Cross-morphemic palatalisation in Getxo Basque: empty positions, bipositionality and place licensing

2.1 Architecture of the process

Getxo is of particular theoretical interest because HB’s analysis of the facts creates something of a paradox. HB note that Getxo palatalisation both seems to apply and not apply at the phrasal level; this is because the palatalisation rule is sensitive to both the leftward phonological context within the word-level and the rightward phonological context of the phrase level. As shown in (5), the trigger and the target of palatalisation need to belong to the same word /in##V/ and /il##V/. If the trigger and target belong to different words (5b) the process does not occur.

The data in (5) seem to suggest that the process does not occur above the word level; however, information about the next word is still needed to determine the outcome: it is palatal when followed by vowel-initial words (as in (3) and (5a)), but a homorganic nasal (6a) or a coronal lateral (homorganic or default) (6b) when followed by consonant-initial ones. So the rule does not appear to be wholly lexical.

The solution HB propose is effective, though it involves a complex derivation. They assume that [ɲ, ʎ] are synchronically allophonic and always derived from UR /in, il/ (cf. non-alternating [ɲ, ʎ] such as /troinu/ [troɲu] ‘knot’ and /akuilu/ [akuʎu] ‘stable’). Then they claim that the palatalisation rule is actually lexical. This accounts for the non-application of the rule when the trigger and target do not come from the same word (5b). Naturally however, this wrongly predicts that all lexical /in/, /jn/ or /il/, /jl/ should palatalise, although there are many lexical exceptions, both old and young loans and even the occasional native item (see back in (4)) (e.g., [arrujna] *[arrujɲa] ‘ruin’, [isilik] *[isiʎik] ‘quiet’).

Moreover, since HB’s rule palatalises all /n/ and /l/ after /i, j/ inside words, it also wrongly predicts the underapplication of palatalisation across words before C-initial words /in##C/ and /il##C/ (shown in (6b)) (e.g., /mutil + bet/ [mutil βet]*[mutiʎ βet]) ‘a boy’).

In order to account for the many lexical exceptions HB simply state that these are lexical exceptions. This is a perfectly common approach, but naturally it is an analysis which is subject to improvement if the exceptions can be explained in some way.

Meanwhile, to explain the underapplication of palatalisation before C-initial words, HB are forced to propose a Duke-of-York derivation. As illustrated in (7), the lexical rule applies palatalisation to these forms only to be corrected by a post-lexical nasal place assimilation and neutralisation rule.

2.2 Areas to improve on previous analysis

The HB account, we claim, misses two observations. Firstly, if analysed correctly, the palatalisation rule is actually exceptionless. There are in fact no exceptions in any loanwords young or old when it comes to word-final /in##/ and /il##/, across Basque morpheme boundaries. All the so called “exceptions” are word-internal. This fact is not captured if they are simply lexical exceptions.

Secondly, and relatedly, the palatalisation rule can in fact be linked to the positional distribution of place features of nasals and laterals, as well as the process of nasal place assimilation across word boundaries.

An alternative account is possible based on the positional underspecification of nasals and laterals, and the spreading of place into derived onsets, those created by bipositional spreading of word-final codas into rightward vowel-initial morphemes (a.k.a. resyllabification). This analysis makes the place-feature distributions that result from the palatalisation alternations and relates them to Getxo Basque’s general rules of nasal place assimilation. The present paper argues that the advantage of such an analysis is that it can fully unify the so called “lexical exceptions”, as well as avoid HB’s Duke-of-York derivation and their necessary ordering stipulations between palatalisation and nasal place assimilation. Moreover, on this account, this process of Getxo palatalisation will no longer require a division between word- and phrase-level phonology; instead, it exploits other general phonological facts about the variety.

In the reanalysis we will show that palatalisation arises in a number of different contexts, but always results in the same bipositional palatal structure. Since Basque does not express phonological quantity as phonetic length, this bipositionality does not result in a phonetic geminate; instead, it is what has been referred to as a “virtual geminate” (see, e.g., Lowenstamm 1996; Ségéral and Scheer 2011). This structure has parallels in the palatalisation of other languages, and we will argue that once the full motivations are elucidated, the process is in fact exceptionless in Getxo Basque.

