Initial lenition and strength alternations (v/b) in Neapolitan: A laryngeal Branchingness condition

In Central and Southern Italian dialects, the word-initial position plays host to a large number of strength-based alternations involving voiced stops. In this paper we look specifically at the v/b pair in Neapolitan. Strikingly, the absolute initial position appears to be weak with the voiced stop patterning phonologically like a singleton in word-medial position (v). After a variety of morphemes however, roots surface with the strong version of the consonant (b). These root-initial consonants (which would otherwise surface as weak) pattern phonologically like word-internal geminates. The strong form of the alternating consonant is a voiced stop. The initial position, although it hosts the weak allophone, is neither prosodically nor positionally weak. Voiceless stops, whose lenition actually is positionally conditioned, surface with their strong variant in this position. Rather than initial positional weakness, we identify a specific condition on the voiced stop variant (b). The voiced stop alternations that characterise initial-weakness are caused by a Branchingness (feature sharing) condition that plays a crucial role in determining strength. The alternating pairs are subject to a condition by which they must be bipositional in order to be strong. The (root) initial position is not weak, indeed only a positionally strong position could host such a contrast. Positional strength and Branchingness weakness have a quasi-morphological interplay that potentially aids the identification of morpheme boundaries. Though the pattern we describe is phonological, we speculate that it could constitute an intermediate step in the morphologisation of initial strong/weak alternations, a typological step toward developing initial mutation.


Introduction
Southern (and Central) Italian dialects such as Neapolitan are characterised by a large set of strength-based alternations in initial position. See, for instance, the Linguistic and Ethnographic Atlas of Italy and Southern Switzerland (AIS) I.67 for Standard Italian (St. Ita.) baciare/bacia. Southern Italian type: [v]asare baciare 'to kiss '. 1 (1) Strong weak alternations in Southern Italian dialects In order to restrict the subject matter, this paper will focus exclusively on the alternations of the b ~ v in Neapolitan. This is because this particular alternation has a special history (known as betacism), which will be of interest to diachronic phonologists. 2 The situation described for b ~ v will be similar for other word-initial strong/weak alternations both in Neapolitan and in other dialects, but the precise details of each alternation in each dialect deserves its own study. 3 In fact, it is especially theoretically interesting that not all the voiced stops have the same lenition pattern (despite them all having strong/weak pairs in these dialects). This fact in itself is worthy of future study.
Even though the discussion is limited to the v/b alternations, the implications for the aid to morphological parsing (and the development of word-initial strength-based alternations) and the implications to the theory of phonological lenition are all sufficiently embodied by discussing just this alternation in this dialect.
As shown in (1), voiced stops alternate between strong or weak forms, depending on their morpho-phonological surroundings; these are presented in (2). Although there are some synchronic lexical exceptions (2b), a large bulk of the lexicon makes up this alternation. 4 Even if synchronic productivity has been sometimes disputed (Andalò & Bafile 1991;Bafile 2014), we believe that this process is still synchronically active (see Fanciullo 1997;2018;Russo & Aprile 2001). However, this debate is not crucial for our point, since the process was synchronically active at some point, so the arguments we are making here are valid independent of that debate. It is also not within the scope of this paper to discuss the impact that loanwords, surface exceptions and inter-speaker variation has on the productivity of a phonological process. Therefore, our diagnosis of the pattern, and all of its phonological implications, diachronically and synchronically, remain unchanged. 5 (2) Morpho-Phonological context (Neapolitan) 5 a. Strong~weak contexts (i) Absolute initial (weak) [ˈvarkə] 'boat' SG-boat (ii) After simple proclitics (weak) [na-ˈvarkə] 'a boat' INDEF.SG-boat (iii) After gemination-inducing proclitics (strong) [tre-bˈbarkə] 'three boats' three-boat.PL (iv) After nasal-final proclitics (strong) [m-ˈbarkə] 'on the boat' Loc.-boat.SG (v) After voiced fricative-final proclitics (strong) [z-ˈbɔta] 'to empty out/Ita. svuotare' 6 rEV.-empty b. Lexical exceptions (in absolute initial position) 7 (i) [ˈbɛlːə] 'beautiful' ADJ.beautful M/F-SG/PL (ii) [ˈb w oːnǝ] 'good' 8 ADJ.good M-SG/PL The same alternation applies to clusters (e.g. [br]) in initial complex onsets, these surface with the weak form in absolute initial context. 678 Voiced stops also surface in the weak form when they occur as singletons intervocalically (3a). However, voiced stops always surface in their strong form when geminated (3b). Voiced stops also appear weak after the non-nasal sonorant coda /r/.
Though all the voiced stops exhibit parallel alternations, we will limit the data/examples to the alternation of the voiced stop [b] and its weak variant [v]. 'beard' N-SG Central and Southern Italian dialects are characterised by an external sandhi rule triggering consonant doubling referred to as Raddoppiamento Fonosintattico or Raddoppiamento Sintattico (RS). There is a rich literature on RS, amongst others we can cite (Chierchia 1986;Serianni 1989Serianni /2006Fanciullo 1983Fanciullo -19862018;Andalò 1991;Russo 2013a;b) the bibliography within the latter and (Passino 2013;D'Alessandro & Scheer 2015).
The process of RS greatly increases the morpheme-specific contexts that show the rootinitial weak/strong alternation. In (4) we present a representative sample of synchronically active morphological contexts that show the strong/weak alternation.

