Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Spell-out of phonological domains : the case of Slovenian Božič, Jurij 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2015_september_bozic_jurij.pdf [ 1.27MB ]
JSON: 24-1.0166385.json
JSON-LD: 24-1.0166385-ld.json
RDF/XML (Pretty): 24-1.0166385-rdf.xml
RDF/JSON: 24-1.0166385-rdf.json
Turtle: 24-1.0166385-turtle.txt
N-Triples: 24-1.0166385-rdf-ntriples.txt
Original Record: 24-1.0166385-source.json
Full Text

Full Text

Spell-Out of Phonological DomainsThe Case of SlovenianbyJurij BožičB.A., University of Ljubljana, 2013A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF ARTSinThe Faculty of Graduate and Postdoctoral Studies(Linguistics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)July 2015© Jurij Božič 2015AbstractNovo mesto Slovenian, a South Slavic language, exhibits a process of un-stressed /i/-deletion that appears to be construction-specific: it applies inverbs and participles, but not in other word classes, such as nouns or adjec-tives under identical phonotactic conditions. This thesis proposes a phono-logical analysis of this deletion process, which determines that the masculineplural exponent /-i/ may attach to a participial stem, as in [ptc]-/i/, andundergo deletion under specific phonotactic conditions, but it may also at-tach to a nominal or adjectival stem, as in [adj]-/i/, where it is preserved inthe same phonotactic conditions in which it deletes with participles. Thisprocess of construction-specific vowel deletion cannot be derived by the stan-dard approaches to construction-specific phonology, such as CophonologyTheory (Orgun 1996; Inkelas et al. 1997; Inkelas & Zoll 2007; Inkelas 2008,2011), or the grammar with phonological levels (Cyclic/Word/Phrase level)as in Embick (2013), which stems from Halle & Vergnaud (1987) and effec-tively mirrors Lexical Phonology (Kiparsky 1982a,b; Mohanan 1986). Theproblem is rooted in the way these approaches define phonological domains.A system with the fixed distinction between Cyclic and Word levels is shownto be inadequate. This thesis subscribes to the research program laid outin Marvin (2002), Marantz (2007), Samuels (2009), Piggott & Travis (2013)and Newell & Piggott (2014), which seeks to interpret the phasal cyclicityof syntax (Chomsky 2001, 2008) as a locality boundary that defines phono-logical domains. It is shown that a simple generalization on the /i/-deletionfacts can be formulated by making reference to phases as phonological spell-out domains. In addition, this thesis proposes a tentative formal solutionto the problem of deriving the deletion of the masculine plural /-i/: sincethe deletion effect is specified in the stem to which the /-i/ attaches, it isproposed that the phonological grammar stores phonology-specification ina buffer, which may persist to the end of the spell-out domain as set bythe phase. Through this, the stem-specified deletion phonology effects themasculine plural suffix /-i/.iiPrefaceThis thesis is original, unpublished, independent work by the author, J.Božič.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Interface: Distributed Morphology . . . . . . . . . . . . . . . . 21.2 Goal of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 142 Novo mesto Slovenian . . . . . . . . . . . . . . . . . . . . . . . . . 172.1 Basic overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.1.1 Vowels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.1.2 Consonants . . . . . . . . . . . . . . . . . . . . . . . . . 212.2 Status of the high vowel [i] . . . . . . . . . . . . . . . . . . . . 222.2.1 Verbs in Slavic morphology . . . . . . . . . . . . . . . . 222.2.2 Verbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.2.3 Participles . . . . . . . . . . . . . . . . . . . . . . . . . . 292.2.4 Interim overview: Verbs and participles . . . . . . . . 442.2.5 Roots and prefixes . . . . . . . . . . . . . . . . . . . . . 452.2.6 Stress and alternations . . . . . . . . . . . . . . . . . . 493 Phonological analysis . . . . . . . . . . . . . . . . . . . . . . . . . 573.1 Schwa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.1.1 Phonotactics and schwa . . . . . . . . . . . . . . . . . . 573.1.2 Cyclic application . . . . . . . . . . . . . . . . . . . . . 683.1.3 Re-evaluating the ban on schwa in open syllables . . . 73ivTable of Contents3.1.4 Residual issues . . . . . . . . . . . . . . . . . . . . . . . 763.1.5 Epenthesis vs. deletion account . . . . . . . . . . . . . 803.2 On the deletion of the high vowel [i] . . . . . . . . . . . . . . . 843.2.1 First generalization: stress and i ∼ ∅ . . . . . . . . . . 843.2.2 Second generalization: CVCR sequences . . . . . . . . 873.2.3 Single consonant roots . . . . . . . . . . . . . . . . . . . 1073.3 Masculine singular . . . . . . . . . . . . . . . . . . . . . . . . . 1093.3.1 Special stress pattern . . . . . . . . . . . . . . . . . . . 1093.4 Nouns and adjectives . . . . . . . . . . . . . . . . . . . . . . . . 1164 At the PF-Interface . . . . . . . . . . . . . . . . . . . . . . . . . . 1214.1 Morphosyntax of NM Slovenian . . . . . . . . . . . . . . . . . . 1214.2 On the nature of Spell-Out at PF . . . . . . . . . . . . . . . . 1284.3 Phonological cyclicity in NM Slovenian . . . . . . . . . . . . . 1324.3.1 Phase status of verbal stems . . . . . . . . . . . . . . . 1334.3.2 π-cycles in the verbal stem . . . . . . . . . . . . . . . . 1354.4 /i/-deletion at the interface . . . . . . . . . . . . . . . . . . . . 1384.5 Proposal: Domains of phonological Spell-Out . . . . . . . . . 1435 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155vList of Tables2.1 Length contrasts according to Toporišič (2000) (from Jurgec2007a: 31) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2 Vowel length contrasts (or, lack thereof) in NM Slovenian . . 192.3 Consonants in NM Slovenian . . . . . . . . . . . . . . . . . . . 212.4 Slavic (finite) verbs . . . . . . . . . . . . . . . . . . . . . . . . . 222.5 Slavic l-participles . . . . . . . . . . . . . . . . . . . . . . . . . . 222.6 Verbs formed with [-a] (a-class) . . . . . . . . . . . . . . . . . 232.7 Imperative verbs formed with [-a] (a-class) . . . . . . . . . . 232.8 Roots suffixed with [-a] to form verbs (a-class) . . . . . . . . 242.9 Verbs formed with [-i] (i-class) . . . . . . . . . . . . . . . . . . 242.10 Verbs formed with [-i] (i-class) . . . . . . . . . . . . . . . . . . 252.11 Roots suffixed with [-i] to form verbs (i-class) . . . . . . . . . 252.12 Verbs formed with [-∅] (∅-class) . . . . . . . . . . . . . . . . . 252.13 Imperative verbs formed with [-∅] (∅-class) . . . . . . . . . . 262.14 Roots suffixed with [-∅] to form verbs (∅-class) . . . . . . . . 262.15 ∅-class roots with with schwa in the infinitive . . . . . . . . . 262.16 Verbs formed with [-e] (e-class) . . . . . . . . . . . . . . . . . 272.17 Imperative verbs formed with [-e] (e-class) . . . . . . . . . . . 272.18 Roots suffixed with [-e] to form verbs . . . . . . . . . . . . . . 282.19 Semelfactive formed with [-e] (e-class) . . . . . . . . . . . . . 282.20 Imperative semelfactive formed with [-e] (e-class) . . . . . . . 282.21 Verbal classes in NM Slovenian (indicative forms) . . . . . . . 292.22 a-classI participle (stress: √"σ − σ) . . . . . . . . . . . . . . 302.23 Roots used to form a-classI . . . . . . . . . . . . . . . . . . . . 312.24 a-classII participle (stress: √σ−"σ) . . . . . . . . . . . . . . . 312.25 Roots used to form a-classII participles . . . . . . . . . . . . 322.26 a-classIII participle: √"σ − σ . . . . . . . . . . . . . . . . . . . 322.27 a-classIII participle: √σ−"σ . . . . . . . . . . . . . . . . . . . 322.28 Roots used to form a-classIII participles . . . . . . . . . . . . 332.29 i-classI participle (stress: √σ−"σ) . . . . . . . . . . . . . . . 332.30 i-classII participle (stress: √σ−"σ, except . . . . . . 34viList of Tables2.31 Roots used to form i-classI participles . . . . . . . . . . . . . 342.32 Roots used to form i-classII participles . . . . . . . . . . . . 342.33 ∅-classI participle (stress: √"σ − σ) . . . . . . . . . . . . . . 352.34 Roots used to form ∅-classI participles participles . . . . . . 352.35 ∅-classII participle (stress: √"σ − σ) . . . . . . . . . . . . . 352.36 Roots used to form ∅-classII participles . . . . . . . . . . . . 362.37 ∅ ∼ i-class participle (stress: √"σ − σ) . . . . . . . . . . . . . 362.38 ∅ ∼ i-class participle (stress: √σ−"σ ) . . . . . . . . . . . . . 362.39 Roots used to form ∅ ∼ i-class participles . . . . . . . . . . . 372.40 ∅-class semelfactive participle (stress: √"σ − σ) . . . . . . 392.41 Roots that form ∅-class semelfactive participles . . . . . . . 392.42 ∅ ∼ i-class semelfactive participle (stress: √"σ − σ) . . . . 392.43 ∅ ∼ i-class semelfactive participle (stress: √σ−"σ) . . . . . 402.44 Roots that form ∅ ∼ i-class semelfactives . . . . . . . . . . . . 402.45 √CVCr-class participle (stress: √"σ − σ) . . . . . . . . . . 412.46 √CVCr-class participle (stress: √σ−"σ) . . . . . . . . . . . 412.47 Participle built on √misl- (stress: √"σ − σ) . . . . . . . . . . 422.48 Roots that form √CVCr-class . . . . . . . . . . . . . . . . . 432.49 Single-consonant roots: √S- ‘to go’ . . . . . . . . . . . . . . . . 432.50 Single-consonant roots: √b- ‘to be’ . . . . . . . . . . . . . . . . 442.51 Roots with an unstressed [i] in ptc . . . . . . . . . . . . . . . 452.52 Prefixes [pRi-] and [p@R-] (verbs and participles) . . . . . . . . 462.53 Prefixes [pRi-] and [p@R-] (nouns) . . . . . . . . . . . . . . . . . 462.54 Prefix [p@R-]: no semantic difference (verbs and participles) . 472.55 Prefix [p@R-]: no semantic difference (nouns) . . . . . . . . . . 472.56 Prefix [iz-] (verbs and participles) . . . . . . . . . . . . . . . . 482.57 Prefix [iz-] (nouns) . . . . . . . . . . . . . . . . . . . . . . . . . 482.58 Probabilistic stress and vowel-deletion . . . . . . . . . . . . . . 532.59 Participial UR’s and SR’s . . . . . . . . . . . . . . . . . . . . . 553.1 Schwa in nouns . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.2 Schwa in adjectives . . . . . . . . . . . . . . . . . . . . . . . . . 583.3 Schwa in loanwords . . . . . . . . . . . . . . . . . . . . . . . . . 603.4 Repairing consonantal nuclei: /ps-∅/ → ["p@`s] . . . . . . . . . 613.5 Repairing consonantal nuclei: /sn-∅/ → ["s@`n] . . . . . . . . . 613.6 Repairing illicit sonority: /"fìlm-∅/ → ["fìl@´m] . . . . . . . . . 623.7 Repairing illicit sonority: /"pO`t-n-∅/ → ["pO`t@´n] . . . . . . . . 623.8 Schwa epenthesis and *@]σ : /ps-∅/ → ["p@`s] . . . . . . . . . . . 633.9 Schwa epenthesis and *@]σ : /sn-∅/ → ["s@`n] . . . . . . . . . . . 633.10 i ∼ ∅-class participles: Root stress . . . . . . . . . . . . . . . 64viiList of Tables3.11 i ∼ ∅-class participles: Theme stress . . . . . . . . . . . . . . 643.12 Schwa in i ∼ ∅-class participles: /"xRan-i-l-a/ → ["xRànlá] . . 643.13 Schwa in i ∼ ∅-class participles: /"xRan-i-l-i/ → ["xRàn@´l]] . . 653.14 √CVCr-class participles: Root stress . . . . . . . . . . . . . 653.15 √CVCr-class participles: Theme stress . . . . . . . . . . . . 663.16 Schwa in √CVCr-class ptc: /"pown-i-l-i/ → ["pòw@nlí] . . . 663.17 Schwa in √CVCr-class ptc: /pow"n-i-l-i/ → [pow"nìl] . . . 663.18 Schwa in semelfactive participles: Root stress . . . . . . . . . 673.19 Schwa in semelfactive participles: Theme stress . . . . . . . . 673.20 Participle built on √ust- ‘mouth’ . . . . . . . . . . . . . . . . . 683.21 Schwa in an open syllable (in ∅-class participles) . . . . . . 693.22 Cyclic application exemplified with √"mR- ‘die’: verbs . . . . 703.23 Cyclic application exemplified with √"mR- ‘die’: participles . 703.24 Cyclic derivation of [u"m@´R@`l]: 1st cycle . . . . . . . . . . . . 723.25 Cyclic derivation of [u"m@´R@`l]: 2nd cycle . . . . . . . . . . . 723.26 Participial cycle of [u"m@R@l] . . . . . . . . . . . . . . . . . . . . 733.27 Participial cycle of [u"m@R@l] with Align(MWd, R; PWd, R) 753.28 Verbs formed with [-∅] (∅-class) . . . . . . . . . . . . . . . . . 763.29 Verbs formed with [-∅] (∅-class) . . . . . . . . . . . . . . . . . 763.30 e∼@-alternation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 793.31 i ∼ ∅-class participles: Root stress . . . . . . . . . . . . . . . 843.32 i ∼ ∅-class participles: Theme stress . . . . . . . . . . . . . . 843.33 Deriving /"xRàn-i-l-i/ → ["xRàn@´l] ( verbal stem . . 873.34 Deriving /"xRàn-i-l-i/ → ["xRàn@´l] ( participle . . . 873.35 in semelfactive participles: Root stress . . . . . . . . . . 883.36 in semelfactive participles: Theme stress . . . . . . . . . 883.37 in √CVCr-class participles: Root stress . . . . . . . . 883.38 in √CVCr-class participles: Theme stress . . . . . . . 883.39 in single-consonant roots: √S- ‘to go’ . . . . . . . . . . . 893.40 Verbal cycle: /"pòwn-i/ → ["pòw@´n] . . . . . . . . . . . . . . . . 913.41 Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí] . . . . . . . . . . . 913.42 Range of data to consider . . . . . . . . . . . . . . . . . . . . . 943.43 Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí] . . . . . . . . . . . 