3.1 Introduction to the framework

Since this analysis is couched in Strict CV (a strand within Government Phonology, GP – Charette 1991; KLV 1985, 1990) and Element Theory, we will present the tools of these frameworks for the unfamiliar reader.

3.1.2 Strict CV

Strict CV (Lowenstamm 1996; Scheer 2004; etc.) is an autosegmental framework where representations are composed of two main tiers, a melodic tier that holds segments/feature bundles and a skeletal tier. The skeletal tier is made up of strictly alternating C and V slots. The two tiers are connected by association lines.

Free combination of these two independent tiers yields the combinations in (10). These are the main shapes that exponents can take – according to the theory.

(10)

Shapes of exponents
a.	Fixed		b.	Floating melody	c.	Empty skeleton		d.	Unfixed
	C	V				C	V		C	V
	|	|
	α	β		α					α	Β

Some of these possibilities might seem to be relatively abstract, however, these configurations follow from general autosegmental principles; namely, the independence of the tiers. If one wants to exclude any of these options it must be marked explicitly in UG. However, as Zimmermann (2017) observes, there is no reason to a priori exclude these configurations if they are useful in linguistic analyses. Very much in line with this, GP/CVCV as representational/autosegmental frameworks have utilised these devices ever since their inception, providing evidence of their theoretical inevitability: floating final consonants in French (e.g., Charette 1991) and English (e.g., Harris 1994); the floating melody of L1 suffixes (Balogné Bérces 1997; Newell 2021); an empty CV unit representing the word-boundary (Lowenstamm 1999) or stress (Larsen 1995); unfixed vowels (i.e., lexically specified vowels that are potential deletion targets – see (12) below) (Scheer 1998, 2004: 87–93 for Slavic; Polgárdi 2000 for Hungarian).

Our analysis will crucially rely on the configurations in (10b), (10c) and (10d): floating melody/floating segments (10b) and empty categories (10c), as well as unfixed/unlinked (10d), the combination of these two in which floating elements come with empty skeletal slots to which they are not associated. If a feature is not associated to the skeleton, it will not receive phonetic interpretation and be effectively “deleted”.

The strict contiguity of C and V slots means that vowel-initial and consonant-final words, long vowels, geminate consonants and consonant clusters all involve some kind of empty skeletal slot (see (11) below, with hypothetical examples). The interpretation of empty positions is constrained by universal principles and some parametric variation. Empty V-slots at the end of domains (FEN – “final empty nuclei”) and within domains are subject to different parameter settings since they can behave differently across languages (Charette 1990). Crucially to our analysis, certain languages, with Basque among them, license FEN by system-specific parametric choice, as a result of which they have surface consonant-final words (similar to the hypothetical form in (11a)).

Several types of clusters are distinguished in GP – the one that is relevant to the discussion below is the traditional “heterosyllabic”, “coda-onset” cluster, which in GP terms is the manifestation of right-to-left government between C1, the dependent/governee, and C2, the head of cluster/the governor. In (11c) above, such a cluster is illustrated, with the government relation between the consonants indicated by ←. Please note that for ease of reference, we continue to use some of the notions of traditional syllabic theory (in quote marks), e.g., “coda”, but those are only meant as descriptive, shorthand labels referring to the Strict CV equivalents (e.g., the “coda” is an (unlicensed) consonant followed by an empty V-slot – cf. /n/ in (11a) and /k/ in (11c)).

The structure in (12) illustrates the basic mechanism silencing empty V positions: V-to-V government (also called Proper Government – PG). In the most wide-spread form of GP it proceeds right-to-left, emanating from an unsilenced/filled V and having the result of satisfying the phonological Empty Category Principle (ECP) for the V-slot it hits.