β/v-b and the diachronic sources of the alternation
The alternation between strong and weak voiced labial obstruents: b and v has a special diachronic origin in Southern Italian. We summarise their development/phonologisation here. 9101112 The Southern Italian weakening of voiced stops seems to go back to the loss of the Latin opposition between /b/ and /w/ (<b> and <v> -/w/ is graphically <v>). This is a historical phase with reflexes across Romance languages including Catalan, Spanish, Occitan. In Occitan today the betacizing area (where <b> replaces <v>) covers Gascony, Languedoc and a large area of Auvergne (see Sauzet & Olivieri 2016: 325).
The Appendix Probi, a manual on the correct pronunciation of certain words (see Asperti & Passalacqua 2014), shows the Latin confusion between /b/ and /w/ in the third century. The Latin orthographic <v> is often replaced by orthographic <b> already in the first century (Adams 2013: 183): Iobe, dibi (Archive of the Sulpicii from Pompeii) and in the Pompeian graffiti: berus (verus). Claudius Terentianus (first century) also presents fourteen cases of b in place of v (Adams 1977: 31;Russo 2013a).
As the Appendix Probi shows, orthographic <b> and <v> were confused for each other in initial, post-consonantal and intervocalic position. This signals a complete phonological merger. This indicates that the pronunciation of Latin <v> and <b> at this stage was fully merged, and probably [β] (see discussion in Adams 2013 Väänänen § 89;Lausberg § 300;366;373;Fanciullo (1997: 38). 14 The number is the order the item appears in the list. After this period, during the 2 nd century (Lausberg § 300), the bilabial fricative [β] became a labiodental /v/ in some of the Romance languages. In Italian (and elsewhere, such as in French) /b/ was lexically established in initial and post-coda position: bacio 'kiss' or erba 'grass' vs. cavallo 'horse' Lat. CABALLU.
As will become theoretically significant later, Tuscan and Neapolitan have non-overlapping patterns of /v/, /b/ in post-coda position. Tuscan typically presents strong allophones in this position, while in post-rhotic-coda position Neapolitan presents the weak allophone (v). Northern and Central Italian dialects innovate the strong form of the voiced stop: /rb, lb/: Ita. car[b]one 'charcoal' or er[b]a 'grass'. However, Neapolitan and Southern Italian dialects have generally opted for a weak consonant in this position: Neap. va[rv]a 'beard', ca [rv]one 'charcoal'. 17 This difference is consistent with our account (even predicted), as we will explain in section 3.
Spanish, Catalan, Occitan (including Gascon), Southern Italian dialects and Sardinian neutralized the contrast between Latin <v> and <b> word-initially. Thus, from the starting point of the confusion between Latin <b> and <v>, the realization of the Latin <v> and <b> has been established in various places by a phonotactically controlled allophony. In Spanish, Catalan, Gascon, and parts of Occitan (Languedoc and Auvergne) and Sardinian, the fricative [-β-] is the reflex of Latin <-v-> and <-b-> in three contexts: intervocalically, word-internally and at morpheme boundaries (Spa. This process and result have been called betacism (Sabatini (1966;1996: 216 and no. 66;Loporcaro 2004).
In Central and Southern Italian dialects, the initial position hosts either [v] (but also [β] see below) or [bː] depending on the phonological environment. This process is already attested in Latin inscriptions, as we have seen above, and see for instance the following early example of strengthening via RS (Sabatini 1966: 214-216;1996: 23-25, no In this intriguing inscription, the author of the graffiti wanted to transcribe the vulgar expression 'loudly' (lit. 'at high voice' with AD as an instrumental). At a second point in time, the author added a second <b> to the inscription in the form of a small <b>: ab b oce. This eliminates the ambiguity in interpretation left by using a single <b>, which could have stood for the bilabial fricative interpretation of the grapheme (Sabatini 1966: 217;1996: 195). This example shows that the vernacular of Rome was already fully committed to this process by this time period. So, Sabatini (1966;1987: 25) suggests that, in the betacistic Central and Southern Italian area, the graphical opposition between <b> and <bb> was a way to make a phonological distinction between weak intervocalic [β] = <b> and strong [bː] = <bb>. 1920 (8) Doubling and RS triggered use of strong form of [bː] a. Proper name frequent in late inscriptions: Calama (Numidia) (iii) abbocatore Campania (10-11 th c.) 20 Meanwhile, the Cassino area of the middle ages shows a phonotactic alternation closer to Spanish, Catalan and Gascon (Western Romance). It concerns at least the quality of the two variants [b] and [β] outside of RS and absolute initial position. The medieval text Ritmo cassinese (13 th century) (Formentin 2007: 74) contains what has been called betacismo cassinese. 21 (9) Cassinese betacism (13 th century) graphic <b> replaces <v> in all the positions a. [β] According to Baldelli (1958: 35) graphical <b> corresponds to [β] (and it is maintained 22 in the area of Montecassino to this day). 23 As suggested by Formentin (2007: 74), the weak [β] is found in initial position (## -Boltiera and # -questa bita) and V -V (fabello), whereas strong [bː] is located in the derived environment of RS, after a lexically-specified trigger: ( ∅c ). 24 Having explained the diachronic circumstances from which Tuscan Italian/Standard Italian and Neapolitan were forming, we now move to a formal phonological explanation of the strong/weak alternations. Specifically, it is not the case that Tuscan/Standard Italian has a positionally strong initial position, while Neapolitan does not. Though this may appear to be the case, because in Neapolitan the initial position was phonologised with the weak variant of the b ~ v pair. However, phonologically there are different types of lenition and, despite all appearances, the so called initial weakening of Neapolitan is not a product positional weakness.