953.44 Participial cycle: /u.("m@R)-l-i/ → [u.("m@.R@l)] . . . . . . . . . 963.45 Participial cycle: /("xRan)-l-i/ → [("xRa.n@l)] . . . . . . . . . . . 963.46 Participial cycle: /("po.w@n)-l-∅/ → [("] . . . . . . . . 973.47 Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí] . . . . . . . . . . . 1023.48 1st cycle: /pown-"i/ → [pow"ni] . . . . . . . . . . . . . . . . . 1033.49 2nd cycle: /pow("niµ)-l-i/ → [pow("niµlµ)] . . . . . . . . . . 1043.50 2nd cycle: /("xRan)-l-i/ → [("xRa.n@l)] . . . . . . . . . . . . . 104viiiList of Tables3.51 2nd cycle: /u.("m@R)-l-i/ → [u.("m@.R@l)] . . . . . . . . . . . . 1053.52 2nd cycle: /("po.w@n)-l-∅/ → [("] . . . . . . . . . . . 1053.53 2nd cycle: /("po.w@n)-l-∅/ → [("pow.n@l)] (no l ∼ u) . . . . . 1063.54 Single-consonant roots: √S- ‘to go’ . . . . . . . . . . . . . . . . 1073.55 Single-consonant roots: √b- ‘to be’ . . . . . . . . . . . . . . . . 1073.56 a-class root and theme stress: √dìx- ‘breathe’ and √pèl-‘drive’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1093.57 a-class variable stress: √jòk- ‘cry’ . . . . . . . . . . . . . . . . 1103.58 i-class participles: √kàd- ‘smoke’ and √dùb- ‘get’ . . . . . . 1103.59 l ∼ w in nouns: presence of alternation . . . . . . . . . . . . . 1113.60 l ∼ w in nouns: absence of alternation . . . . . . . . . . . . . . 1113.61 l ∼ w in adjectives . . . . . . . . . . . . . . . . . . . . . . . . . . 1123.62 l ∼ w alternation in feminine and neuter nouns . . . . . . . . . 1123.63 l ∼ u alternation in nouns . . . . . . . . . . . . . . . . . . . . . . 1133.64 2nd cycle: /("po.w@n)-l-∅/ → [("] . . . . . . . . . . . 1153.65 2nd cycle: /("po.w@n)-l-∅/ → [("pow.n@l)] (no l ∼ u) . . . . . 1153.66 Noun system: masculine nouns . . . . . . . . . . . . . . . . . . 1173.67 Noun system: neuter nouns . . . . . . . . . . . . . . . . . . . . . 1173.68 Noun system: feminine nouns . . . . . . . . . . . . . . . . . . . 1173.69 Participle √pòwn-í-l- ‘to fill’: gender-number exponents . . . 1183.70 Nouns in nominative case: gender-number exponents . . . . . 1193.71 Adjectives in NM Slovenian (nominative case) . . . . . . . . . 1194.1 -uje∼(u)va and -eva as scnd.imperf in NM Slovenian . . . 1244.2 Imperfective (default: -a) vs. semelfactive -ni∼ne . . . . . . . 1254.3 Imperfective (default: -a,-e) vs. scnd.imperf -uje∼(u)va . . 1254.4 Imperfective (default: -i,-e) vs. scnd.imperf −eva . . . . . . 1254.5 √b- ‘be’ and √S- ‘go’ under two cycles ( . . . . . . . . . 1364.6 /i/-deletion in word-final position ( participles) . . . . . 1384.7 No /i/-deletion ( nouns/adjectives) . . . . . . . . . 1394.8 De-participial adjective . . . . . . . . . . . . . . . . . . . . . . . 145ixList of Figures1.1 Y-model of grammar . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 [[[√root] x] y] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.3 [[[[√root] v0] Asp0] T0]: Before head movement applies . . 51.4 [[[[√root] v0] Asp0] T0]: After head movement applied . . . 61.5 PF-branch modularity . . . . . . . . . . . . . . . . . . . . . . . . 81.6 Building [[[[√root] v0] Asp0] T0] . . . . . . . . . . . . . . . . 91.7 [[[[√root] v0] a0] n0] . . . . . . . . . . . . . . . . . . . . . . . . 101.8 [[[[[√root] v0] Asp0] Ptc0] a0] . . . . . . . . . . . . . . . . . . 112.1 Vowel system of NM Slovenian . . . . . . . . . . . . . . . . . . 183.1 Structure of ["xRànlá] . . . . . . . . . . . . . . . . . . . . . . . . 983.2 Structure of ["xRàn@´l] . . . . . . . . . . . . . . . . . . . . . . . . 983.3 Structure of ["pòw@nlí] . . . . . . . . . . . . . . . . . . . . . . . 993.4 Cyclic derivation of ["pòw@nlí] . . . . . . . . . . . . . . . . . . . 993.5 Cyclic derivation of *["pòw@n@´l] . . . . . . . . . . . . . . . . . . 1003.6 Cyclic derivation of [u-"m@´R@`l] . . . . . . . . . . . . . . . . . . . 1004.1 Verb formation: [[[[√root] v0] Asp0] T0] . . . . . . . . . . . . 1224.2 Participial formation: [[[[√root] v0] Asp0] Ptc0] . . . . . . . 1224.3 Participle /√xRan-i-l-i/ ‘having fed (’ . . . . . . . . 1274.4 Cyclic application in Cophonology Theory . . . . . . . . . . . 1294.5 Embick’s phonological grammar . . . . . . . . . . . . . . . . . 1304.6 Constraint ranking (/i/-deleting phonology) . . . . . . . . . . 1394.7 Constraint ranking (/i/-preserving phonology) . . . . . . . . . 1404.8 Deriving /"xRan-i-l-i/→["xRa.n@l] in Cophonology Theory . . . 1434.9 Single phase cycle with no π-cycles . . . . . . . . . . . . . . . . 1484.10 Single phase cycle with a π-cycle (π1) . . . . . . . . . . . . . . 1484.11 Deriving the participle ["√xRan-@l] ‘having fed’ . . . . . . . . . 149xAcknowledgementsWhat pushed me to analyze the distribution of the high vowel /i/ and schwa,and their intricate connection with cyclicity in Novo mesto Slovenian was asudden, personal realization that I had no adequate understanding of eventhe basic processes that govern the phonology and morphology in my nativelanguage, Novo mesto Slovenian. This recognition of my ignorance first trig-gered a basic survey of the morpho-phonological properties of the language,which turned out to be of substantial interest for the morphology-phonologyinterface (the ‘PF-interface’, in my case), culminating in the present thesis.I first need to state that this thesis could not have been produced in itscurrent form and in such a timely fashion were it not for my supervisor,Gunnar Hansson. I am very grateful to him for his guidance in my research,and especially for helping me organize my second year at UBC so as toleave enough time for thesis writing. I will miss our meetings and, espe-cially, the discussions on the phonology-morphology interface and cyclicity.A big thanks also needs to go out to my remaining committee members:Doug Pulleyblank always pushed me to consider a broad range of theo-retical approaches so that I would ultimately arrive at the most objectivepicture possible, whereas Joe Stemberger very helpfully supplied me withjust the right counter-examples from Slovenian that substantially improvedmy analyses. I have also had several discussions on Slovenian phonologywith Peter Jurgec, for which I am grateful. In my time at UBC, I wasgreatly influenced by Michael Rochemont’s course on syntax, which enabledme to properly understand and appreciate the Minimalist Program. Also, Iwould like to thank Michael for the discussions we have had on Phase Theoryand the grammar in general, which has had an impact on the present the-sis. I am also grateful to Rose-Marie Déchaine for all the morpho-syntacticdiscussions that we had when I initially became interested in the topic.I want to express how grateful I am to my wife, Staša, who was stuckwith me throughout my thesis-writing days and helped me keep in touchwith reality by pointing out the small things in life that ultimately matterthe most. Finally, I want to extend a warm thanks to my family in Sloveniafor their continuous support in my academic pursuits abroad.xiChapter 1IntroductionThe idea that phonological processes may be specific to some morpholog-ical constructions but not others is not novel: in the founding documentof generative phonology, Chomsky & Halle (1968) needed to make a dis-tinction between affixes that trigger stress re-assignment in English andthose that do not, which was formally implemented by phonological rule-indexation to morphological contexts. This general topic of ‘construction-specific’ phonology has remained with the field throughout its develop-ment: formal answers for this phenomenon have been sought in LexicalPhonology (Kiparsky 1982a,b; Mohanan 1986), and later in Stratal Op-timality Theory (Kiparsky 2000; Bermúdez-Otero 2011), in CophonologyTheory (Orgun 1996; Inkelas et al. 1997; Inkelas & Zoll 2007; Inkelas 2008,2011) and Constraint Indexation within Optimality Theory (Itô & Mester1995, 1999; Pater 2000, 2007, 2010). A recent view that attempts to dealwith construction-specific phonological processes takes the specific view ofthe morphology-phonology interface put forth by Distributed Morphology(Halle & Marantz 1993); an example of this is Embick (2013),In this thesis, we will also adopt the view of the interfaces advocatedby Distributed Morphology. We will discuss data from Novo mesto Slove-nian, a South Slavic language, which reveals a construction-specific processof unstressed vowel deletion. These data are of much interest for theoriesof construction-specific phonology because they reveal that the same word-final affixal vowel, viz. masculine plural /-i/, deletes in some morphologicalconstructions but not in others. This generates the question of what the‘domain’ for a construction-specific phonological process should be. We willdiscuss how the standard approaches fail in modelling this process of voweldeletion, which should set specific guidelines for future research on this andrelated phenomena. At the very end of the thesis, we also show that itis possible to formulate a plausible generalization of this problematic pro-cess of vowel deletion by referring to syntactic locality, viz. Phase Theory(Chomsky 2001, 2008). A tentative formal implementation of this idea isoffered, which seeks to define the spell-out of phonological domains primar-ily through syntactic phases. This thesis places emphasis on the hope that11.1. Interface: Distributed Morphologylocality boundaries for phonological processes may be found in the indepen-dently motivated locality constraints on syntactic computation, making fora modularly distinct, but unified theory of grammar.This introductory chapter has two sections. Section 1.1 introduces thecrucial concepts of Distributed Morphology that are followed in this thesis,most of which a reader familiar with the theory may skip, while section 1.2lays out the goal of the thesis in more detail.1.1 Interface: Distributed MorphologyIn the chapters that follow, this thesis adopts the general outlook providedby the Minimalist Program (Chomsky 1993, 1995, 2000, 2001), togetherwith a specific view of the interface between morphosyntax and phonology,viz. Distributed Morphology (Halle & Marantz 1993, 1994; Embick 2010).This section is dedicated to a general description of the basic assumptionsthat will be made; this seems especially necessary since there is no fullyunified theory of Minimalist grammar or Distributed Morphology. The as-sumption that will underly all the analyses to follow is that natural humanlanguage is the reflection of a cognitive language faculty, essentially a gram-mar, that can be teased apart into several modularly distinct components.We will assume a ‘Y-model’ of grammar, which stems from work in theMinimalist Program:NumSpell-OutPF LFFigure 1.1: Y-model of grammarIn a Y-model of grammar, syntactic computation provides the ‘central’linguistic structure, which is then ‘spelled out’ to the two interfaces: thePhonetic Form (PF) is used to interface syntax with the ‘sensory-motor’ sys-tems, while the Logical Form (LF) interfaces syntax with the ‘conceptional-intentional’ systems (Chomsky 1995). The grammar is fed by a Lexicon,which includes the lexical information which syntax requires to project struc-ture. The ‘Numeration set’ Num represents the input where the lexicalcomponents for a particular sentence are collected; out of these lexical prim-itives, the syntax projects structure, which is subsequently spelled out to thetwo interfaces. As a terminological point, it should be noted that ‘Phonetic21.1. Interface: Distributed MorphologyForm’ does not involve anything ‘phonetic’, but its role is to convert syntac-tic information into phonological representations, which only then interfacewith the sensory-motor system, giving rise to articulatory (motor) move-ments. This thesis will be concerned with the PF-branch of computation.Distributed Morphology subsumes the notion of a Y-model of grammarand lays out very specific claims about the interface between morphosyn-tax and phonology. It crucially assumes Late Insertion (Halle & Marantz1994: 275), which implies that morphemes are not yet associated with anyphonological features in the syntax or the lexicon, and that any phonologicalfeatures are mapped onto the morphemes at the level of PF; before that,the morphemes are only represented by bundles of morphosyntactic features.This process of mapping phonological features to the morphemes is termedVocabulary Insertion (often abbreviated as ‘VI’). The process of mappingphonological features to morphosyntactic ones is formalized by a system ofVI-rules. Consider the following two VI-rules that insert /d/ for the Englishpast tense and /z/ for the English third person singular present tense suffix:(1) Vocabulary Insertion[past] ↔ d[] ↔ zThis way, syntax does not need to compute any features that are not re-quired in that part of the grammar. Distributed Morphology also contendsthat VI-rules may be underspecified (Halle & Marantz 1994: 276) for mor-phosyntactic features. This means that a rule may specify only [present],or it may specify [], which means that the more underspecifiedexponent