Government is in Strict CV (to be precise, in a subtheory called Coda Mirror – Ségéral and Scheer 1999, 2001, etc.) conceived as a force that inhibits segmental expression of its target: in the case of nuclear positions, this amounts to granting it non-expression in phonetic realisation. In the Hungarian example in (12), V2 is an unfixed vowel: its empty V-slot can only remain empty (leaving the floating melody unlinked and hence phonetically unrealised) if the following V3 is filled and able to govern it (12b). Elsewhere, as in (12a), it is not silenced by government and will therefore need the floating melody to attach to it. Our analysis of the alternation between the absence and the presence of what HB call “glide absorption” (see the unabsorbed forms in (3) versus absorption in (1)) interprets the phenomenon as a case of vowel∼zero alternation similar to the Hungarian data, as such vitally hinging on the governing capacities of the following vowel.

According to Coda Mirror, the second of two antagonistic forces in phonology is licensing, whose function is just the opposite: to comfort segmental expression of its target (both definitions cite those in Ségéral and Scheer 1999: 20). In the analysis below, it will be crucial that filled/nonempty vowels are able to both govern and license (therefore, the preceding consonant will be licensed, and a preceding empty V may remain phonetically unexpressed – see V2 in bokrot in (12)), while the FEN (in Basque) is unable to either license or govern.

3.2 Distributional facts of place in Getxo Basque

We take the following facts and observe that place features are highly asymmetrically distributed in Basque. Word-final consonants are restricted to coronals (sonorants and voiceless obstruents)^[4] – we interpret this as implying that word-final consonants are underspecified for place and coronal is the default specification. In addition, closed syllables in general do not permit place contrasts: “coda” laterals and nasals show strong place of articulation agreement effects with ensuing consonants (e.g., asal + ɟana [asaʎ ɟana] ‘eaten peel’, gison + barri [gisom bari] ‘new man’). This again indicates place underspecification, this time of “coda” laterals and nasals.

Place features of laterals and nasals are subject to many further asymmetries. As we have seen, the two coronal sonorants that are affected by palatalisation in Getxo are /n/ and /l/. Their palatal variants, [ɲ] and [ʎ] have this highly asymmetric place distribution. For historical reasons, as they developed from historical /n/ and /l/, they are vanishingly rare in word-initial position (with a few marginal counter-examples, e.g. interjection ño with [#ɲ] and the one single word with [#ʎ] lloba ‘nephew; grandchild’, which in fact alternates with illoba, see HB p. 7) or post-consonantally. Moreover, word-finally, the place of articulation of both /n/ and /l/ is neutralised to coronal, whereas, syllable-finally (incl. across words) both /n/ and /l/ take the place of articulation from the following consonant.

The syllable-final place assimilation process is somewhat more extensive for the nasal than for the lateral, which assimilates only to coronals (il + san [il san] ‘he died’) and palatals (asal + ɟana [asaʎ ɟana] ‘eaten peel’), but preserves its PoA before a labial or velar (mutil + bet [mutil βet] ‘a boy’; Algorta [alɡorta] ‘ibid. (place)’). This may be because the lateral is complex for place (see Backley 2011; Polgárdi 2019) – as this issue is beyond our present scope, we will take no stance regarding this aspect of melodic representation; we simply observe that only |I| and |R| (palatality and coronality) spread to the lateral, and it acts as a non-assimilating consonant when followed by a labial or velar.^[5]

Therefore, these asymmetries make both sonorants restricted in their occurrence to the VCV environment, in which, accidental gaps aside, laterals and nasals are contrastive on the surface: [ama] ‘mother’, [sana] ‘vein’, [troɲu] ‘knot’.