Types of lenition
Considering Neapolitan/Central and Southern Italian dialects from a typological and theoretical perspective, the pattern of strength and weakness seems to fall across an unexpected boundary.
The root-initial position appears to be weak, patterning with intervocalic singleton consonants and some post-coda positions. However, there is a typological tendency for a consonant in Initial and post-coda position to be strong relative to the intervocalic singleton position. However, there are various sources of phonological strength. In representationally-minded models of phonology there are in fact three different kinds of 'strength'. (10) Factors in lenition (a) (Pure) Positional (Charette 1990;Harris 1990;Kaye et al. 1990;Ségéral & Scheer 2001) (b) Prosodic (Harris 1997) (c) Feature sharing/Bipositionality (Scheer 2000;Honeybone 2001;Bucci 2013) Theoretical models of positional strength and weakness have long taken the initial position to be inherently strong. This finding has been confirmed experimentally (Becker et al. 2012). Also, there are many phenomena that speak to this conclusion. In English for instance, even though aspiration looks like it ought to be foot-initial (cf. Anderson & Ewen 1987), non-pretonic initial consonants are also aspirated: Harris 1994).
Prima facie, Neapolitan voiced-stop lenition of /b/ looks like it hinges on positional factors. This, however, will be shown to be false, even though positional terms have so far been crucial to describing the pattern (absolute initial position, intervocalic etc…). Moreover, the pattern of lenition is completely at odds with prosodically-driven lenition.