could be inserted for several different morphemes, as long as theycontained the feature [present]. This is an important mechanism that isvery valuable for overcoming the danger of freely positing homophonous ex-ponents. As regards homophony throughout this thesis, we will subscribeto the following principle laid out by Embick (2003):(2) Avoid Accidental Homophony (Embick 2003: 156)Learners seek to avoid accidental homophony; absent evidence to thecontrary, identities in form are treated as systematic.Positing an accidentally homophonous exponent must generally be groundedin good independent (morphosyntactic and morphosemantic) reasons. Butlet us turn back to the two rule specifications, [present] and [].Above we mentioned that both are valid exponent specifications for VI-rules, but they make different predictions, viz. the more underspecified ex-31.1. Interface: Distributed Morphologyponent would be inserted for more different morphemes that bear the fea-ture [present]. Note, however, if both these exponents existed in the samegrammar only one of them could apply for a given morpheme. In case ofa morphosyntactic node that would contain a third person singular presenttense morpheme, only the rule which specifies [] could apply,crucially because the rule which specifies just [present] is less specific, andso cannot apply. This is regulated by the Subset Principle (Halle 1997):(3) Subset Principle (Halle 1997: 128)The phonological exponent of a Vocabulary item is inserted into amorpheme in the terminal string if the item matches all or a subsetof the grammatical features specified in the terminal morpheme. In-sertion does not take place if the Vocabulary item contains featuresnot present in the morpheme. Where several Vocabulary items meetthe conditions for insertion, the item matching the greatest numberof features specified in the terminal morpheme must be chosen.A key tenet of Distributed Morphology is also its explanation of word-formation. Distributed Morphology denies the ‘morphological word’ the sta-tus of a theoretical primitive, and rather seeks to explain all word-formationas the reflection of independently motivated syntactic processes: specifi-cally, morphological words are complexes of syntactic heads, adjoined toone another through syntactic movement operations (Embick & Noyer 2001;Embick 2007, 2010), or also lowering operations (Skinner 2009). For in-stance, consider the morphological structure [[[√root] x] y]:yPy0 xPx0 √root→ head-movement → yPy0x0√root x0y0xPx0 √rootFigure 1.2: [[[√root] x] y]On the left-hand side is the syntactic structure prior to any movement hasapplied. On the right-hand side, is the resulting structure. To derive thisresult from the syntactic structure on the left, a process of successive cyclichead movement has to apply. The root of the structure moves and adjoins tothe head x0, which yields the following adjunction structure: [[√root] x].After that, the newly created adjunction structure moves again and adjoins41.1. Interface: Distributed Morphologyto the higher head y0, yielding [[[√root] x] y]. Since the morphemes inthis structure are essentially syntactic heads, they must abide by the HeadMovement Constraint (Travis 1984), which states that no intermediate headmay be skipped when moving to a higher head; this also explains why theentire adjunction complex of heads is always pied-piped to the higher tar-get of head movement. As a side-effect, this syntactic phenomenon explainswhy rigid orderings of morphemes may be observed in words. Such an ap-proach to morphology makes Distributed Morphology a syntactic approachto word-formation along the lines of Lieber (1992), and contrasts it heavilywith lexicalist approaches (Halle 1973; Aronoff 1976; Lieber 1980; Anderson1982; Kiparsky 1982a), where word-formation more or less takes place inthe lexicon; see Marvin (2002: 11) for an overview and comparison withDistributed Morphology.This brings us to the discussion of roots. Roots are treated as acat-egorial heads (Marantz 1997; Embick & Halle 2005; Embick 2010; Harley2014) which need to be categorized by a categorial head. The basic cate-gorial heads are the nominalizer n0, the verbalizer v0 and the adjectivizera0. Heads that are used to construct tense, participles, mood constructions,and so on, are treated as non-categorial functional heads. Let us considera sample derivation that will shed light on these relations between differentheads and will also illustrate the head movement that was discussed above.TPT0 AspPAsp0 vPv0 √rootFigure 1.3: [[[[√root] v0] Asp0] T0]: Before head movement applies51.1. Interface: Distributed MorphologyTPT0Asp0v0√root v0Asp0T0AsPAsp0 vPv0 √rootFigure 1.4: [[[[√root] v0] Asp0] T0]: After head movement appliedThis example illustrates how the morphological word [[[[√root] v0] Asp0]T0], a verb, is derived. In the first syntactic tree above, the root is in a po-sition where one would usually find the verbal head V0 in other approachesto syntax, and indeed it performs the same role, with the exception that itis an acategorial head; it may even project up to √P, as it may c-commandan object. But the crucial aspect of such an analysis for us is that it isacategorial, and it needs to be categorized. It moves to the verbalizer v0,which, on the other hand, is categorial, and it categorizes the root, makingit verbal. The movement then proceeds to the aspectual head Asp0 and sub-sequently to the tense head T0, both of which are non-categorial functionalheads. It is usually assumed that the agreement head Agr0, expressing theφ-features (gender, number, person) is adjoined to the complex of heads (ifthey are clausal predicates that seek agreement such, as verbs and partici-ples) after syntactic spell-out, at PF (Halle & Marantz 1993; Embick 2010;Arregi & Nevins 2012), and we will make that assumption, as well.Nouns or adjectives are formed in exactly the same way. Take the nounbreakability, which can be decomposed into break-able-ity, essentially a verb,an adjective and finally a noun, where only the verb operates on the root.Such a formation starts out as the syntactic structure [nP n0 [aP a0 [vPv0 √break]]], and emerges as the adjunction complex [[[[√break] v0] a0]n0] after head movement has applied. What must happen at this pointis the process of converting this adjunction complex, which is a syntacticformation, to information that can be processed by phonology. The firststep is that of linearization: Embick (2010: 32-35) assumes a process oflinearization that is in line with Sproat (1985) and Marantz (1984, 1988),which means that the heads must be concatenated so that they can be inter-preted as a linear string. We will follow Embick & Noyer (2001) and Embick(2007) by using two concatenation operators, viz. ⊕, which concatenates61.1. Interface: Distributed Morphologyheads in an adjunction complex (used for ‘word internal’ concatenation),and ⌢, which concatenates heads not merged in an adjunction complex(used for ‘word external’ concatenation, for ordering ‘words’). What is im-portant to note is that these are binary operators, which means that theyyield maximally binary concatenation statements. The concatenation of theadjunction complex [[[[√break] v0] a0] n0], therefore, yields the followingconcatenation statements:(4) Concatenating the heads in [[[[√break] v0] a0] n0][[[[√break] v0] a0] n0] → √break⊕v0, v0⊕a0, a0⊕n0Such a process of concatenation might function in binary steps, as repre-sented above, but crucially it outputs a list of heads that reflect a linearprecedence relation. For more on this see subsection on the locality of expo-nence below. The structure that we have produced in (4) may be precededor followed by other heads (or, adjunction complexes) in the syntax. Forinstance, if we wish to capture the fact that it is preceded by, say, a deter-miner head, D0, we use the word external operator ⌢ to do this, as follows:D0⌢[√break⊕v0, v0⊕a0, a0⊕n0].The next step is that of Vocabulary Insertion, as mentioned before.Embick (2010: 42) explains that VI-rules apply in an ‘inside-out’ fashion,which means that they proceed derivationally (or rather, as a reflection ofthat derivation) and start with the most embedded head and the work theirway up. In this case, they have to start with the root:(5) VI-rules applying to [√break⊕v0, v0⊕a0, a0⊕n0]a. Step #1: [√break] ↔ breIkb. Step #2: [v0] ↔ ∅c. Step #3: [a0] ↔ @bIl /v0d. Step #4: [n0] ↔ Iti /a0In these four steps, the exponents are inserted for their corresponding mor-phemes (syntactic heads). Vocabulary Insertion and head concatenationbefore it are operations that illustrate the crucial task of the PF-interface,which is one of converting morphosyntactic information to the form thatcan be processed by phonology.After Vocabulary Insertion has inserted the relevant exponents, the stringof exponents is ready for phonological processing. The output of VI-rulesessentially constitutes an input to the phonological component. This createsthe following overall conception of the PF-branch computation in terms ofmodularity:71.1. Interface: Distributed MorphologySpell-Out of syntax↧▸ Linearization▸ Vocabulary Insertion z→ ▸ Phonological processingFigure 1.5: PF-branch modularityA note on terminology is on order: it is important to distinguish between‘spelling out syntactic structure’ and the ‘spell-out of a morphosyntactichead’. The former refers to sending the syntactic structure to the interfacesfor processing (linearization, etc.), while the latter simply refers to VI-rulesinserting a certain exponent for a certain morphosyntactic head; this lastdefinition has nothing to do with ‘spelling out’ syntactic structure to theinterfaces, but it is used nonetheless.Much work on Distributed Morphology assumes a system of phono-logical re-write rules to formalize phonological processing, in the sense ofChomsky & Halle (1968), including Halle & Marantz (1993) and Embick(2010). However, in this thesis, we shall assume that the phonologicalcomponent is in fact an Optimality-Theoretic grammar, in the standardsense (Prince & Smolensky 2004), following approaches such as that bySvenonius & Bye (2010) and Bye & Svenonius (2012), which subsume someversion of Distributed Morphology, but with an OT-based phonologicalgrammar. We shall also assume, in the spirit of Cophonology Theory (Orgun1996; Inkelas et al. 1997; Inkelas & Zoll 2007; Inkelas 2008, 2011), that phono-logical computation is generally free of morphosyntactic information, such asreference to individual morphemes, with the exception of ‘telling apart’ themorphological primitives, such as exponent boundaries and root-affix dis-tinctions, which for some processes still seem necessary. Though even thesewill be subject to morphosyntactic locality constraints (see the following sub-section). In this way, such an approach by default argues against lexically in-dexed constraints, be they merely faithfulness constraints or faithfulness andmarkedness constraints (Itô & Mester 1995, 1999; Pater 2000, 2007, 2010).The approach advanced here will, on the other hand, make use of phono-logical cycles, as also employed in Cophonology Theory, Lexical Phonol-ogy (Kiparsky 1982a,b; Mohanan 1986) and Stratal OT (Kiparsky 2000;Bermúdez-Otero 2011), and are subsumed in most work on Distributed Mor-phology, stemming back to Halle & Vergnaud (1987) and Chomsky & Halle(1968). It should nevertheless be pointed out that the conclusions with re-spect to the ‘domains of phonological spell-out’ (see sections 4.2 and 4.5)could be reached either under an OT-based or rule driven phonological com-81.1. Interface: Distributed Morphologyponent, and are hence relevant for both approaches.Now that we have introduced the basic assumptions that will be sub-sumed throughout this thesis, it is important to clarify one more thing thatpertains to the treatment of roots. Marantz (1997), Embick & Halle (2005)and Embick (2010) assume that roots, unlike all other morphemes, are notsubject to Late Insertion, but enter the syntactic derivation with a pre-specified phonological make-up. This is an assumption that is rooted inthe idea that root suppletion does not exist. This is far from uncontro-versial: Bonet & Harbour (2012) and Merchant (2015) convincingly demon-strate that root suppletion is a reality, and Merchant (2015) together withHarley (2014) argues that roots must be subject to Late Insertion, just likeany other morpheme. In this thesis, we will side with Merchant (2015)and Harley (2014), in particular, we may follow Harley and assume thatroots are nothing more than sets of random integers (e.g. √23875, √98045,etc.), which receive their phonological make-up at PF and their semanticinterpretation at LF. But for ease of exposition, we will represent each rootby the most appropriate English translation of its semantics: for instance,the Slovenian root √xRan- ‘feed’, will be represented as √feed in the mor-phosyntax. While this is an important point to make, it is not crucial tothe overall argument of the