As a consequence of the defective distribution and the palatalisation facts, HB chose to analyse the palatal nasal in every [Vɲ] sequence as deriving from underlying /Vin/, and every [VʎV] as derivable from /VilV/. As noted above, this leads to many “lexical exceptions” (recall the examples in (4) and the discussion in Section 2.2). However, from a surface perspective there are onset palatal laterals and nasals (e.g., [troɲu] ‘knot’). This fact is important since we link it to our reanalysis of palatalisation. Specifically, we note an apparently unnoticed generalisation in the so called “exceptions”: recall that word-finally there are no exceptions, not even in loanwords. All /in+V/, /il+V/ contexts produce alternations: /makin/ [makin] versus [makiɲa] uninflected versus ‘machine-ABS.SG’ and /fusil/ [fusil] versus [fusiʎ-e] uninflected versus ‘rifle-ABS.SG’.

We take this to mean that the exceptions (all exponent-medial and in onset/prevocalic position) are actually underlyingly palatal, whereas all /l/ and /n/ in non-prevocalic position (including all exponent-final instances) are underlyingly placeless and only gain coronality through it being the default place of articulation. This is expressed in Strict CV terms beneath, showing that in order to have a place feature, /n/ and /l/ must be licensed (i.e., hit from the right by the supportive/comforting force, as explained in Section 3.1.2 above).

These conditions mean that in Getxo the word-internal intervocalic position requires nasals and laterals with a place feature. In fact, all three place features are licensed in this position: bilabial /m/, alveolar /n/ and palatal /ɲ/ (e.g., ama ‘mother’, sana ‘vein’, troñu ‘knot’). The consonant (whose melodic make-up we ignore here and simply symbolise with N, a placeless nasal) is in +Lic position and is therefore able to maintain any of the available place elements: |U|, |R|, |I| – see (14).

(14)

Place contrasts in licensed positions[6]

In “coda” position, the nasal and lateral would not be able to be licensed because they precede an empty V-slot: the nasal (symbolised by N) and the lateral (symbolised by L) are underspecified for Place lexically – see (15a), where we indicate placelessness with empty brackets [ ]. However, in these positions, where possible, the nasal and lateral would obtain a place feature from place assimilation with the ensuing consonant (15b), which itself is in a licensed position. The only place of articulation available in coda position that does not result from spreading is the default place: Coronal, which eventually fills the underlyingly empty place [ ] (compare (15a) and (15c)).

(15)

Place in “coda” positions[7]

Note that meanwhile, word-finally, the nasal and lateral would be before an empty nucleus, which cannot provide licensing to the position. As such, no underlying contrast in this (or in “coda” position) is permitted. The resultant effect is that word-finally all nasal and lateral consonants are underlyingly placeless.

The place they obtain will depend on their phonological environment, and it is these contexts that drive the alternations. In fact, from this perspective, the whole of cross-morphemic palatalisation is driven by the various ways in which these positions derive place.

3.3 Palatalisation across morphemes

Across morphemes these placeless word-final nasals and laterals are derived in various ways resulting in various phonological contexts. Returning to the cases that we list in Section 1, cross-morphemic palatalisation occurs in the following environments: with clitics and certain suffixes (within the word) and across words (before a vowel).

3.3.1 Clitics and suffixes

It is a commonly held view in representational (morpho)phonology that the exponents of functional items are structurally incomplete: they contain underspecified, unfixed or floating material, which very often accounts for the behaviour they exhibit and the modifications they impose on the root. As is apparent below, the present analysis heavily draws on this idea.

We start with vowel-initial clitics and suffixes (i.e., the data in (1) above). These vowel-initial forms contain floating melody, which docks to the final V-slot of the stem. Consonant-final stems have their final empty V-slot filled by this operation, whereas vowel-final roots (which contain an underlying vowel in their final V-slot), fuse with the suffix into the stem’s final V-slot, as in /neska + a/ [nesk-e] ‘girl-ABS.SG’. These suffixes are, in a sense, internal to the stem. We propose that the Strict CV mechanism for vowel-initial clitics and suffixes in Getxo is very similar to that used by, e.g., Balogné Bérces (1997) and more recently by Newell (2021) for L1 affixes in English.