Ruling out prosodic lenition
Licensing Inheritance (Harris 1997) is a framework of strength and weakness that focuses on prosodic factors. In this model, initial positions are not inherently strong, but foot-initial positions certainly should be. The diagram in (11) shows how phonological strength, the ability to license features is proportional to the proximity to the head of the domain (the stressed nucleus). (11) Strength and weakness according to Licensing Inheritance However, this model cannot account for this pattern of weakening in Neapolitan because the weak alternant is found in foot-initial position.

Ruling out positional lenition
Taking pure positional factors into account, typically strength and weakness are distributed as shown in (13). Initiality, either of the word or syllable, is taken as being inherently strong. Word-initial onsets, particularly those in absolute-initial position, will be taken to be strong. Meanwhile, the intervocalic onset context is assumed to be weaker, and the disjunctive context of being pre-consonantal and word-final { -.C, -#} is taken to be weakest of all. 25 (13) Strong and weak positions (a) Strong positions (onsets)

.1 Introducing strength and weakness in Strict CV
Strict CV phonology, sometimes called CVCV or Lateral phonology (Lowenstamm 1996;Ségéral & Scheer 2001;Scheer 2004;2012), is an offshoot of the framework of Government Phonology (Kaye et al. 1985;Charette 1990;Harris 1990;1994). Strict CV built these strength and weakness relations into its theory. The following paragraphs serve as an introduction to the principles of representation of the framework, with special emphasis on how it analyses strength and weakness. In Strict CV, positional strength derives from its syllabic environment. Specifically, it obtains from a calculation of the contradictory forces of strength (licensing) and weakness (government) (Ségéral & Scheer 2001). Fittingly, positions receiving the strong force will be strong and positions receiving the weak force will be weak. 26 These conditions give rise to the following constellation of strength and weakness contexts, the outcomes are listed in (14), while the structures that give rise to the outcomes are shown in (15). 27   (14) Summary of phonological strength in Coda Mirror a. C is weak when it is: (2010), it is further understood that a position which could in principle be both licensed and governed is only governed (only weak), (but see Balogné Bérces & Honeybone 2012). 27 There is a slight complication because word-final codas can be stronger than word-internal codas, but this has no bearing on Neapolitan.

b. Strength in Coda-Mirror
In this model, the post-coda position can never be weak because Gov will always be needed to silence the preceding empty V. However, the presence of the initial empty CV (15b i) is parametric, which means that the initial position can be either strong or weak (from a positional perspective). However, there is an important typological implicational hierarchy where the post-coda position will either be equal in strength to the initial position or it may be stronger than the initial position (when the initial position is not preceded by an empty CV). Wherever there is an empty CV, the initial position and post-coda position must be of equal strength (from a positional perspective). As we will show in the next section, though Neapolitan does have an empty initial CV, when it comes to the b ~ v alternation we still see (a) the weak alternant in initial position (when this should be strong) and (b) either weak or strong alternants in post-coda position depending on the nature of the coda consonant. From this it will be concluded that positional factors are not sufficient to handle the Neapolitan b ~ v alternations.

Positional lenition and the Initial CV are not enough
The structure in (15b i) shows the initial position. It is strong because it is preceded by an Empty CV. This Empty CV is inserted into the representation at the beginning of each phase/string (Scheer 2012;cf. Lowenstamm 1999 for a precursor). The Initial Empty CV (henceforth the Initial CV) absorbs the destructive force of Government, leaving the initial consonant in a strong position (14b i). However, the possibility of having an Initial CV to mark the beginning of a domain is parametric. Therefore, it is conceivable to have either strong or weak initial positions (the weak is shown in (15b ii)). In Tuscan, the presence of the Initial CV is confirmed by other properties of the language (16a). Spirantisation and de-affrication do not occur in utterance initial position, though they do follow vowel-final proclitics (Agostiniani & Giannelli 1983;Marotta 2008;Ulfsbjorninn 2017). Elsewhere both Spirantisation and De-affrication are blocked in post-coda position (after an empty V slot) creating the natural class between initial and post-coda position expected according to the Coda-Mirror (Ségéral & Scheer 2001).
However, the fact that Neapolitan had initial weakening does not automatically imply that it does not have an initial CV and that (16b) is the right representation. After all, the external sandhi effects are very similar in the two varieties, so it would be hasty to merely assume (16b).
(16) a. Tusc: CV + /dito/ Gov In fact, there is a phonological process that strongly suggests that (16b) is not the correct representation for Neapolitan. This comes from the pattern of the voicing of voiceless stops (Andalò & Bafile 1991: 250).  Gov Gov

