Palatalization, fortition and deletion in the high vowels of Arcadian Greek

The present study discusses the status of the high vowels /i/ and (secondarily) /u/ in Arcadian Peloponnesian, an entirely unstudied variety of Greek. The high vowels of Arcadian Greek undergo different processes or are altogether deleted in an unstressed environment. In particular, unstressed [i] palatalizes or undergoes fortition (depending on the target consonant) or gets deleted. Kochetov (2016) argues that palatalization and fortition work in parallel as repair strategies in languages affecting different target consonants, so that marked C+palatal sequences are avoided. This seems to be the motivation behind the widely attested palatalization and fortition in nonstandard Greek varieties more generally, which may need to be unified under one pandialectal study in the future. Until now, high vowel deletion has been thought to be restricted to northern dialects of Greek and has constituted a defining criterion for the classification of Greek dialects in the literature.


The dialect of Arcadian Greek
The data of the present study comes from the mountainous part of the region of Mantineia 4 , situated in the northeastern part of Arcadia, on the slope of Mt Mainalon and bordering on mountainous Gortynia. The study is empirical as the author has a solid passive knowledge of the dialect. As previously mentioned, the phonology of Arcadian Peloponnesian (or, more accurately, Arcadian Peloponnesian varieties) is entirely unstudied. The degree of regional variation within Arcadia remains to be investigated.

The current status of Arcadian Greek
Arcadian Greek (hereafter ArcGr) has been undergoing significant levelling for decades and is still continuously losing ground to SMG. This is certainly true of most other nonstandard varieties of Greek. Arcadian Greek with a fully-fledged phonology, i.e. unaffected by SMG, is still spoken by old people, almost all of whom are bidialectal in ArcGr and SMG. This dialect would have been still fully vibrant at around the end of the 19th century and the beginning of the 20th century and would have been spoken most probably exclusively by monodialectal speakers. Trudgill's (2003) classification and cartographical representation of the Greek dialects also refers to this period, during which, regional dialects were still fully vibrant all over Greece.
According to Pantelidis (2015), the levelling of the Peloponnesian dialects started earlier than in other regions of Greece due to the geographical proximity of the Peloponnese to Athens, the capital of the new State, the increase of language contact between the two through commercial interaction, its advanced infrastructure compared to other regions of Greece etc.
As a result, it seems possible that Peloponnesian varieties have undergone more levelling than other Greek varieties. This would make future research a rather difficult task. After all, the Peloponnese is a large area of Greece with urban centres (chiefly Patras in the north) but also a multitude of towns 3 A dispute around the establishment of an official language (katharevousa vs. demotic) for the Greek State that lasted for well over a century (see Mackridge 1990Mackridge , 2004Mackridge , 2009). Katharevousa is a purified artificial form of Greek "based on modern Greek (that is, on the spoken language of educated people and on the written language of the pre-Independence era), but with its orthography, morphology, vocabulary and (to some extent) syntax adjusted, rather unsystematically, so as to conform in part to the rules of ancient Greek grammar." Mackridge (2004: 118). 4 This municipal unit of Arcadia is generally not particularly mountainous, with the exception of its northeastern part (where the present data stems from) that borders on very mountainous Gortynia.
3 and villages in its very diverse geographical landscape. However, as long as there still are native speakers around (albeit old and bidialectal) it will not have been left too late. It should be noted here that the present study is not so much interested in the current phonological situation of levelled ArcGr but rather in what its phonological system looked like prior to any SMG interference.

The vowel system of Arcadian Greek
ArcGr has a basic five-vowel system /i e a o u/, i.e. five monophthongs and no diphthongs. As such, Arcadian Greek and SMG (Mackridge 1985;Joseph and Philippaki-Warburton 1987;Arvaniti 1999;Sfakianaki 2002;Themistocleous 2017) share the same vowel system 5 , possibly with minor acoustic differences. Most Greek varieties share this basic vowel system with SMG (see Trudgill 2009 for exceptions).