The fact that V-initial suffixes float into the final V-slot of stems has a particular significance for Getxo given the place asymmetries described above because it means that the root-final /n/ and /l/ that are underlyingly placeless due to (13) go from being −Lic to +Lic and, consequently, now need a place feature. We propose that in these cases they gain it through palatalisation from the local palatal vowel and through the spreading of |I|. This process is shown beneath, taking [oɲé] ‘the foot-ABS.SG’ as a representative example. As shown in this derivation, the root vowel is a diphthong (a long vowel, regularly occupying a VCV sequence, i.e., two CV units, in Strict CV) whose second term is lexically floating (therefore C2V2 is underlyingly empty) – this floating palatal element is fully absorbed into the nasal or lateral’s C-slot upon palatalisation leading to “glide absorption” (16c and d).

As shown in (16a and b), a word-final nasal (an underspecified N) remains unlicensed and its place (lexically empty [ ]) is filled by default with the coronal element (the |R|). In contrast, when the affix is attached to the root-final V-slot (16c), the nasal finds itself in a +Lic position and, according to (13), it is now able to receive the |I| spreading from the left (16d). A more detailed exposition of glide absorption (and where it does not occur) will be discussed in Section 3.4.

The data in (2) illustrate that consonant-initial suffixes also participate in this cross-morpheme palatalisation process. We must propose that the initial consonants of these affixes are underspecified for place, unlike root-initial /n/ and /l/.^[8] Whenever no other PoA source is available, the default will grant coronality eventually (17a and b); however, the consonant now sitting in a +Lic position can gain place by spreading from the left if an |I| is there (17c and d).

(17)

Morpheme-initial palatalisation[9]

What is clear from both (16) and (17) is that the default mechanism supplying the coronal feature is indeed – as the name indicates – default, meaning that it is a last resort mechanism (default phonetic interpretation) and is unavailable as long as other modes of acquiring place are accessible. When the preceding vowel, however, does not contain |I|, this (originally) final/placeless sonorant is still doomed to remain placeless and be interpreted with the default PoA.

3.3.2 Place and palatalisation across words

Post-lexically, palatalisation only occurs before vowel-initial words because the consonant in C-initial words imposes its place feature on the root-final sonorant through the regular process of place assimilation whenever it can, i.e., for three places in nasals (spreading |U|, |R|, |I|) and two for laterals (|R| and |I|). Recall from Section 3.1.2 that according to our model, a “coda-onset” cluster involves a dependency relation between the two consonants: there is a dependent and the head of the cluster. The head of the cluster is in the licensed (prevocalic) position. In Getxo (and in many languages) this head imposes its Place feature on its dependent, leading to partial gemination/place assimilation – i.e., a bipositional configuration. In most cases, this relationship provides the placeless nasal and lateral with a place feature; as such it does not require one from palatalisation.

(18)

“Ordinary” place assimilation[10]

As long as the root-final consonant is followed by a FEN and then a filled C-slot, palatalisation is blocked even when the preceding vowel is palatal. When the lateral is unable to receive the place of the second consonant in a potential “coda-onset” cluster (as in (19)), it remains placeless – eventually filled with coronal by default.

(19)

Preconsonantal palatalisation blocking

/mutil + bet/ → [mutil βet] (cf. *[mutiʎ βet]) „a boy‟

As we will see later, the representation in (19) is incomplete: part of the explanation of why palatalisation does not take place in such cases concerns the inability of the |I| to attach anywhere else but its V-slot. See (22) and (26) below.

Now we turn to vowel-initial words (see the data in (3) above). What makes a crucial difference is that vowel-initial words start with an empty C-slot. We assume that this forms a bipositional structure with the previous consonant: the ensuing configuration has the same properties as a “coda-onset” cluster (it is effectively a geminate – though Basque does not interpret it by phonetic duration), whose head is an empty C-slot. Being empty, it has no place of articulation, and neither does the final /n/ or /l/ have one underlyingly. In these cases, if an |I| element is available, it will spread into the bipositional construct, whose place feature comes from the palatalisation.

