proto-Murutic voiced stops in the word-final position are illustrated in (20).

In sum, three out of the four Austronesian languages with final nasalization reported in Blust (2016) show strong evidence for an interstage in which voiced stops develop into fricatives. In two of these (Kayan-Murik and Berawan), the evidence for a stage with fricativization is independent of final nasalization data. In Kalabakan Murut, word-final nasals still alternate with an approximant r (< *ð) and a voiceless fricative h (< *x < *ɣ), pointing directly to a stage of fricativization. Thus, the three Austronesian languages—we argue—had a stage of word-final voiced stops surfacing as fricatives, which is precisely the environment where spontaneous nasalization is phonetically motivated (§2). We propose that the fricatives devoice word-finally (if not already devoiced as in Berawan). Next, word-final rimes nasalize (VS > ṼS̬̃) due to the perceptual similarity between fricative-adjacent and nasalized vowels (Ohala & Amador 1981). Finally, the nasalized fricatives occlude to nasal stops.

3.4 Karo Batak

Finally, we turn to discuss the case of Karo Batak, a Northern Batak language spoken by the Karo people in North Sumatra. Blust (2016) reports that Proto-Batak (PB) (21a) word-final voiced stops devoice in Southern Batak (SB), including Toba Batak (TB) (21b), but nasalize in Northern Batak (NB), including Karo Batak (KB) (21c). However, in Karo Batak, no direct evidence for an interstage with frication exists.

There are three possible explanations for final nasalization in Karo Batak. First, final nasalization in Karo Batak could have resulted from a single sound change (22) (as proposed by Blust 2016), which would distinguish Karo Batak from the other languages discussed above, where evidence for intermediate stages is present.

Second, Karo Batak FN could have arisen from the same combination of sound changes as in the other Austronesian languages (23). In this familiar scenario, post-vocalic stops fricativize (23i), devoice word-finally (23ii), spontaneously nasalize (23iii), and lastly occlude (23iv).

In other positions, KB fricatives undergo occlusion without devoicing or nasalization, i.e. *-b- > *-β- > -b-, *-d- > *-ð- > -d-, and *-g- > *-ɣ- > -g-. We reconstruct this sequence of sound changes in Berawan, Kalabakan Murut, and Murik as well. For example, in Murik and Kalabakan Murut, the intervocalic labial fricative *β (< *b) occludes back to b. In Berawan, occlusion occurs in the velar series (*b, *g > *x > k). What distinguishes Karo Batak from the other languages is that in KB, fricatives occlude in all places of articulation, blurring the evidence for the initial development of frication.

While we are unaware of direct evidence for the sequence of changes outlined above, languages closely related to Karo Batak exhibit spontaneous nasalization of fricatives, which is precisely the step we are proposing in the history of Karo Batak’s final nasalization. These developments further support our incremental scenario. For example, NB picat ‘to squeeze sth., to massage breasts’ corresponds to TB, Simalungun pisat and and Angkola Mandailing (AnMa) piñat (Adelaar 1981: 15). The AnMa mañabia, manabia is a borrowing from Sanskrit vāyavya (Adelaar 1981: 16). Sanskrit lakṣa ‘hundred thousand’ has been borrowed into NB as laksa ‘ten thousand’ and becomes TB laso and in AnMa laño (Adelaar 1981: 15). These examples reveal the directionality of change—in AnMa, fricatives (and approximants) have undergone spontaneous nasalization.

Finally, Karo Batak’s final nasalization could also have resulted from analogical extension. While the Proto-Batak final voiced stops *-b, *-d, and *-g correspond to Karo Batak -m, -n, and -ŋ, these nasalized final stops do not alternate (21d). For example, PB *tərəb ‘mass of people’ yields KB tərəm, but when the stop appears in a non-final position, it also surfaces as a nasal, e.g. kə-tərəm-ən ‘receive many guests at once.’ This stands in contrast with Kalabakan Murut which preserves a synchronic alternation (17b-c), e.g. aŋ-ipaŋ ∼ ipah-in or an-akon ∼ tokor-on.

While by itself, this piece of evidence does not necessarily mean nasalization has to be analogical, evidence from the related Toba Batak points to a stage which could have been its precursor. Toba Batak has a sandhi phenomenon whereby word-final nasal (24a) and oral voiced stops merge to oral stops before word-initial stops (24b-c).

If the same sandhi phenomenon was a stage in Karo Batak, the word-final nasal variant could have arisen via generalization. In this scenario, *N D- and *-D D- both merge to *-D D- in sandhi. Then, nasal articulation gets analogically extended from the sandhi merger position. This hypothesis is illustrated in (25). The analogically extended form is marked with .

In sum, it is difficult to assess which of the three scenarios actually gave rise to final nasalization in KB. Nonetheless, given the evidence from related languages and the plausibility of analogical extension, it is likely that Karo Batak FN did not arise from a single sound change.

Finally, while we propose that the sequence of sound changes in (1) took place in three (or four) Austronesian languages, we note that sequence (1) is not shared by their common ancestor. We also do not have any evidence suggesting that final nasalization is an areal feature. As such, we consider final nasalization to be an independent development in each of the four Austronesian languages in question.