The status of high vowels in Arcadian Greek
SMG and ArcGr differ in that the latter has a plethora of light diphthongs: /i/ semivocalizes and forms a light diphthong with the vowel of the preceding syllable. Semivocalization of /i/ in this position applies intermorphemically, across morpheme boundaries and across word boundaries. Light diphthongs also feature in SMG but are not common.
The most significant difference in the phonological systems of SMG and ArcGr is the status of their high vowels, /i/ and /u/. In ArcGr, unlike SMG, these appear to be less stable than the other three vowels when unstressed and therefore more vulnerable to their phonological environment.

High vowel deletion in Greek
One can only find short and generic references to the phonology of Peloponnesian in the literatureusually in studies of another dialect or dialectswhich are not supported by specific data. In particular, unlike in northern dialects, high vowel deletion in Peloponnesian has been described as a fast or casual speech phenomenon: "VD [vowel deletion] is attested in some southern dialects such as SMG and Peloponnesian Greek Dauer 1980;(Baltazani 2006(Baltazani , 2007, but to a much lesser degree and typically in "quick casual speech" (Loukina 2008: 323). As a result, it is not considered a feature of the Southern Greek varieties." (Baltazani, Kainada, Revithiadou, and Topintzi 2016).
Changes in the acoustic quality of vowels, such as devoicing, reduction, etc. are considered common in languages and are naturally attested in SMG and dialects other than the northern ones (see Loukina 2009: 38 for a summary of experimental and non-experimental studies discussing vowel reduction outside of the northern dialects of Greek). A more recent experimental study by Baltazani (2006) has shown that vowel duration in Greek is affected by stress, word length, and the position of the vowel relative to word stress. Thus, vowels have been found to reduce more in longer words and after (rather than before) a stressed syllable. Prior to this, Dauer (1980) had also demonstrated that vowel reduction in SMG predominantly depended on the phonetic environment and the position relative to the stressed syllable.
Nevertheless, fast speech processes are optional and have been claimed to occupy a different place in the phonological component of a language (Kaisse 1985). High vowel deletion in ArcGr, as in northern dialects, is a regular process, central to its phonology. Unstressed [i] loses its syllabicity through semivocalization, palatalization or fortition or is deleted. This entirely depends on the preceding and/or following segment in its phonological environment:

3) (a)
Post-consonantally and word-finally (Ci#), /i/ triggers either full palatalization in the preceding consonant or is deleted, triggering secondary (or strengthened -see Baltazani et al. 2016) palatalization in the preceding consonant.
(b) When flanked by two consonants (CiC), /i/ either triggers full palatalization in the preceding consonant or is deleted.
We will now look at each case separately: 1) Post-vocalically (ViC, Vi#), /i/ semivocalizes and forms a light diphthong with the vowel of the preceding syllable, thus resolving the undesired hiatus.
Each example below is chosen to include /i/ following each one of the five vowels of ArcGr, with which it forms a light diphthong. In SMG, the words below (i.e. the shared ones: 2b, 2d, 2e) would have one extra syllable: (2) Post-vocalic /i/ across morpheme boundaries: As can be seen, adjacency of two high front vowels as in (3d) and (4e) will result in the deletion of the first of the two, rather than semivocalization of the second. It should be noted that these sequences are quite rare in the first place. Also, [ij] (as in 2a) is only attested in word-final position in masculine plural nouns, adjectives and pronouns ending in -ios. Therefore, hiatus is resolved with deletion in (3) and (4) and semivocalization in (2). Hiatus is resolved with the help of an epenthetic j across i# #V (5a-d) or iV word-medially (5ef): (5)
The underlying singular forms, where stressed syllabic [i] appears, are provided simply to justify the existence of an underlying /i/ in the plural forms. There are also other morphophonological alternations that justify this.