In (20), “Resyllabification” creates the same structure as the previous C1 and C2 in (18), a bipositional structure where the consonant now occupies a +Lic position. As a result, it cannot remain placeless: the palatalising feature |I| will spread into the bipositional complex. This unifies nasal and lateral place assimilation and cross-word palatalisation with the same structure, having the distinctive property of bipositionality. In Basque, bipositionality may produce underlying geminates because in this language (with the sole exception of Arbizu dialect), neither vowels nor consonants contrast phonetically for length.^[11]

A significant element of the analysis lies in the drive that motivates the alternations: the unlicensed sonorants need Place and the one they obtain depends on their phonological context. As a result, the whole of cross-morphemic palatalisation (as well as, in fact, nasal/lateral place assimilation) is driven by the various ways in which these positions derive place.

In addition, it is worthy of note that under the present analysis, the differences in phonological behaviour of affixes and words as they are concatenated with the root is fully encoded in the phonological representation without any non-modular reference or diacritic. Following previous proposals like Balogné Bérces (1997) or Newell (2021) for English, Level 1 clitics and suffixes are marked by their phonological structure. On their left edge, the vowel is floating or the consonant is underspecified. In GP/Strict CV, V-initial words are also special, as they start with an empty onset. Given these representations, it comes as no surprise that such morphemes act as triggers of morpho-phonological processes.

3.4 The representation of vowels/diphthongs and glide absorption

Getxo Basque has a five vowel system, with /i, e, a, o, u/. In addition, it is said to contain four i-final diphthongs (/ai, ei, oi, ui/, which we represent as [aj, ej, oj, uj]) and two u-final ones (/au, eu/) (HB p. 9). From the perspective of the palatalisation process under discussion here, the relevant (trigger) vowels are /i/ (seemingly a short monophthong) and the i-final diphthongs.

HB argue that the diphthongs are derived by a rule that makes an unstressed high vowel into a glide immediately after another vowel (/Vi/ > [Vj]). In what follows, we accept this analysis: element |I| in an unstressed postvocalic V position is phonetically interpreted as a [j] offglide. We also make use of the following crucial facts (generally true in Basque): (i) the fifth i-final diphthong /ii/ (= [ij]) seems to be missing from surface forms; and (ii) neither vowels nor consonants contrast phonetically for length. These two facts are significant because, as shown in (21), a preceding /i/ triggers palatalisation in exactly the same way as [j]-final diphthongs do.

Consequently, we assume that the two have identical compositions: /i/ also has an offglide, i.e., it is also a long vowel (diphthong): /ii/ (the missing i-final vowel of the inventory).

In Strict CV terms, this means that all palatalising vowels share two common properties: (i) all occupy two V-slots (that is, a VCV sequence); and (ii) all have the second V occupied by an |I| element which serves as the second term of palatalising diphthongs as well as of the bipositional structure of the long /ii/. The examples in (21) can accordingly be modified as (21′). The apparent difference between (a) and (b) comes from the fact that in Basque, bipositionally linked melody is never interpreted phonetically as long. Underlying /ii/ is realised as [i]. Note, however, that phonologically the trigger vowels comprise the natural class of all V1CV2’s where V2 contains |I|.

Another apparent difference is found in how palatalisation is accompanied by glide absorption, which seems to follow either of two routes: if the vowel before the /n, l/ is /i/, plain palatalisation happens in all contexts (21/21′b); if it is any one of the i-final diphthongs, palatalisation takes place with absorption before a suffix/clitic and without absorption across words (21/21′a). If, however, glide absorption is conceived as i∼zero alternation, the difference disappears since both /i/ and /ii/ are interpreted as [i] so ii+L/N sequences as in /mutiil-e/ can undergo glide absorption without a visible trace on the surface.