= [kɔrpə] 'body'
Russo and Ulfsbjorninn: Initial lenition and strength alternations (v/b) in Neapolitan Art. 11, page 12 of 27 b. Post-vocalic and intervocalic positionally weak [+gov] Gov The fact that Neapolitan can be demonstrated to have an initial CV shows that whatever caused initial weakening in the language, it was not positional lenition. The b ~ v distribution does not follow the distribution of Gov, unlike voiceless stop voicing which clearly does (the strong alternant appears in initial and post-coda positions). This is confirmed by the fact that the voiced stop lenition (b ~ v) in Neapolitan does not pattern either like voiceless stop lenition in Neapolitan nor Spirantisation in Tuscan. The latter two processes are clearly tied to the distribution of Gov. The b ~ v alternation that we see in initial position is recapitulated in both intervocalic position and post-rhoticcoda position. 28 The post-rhotic-coda position of Neapolitan is a context of voiced stop weakening -akin to the initial position *β: *βarβa > Tusc. barba and Neap. *βarβa > varva 'beard'. 29 This shows that the b ~ v alternation cannot be positional. In this context, Tuscan and Neapolitan (and voiceless stop voicing in Neapolitan) ought to have identical licensing conditions but they lead to different strength outcomes. (19) Tuscan and Neapolitan, difference in post-coda stops a. Tusc. barba 'beard' (Initial CV omitted)

Gov
This diagnoses the Neapolitan b ~ v weakening alternations as being the third type of lenition, not prosodic, not positional, but Bipositional lenition (cf. 10c). 30 This conclusion is shown schematically in the diagram beneath. It shows the overlapping spaces in the formal characterisation of the diachronic path from Latin's single phoneme /β/ to Tuscan and Neapolitan /b vs. v/. In (20)

Strong positions Weak positions α
The whole diagram is the distribution of the original Latin /β/, as a single phoneme. The environments it occurred in are listed with an example word. Line (α) the diagram separates the space according to what is understood to be positional strength. To the left of line (α) are the positionally strong contexts, and to the right of line (α) are the positionally weak ones.
As the diagram shows, Tuscan generally conforms to this positional split. Though some /rb/ clusters were lexicalised as /rv/: corvo, nervo, servo… This was not categorical or uniform in Tuscan and, as we will see, synchronically the post-rhotic-coda environment is categorically strong yielding synchronic [rb] sequences and not the weak alternants: *[rβ]. 31 For Tuscan, the environments to the left of line (α) (labelled as area (ζ)) often resulted in the strong form: /b/ and correspond synchronically with a positionally strong position.
Meanwhile, the Neapolitan split does not conform to notions of positional-markedness. Its /b/ (demarcated by area ψ) does not align with positional-lenition, therefore, a different kind of factor must be responsible for the pattern.

Morphological consequences of the strength alternations
Before we turn to formally explaining the phonological factors involved in this bipositional lenition, we wish to briefly illustrate an interesting implication that this pattern has.
Specifically, there is an interplay between bipositional lenition and classical positional strength that acts as a reinforcement to the morphological parsing, in what looks very much like a beginning step in the morphologisation of the phonological pattern into one of initial strength mutation.
Though it may at first appear paradoxical, the abundance of utterance initial weakinitial forms in a positionally strong position could actually be a manifestation of the inherent strength of that initial position because it is only a strong position that can host a weak/strong contrast.