Underlying form Surface form Gloss Surface form
Plural forms of neuter nouns ending in -i Singular forms Heteromorphemic CiV sequences identical to the ones in (7) and (8)  Tautomorphemic CiV sequences are not as frequent (e.g. /piat-o/ [pçáto], 'plate' etc.) and are mostly found within the common and quite productive prefix /ðia/ > [ðʝa], 'through, trans-' when this is attached to a consonant-initial stem (9a-c). This prefix drops its final vowel when attached to a vowel-initial stem (9d-f). Some examples with i-ending prefixes attached to a vowel-initial stem are also provided in (9g-j) below. A similar picture to that in (7) and (8) (7) and does not always undergo fortition in the environment of (8) and (9). Its behaviour is not at all uniform in the above environments. Katharevousa rules, which demand that [i] stays syllabic as much as possible, frequently interfere with the application of palatalization and fortition. For example, despite the fact that [paɲá] < /pania/ and [evʝénia] < /evɣen-ia/ share the same /CiV/ environment, the former features palatalization while the latter features a syllabic [i] in SMG. In examples (10g-i) below, two outputs are possible in SMG, with the second being more likely to appear in more formal registers of speech (e.g. the news broadcast  Baltazani et al. (2016) propose that high vowel deletion in this context triggers strengthened palatalization (and not secondary palatalization) in the preceding consonant in Kozani Greek. This extension to the current palatalization typology (see Kochetov 2016) is said to be justified on both phonetic and phonological grounds: "The need for this addition is justified on both phonetic and phonological grounds; its acoustic output is dissimilar to both full and secondary PAL [palatalization], while the triggering force is not the presence of a front vocoid but rather the loss of it in unstressed positions." (Baltazani et al. 2016: 5). Therefore, they take secondary and strengthened palatalization to be two distinct processes: in the former, the high front vocoid is present in the output as a secondary gesture, while in the latter it is not. Additionally, they found that no F2 raising was detected in the vowel of the relevant VC sequences in the surface forms of (13), while frication, aspiration and lengthening were detected instead. In ArcGr, the final consonants in (12) are auditorily distinct from corresponding ones in SMG that have not undergone secondary (or strengthened) palatalization. The loss of the nuclear vowel becomes particularly clear, even for the untrained ear, when the names above are called out, i.e. when used in vocative case. Here they carry the notation of secondary palatalization, although an acoustic analysis would determine the kind of palatalization that these consonants undergo more accurately.
Palatalization and deletion in (11) and (12) give rise to a number of marked consonant clusters and prosodic structures which are not found in SMG and which differentiate the two systems quite dramatically. Topintzi and Baltazani (2012: 2) found that corresponding clusters in Kozani Greek were "less stable in duration than underlying ones". [paltó] 'coat'

The high back vowel /u/
Before we discuss /u/, we must bear in mind that it is much less frequent than /i/ and by far the least frequent of all five vowels of Greek (Protopapas, Tzakosta, Chalamandaris, and Tsiakoulis 2012). In that sense, we are presented with a small challenge, as not all possible phonetic contexts are amenable to examination.
To begin with, stressed [u] retains its syllabicity intermorphemically and across morpheme boundaries (e.g. /surn-o/ > [súrno] 'drag', /lem-u/ > [lemú] 'neckgen.' etc.). As with unstressed [i], unstressed [u] is susceptible to its environment. As already mentioned, we will have to bypass certain contexts: VuC, Vu# and CuV. The first two are rare and can mainly be found in the genitive case of a handful of masculine nouns and adjectives, some of which are not even dialectal. At the same time, CuV context is extremely rare and can be found in very few non-dialectal loanwords.
We are therefore left with Cu# and CuC. Word-finally (Cu#), unstressed [u] can be found in the genitive case of masculine and neuter nouns and adjectives ending in -os and -o respectively (16). It can also be found at the end of some adverbs (17). In this vulnerable word-final position, [u] is deleted and there is a reduction in syllable number by one in the relevant words.
Impressionistically, the final consonants (in bold type) in the second column in (16) are produced differently from their counterparts in SMG. An acoustic analysis would confirm whether labialization as a secondary articulatory gesture is in place. The status of the high back vowel /u/ has occupied a shorter space in this study and needs to be further investigated in the future.