It has been often observed that related languages tend to undergo the same changes independently of each other (Bowern 2015). While we do not have a definitive answer as to why, we speculate that cases of similar developments across related but non-interacting languages might be due to information-theoretic (Cohen Priva 2017), phonological (Fruehwald 2013), or phonetic (Ohala & Ohala 1993) precursors already present in the mother language, which constitute preconditions favoring certain pathways of change.

5.1 Hueyapan Nahuatl

In Hueyapan Nahuatl (Southern Uto-Aztecan), the bilabial fricative β (< Classic Nahuatl w) is realized as h word-finally after (historically) short vowels, but as ŋ after (historically) long vowels (Pharao Hansen 2018). For example, β alternates with h in kioβi ‘it rains’ ∼ oːgioh ‘it rained,’ where the preceding vowel is short (p. 3), but with ŋ in tɛlɑ:βi ‘it rains heavily’ ∼ oːtɛlɑːn ‘it rained heavily,’ where the preceding vowel is long (p. 4). We propose that β developed into h and ŋ through an interstage with the voiceless bilabial fricative *ɸ. After short vowels, the voiceless fricative debuccalized to h. After long vowels, the fricative additionally nasalized to ŋ. Our proposal is supported by dialectal data. For example, not͡ɬɑːgah ‘my man/husband’ is realized by some speakers from Atlahuilco as not͡ɬagaɸ (p. 7), directly testifying to the stage of final devoicing.

5.2 Bribri

In section 3, we apply our incremental sound change explanation to word-final nasalization. However, nothing in our account precludes the same sequence of changes from taking place word-internally. Another case where nasal stops plausibly developed along a path similar to (35) includes codas in Bribri (Chibchan). In Bribri, the voiced bilabial stop /b/ is realized as a nasal [m] in the coda position, e.g. /ǰéb/ [ǰém] ‘espavel (type of tree),’ /bâbbǎ/ [bâmbǎ] ‘hot’ (Constenla 1981: 112). The voiced dental stop /d/ is realized as a nasal [n] before voiced consonants, e.g. /dîddǐ/ [dîndǐ] ‘sharp’ (p. 112–113). The voiceless velar stops /k/ is realized as a nasal [ŋ] before voiced stops, e.g. /tkʊ̂kdʊ̀/ [t͡k̯^x̯ʊ̂ŋdʊ̀] ‘to sit’ (p. 112). Language-internal and dialectal evidence both point to earlier stages of lenition, including fricativization. For example, the Bribri /d/ is realized as a voiced alveolar tap or trill [r] (possibly from an earlier *ð) before voiceless consonants and pause, e.g. /iũ̂dkẽ̀/ [iũ̂rkẽ̀] ‘it flies,’ /uʊ̂kìd/ [wʊ̂kìr] ‘head’ (p. 112–3). In many related languages, Proto-Chibchan (PCh) stops are realized as fricatives. For example, the PCh *p and *b become Térraba ɸ in word-initial position (p. 219, 221). PCh *b becomes Muisca β (p. 222). PCh *d becomes s in Boruca intervocalically before an oral vowel, and in Muisca before oral vowels in non-morpheme-final position (p. 225–6). PCh *g becomes Cabécar /h/ before i and u (presumably via an interstage with *x < *ɣ) (p. 230).

5.3 Southeastern Tepehuan

We suggest that the development of final nasalization in several Austronesian languages (§3), Dakota (§4), and plausibly Hueyapan Nahuatl as well as Bribri, proceeded through an interstage with fricativization. Nonetheless, other pathways have also been attested. In Noon (§1), for example, final nasalization developed from the deoralization of prenasalized stops (Merrill 2023; 2015). A third trajectory—different yet—is suggested by Southeastern Tepehuan (Uto-Aztecan). In Tepehuan, voiced stops (and an affricate) /b/, /d/, /dž/, and /g/ are realized as preglottalized nasals [ʔm], [ʔn], [ʔñ], and [ʔŋ] in codas (Willett 1991: 13), e.g. ˈkai.baʔ ‘it will ripen’ but kaiʔm ‘it (has) ripened’ (p. 17). We suggest that a likely course of events involved, first, the preglottalization of coda stops (*-VD > *-VʔD), and second, post-glottal nasalization (> -VʔN). Both changes are well motivated—the glottalization of coda stops is independently attested, and so is spontaneous nasalization in the vicinity of glottals (rhinoglottophilia in Matisoff 1975). Thus, while many a pathway may lead to the phonetically unmotivated word- (or syllable-)final nasalization, each of them can be explained as proceeding through a series of phonetically grounded steps.

In sum, a preliminary look at each of the three additional languages suggests that final nasalization developed via a sequence of natural sound changes. In Hueyapan Nahuatl and Bribri, word-final and coda nasalization plausibly developed via interstages with fricativization. In Southeastern Tepehuan, final nasalization likely involved a stage of stop preglottalization followed by post-glottal nasalization.