Indirect signaling of morpheme boundaries
Despite what appears to be a general ban on initial voiced stops, these sounds are, in a sense, still allowed in initial position (in the first phonological onset of the word), but only under certain morphological contexts. Consider this from the perspective of contrast. Once the full set of morpho-syntactic contexts has been listed, the first onset of roots can hold any consonant phoneme of the language, including voiced stops. Therefore, all consonant contrasts are permitted in the initial onset position once the full set of word-initial environments have been considered. This is sharply different from an actually positionally weak syllabic position, such as the word-internal coda. Typically, an internal coda hosts a dramatically reduced set of consonant contrasts, and this remains true when all the morphological contexts are considered. That is because this is a truly positionally weak position and it leads to extensive neutralisation (via devoicing, sonorisation, debuccalisation, limit to coronality etc…).
In Neapolitan, the conditions by which voiced stops are allowed in the first onset of a root are as follows: (a) they form a complex with a voiced fricative or a nasal consonant, (b) they form a geminate. These conditions are invariably created by preceding morphemes. Therefore, the strong variant of a voiced stop in root-initial position reinforces the presence of a morpheme boundary to its left. This creates alternations where a morpheme in isolation surfaces with the weak form of a voiced stop and the multi-morphemic form shows up with the strong form. The fact that C 1 is a strong position allows it to be a good position for expressing patterns of morphology that are signalled exclusively by the amount of phonological material in C 1 . The argument is as follows: if a morphological paradigm is defined by melodic complexity in the form of a large amount of phonological material, this necessarily requires that elsewhere, related stems do not contain those same feature combinations. Therefore, in such systems, C 1 will be simple in paradigm alpha (i.e. basic, unsuffixed word-initial) and complex in beta (i.e negation). Importantly then, even a simplex alternanting root such as [r]ito 'finger' (weak), systemically the C 1 position is exercising its positional strength in that: C 1 is simplex in a quasi-paradigm that helps signal the morpheme boundary by C 2 . 3233 (21) Initial strength alternations in Neapolitan (repeated from 3) a. The contrast between Count and non-feminine Mass nouns (21h) shows some of the strongest evidence for root-initial consonants alternating according to a strong and weak paradigm. These Count vs. non-feminine Mass nouns (and certain clitics and determiners) are only distinguished by the root-initial consonant (C 1 ). C 1 must be geminated, leading to a contrast for the v/b alternating pairs based on the weak (v) and the strong (bː).
The details of the syntactic structure are not crucial for the paper, we suppose the number and gender features are introduced by the appropriate syntactic heads, however these are not represented in (22). What is crucial is that when Det is valued for Count the exponent is /o-/ which does not trigger RS, while if it is valued for Mass [-fem] then it will be exponed by /oCV-/. … √ 32 Medieval Neapolitan (but also modern), text De Rosa, Formentin (1998: 190, no. 526 The Locative is another domain where the phonology appears to contain a nearly fully morphologised weak/strong alternation. The exponent of the Locative is presumably just a piece of syllable structure linked to a nasal feature. This allows the root-initial consonant to surface in its strong form. 34 (24) Derivation of Locative + nouns a. UR /CVnas/ + /√bokː + a/ To a lesser degree, this is also the case with the (quasi)prefix S-: [z-ˈbiɲːə] 'escape'. The S-forms a semi-productive quasi-paradigm, the semi-fossilised inheritance from Latin EX-'out of'. This alternation can only be maintained because C 1 is a phonologically strong position and it is able to license the contrast and its strength reinforces the parsing of a morpheme boundary to its left. Although, Neapolitan has not made the final step toward a true morphological pattern based on the strength of C 1 , and maybe never will be given the influence from Italian (Andalò & Bafile 1991), there are some striking parallels here with languages that did.