Discussion
The behaviour of high vowels, and /i/ in particular, in ArcGr is interesting for various reasons. Firstly, it bears striking similarities to other nonstandard Greek dialects, which, to the best of my knowledge, do not retain a syllabic [i] in an unstressed position, at least in CiV context (see Newton 1972 for the most comprehensive theoretical work on Modern Greek dialects). Secondly, it improves our understanding of the mechanisms that these vowels deploy in a vulnerable unstressed context. It would be simply inadequate to put deletion down to the fact that /i/ and /u/ are the shortest unstressed vowels in Greek (Fourakis et al. 1999;Dauer 1980). Viewing their behaviour in ArcGr through the prism of the theory of markedness would help us understand it in more depth. To begin with, the fact that /i/ is unable to retain its syllabicity in an unstressed position can easily be explained through markedness, as C+palatal sequences (C+consonant, C+glide and C+vowel) are considered marked cross-linguistically (see Kochetov 2016). According to Kochetov (2016), the phonetic motivation for this lies in the fact that the palatal gesture always conflicts with other lingual gestures produced further front or further back in the oral cavity. Additionally, these sequences are acoustically and perceptually problematic.
In that respect, SMG is significantly more marked than ArcGr as it often favours a highly marked syllabic /i/ in unstressed positions, even in the very susceptible post-consonantal pre-vocalic environment (CiV). This is hardly surprising if we take into consideration the fact that SMG evolved less than naturally and was heavily affected by cumbersome katharevousa rules. Kochetov (2016), using data from Kirundi and, amongst other languages, the Greek dialect of Kos (strikingly similar to ArcGr /i/ data), argues for a 'functional and formal affinity' between palatalization and fortition as repair strategies in languages, proposing that they work in parallel, affecting different target consonants with the ultimate goal that marked C+j sequences are avoided. Kochetov (2016) therefore argues that unless we unify these two seemingly unrelated phenomena, we may miss important generalizations in the behaviour of C+j sequences.
Furthermore, vowel deletion should not be seen separately from palatalization and glide fortition either as it also seems to be strongly related to them. In other words, /i/ first takes recourse to palatalization or fortition that ensure that it will 'survive' (or at least some of its features) through the coalesced palatal output or, even better, through the strengthened output of fortition. As shown here, in ArcGr, full palatalization is triggered after dorsals and alveolar coronals, while fortition is triggered after labials and dental coronals. Deletion, which is also one of the strategies deployed to avoid C+j sequences (along with palatalization, fortition, epenthesis and metathesissee Kochetov 2016), always seems to be the last resort.
Finally To sum up, translated into optimality theory terminology, the proposed contextual markedness *C+palatal constraint proposed by Kochetov (2016) seems to be the motivation behind glide palatalization and fortition in nonstandard Greek varieties, which we may need to unify under one pandialectal study in the future (see also Baltazani and Topintzi 2012).

Sociolinguistic implications
High vowel loss, as it is traditionally termed, is perhaps the most basic or defining criterion used throughout the history of Greek dialectology for the classification of Greek dialects (amongst others, Hatzidakis 1905;Triandaphyllidis 1938;Tzitzilis 2001;Argiriadis 1990;Newton 1972). For example, Hatzidakis (1892) was the first one to use high vowel loss as the diagnostic criterion for the classification of Greek dialects, and his division of them into northern and southern is based solely on it. Trudgill (2003) also deploys high vowel loss as one of his criteria for the classification and first ever cartographical representation of the Greek dialects, a notoriously difficult task for various linguistic and non-linguistic reasons. He proposes fifteen areas on the dialectal map of Greek. On that map, Peloponnesian is classified along with the Heptanese and North Epirus with the Central group of Greek dialects: "Western Epirus, Corfu, Kephalonia, Zakinthos, the Peloponnese. This area corresponds more or less to Newton's Peloponnesian-Ionian area and has none of the six features we have been employing as criteria. This is not surprising in view of the fact that the Ionian islands and the Peloponnese are generally agreed to have supplied most of the input into Standard Greek (Horrocks 1997: 300), although this has been disputed, at least for the Peloponnese, by Pantelidis (2001b)." (Trudgill 2003: 58-59).
However, as shown in the present study, high vowel loss (a) also features in Peloponnesian Greek, and (b) should be viewed in conjunction with palatalization and glide fortition (in relation to /i/). The sociolinguistic implications of this study could be significant. If high vowel loss is found anywhere from northern to southern dialects of Greece it fails to do its job. One way around this problem would be to abandon high vowel loss as a criterion for the classification of the Greek dialects and use mid vowel raising instead, which is restricted to northern dialects alone.