thesis.Syntactic localityA common claim in Minimalist syntax is that syntactic structure is notspelled out globally, but rather in cycles, which are termed phases (Chomsky2000, 2001, 2008). According to Chomsky, this is a reasonable way of en-coding locality as such cyclic spell-out must be rooted in principles of com-putational efficiency. Certain morphosyntactic heads are ‘phase heads’ andtrigger spell-out to the interfaces.TPT0 AspPAsp0 vPv0 √P√root DPÐ→ TPT0 AsPAsp0 vPv0√root v0√P√root DPFigure 1.6: Building [[[[√root] v0] Asp0] T0]91.1. Interface: Distributed MorphologyIn the structrure in Figure (1.6), we have the syntactic projections whoseheads make up a typical verb – the root also takes a DP object complement.Note that on the left-hand side, the root undergoes movement and adjoinsto v0; this step is shown on the right-hand side. Notice also that the DPobject is spelled out (represented by a frame-box) as soon as that happens.This is because v0 is a ‘phase head’. The initial idea was that there weretwo phase heads, v0 and C0, so that every clause would be computed intwo cycles. However, this assumption was extended to all categorial heads(Marantz 2001, Marvin 2003, Marantz 2007, Embick and Marantz 2008),which means that v0 as well as n0 and a0 trigger phase spell-out to theinterfaces. This, however, does not hold for non-categorial functional heads:in Figure (1.6) above, only v0 triggers phase spell-out, while Asp0 and T0will not do so when the derivation continues. Phase heads generally triggerspell-out of their complements, and this only happens once the final maximalprojection of the phase head has been constructed. It should be understoodthat in Figure (1.6), the DP object was only spelled out when vP was built.This means that any elements in the specifier of vP, or adjoined to vP canescape spell-out in this phase, and are so called the edge of the phase.Notice that the root itself is also the complement of the phase head v0,and yet it escapes spell-out in this phase. We will subsume the assumptionson phases that are outlined in Embick (2010, 2013) and Marantz (2013).(Marantz 2013: 98-99) notes that the root must always be processed in thesame cycle as the categorial head that categorizes it. If we, therefore, havea case of a noun that is derived from an adjective and the adjective from averb (like break-able-ity), we predict the following phase spell-outs:nPn0 aPa0 vPv0 √rootnPn0a0v0√root v0a0n0aPa0 vPv0√rootFigure 1.7: [[[[√root] v0] a0] n0]In the example above we see three phases: the verbal phase, the adjectivalphase and the nominal phase. The construction of such structure of courseproceeds derivationally: first the root is adjoined to v0, but the first phase101.1. Interface: Distributed Morphologyspell-out only occurs when they are merged with a0. In the second phase,a0 is spelled out when it is merged with n0, and n0 undergoes spell-outas the final phase.1 Embick (2010: 18) states that ‘when a cyclic head ismerged [to the structure], it triggers the Spell-Out of cyclic domains in itscomplement’, which explains why the root itself is not spelled out, as rootsare not cyclic (i.e. phase) heads.2 We can now formally refer to this as theDomain of Phase Spell-Out:(6) Domain of Phase Spell-Out (Embick 2010, 2013; Marantz 2013)A phase head will trigger spell-out of domains that host a phasehead.It is quite crucial to observe what this principle predicts for the spell-out ofnon-phase heads, as Embick (2010, 2013) notes. Consider an adjective thatis built on a verbal phase, where the verbal phase also hosts an aspectualhead and a participial head (so the adjective is built on a participial base):a0Ptc0Asp0v0√root v0Asp0Ptc0a0Figure 1.8: [[[[[√root] v0] Asp0] Ptc0] a0]Notice that the only two phase heads in this adjectival formation are theverbalizer v0 and the adjectivizer a0, simply because the other heads are notcategorial. The domain of the first, verbal phase is the root and anythingbetween itself and the next phase head, which constitutes the heads Asp0and Ptc0. Embick (2010: 51) terms such heads that fall into the phasedomain of a preceding phase head as the edge+ of a phase domain.1Not only parts of the adjunct complex of heads are spelled out through phases, butthe entire syntactic structure is, as well, but this is difficult to represent graphically inFigure (1.7), so it is avoided.2The phase heads, such as a0 and n0 above, are not strictly speaking in the complementposition once head movement applies, but see Marvin (2002: 26), who explains that the‘complement’ is too specific a term and proposes that this should be defined as a ‘phasedomain’ in some other way.111.1. Interface: Distributed MorphologyThe application of phase spell-out also has direct repercussions for thedefinition of Bracket Erasure in Distributed Morphology. Bracket Erasureessentially ‘erases’ any trace of the morphological affiliation of exponentsas the derivation proceeds higher up the tree, and it has been present, inone form or another, in most theories of the phonology-morphology inter-face (Chomsky & Halle 1968; Kiparsky 1982a,b; Inkelas & Zoll 2007; Inkelas2008). Embick (2013: 8) argues for a principle of Phase Impenetrability forPhonology, which in essence dictates that all the heads (i.e. morphemes)present in the same phase cycle may be identified as roots or affixes. Giventhe structure in Figure (1.8), the first phase cycle would include √root,v0, Asp0, and Ptc0, but not a0. When a0 would be merged to the previouslyconstructed structure, √root, v0, Asp0, and Ptc0 would be spelled out tothe interfaces and would undergo linearization and Vocabulary Insertion atPF (and also phonological processing; see section 4.2 on this). When a0would be spelled out, the root would lose its identity as a morpheme andwould only be identifiable as a string of phonological segments, while v0,Asp0, and Ptc0 would still be able to interact with a0, until a new phasehead is spelled out. For details on this, see Embick (2013). Since all ourdata will only represent formations within a single phase cycle, this is notdiscussed further here. But Embick’s Phase Impenetrability for Phonologyis essentially a type of Bracket Erasure principle.Probably the most interesting insight of phase cyclicity is the domain inwhich spell-out proceeds: while structure is built in purely derivational steps,as dictated by the operation Merge, spell-out is not purely derivational. Asnoted by Marantz (2001), the idea that syntax is spelled out in phases, asdefined by Chomsky, implies that there is no rigorous overlap of the strictlyderivational mechanism of Merge and the spell-out of the structure it builds,which makes human syntax less like the spell-out employed in theories thatclosely follow the principles of formal logic, such as Montague Grammar(Montague 1970): Marantz explains that every structure building operationis accompanied by immediate spell-out in such theories, which is unlikephases, which must ‘delay’ spell-out. If phase spell-out is the best way ofcapturing locality in human syntax, it will represent a point of departure(among many) between natural and artificial language.Locality of exponenceIn the previous section, we discussed an important locality constraint on syn-tactic computation, viz. phase spell-out. Here, on the other hand, we willbriefly define the assumptions on the locality of exponence; in other words,121.1. Interface: Distributed Morphologythe locality of Vocabulary Insertion. This will touch on the topic of allo-morphy. To begin with, it should be noted that, throughout this thesis, theterm ‘allomorphy’ refers to different exponents of the same morphosyntactichead, which cannot be predicted phonologically. In that sense, an allomorphwill refer only to a morphologically determined exponent, unless specifiedotherwise. An example of such allomorphy would be the use of a specialexponent for the second person singular morpheme in the perfect verbalforms in Latin – this example is taken from Embick (2010: 70-75). Latinsystematically employs [-s] as the exponent of the second person singularagreement morpheme, viz. Agr0[]: consider the present tense indicative[√am-a:-s] ‘you love (’ and the corresponding perfect form [√am-a:-v-isti:], where the second person singular exponent is [-isti:], different fromthe exponent used in the present indicative form. The morphosyntacticstructure for the perfect form supplied by Embick is the following:(7) [[[[[√root] v0] Asp0[perf]] T0] Agr0[]]Crucially, Asp0[perf] is exponed by [-v], T0 by [-∅] and Agr0[] by [-isti:].A typical analysis in Distributed Morphology would say that VI-rules aresensitive to the context of the Asp0[perf] head, and in its presence insertthe exponent [-isti:] for Agr0[], instead of the default [-s]. The localityof VI-rules now comes into play, as it becomes important to define in whatway VI-rules ‘see’ the presence of neighbouring heads in the string.In this thesis, we will subscribe to the Span Adjacency Hypothesis, asdefined by Merchant (2015), who takes the notion of a ‘span’ from the workof Svenonius (2012, 2013). This approach defines VI-rule locality in termsof ‘spans of heads’ that are adjacent to the point of insertion. A span ofheads is defined in the following way by Merchant (2015):(8) Span (Merchant 2015: 288)3Let T be an ordered n-tuple of terminal nodes ⟨t1, ..., tn⟩ such thatfor all t ∈ T, t = t1 or t is an element of the extended projection of t1.a. For all k = 1...n, tk is a span. (Every node is a trivial span.)b. For any n > 0, if tk is a span, then ⟨tk, ..., tk+n⟩ is a span.This definition of a span crucially expresses that a single head or a numberof different heads may form a span, but that, in the former case, the span3A quick note on terminology is required: a (terminal) node is equivalent to any lin-earized head here, and an extended projection may simply be understood as any collectionof heads between two points in a linear string for our purposes.131.2. Goal of the thesisis trivial. It also expresses that a span between two heads in a linear stringmust by default include all the heads in between, which entails that no headscan be ‘skipped’ when referring to heads in spans. According to Merchant(2015: 294), ‘allomorphy is conditioned only by an adjacent span’. Returningto the Latin example that we discussed above, this approach to the localityof exponence predicts that the VI-rule inserting [-isti:] in the perfect formwill only do so when adjacent to the span encompassing Asp0[perf] and T0.We could represent this span through the concatenation statement with theword-internal concatenation ⊕-operator, as in ‘Asp0[perf]⊕T0’, but sincethe application of this operator essentially yields a list of linear precedencerelations, we may represent spans as n-tuples. This is something that wewill follow throughout this thesis. In this case, we only need to state theordered pair ⟨Asp0[perf], T0⟩:(9) Latin allomorphy[Agr0[]] ↔ s[Agr0[]] ↔ isti: / ⟨Asp0[perf], T0⟩The default VI-rule that inserts /-s/ as the second person singular expo-nent specified no conditioning environment, while the more specific VI-rulespecified the span ⟨Asp0[perf], T0⟩ as the context in which /-isti:/ will beinserted for the second person singular agreement head. This is the wayallomorphy will be derived throughout this thesis.Embick (2010: 49) proposes a different locality constraint on exponence,one which only permits VI-rules to see immediately adjacent single nodesand not spans, which Merchant (2015) terms the Node Adjacency Hypoth-esis. Embick’s approach does appear more restrictive, but it requires theuse of additional stipulative operations such as ‘node deletion (Pruning)’and the use of ‘Readjustment Rules’. Both these operations are relativelyunconstrained and will not be adhered in the analysis of this thesis. Whilethese assumptions need to be specified, it should be noted that the localityof exponence has no relevant bearing on the key proposal of this thesis dis-cussed in section 4.5, and Embick’s view on VI-locality through immediatehead adjacency could very well be adopted there.1.2 Goal of the thesisAs briefly noted in the introduction to this chapter, this thesis is primarilyconcerned with construction-specific phonology. In particular, we will ex-amine how the domains of phonological computation need to be defined in141.2. Goal of the thesisorder to adequately derive construction-specific phonological processes. Thedata for the crucial analyses in this thesis are from Novo mesto Slovenian,a dialect of Slovenian, a South Slavic language. We will show that Novomesto Slovenian contains a process of unstressed /i/-deletion in verbs andparticiples, but not in other word classes such as nouns or adjectives. Forinstance, we will show that the masculine plural ( agreement suffix/-i/ may attach to a participial stem, as in [ptc]-/i/, where it will haveto undergo deletion under specific phonotactic conditions. But the suffix may also attach to a nominal or adjectival stem, as in [adj]-/i/,where it will be preserved in the surface form under the exact same phono-tactic conditions in which it deletes with participial stems. In essence, weare dealing with an affix that attaches to different stems and ‘inherits’ thephonological effects to which that stem subscribes.We will show that the phenomenon described above is problematic forthe standard accounts of construction-specific phonological processes. Forinstance, Embick (2013) assumes Distributed Morphology with a phono-logical grammar that stems from Halle & Vergnaud (1987), which meansthat there is a ‘Cyclic level’ of phonology, feeding a ‘Word level’, whichin turn feeds the ‘Phrasal level’ – a system that in effect mirrors the lev-els in Lexical Phonology (Kiparsky 1982a; Mohanan 1986). In such a sys-tem, construction-specific phonology is tied to the Cyclic level, where ex-ponents may introduce diacritics that trigger specific phonological effects.In [[[√root] x] y ], x can trigger a pass through some phonology, where/root-x/ are processed to the exclusion of y – this is because VocabularyInsertion proceeds from the root outwards, and when the exponent for xis inserted it triggers a pass through the phonology, but no exponent hasbeen inserted for y at this point. In our scenario of [ptc]-/i/ vs. [adj]-/i/above, it is clear that a phonological process, triggered by some exponentbefore /-i/ is inserted, cannot effect /-i/ itself. After /-i/ is inserted,the whole set of exponents is sent to the Word level phonology, a level ofphonology fixed for the entire grammar, where construction-specific phono-logical effects cannot be derived. The idea that /-i/ can ‘inherit’ theconstruction-specific phonology of its stem cannot be derived in such a sys-tem. Cophonology Theory (Inkelas & Zoll 2007; Inkelas 2008, 2011) cannotreadily derive this phenomenon either – see section 4.4 on specifics.The problem itself appears to stem from the way the spell-out of phono-logical domains is defined. The distinction between the Cyclic and Wordlevels under-generates and over-generates at the same time: situating /i/-deletion on the Cyclic level would fail to delete the /-i/, while situatingit on the Word level would trigger /i/-deletion across all word classes in the151.2. Goal of the thesisgrammar. Section 4.4 defines this problem more explicitly. To begin toaddress this problem in section 4.5, we subscribe to the over-reaching re-search program laid out by Marvin (2002), Marantz (2007), Samuels (2009),Piggott & Travis (2013) and Newell & Piggott (2014), which seeks to in-terpret the phasal cyclicity of syntax (Chomsky 2001, 2008) as a localityboundary that defines phonological domains. We will demonstrate that avery simple generalization on the /i/-deletion facts can be formulated if ref-erence to phases is invoked. We will suggest that phasal cycles should replacethe rigid distinction between Cyclic and Word level phonology as phonologi-cal spell-out domains. We will also formulate a brief tentative proposal thatoffers a formal explanation of why the is allowed to ‘inherit’ its stem’sphonology, which will heavily rely on the locality as set by phase cycles.In this way, the contribution of this thesis is twofold. It presents an em-pirical challenge for the standard accounts of construction-specific phonolog-ical processes and, at the same time, it provides support for Phase Theoryas a crucial model of locality that has repercussions ‘all the way down’. Morespecifically, this thesis demonstrates that phase cycles seem to represent acrucial locality boundary for construction-specific phonological processes.This thesis has four remaining chapters. In chapter 2 of this thesis, wedevote a generous amount of time to discussion of the phonological general-izations that the Novo mesto Slovenian data offer. It will be shown that theverbs and participles undergo a process of unstressed /i/-deletion. Chapter3 presents a phonological analysis of this process of /i/-deletion and howit interacts with other relevant processes in the phonology of Novo mestoSlovenian, most prominently with schwa epenthesis. In Chapter 4, we turnto examine the PF-interface: some crucial points about the morphosyntaxof Novo mesto Slovenian are made, and the spell-out of phonological do-mains is discussed. It is shown that the data on /i/-deletion from Novomesto Slovenian cannot be accounted for by the standard approaches toconstruction-specific phonology, which is followed by a brief proposal onhow to derive such processes. Chapter 5 concludes the thesis.16Chapter 2Novo mesto SlovenianNovo mesto Slovenian is a dialect of Slovenian, a South Slavic languagesituated in Central Europe in Slovenia. It is spoken in the area of thetown Novo mesto, which is located in the Dolenjska region in the south-eastof the country, also referred to as Lower Carniola. While the Dolenjskadialects have received several treatments, which are mainly concerned withthe diachronic aspects of their phonological processes (Ramovš 1995; Lenček1982; Toporišič 1992; Logar 1993; Greenberg 2000, 2006), the dialect of Novomesto has remained undiscussed. In terms of synchronic studies, StandardSlovenian, which is the language used in formal setting, has been studiedthe most (Toporišič 2000; Herrity 2000; Jurgec 2007a). Overall, Novo mestoSlovenian has eluded a basic description.The following sections present a basic overview of the segmental inven-tory of Novo mesto Slovenian, but, more importantly, also reveal data thatshed light on the distribution of the vowels [i] and [@]. In particular, we exam-ine how their distribution correlates with different morphological contexts.Section 2.1 gives a basic overview of the segmental inventory of Novo mestoSlovenian, and 2.2 discusses the distribution of [i] in verbs and participles.A detailed discussion of schwa is postponed until chapter 3.The data presented in this chapter were selected by the author who isa native speaker of Novo mesto Slovenian. In addition, all the data werealso checked against grammaticality judgements of two to three other nativespeakers of the language.2.1 Basic overviewThis section presents a basic overview of the vowel and consonantal inven-tories of Novo mesto Slovenian (henceforth, NM Slovenian). Section 2.1.1discusses the basic distinctive properties of vowels in NM Slovenian, includ-ing quality, quantity and tone. Section 2.1.2, on the other hand, presentsthe basic consonantal distinctions.172.1. Basic overview2.1.1 VowelsNM Slovenian is characterized by an eight-vowel system. This system canroughly be captured by the approximated vowel-values, representing thepossible surface forms of vowels below in Figure (2.1):aEeiOou@Figure 2.1: Vowel system of NM SlovenianThis system of vowel qualities is much like that of Standard Slovenian(Toporišič 2000; Šuštaršič et al. 1995, 1999; Herrity 2000). At this point,two more distinctive properties of vowels in NM Slovenian need to be dis-cussed: vowel length and tone specification.In Standard Slovenian, as described in Toporišič (2000: 60), length is acontrastive property of vowels. It can only be contrasted in stressed syllablesand not in unstressed syllables. Recent studies have, however, performedphonetic studies of Standard Slovenian as spoken in the Ljubljana region(the capital city of Slovenia) (Srebot Rejec 1988; Šuštaršič et al. 1995, 1999;Petek et al. 1996; Jurgec 2007a, 2011), and they all show that length is nolonger a contrastive property of Standard Slovenian, at least not the way itis spoken in the Ljubljana region. NM Slovenian seems to reflect a similarstate of affairs in terms of vowel length: NM Slovenian correspondents of thevowel length contrasts that are found in Standard Slovenian, as discussed inToporišič (2000), are either (i) encoded through a quality contrast, or (ii)they are not encoded at all. To illustrate this, I first list the typical vowellength contrasts illustrated by Toporišič (2000) (but the actual examples aretaken from Jurgec (2007a)):short longVas ‘you (’ Va:s ‘village (’Rat ‘fond of (’ Ra:t ‘radian (’slap ‘bad (’ sla:p ‘waterfall (’sit ‘sated (’ si:t ‘sieve (’bit ‘be (inf.verb)’ bi:t ‘being (’kup ‘pile (’ ku:p ‘cup (’Table 2.1: Length contrasts according to Toporišič (2000) (from Jurgec2007a: 31)182.1. Basic overviewThe table in (2.1) presents a list of minimal pairs that illustrate vowel lengthcontrast in Standard Slovenian according to Toporišič (2000). Now, observethe corresponding examples from NM Slovenian listed in the table below:short longv@s ‘you (’ vas ‘village (’R@t ‘fond of (’ Rat ‘radian (’sl@p ‘bad (’ slap ‘waterfall (’sit ‘sated (’ sit ‘sieve (’bit ‘be (inf.verb)’ bit ‘being (’kup ‘pile (’ kup ‘cup (’Table 2.2: Vowel length contrasts (or, lack thereof) in NM SlovenianNotice that the supposed length contrast for [a] is always encoded throughquality in NM Slovenian: [@] contrasts with [a]. For the high vowels [i]and [u], however, no length contrast seems to occur at all: the pairs listedabove are completely homophonous.4 The mid vowels [e], [o] and [E], [O], onthe other hand, can appear to be generally longer in duration than all theother vowels (which has also been claimed for Standard Slovenian (Toporišič2000)). Furthermore, a slight difference between the mid vowel segmentsseems to occur depending on whether they occur in an open or closed sylla-ble, cf. ["pet] ‘five (num)’ vs. ["ne:.xam] ‘stop (’. No contrastivelength can, however, be found in identical syllable structures containing themid vowels, which means that the ‘lengthening effect’ in the mid vowelscould possibly be an entirely phonetic, perceptual matter, and perhaps hasno phonological correlate (though this does warrant a further, but indepen-dent phonetic study). Let us also mention that, like Standard Slovenian,NM Slovenian has no length contrast in unstressed syllables, which meansthat no length interaction is possible in the absence of stress.Whether the mid vowels exhibit actual phonological lengthening or notis not an issue for the present study. It should be noted that the phenomenadiscussed in the sections to come cut directly across the vowel distinctionsbased on quality, and, hence, also any potential length differences. Due tothis and the uncertain status of phonological length of the mid vowels, Iwill not retain the notation of (potential) length in the data presented from4In some cases, it seems that there might be a slight difference in vowel length, but theeffect is most likely due to different tonal specification: cf. ["dìxát] ‘breathe (inf.verb)’vs. ["dvígàt] ‘lift (inf.verb)’, where [ì] in ["dìxát] seems longer than the [í] in ["dvígàt].This is probably a side-effect of the low tone on [i] in ["dìxát], which is phonetically rising.See next page for more on tone.192.1. Basic overviewnow on. Vowel length is, at best, a marginal phenomenon in NM Slovenianphonology, and I, therefore, leave further examination of it to future studies.The second distinctive property of vowels that needs to be discussed istonal specification. Vowels may contrast for tone: specifically, high tone andlow tone. Similar to Standard Slovenian, NM Slovenian seems to place a toneon the stressed syllable of the word, but the final post-stressed syllable in theword bears a ‘boundary tone’, viz. it receives the opposite tone value of theone realized on the stressed syllable, as discussed in Toporišič (1968, 2000),Jurgec (2007a: 71–94) and Becker & Jurgec (2015).5 This yields patternssuch as ["σ´σ`] ∼ ["σ´σσ`] and ["σ`σ´] ∼ ["σ`σσ´], where the final, boundary tone isdependent on the tone of the stressed syllable and, hence, fully predictable.Consider the following examples of verbs, in which the (boundary) tone onthe final syllable is determined by the stressed syllable (the root syllable inthese cases):(10) Tones in NM Sloveniana. L-tone on √dix-["dìx-á-t] ‘breathe (inf.verb)’b. H-tone on √dvig-["dvíg-à-t] ‘lift (inf.verb)’Jurgec (2007a) and Becker & Jurgec (2015) also show that, when the finalsyllable of the word receives stress (and hence a tone), the boundary toneis realized on that same syllable, creating a contour tone. It does not seemthat this is the case in NM Slovenian. Instead, the tone on the final stressedsyllable of a word seems to be a simple high or low tone.6 Consider thefollowing example of participles with final stress (in these cases, the themevowel [-i] is stressed):(11) Tones on final stressed syllables[√xRa"n-ì-l] ‘feed (’[√va"b-ì-l] ‘invite (’[√pow"n-ì-l] ‘fill (’[u-√gas"n-ì-l] ‘turn off (’Since the sections to come will mainly be concerned with processes thataffect suffixes, we shall chiefly be dealing with a fixed set of suffixes. This5Strictly speaking, in Toporišič (1968, 2000) only the immediate post-stressed syllableof any word is discussed, while Jurgec (2007a) and Becker & Jurgec (2015) show theexistence of ‘boundary tones’ on the very final syllables of words.6I urge future phonetic studies to examine this in detail.202.1. Basic overviewmeans that no tonal variation will be found on stressed suffixes (as illus-trated by the suffix [-i] above), as the same suffix will always be specifiedfor the same tone. Because of this, only low-toned examples are given above(but high-toned ones can be found in constructions other than participles).However, tonal variation will be possible when an unstressed suffix carries aboundary tone determined by the tone on the (stressed) root syllable, sincedifferent roots may be specified for different tones. The notation of tonesin this thesis is primarily added for a more complete representation of thedata, as the processes examined further is insensitive to tonal specification.2.1.2 ConsonantsIn NM Slovenian, the following consonants may appear in the surface rep-resentations of phonological forms:BilabialLabio-dent.DentalAlveolarPalat.