Typological connections
Typologically, this initial strength-based alternation is strikingly similar to a family of fully morphologised initial-strength alternations called consonant mutations.
These are found in many unrelated language families and are not instances of positional lenition. Instead, they constitute a paradigmatic effect based on whether C 1 is simple or complex (for a certain feature): Bantu (Kula 2002), Nivkh (Shiraishi 2006), West Atlantic (McLaughlin 2000 and Celtic (Breit 2015).
The following data show how paradigms can be defined in this way. Roots undergo weak-initial vs. strong-initial alternations depending on their morphemic context. These patterns occur in a wide range of language families from all over the globe, including isolates such as Nivkh spoken in North-Eastern Asia, or Niger-Congo languages from Africa, especially West-Atlantic such as Seereer-Siin, and Celtic languages such as Manx from North-Western Europe.
Clearly this type of morphology recurs in human languages (it is not a product only of relatedness or areal effects) and our discussion of the Neapolitan data shows a case where historical and phonological processes could have led (or perhaps are still leading) to this pattern. We hope this could trigger interest in the formal diachronic stages of consonant mutation in language families where it is less well understood. (25) Strong weak Initial C 1 alternations in other languages -Bantu (Kula 2002)  In all these cases, we see that the inherent strength of the initial position is recruited as the property that expones or reinforces a certain morphological pattern. The fact that C 1 is recruited for this purpose is probably not coincidental, as only (a) a positionally strong position or (b) a position that is optionally/parametrically strong ~ weak could bear this alternation. The difference between Neapolitan and the consonant mutation systems listed above is that the consonant mutation patterns are always phonologically irregular. In Neapolitan that is not the case. The alternations we see word-initially are basically the same phonological conditions that hold word-medially. However, this does not change the fact that the phonological system helps identify morpheme boundaries (on the left edge -and only the left edge, because it is a strong position). The less regular the Neapolitan phonological pattern becomes, the more likely it is that the morphology will exploit the huge set of existing root-initial alternations.

Strength and bipositionality
Returning to the phonology of Neapolitan and Central/Southern Italian, we will now show what kind of specific phonological process is responsible for the pattern.
As we concluded in section (3), the clue to understanding the phonological cause of the pattern is the post-rhotic coda context: Tusc. barba vs. Neap. varva 'beard'. This datum is not caused by positional nor prosodic factors, it is caused by the third type of phonological strength and weakness: feature sharing.
Feature sharing, or branchingness/bipositionality, is an effect whereby the features/segments that are bipositional (shared across syllable structure) have a far greater chance of remaining in the phonological forms over time and being phonologically 'strong' (Scheer 2000(Scheer , 2004Honeybone 2001;Bucci 2013;Ulfsbjorninn & Lahrouchi 2016). Synchronically, these bipositional features/segments are also more resistant to phonological tampering/manipulation.
There is both a substantive and formal component to the Honeybonian principle: sharing makes us stronger. Substantively, features in multiple positions are more likely to receive strong cues in the phonetic form. As such, they are more likely to be perceived by listeners, therefore lexicalised and preserved over time. Formally, interrupting segmental material that is shared across two positions requires intervening between association lines. Disrupting association lines brings its own formal difficulties and so it is also avoided (the no crossing condition/geminate inalterability (Goldsmith 1976;Hayes 1986;Hammond 1988;Blevins 2004;Russo & Ulfsbjorninn 2017).

Bipositionality of voiced stops in Neapolitan
The strong/weak alternations in the Neapolitan data can be understood to be caused by a specific phonotactic on voiced stops.
These sounds are subject to a phonotactic called a Melody-to-Structure Licensing Constraint (MSLC). These are a family of licensing constraints (cf. Charette & Göksel 1998) that specifically regulate the associations between melodic features and certain types of syllable structure (Ulfsbjorninn & Lahrouchi 2016;Lahrouchi in prep.).
According to the relevant MSLC of Neapolitan, in order to be phonetically interpreted, voiced stops (or at least /b/) must be bipositional. This means that the features of a voiced stop are obliged to attach to two pieces of skeletal structure.

(28)
Branchingness condition (to be elaborated) Voiced stops must branch across two C positions The Branchingness condition in (28) means that, with the exception of some recent loans, underlying singleton voiced stops cannot be phonetically interpreted. Geminate voiced stops on the other hand are permitted because they are bipositional (connected to two pieces of syllable structure) (cf. stop lenition in Ulfsbjorninn & Lahrouchi (2016) The data in (29) shows that full gemination is not necessary to license the voiced stop. Instead, even some kinds of partial gemination are sufficient. The features that are required for this partial geminate hypothesis are surprisingly simply to state in Element Theory.
(30)  Halle & Stevens 1971). Therefore, a rhotic such /r/ does not share a phonological voicing feature with voiced stops. 35 The implication of this for Neapolitan is that nasals and voiced fricatives will be able to share a phonological feature with voiced stops (|L|), and as such they can come together to form a partial geminate. The rhotic, however, does not share a voicing feature with voiced stops and, therefore, it cannot form a partial geminate with this class of sounds. This leaves voiced stops completely monopositional after rhotic-codas (just as they are as singletons in an intervocalic context).
It seems, therefore, that the Branchingness condition in Neapolitan is specified for the branching of a specific feature: |L| (voicing in obstruents, nasality in sonorants).