Glide absorption, then, is but a form of vowel∼zero alternation, as it is understood in the GP tradition (and illustrated in (12) above) involving a floating element |I|. Therefore, we assume that glide /i/ is a floating |I| element, whose phonetic interpretation depends on whether its designated V-slot is silenced by Proper Government (in which case it is unable to dock onto it) or is left ungoverned (in which case |I| links to it to satisfy its ECP). This latter is what happens when the following consonant is word-final, as in (22). Since the FEN is unable to govern, |I| links to V2; the final sonorant remains monopositional and receives COR by default.

Moving now to the boundary between words where the second begins with a vowel, we see that in these strings, V2 is ungoverned just as it is in (22). This is because the V-slot that corresponds to V_FEN in this string is silenced by V3 (a filled V-slot). The inability to silence V2 leads to the floating |I| surfacing in that position and the interpretation of a glide. For this reason, glide-absorption is blocked again.

As shown in (23b), since the second word starts with an empty C-slot (C3), a bipositional structure is created by spreading the final (palatalised) sonorant into C3.

In both these contexts, where a floating |I| is preceded by a FEN or a vowel-initial word, the V2 that would host the glide fails to be silenced. Crucially, this is different in affixation/cliticisation. As we have explained already in Section 3.3.1, we assume that suffixation involves the linking of a floating vowel to the final V-slot of the stem. If the stem is consonant-final (as with all the stems that undergo palatalisation), suffixation changes the status of the final V-slot. It goes from being a FEN to being a filled V-slot.

This change of status allows the final V-slot of the stem (when suffixed) to silence V2, the position that hosts the glide. This means that when a stem undergoes affixation, the V-slot that allows the glide to surface is silenced. The surface effect of this will be glide absorption but only in cases of suffixation.

It needs to be pointed out that, as argued above, /ii/ and the i-final diphthongs have parallel underlying representations, and as a result they contribute to the derivation of the output structures identically (compare /ii/ in (22) and /oi/ in (23–24)), therefore no separate illustrations are needed for the mechanisms leading to or preventing absorption with the two vowel classes. Also, the discussion above aims to shed light on the rather enigmatic earlier representations of the palatalising vowel with a floating |I| in (16) and/or [i] as VCV in (17), (19) and (20a) in the previous sections.

A further observation to make is that the palatal output is always a shared/bipositional structure (a virtual geminate),^[12] but it comes in two different forms: it either branches to the right (on the initial empty onset of V-initial words – e.g., (20), (23)) and there is no absorption, or to the left (on the C-slot accessible because V2 was vacated by PG – (24)) and absorption ensues. It may even be claimed that the motor of the entire phenomenon is the government that determines the fate of the floating |I|.^[13]

There is in fact one more situation in which this left-branching construction is derived: the palatalisation that takes place with [i/j]-final roots and underspecified initial sonorants in affixes. In (25) we give the updated version of (17c and d).

For a final illustration, let us consider cross-word place assimilation. As normal, in the cross-word situation, |I| links to V2 because this V-slot is not silenced. This leads to pronunciation of the glide (the lack of glide absorption).

In cases like /mutil + bet/ → [mutil βet] ‘a boy’ (cf. (19) above), where C-to-C place assimilation cannot happen but a palatalising vowel is available, nevertheless palatalisation will not take place (*[mutiʎ βet]) and instead we get a default. As shown in (26), the floating |I| cannot dock to the underspecified nasal or lateral in C3.

(26)

Across a word before a non-assimilating consonant[14]

This is because it would lead to a place of articulation in a non-licensed position (since V3 is silenced by government). Place of articulation is not permitted before a silenced V-slot (such as V3). Moreover, since (for independent reasons) the lateral does not assimilate to the Labial or Dorsal PoA, this source of Place is also denied to the consonant in C3 (shown with the blunted arrow end). As such, in this context the only option is the default PoA: coronal.

Instead, in this same context, where place assimilation is allowed (such as with a nasal and an ensuing stop), this is how the consonant in C3 obtains Place. This is possible because the Place of the ensuing stop is in a +Lic position and allowed to be part of the representation (which is not the case with rightward spreading into C3 from V2, so long as V3 is silenced). This outcome is shown in (27).