-alveo.PalatalVelarStop p b t” d” k gFric. f v s z S Z xAffric. Ù dZNasal m n” ŋTap RApprox. w l jTable 2.3: Consonants in NM SlovenianIt should be noted that the labio-dental fricative [v] is often described as alabio-dental approximant, viz. [V], in the descriptions of Standard Slovenian(Šuštaršič et al. 1995, 1999; Jurgec 2007a), but this does not seem to holdfor NM Slovenian. This is in no way connected to the data examined in thisthesis, which is why I defer an investigation of the status of this segment tofuture work. I also relax the notation of the dental segments in the sections(and chapters) that follow, by simply notating them without any subscriptdiacritics, e.g. [t”] as [t], etc.212.2. Status of the high vowel [i]2.2 Status of the high vowel [i]This section discusses the distribution of the high vowel [i]. First, a verygeneral overview of the structure of Slavic verbs and participles is given in2.2.1. In 2.2.2 and 2.2.3, the status of [i] is discussed in the verbal andparticipial paradigms of NM Slovenian, with an interim overview in 2.2.4.In section 2.2.5, the status of [i] inside prefixes and roots is discussed, andsection 2.2.6 discusses how the alternations found in verbs and participlesconnect with stress.2.2.1 Verbs in Slavic morphologyA rather standard ‘template’ for Slavic verbal morphology exists, as dis-cussed in various works, including Rubach (1984: 35-39), Svenonius (2004a,b),Manova (2011: 13), Biskup (2012), where the root is followed by a themesuffix, typically a vowel, which is followed by the agreement suffix codingperson and number:√root – theme – agrTable 2.4: Slavic (finite) verbsOn the verbal stem, the ‘l-participles’ can be constructed, which are par-ticiples derived with the /-l/ suffix. This suffix is again followed by anagreement suffix, just that it here codes gender and number:√root – theme – ptc-l – agrTable 2.5: Slavic l-participlesStandard Slovenian (Toporišič 2000; Herrity 2000) follows this general tem-plate, as does NM Slovenian. In Standard Slovenian, different theme vowelsmay occur in the theme position, such as /-a/, /-i/ and /-e/, which are alsosubject to different stress assignment. Stress in Standard Slovenian may oc-cur on the root or on the theme, which Toporišič (2000: 378-380) uses as acriterion to distinguish between different verbal and participial classes. Wewill witness the same in NM Slovenian, and we will also focus on an inter-esting phenomenon in connection with stress: when constructing participlesin NM Slovenian, some roots may either have stress on the root or on thetheme vowel, which means that stress is variable with some roots.222.2. Status of the high vowel [i]2.2.2 VerbsThe purpose of this section is to present the most typical and relevant pat-terns of NM Slovenian verbal morphology. The data in this section do notsay anything crucial with respect to the distribution of the high vowel [i].However, they will be of use to us in several of the sections to come, wherewe will have to refer to them to construct a complete morphological analysisof verbs and participles, and discuss their phonological properties.All the verbs in NM Slovenian are constructed by following the sametype of morphological pattern, as discussed in 2.2.1. Verbs in NM Sloveniancode person in verbs through expressing a three-way contrast: first, secondand third person. The number system expresses a three-way contrast so thatsingular, dual and plural number are distinguished. NM Slovenian is char-acterized by a fusional system of morphology, which means that person andnumber are coded through a single phonological exponent. In what follows,the infinitive, though a non-finite form, is given in the verbal paradigms aswell, for convenience.Verbs that are suffixed with the theme vowel [-a] form a representativeexample of how NM Slovenian verbal morphology works. Consider the root√del- ‘work’ whose verbal paradigm in the indicative mood is given, as isthe paradigm in the imperative and infinitive:indic. du pl1st.p "dél-à-m "dél-a-và "dél-a-mò2nd.p "dél-à-S "dél-a-tà "dél-a-tè "dèl-á-t3rd.p "dél-à-∅ "dél-a-tà "dél-a-jòTable 2.6: Verbs formed with [-a] (a-class) du pl1st.p "dèl-ej-vá "dèl-ej-mó2nd.p "dèl-éj-∅ "dèl-ej-tá "dèl-ej-té3rd.pTable 2.7: Imperative verbs formed with [-a] (a-class)The imperative forms of verbs generally contrast the same three-way num-ber and person categories, but not all persons are contrasted. The infinitive,on the other hand, does not contrast number and person. Notice that the232.2. Status of the high vowel [i]imperatives of this verbal class are suffixed with [-ej].7 It is not certainwhether this could perhaps be theme vowel [-a], suffixed with [-j], an imper-ative derivational morpheme, which would imply that [-a] assimilates to [-j],yielding [e] before [j]. Since [e] is always followed by [j] in the imperativeforms of [a]-stem verbs, I treat [-ej] as a single morpheme that is used tocode the imperative – perhaps an imperatival theme suffix. Infinitives, onthe other hand, are simply signalled by the suffix [-t]. The final inflectionalmorphemes coding person and number all have overt exponents, except forthe third person singular inflection, which reveals a zero-exponent.Verbs that belong to this class that employs the theme vowel [-a] includethe following members (the list is not exhaustive):root infinitive meaningjok- "jòk-á-t ‘cry’pix- "pìx-á-t ‘blow’dix- "dìx-á-t ‘breathe’dvig- "dvíg-à-t ‘lift’ruk- "rúk-à-t ‘hit’tsuk- "tsúk-à-t ‘tug’kix- "kíx-à-t ‘sneeze’Table 2.8: Roots suffixed with [-a] to form verbs (a-class)This class of verbs is formed exactly as laid out in Table (2.6): the themevowel is [-a], and the root is the one that always receives stress.Another class of verbs employs a different theme vowel, viz. [-i]. Theinflectional morphemes are precisely the same, but there are some differencesin the stress and imperative formations. Consider the verbal paradigm ofthe root √kad- ‘smoke’:indic. du pl1st.p ka"d-ì-m ka"d-ì-vá ka"d-ì-mó2nd.p ka"d-ì-S ka"d-ì-tá "ka"d-ì-té "ka"d-í-t3rd.p ka"d-ì-∅ ka"d-ì-tá ka"d-ì-jóTable 2.9: Verbs formed with [-i] (i-class)7An interesting aspect of the imperatival paradigm is that it seems to employ a differenttonal pattern then the one found in the indicative paradigm.242.2. Status of the high vowel [i] du pl1st.p "kád-∅-và "kád-∅-mò2nd.p "kát-∅-∅ "kát-∅-tà "kát-∅-tè3rd.pTable 2.10: Verbs formed with [-i] (i-class)The i-class paradigm is different from the a-class paradigm in term ofstress assignment: the [-i] theme vowel must be stressed in this verbal class,and not the root syllable as in the a-class. Notice that the imperativederivational morpheme is realized by a zero-exponent and that the stressdoes not occur on the theme vowel in these forms, but rather remains onthe root. The infinitival form employs the [-i] theme vowel, which alsoreceives stress, and the final inflection is [-t] as in the a-class verbs. Somemembers of this verbal class comprise the following roots:root infinitive meaningdel- de"l-í-t ‘deal’t@Rd- t@R"d-í-t ‘claim’dob- do"b-í-t ‘get’sled- sle"d-í-t ‘follow’bud- bu"d-í-t ‘awaken’tsed- tse"d-í-t ‘strain’Table 2.11: Roots suffixed with [-i] to form verbs (i-class)The next class of verbs is slightly different from the a- and i-classes.In this class, we find verbs that have no overt theme vowel, though in somepositions a schwa occurs instead. In terms of stress assignment, these verbsalways have stress on the root syllable, precisely like the a-class verbs. Letus call this class the ∅-class of verbs. Consider the paradigm of the root√xRan- ‘feed’:indic. du pl1st.p "xRàn-∅-@´m "xRàn-∅-vá "xRàn-∅-mó2nd.p "xRàn-∅-@´S "xRàn-∅-tá "xRàn-∅-té "xRàn-∅-t3rd.p "xRàn-∅-∅ "xRàn-∅-tá "xRàn-∅-jóTable 2.12: Verbs formed with [-∅] (∅-class)252.2. Status of the high vowel [i] du pl1st.p "xRán-∅-và "xRán-∅-mò2nd.p "xRán-∅-∅ "xRán-∅-tà "xRán-∅-tè3rd.pTable 2.13: Imperative verbs formed with [-∅] (∅-class)As mentioned above, this class of verbs has a zero theme suffix. Noticethe schwa in the first and second person singular: it may be possible thatthis schwa is phonologically motivated and, hence, has no morphologicalaffiliation, especially since the first and second person singular inflectionsoccur as [-m] and [-S] in the other verbal classes, crucially without schwa.Such an assumption is discussed in section 3.1 and can be set aside for now.Roots that form verbs according to this pattern that includes a zero themesuffix include the following (the list is not exhaustive):root infinitive meaningplan- "plàn-∅-t ‘leap’xlin- "xlìn-∅-t ‘fake’pil- "pìl-∅-t ‘file’mam- "màm-∅-t ‘tempt’kuR- "kùR-∅-t ‘burn’vol- "vO`l-∅-t ‘vote’mol- "mO`l-∅-t ‘pray’Table 2.14: Roots suffixed with [-∅] to form verbs (∅-class)However, some roots follow the ∅-pattern as laid out in Table (2.12), butwith a difference in the infinitival form: their infinitive also contains a schwa.These roots always end in an obstruent. Since the roots that form an in-finitive with zero always end in a sonorant (Table 2.14), this phonologicalproperty seems to be correlated with the presence or absence of schwa inthe infinitives of ∅-verbs. Roots ending in an obstruent include these:root infinitive meaningstop- "stòp-∅-@t ‘step’potS- "pòtS-∅-@t ‘crack’pRos- "pRO`s-∅-@t ‘ask’slut- "slùt-∅-@t ‘suspect’Table 2.15: ∅-class roots with with schwa in the infinitive262.2. Status of the high vowel [i]Up to this point we have discussed the three predominant classes of verbsin NM Slovenian, viz. a-class, i-class and ∅-class. Another minor classof verbs exists: in this class, verbs are formed by suffixing the theme [-e] tothe root. Consider √pis- ‘write’:indic. du pl1st.p "pìS-@´-m "pìS-e-vá "pìS-e-mó2nd.p "pìS-@´-S "pìS-e-tá "pìS-e-té "pìs-á-t3rd.p "pi`S-é-∅ "pìS-e-tá "pìS-e-jóTable 2.16: Verbs formed with [-e] (e-class) du pl1st.p "píS-∅-và "píS-∅-mò2nd.p "píS-∅-∅ "píS-tà "píS-∅-tè3rd.pTable 2.17: Imperative verbs formed with [-e] (e-class)Several things need to be said about this paradigm. While the indicativeverb forms are suffixed with [-e], the imperative forms are derived by using azero theme suffix, instead. Also, notice that the root-final consonant in theindicative and imperative forms is [S], but [s] in the infinitive. The themevowel [-e] generally has a palatalizing effect on verbal roots, which neednot receive further discussion here, e.g. also ["kàZ-é-∅] (3rd.p) vs. ["kàz-á-t](inf) ‘show’, ["màZ-é-∅] (3rd.p) vs. ["màz-á-t] (inf) ‘smear’, etc. Anothermatter needs to be mentioned in connection with the infinitives: thoughthe indicative verbal forms are formed with [-e] in this class, the themevowel used in the infinitive is either [-a] or [-∅], as shown in the list of rootsbelonging to this class in Table (2.18), where both themes are represented.The most curious aspect of the verbal e-class is, however, the absenceof the theme [-e] in the first and second person singular, and the presenceof schwa. Notice that I have notated the schwa as a regular theme vowelmorpheme. More shall be said about this matter in section 3.1, which ison schwa in NM Slovenian, where we will speculate that this schwa couldbelong to the following inflection.88Note, also, that the theme [-e] can occur as [-E] when stressed. This happens with asmall group of roots, see the following footnote. This is an expected alternation as the272.2. Status of the high vowel [i]The root-members belonging to the verbal class formed with the theme[-e] are the following (the list is not exhaustive):9root infinitive meaningpel- "pE`l-á-t ‘drive, lead’kaz- "kàz-á-t ‘show’maz- "màz-á-t ‘smear’petS- "pètS-∅-t ‘bake’RetS- "RètS-∅-t ‘pray’tetS- "tètS-∅-t ‘run’Table 2.18: Roots suffixed with [-e] to form verbsThe theme vowel [-e] is not only used to form the indicative verbal formsof the e-class roots (listed above), but also to mark a specific semanticeffect: verbal roots can be suffixed with an aspectual morpheme, viz. [-n],which seems to yield a ‘semelfactive’ reading; i.e. it denotes an action thatis perfective and instantaneous (Comrie 1976). The indicative verbal formsof roots suffixed with [-n] are always accompanied by the theme vowel [-e]. Consider the root √pix- ‘blow’, which otherwise belongs to the verbala-class, but when suffixed with the semelfactive [-n], it selects for the [-e]theme vowel, and the reading can then be translated as ‘give a quick blow’:indic. du pl1st.p "pìx-n-@´-m "pìx-n-e-vá "pìx-n-e-mó2nd.p "pìx-n-@´-S "pìx-n-é-ta "pìx-n-e-té "pìx-@´n-∅-t3rd.p "pìx-n-é-∅ "pìx-n-é-ta "pìx-n-e-jóTable 2.19: Semelfactive formed with [-e] (e-class) du pl1st.p "píx-@n-∅-và "píx-@n-∅-mò2nd.p "píx-@`n "píx-@n-∅-mò "píx-@n-∅-tè3rd.pTable 2.20: Imperative semelfactive formed with [-e] (e-class)mid vowels [e], [o] do not contrast with [E], [O] in unstressed position, a pattern also seenin Standard Slovenian (Jurgec 2011).9Note that a small list of roots, such as √mr- ‘die’ and √dR- ‘knock over’, require thestress to occur on theme [-e]. See sections 2.2.3, 3.1 and 3.2 for discussion of this class.282.2. Status of the high vowel [i]Notice that some additional schwa-vowels occur in the imperative forms ofthe semelfactive. Along with other instances of schwa, these will be discussedin section 3.1.Now that we have covered all the verbal classes in NM Slovenian, we canrepresent them in the following table:roots theme stressindicative⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩√del-√pix-√dix-...⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭[-a] √"σ − σ⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩√kad-√dob-√sled-...⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭[-i] √σ−"σ⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩√xRan-√stop-√xlin-...⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭[-∅] √"σ − ∅⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩√pel-√maz-√petS-...⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭[-e] √"σ − σ{ √Root+semelf. } [-e] √"σ − σTable 2.21: Verbal classes in NM Slovenian (indicative forms)2.2.3 ParticiplesNow that we have given an overview of the NM Slovenian verbal morphology,we can move on to a discussion of participial morphology, which is cruciallybased on that of the verbs. In this section, we shall discuss the generalmorphological properties of participles and their formation patterns. In ad-dition, the data in this section will enable us to demonstrate that unstressed[i] synchronically alternates with [@] or [∅] in NM Slovenian.The participles we shall discuss are usually termed l-participles in mostliterature on Slavic morphology simply because they are formed with themorpheme [-l]. In NM Slovenian, [-l] is the participial derivational morphemethat is suffixed to a verbal stem, i.e. [-l] is attached to a √root that is292.2. Status of the high vowel [i]followed by a theme vowel. This constitues the participial stem. Considerthe root √jok- ‘cry’, which is suffixed with the theme [-a], yielding theverbal stem [√jok-a-], to which we add [-l] to create the participial stem[√jok-a-l]. The participial stem is then followed by an inflection, e.g. thefeminine singular inflection [-a], to yield [√jok-a-l-a]. Participles in NMSlovenian code gender and number, but not case nor person. They also usedifferent exponents of inflection than verbs (which code person and number).l-participles are essentially ‘active’ participles: they are used to constructfuture and past tenses (periphrastically) together with the auxiliary biti‘to be’, and they also participate in forming conditionals together with theparticle bi (Marvin 2002).Let us begin by discussing the first class of participles. Like verbs, Iterm the participles that are based on a verbal stem derived by the themevowel [-a] a-class participles. Such a classification will prove necessarybecause participles sometimes select different verbal theme vowels than theverbs themselves do. Furthermore, participles are fraught with differentstress patterns, much more so than their verbal counterparts. To distinguishbetween different stress patterns within a participial class, I designate themwith a roman numeral next to the class designations: for instance, the classof participles that are built on the verbal stem which employs the themevowel [-a], and have stress on the root, are classified as a-classI participles.Consider √dix- ‘breathe’, which represents a-classI participles:sg du plmasc "dìx-ó-w-∅ "dìx-a-l-á "dìx-á-l-∅fem "dìx-a-l-á "dìx-a-l-é "dìx-a-l-éneut "dìx-a-l-ú "dìx-a-l-á "dìx-a-l-áTable 2.22: a-classI participle (stress: √"σ − σ)The stress is always on the root syllable in this class of participles. Noticethat instead of the participial [-l], this class shows [w] in the masculinesingular form, and the theme [-a] is replaced by [o].10 All the participlesclasses will either show [w] or [u] instead of [l] in the masculine singularform. However, we can set these differences aside for now, as they will bediscussed in the sections to come.Let me also take this opportunity to briefly comment on the tonal speci-fication of participles in general: participles follow the tonal pattern laid out10This seems to be a purely phonological process of rounding [a] when it precedes [w].302.2. Status of the high vowel [i]in section 2.1.1, which means that either the vowel in the root syllable deter-mines the final boundary tone (if it is stressed), or the suffix determines it(again, if it is stressed). The tone patterns indicated on the relevant vowelsin this section represent the most typical tonal realization in NM Slovenian:sometimes tone specification can be somewhat variable, viz. a low or a hightone on the root syllable may be acceptable. In the majority of the cases,NM Slovenian speakers opt for a low tone on most stressed roots and all thestressed verbal and participial suffixes discussed in this section. In cases likethat, I notate the preferred specification, viz. low tone. When a high tone isin fact preferred, I notate a high tone. The crucial point of this brief tonalinterlude is to note than tone assignment may not be as rigid as reflected inthe notations in this section and it may vary to some extent – I have onlyattempted to capture the relevant preferences.Let us now return to discussing the a-classI participles. To recapitulate,these participles have the root syllable stressed and are formed on the verbalstem containing the theme [-a]. Some roots that are used to build suchparticiples are the following (the list is not exhaustive):root verbal classdix- "dìx-á-l-∅ a ‘breathe’dvig- "dvìg-á-l-∅ a ‘lift’tsuk- "tsùk-á-l-∅ a ‘tug’nex- "néx-à-l-∅ a ‘stop’Table 2.23: Roots used to form a-classINote that the third column, ‘verbal class ’, indicates what class the verb,corresponding to the participle, belongs to based on the theme vowel itemploys. All the roots that form a-classI participles also form verbs withthe theme vowel [-a].A different set of roots may also be used to form participles with theverbal stem containing [-a], but these have a different stress pattern: stressin this group of participles is always on the theme vowel. Let us call thisgroup a-classII participles. Consider the example of √pel- ‘drive, lead’:sg du plmasc "pE`l-ó-w-∅ pe"l-à-l-á pe"l-à-l-∅fem pe"l-à-l-á pe"l-à-l-é pe"l-à-l-éneut pe"l-à-l-ú pe"l-à-l-á pe"l-à-l-áTable 2.24: a-classII participle (stress: √σ−"σ)312.2. Status of the high vowel [i]Notice that there is an exception to the overall stress pattern: the masculinesingular form has stress on the root syllable and not on the theme vowel.Roots that form participles of a-classII are the following:root verbal classpel- pe"l-à-l-∅ e ‘drive, lead’pis- pi"s-à-l-∅ e ‘write’d@RZ- d@R-"Z-à-l-∅ i ‘hold’beg- be-"Z-à-l-∅ i ‘run away’Table 2.25: Roots used to form a-classII participlesRoots that are used to form this class of participles may use different themevowels to form verbs. Consider [√pel-a-l-∅]: the verbal stem upon whichthe participle is built is indeed [√pel-a-], but this same stem cannot be usedto form verbs. A verb can only be formed with √pel by suffixing the theme[-e] to it, which makes the verbal stem upon which verbs are built [√pel-e-]. This, in essence, implies that two different verbal stems may exist perroot, but only one of them will actually be used to construct a verb. Thesame applies for roots that use the theme [-i] to form verbs, but [-a] to formparticiples. It seems that this class of participles can only be formed fromverbal stems that take the themes [-i] or [-e] when forming verbs.A third set of roots may employ the theme vowel [-a] in the stem thatis used to construct a participle. Participles built from this third set ofroots can be stressed variably: stress may occur on the root syllable oron the theme-vowel. Speakers of NM Slovenian might have an individualpreference for one of the stress types, but most usually they indicate thatthey would use either of them. Let us call this third group of participles thea-classIII. Consider the root √jok- ‘cry’:sg du plmasc "jòk-ó-w-∅ "jòk-a-l-á "jòk-á-l-∅fem "jòk-a-l-á "jòk-a-l-é "jòk-a-l-éneut "jòk-a-l-ú "jòk-a-l-á "jòk-a-l-áTable 2.26: a-classIII participle: √"σ − σsg du plmasc "jòk-ó-w-∅ jo"k-à-l-á jo"k-à-l-∅fem jo"k-à-l-á jo"k-à-l-é jo"k-à-l-éneut jo"k-à-l-ú jo"k-à-l-á jo"k-à-l-á322.2. Status of the high vowel [i]Table 2.27: a-classIII participle: √σ−"σThe variable nature of stress assignment in this participial class means that,for instance, the feminine singular participle of √jok- may be either realizedas ["jòk-a-l-á] or [jo"k-à-l-á]. However, observe that, like with the a-classIIparticiples, the masculine singular again fails to host stress on the themevowel. Roots that are used to form a-classIII participles are the following(the list is not exhaustive), and they all correspond to verbs that are formedwith the theme vowel [-a]:root verbal classjok- "jòk-á-l-∅ / jo-"kà-l-∅ a ‘cry’pix- "pìx-á-l-∅ / pi-"xà-l-∅ a ‘blow’val- "vàl-á-l-∅ / va-"là-l-∅ a ‘roll’kop- "kòp-á-l-∅ / ko-"pà-l-∅ a ‘bathe’Table 2.28: Roots used to form a-classIII participlesWe have now exhausted the list of roots that form participles throughemploying [-a] as the theme vowel in their verbal stem. Let us, therefore,turn to participial classes that make use of the theme [-i] in their verbalstem. Most participles that use the theme [-i] assign stress to this themevowel, even in the masculine singular form (recall that this does not happenin the a-classII of participles, which also contain a stressed theme for allthe other forms). Participles constructed in this fashion are termed i-classIof participles. Consider √kad- ‘smoke’:sg du plmasc ka"d-ì-w-∅ ka"d-ì-l-á ka"d-ì-l-∅fem ka"d-ì-l-á ka"d-ì-l-é ka"d-ì-l-éneut ka"d-ì-l-ú ka"d-ì-l-á ka"d-ì-l-áTable 2.29: i-classI participle (stress: √σ−"σ)Notice that stress is on the verbal theme vowel [-i], throughout this paradigm.However, not all participles that employ the verbal theme [-i] act in this way.Some have the theme vowel stressed in all the forms but the masculine singu-lar forms (like the a-classII participles). We term the group of participlesthat fail to assign stress to the theme [-i] in the masculine singular i-classIIparticiples. Consider the example √dub- ‘get’:332.2. Status of the high vowel [i]sg du plmasc "d-ùb-∅-ú-∅ du"b-ì-l-á du"b-ì-l-∅fem du"b-ì-l-á du"b-ì-l-é du"b-ì-l-éneut du"b-ì-l-ú du"b-ì-l-á du"b-ì-l-áTable 2.30: i-classII participle (stress: √σ−"σ, except this group of participles, stress is assigned to the theme vowel [-i] in allthe forms, e.g. [du"b-ì-l-á] (, [du"b-ì-l-∅] (, etc., except in themasculine singular, where it remains on the root, cf. ["d-ùb-∅-ú-∅] (,and the theme [-i] fails to show up. This alternation looks suspiciously likethe result of a vowel-deletion process (which we will ultimately propose).Roots that make use of the verbal theme vowel [-i] and have stress onthe theme throughout the participial paradigm (viz. i-classI participles)are the following (the list is not exhaustive):root verbal classkad- ka"d-ì-l-∅ i ‘smoke’t@Rd- t@R"d-ì-l-∅ i ‘claim’Table 2.31: Roots used to form i-classI participlesMost of these roots use the same theme vowel and the same stress-pattern intheir verbal formations. Roots that, on the other hand, make up i-classIIparticiples, are the following (the list is not exhaustive):root verbal classdub- du"b-ì-l-∅ i ‘get’mol- mo"l-ì-l-∅ ∅ ‘pray’voz- vo"z-ì-l-∅ ∅ ‘drive’Table 2.32: Roots used to form i-classII participlesThese roots can belong to the ∅-class (e.g. √mol-), which implies that, intheir verbal forms, they assign stress on the root syllable, or they can belongto the i-class of verbs (e.g. √dub-), which in turn implies that they formverbs with the theme vowel [-i], which also carries stress. When formingparticiples, however, they all use the verbal theme [-i] and assign stress toit, except for the masculine singular form where the theme [-i] fails to showup and the stress is on the root syllable, as shown in Table (2.30).The next class of participles that we discuss has no overt verbal themevowel, i.e. the theme employed in the verbal stem that is used to form342.2. Status of the high vowel [i]participles is [-∅]. We term this class of participles ∅-class. However, thisclass can be further subdivided into ∅-classI and ∅-classII. Let us