(31)
Branchingness condition (revised) The |L| of a voiced stop (|ʔ, H, L|) must branch across two C positions In the absolute-initial position (## -) of a monomorphemic form, a Neapolitan voiced stop-onset has no choice but to be monopositional. Therefore, utterance-initial voiced stops will always surface in their weak version (32b). Conversely, the condition in (31) is straightforwardly satisfied by the structure in (32a) where the whole segment branches and forms a voiced-stop geminate. This is why any RS trigger that induces gemination will allow the strong from of the root-initial consonant to surface.
Neapolitan does have an initial-CV in utterance initial position, however, just as in Italian this does not lead to spreading. Therefore, while the initial position is positionally strong, it does not allow the Branchingness that is required for the survival of the voiced stop.
However, Branchingness does not require full gemination to interpret the voiced stop. A nasal or voiced fricative occurring to the left of a voiced stop will allow the branching of the feature |L|, thereby the whole structure will be licensed. These will be supplied by various morphemes. As also stated in the table (30), in ET sonorants are taken to be devoid of laryngeal features. Their voicing is automatic, non-contrastive and therefore not featurally represented (cf. passive voicing Halle & Stevens 1971;Dresher 2009). Therefore, even word-internally, Neapolitan /r/ will have no |L| to share with a voiced stop that follows it. So items like /barba/ 'beard', shown in (36) result in an ill-formed sequence from the perspective of the Branchingness condition. Because the complex (ʔ, H, L) does not share its |L| feature. Consequently, the /b/ is doomed to be monopositional and therefore surfaces as its weak equivalent. 3738 (36) Equivalence of utterance-initial 38 and post-rhotic coda No spreading of |L|, so Branchingness condition is not met [v]ar[v]a *barba 'beard' 37 These structures show the phonological result after any OCP of identical underlying features has applied. There is a further phonetic condition that makes nasal + voiced stop into a geminate nasal. In this structure, the /b/ was not reduced to v: *[mˈvarkǝ]. 38 Neapolitan has an initial CV that makes the initial position positionally strong. However, like in Standard Italian and Tuscan, consonants do not spread into the utterance initial CV. Therefore, even if there is an initial CV it will not trigger RS, unlike an empty CV in other contexts.

Conclusion
The initial position of Neapolitan hosts a large number of strength-based alternations involving voiced stops. In this paper, we focused on the b ~ v alternation as a strong ~ weak pair. In what is traditionally described as initial weakening, we show that it is the weak alternant that appears in absolute initial position. In fact, the initial position patterns like a singleton in word-medial position and one of the post-coda positions (after a rhotic). However, the initial position of Neapolitan is not positionally weak, as can be demonstrated by the distribution of underlyingly voiceless stops which are strong in initial and all post-coda positions. Instead, the voiced stop alternations that characterise initial-weakness are diagnosed as a case of a Branchingness (feature sharing) condition, that plays a crucial role in determining the distribution of underlyingly voiced stops in Neapolitan (Honeybone 2002;. The Branchingness condition is trivially met in cases of gemination (supplied by processes such as Raddoppiamento Sintattico), however, when analysed in Element Theory, we see that voiced stops are preserved in Neapolitan not only when they geminate, but also when they share their voicing feature with a preceding consonant: [ m ˈbɔːlə] 'in flight' and [zˈbiɲːə] 'escape' vs.
As a consequence of the data, the Branchingness condition that enforces bipositionality of a certain feature in branching onsets, potentially plays a role in the identification of morpheme boundaries. As the strong version of the alternant in root-initial position will always coincide (and therefore potentially reinforce the parsing of) a morpheme boundary to its left. This is interesting because it is only a positionally strong position that can host such a contrast. Though these alternations are still phonological in Neapolitan, the less phonologically regular the more likely it is that the pattern could become morphologised. We speculated that this stage of Neapolitan could be an intermediate step in the morphologisation of initial strong/weak alternations (cf. initial mutation).