firstdiscuss the former. Consider the example of √xlin- ‘fake’:sg du plmasc "xlìn-∅-ú-∅ "xlìn-∅-l-á "xlìn-∅-@´l-∅fem "xlìn-∅-l-á "xlìn-∅-l-é "xlìn-∅-l-éneut "xlìn-∅-l-ú "xlìn-∅-l-á "xlìn-∅-l-áTable 2.33: ∅-classI participle (stress: √"σ − σ)The stress in this group of participles is always realized on the root syllableand the verbal theme is [-∅]. The curious aspect of this group of participlesis the schwa-vowel that occurs in the masculine plural form (the precisemorphological status of this schwa will be determined in the forthcomingsections; for now, we can group it together with the participial [-l]). Hereare some roots that follow this pattern:root verbal classxlin- "xlìn-∅-@´l-∅ ∅ ‘fake’pil- "pìl-∅-@´l-∅ ∅ ‘file’pad- "pàd-∅-@´l-∅ e ‘fall’Table 2.34: Roots used to form ∅-classI participles participlesThe second subgroup of the ∅-class participles, viz. ∅-classII containsa schwa as the stressed root vowel. One such example is (u-)√"mR- ‘die’;notice that we specified no vowel in the root itself here, the reason for whichwill become clear du plmasc u-"m@´R-∅-ù-∅ u-"m@´R-∅-l-à u-"m@´R-∅-@`l-∅fem u-"m@´R-∅-l-à u-"m@´R-∅-l-è u-"m@´R-∅-l-èneut u-"m@´R-∅-l-ù u-"m@´R-∅-l-à u-"m@´R-∅-l-àTable 2.35: ∅-classII participle (stress: √"σ − σ)In this subgroup of the ∅-class participles, we can observe that the stressedsyllable is always contained in the root and it is always realized by schwa.The theme suffix is realized by zero and another schwa occurs in the mascu-line plural form, between the position for the zero suffix and the participialform, exactly like in ∅-classI participles. Some of the roots that form thisclass are the following:352.2. Status of the high vowel [i]root verbal class(u-)"mR- u-"m@´R-∅-@`l-∅ e ‘die’"tsvR- "tsv@´R-∅-@`l-∅ e ‘fry’(u-)"pR- u-"p@´R-∅-@`l-∅ e ‘resist’Table 2.36: Roots used to form ∅-classII participlesThe very reason for distinguishing between ∅-classI and ∅-classII is theroot structure of the latter. Notice that in the roots given in Table (2.36)above, no schwa vowel is indicated, but a schwa does in fact appear in theparticiples built on these roots, and the schwa is actually the stressed vowel.However, these roots form verbs in a more peculiar way: they form verbswith the theme vowel [-e], as indicated above, but it is the theme [-e] that isstressed in the verbal forms and no schwa occurs inside the root. Considerthe verbal form of u-√"mR-, which is [u-"mR-è-m] ‘I die (’, and theverbal form of √"tsvR-, which is ["tsvR-è-m] ‘I fry (’. Since no schwais to be found within the roots of these verbal forms, we can assume thatit is specific to the participial formations (later on we will explain it asan expected consequence of NM Slovenian phonology). And if no schwa isavailable, these roots seem to stress whatever vowel is closest to the root;in the case of verbs, this is the theme vowel [-e]. In this sense, these rootshave no vowel of their own, and it is this characteristic that sets them apartfrom ∅-classI participles.The next group of participles is subject to variable stress assignment: thestress in this group can either appear on the root or on the theme vowel (withthe exception of the masculine singular, which only has stress on the rootsyllable). This makes them a direct equivalent to the a-classIII participles,except that this group does not employ the theme vowel [-a], but actuallyexhibits alternation between [-i] and [∅]. This is why we conveniently termthis class of participles the ∅ ∼ i-class. Consider √xRan- ‘feed’:sg du plmasc "xRàn-∅-ú-∅ "xRàn-∅-l-á "xRàn-∅-@´l-∅fem "xRàn-∅-l-á "xRàn-∅-l-é "xRàn-∅-l-éneut "xRàn-∅-l-ú "xRàn-∅-l-á "xRàn-∅-l-áTable 2.37: ∅ ∼ i-class participle (stress: √"σ − σ)sg du plmasc "xRàn-∅-ú-∅ xRa"n-ì-l-á xRa"n-ì-l-∅fem xRa"n-ì-l-á xRa"n-ì-l-é xRa"n-ì-l-éneut xRa"n-ì-l-ú xRa"n-ì-l-á xRa"n-ì-l-áTable 2.38: ∅ ∼ i-class participle (stress: √σ−"σ )362.2. Status of the high vowel [i]Notice that if the stress occurs on the root syllable, as in (2.37), the themeis realized by [∅], but it is also accompanied by [@] immediately before theparticipial [-l] in the masculine plural form. However, if stress is assigned onthe theme, as in (2.38), the theme surfaces as [i], even in the masculine pluralwhere no schwa occurs then. The masculine singular is again exempt fromstress variability, as it always occurs with stress on the root syllable and thetheme [-i] always fails to show up. These data show that [-i] alternates withzero under different stress assignments: we are dealing with two differentstress-based realizations of the same paradigm, and within this pattern thevowel [i] appears to be in perfect complementary distribution with zero in away the is correlated with stress. If the theme is stressed it shows up as [i],but if not it shows up as zero. This suggests that a process of /i/-deletion isactive in the phonology and that we are dealing with segmentally identicalunderlying forms for the theme vowel – we come back to this below. Addi-tionally, the absence of [-i] seems correlated with the presence of the schwain the masculine plural in (2.37) – see 3.1 for more on schwa. Roots thatform ∅ ∼ i-class participles are the following (the list is not exhaustive) "xRàn-∅-@´l-∅ / xRa"n-ì-l-∅ ∅ ‘feed’kuR- "kùR-∅-@´l-∅ / ku"R-ì-l-∅ ∅ ‘burn’mam- "màm-∅-@´l-∅ / ma"m-ì-l-∅ ∅ ‘tempt’govor- go"vóR-∅-@`l-∅ / govo"R-ì-l-∅ i ‘talk’(o-)tvoR- (o-)"tvóR-∅-@`l-∅ / (o-)tvo"R-ì-l-∅ i ‘open’Table 2.39: Roots used to form ∅ ∼ i-class participlesThe roots above all follow the pattern given in Tables (2.37) – (2.38). Theirverbal correspondents are formed either by following the patterns repre-sented by the ∅-class of verbs or the i-class of verbs.What is particularly striking about this class of participles is that itis directly comparable with the a-classIII participles in that it exhibitsvariable stress realization. But it is also crucially different from a-classIII:notice that in the i ∼ ∅-class the alternation of the theme [-i] with zero isperfectly correlated with stress realization, as observed above: if stress is onthe root syllable, the theme is zero, but if stress is on the theme, the themeis realized as [-i]. This correlation of phonological factors seems to suggest372.2. Status of the high vowel [i]that something more than morphology is at play here; it seems that weare dealing with a phonologically conditioned alternation, one involving thesegment [-i], which realizes the theme suffix for this entire class. This mustbe the case because no such correlation of theme realization and stress canbe uncovered in the a-classIII participles, where the theme is realized as[-a], regardless of stress placement. Furthermore, it is a distinct possibilitythat even the ∅-class participles are a part of this alternation.11 This isa possibility because they uniformly have stress on the root syllable andnever on the theme syllable, under which condition we expect the themeto be realized as zero (which it is), precisely as in the ∅ ∼ i-class. Thiscan also be extended to the i-class participles, which have stress uniformlyon the theme, which is always realized as [-i]. This implies that the theme[-i] is in perfect complementary distribution with zero with respect to stressassignment. It is thus possible to entertain the option that these classes areall derived from a common ‘i based class’: this would be a welcome solution,as it would lead to a three-way bifurcation involving a constant root stressed‘i based class’ (∅-class), a constant theme stressed ‘i based class’ (i-class)and a variable stress ‘i based class’ (i ∼ ∅-class), and this would directlymirror the a-class participles, which have this same three-way split. Thiswould not only unify, but also significantly simplify the morphology of NMSlovenian.We have now discussed the major classes of participles in NM Slove-nian. Let us, also, examine the participial formations that are based onsemelfactives. Recall from section 2.2.2 that semelfactive verbs are derivedby suffixing the root with the semelfactive suffix [-n], which is in turn fol-lowed by the theme [-e]. Semelfactive participles also suffix the root directlywith [-n], but the theme that follows is either [-i] or [-∅] (and, crucially, not[-e]). The theme is followed by the participial [-l], and then the inflectionsfollow. Consider for instance, √max- ‘slap’, which forms the semelfactiveverbal stem [√max-n-i-]; to this stem the participial [-l] is added, yielding[√max-n-i-l-], to which we can add the inflection [-a] to express the femininesingular semelfactive participle, resulting in [√max-"n-i-l-a].Semelfactive participles can also be subdivided into two classes depend-ing on what stress-pattern they follow. First, we discuss the ∅-class ofsemelfactive participles, which occur with the theme [-∅]. This class ofsemelfactive participles is here termed ∅-class. Consider √but- ‘hit’:11At least ∅-classI, but not ∅-classII participles. See section 2.2.6 for a more detaileddiscussion on this interplay of phonological factors382.2. Status of the high vowel [i]sg du plmasc "bùt-n-∅-ú-∅ "bùt-@n-∅-l-á "bùt-@n-∅-l-ífem "bùt-@n-∅-l-á "bùt-@n-∅-l-é "bùt-@n-∅-l-éneut "bùt-@n-∅-l-ú "bùt-@n-∅-l-á "bùt-@n-∅-l-áTable 2.40: ∅-class semelfactive participle (stress: √"σ − σ)The stress in this class of semelfactive participles is always on the rootsyllable and the theme following the semelfactive [-n] is always ∅; thesesemelfactives are exactly like the ∅-class pattern in Table (2.33) in thatthey always show root stress and the theme is realized as zero. In all casesbut the masculine singular, the semelfactive [-n] is also preceded by a schwa-vowel. Notice, also, another curious aspects of these forms: the masculineplural inflection is here exponed by [-i] and not [-∅], which is the exponentused in all the regular (non-semelfactive) participles. Some of the roots thatbelong to the ∅-class of semelfactive participles are the following:root "bùt-@n-∅-l-í ‘hit’b@Rts- "b@`Rts-@n-∅-l-í ‘kick’tsuk- "tsùk-@n-∅-l-í ‘tug’kap- "kàp-@n-∅-l-í ‘trickle’Table 2.41: Roots that form ∅-class semelfactive participlesSome semelfactive participles can, however, undergo variable stress as-signment, i.e. they correspond to the stress patterns of a-classIII and ∅ ∼ i-class of regular participles. This means that stress can either be assignedto the root syllable or to the theme. If the stress is assigned to the theme,the theme occurs as [-i], but if stress is assigned to the root, the theme oc-curs as [-∅]; such semelfactive participles are termed ∅ ∼ i-class. Consideru-√gas- ‘turn off, put out’:sg du plmasc u-"gàs-n-∅-ú-∅ u-"gàs-@n-∅-l-á u-"gàs-@n-∅-l-ífem u-"gàs-@n-∅-l-á u-"gàs-@n-∅-l-é u-"gàs-@n-∅-l-éneut u-"gàs-@n-∅-l-ú u-"gàs-@n-∅-l-á u-"gàs-@n-∅-l-áTable 2.42: ∅ ∼ i-class semelfactive participle (stress: √"σ − σ)392.2. Status of the high vowel [i]sg du plmasc u-"gàs-n-∅-ú-∅ u-gas-"n-ì-l-á u-gas-"n-ì-l-∅fem u-gas-"n-ì-l-á u-gas-"n-ì-l-é u-gas-"n-ì-l-éneut u-gas-"n-ì-l-ú u-gas-"n-ì-l-á u-gas-"n-ì-l-áTable 2.43: ∅ ∼ i-class semelfactive participle (stress: √σ−"σ)The stress properties of the semelfactive participle in (2.42) are identical tothose of the ∅-class semelfactive participles given in (2.40): the stress isalways on the root syllable. The theme that follows the semelfactive [-n]is always [-∅] and the semelfactive is preceded by a schwa-vowel, except inthe masculine singular form. The masculine plural inflection is exponed by[-i]. The semelfactive participle in (2.43), on the other hand, hosts stress onthe theme. The theme, following the semelfactive [-n], is always [-i], but noschwa-vowel occurs before the semelfactive, and the masculine plural is nowexponed by [-∅]. The expected exception to this is the masculine singularagain: it shows root-stress and [-∅] as the theme. The realization of thetheme as [-i] when it receives stress and its realization as zero when it is notstressed is identical to the pattern of stress and theme realization in i ∼ ∅-class participles shown in (2.37); the inability of the masculine singular toexhibit theme stress is also precisely like in i ∼ ∅-class participles, and alsoas in i-classII as shown in (2.30).∅ ∼ i-class of semelfactive participles also shows that the high vowel[-i], [-∅] and schwa alternate, depending on what stress pattern is used: ifthe stress is on the root syllable, a schwa occurs before the semelfactive [-n],the theme is [-∅], while the masculine plural inflection is [-i]. On the otherhand, if the stress is on the theme, no schwa occurs before the semelfactive,the theme is [-i] and the masculine plural is [-∅]. Roots that exhibit suchbehaviour when forming semelfactive participles are the following (the listis not exhaustive):root (u-)"gàs-@n-∅-l-í / (u-)gas-"n-ì-l-∅ ‘turn off’(od-)dax- (od-)"dàx-@n-∅-l-í / (od-)dax-"n-ì-l-∅ ‘catch breath’(na-)tak- (na-)"tàk-@n-∅-l-í / (na-)tak-"n-ì-l-∅ ‘spike’max- "màx-@n-∅-l-í / max-"n-ì-l-∅ ‘slap’(o-)g@R- (o-)"g@`R-@n-∅-l-í / (o-)g@R-"n-ì-l-∅ ‘wrap’Table 2.44: Roots that form ∅ ∼ i-class semelfactivesHowever, not only semelfactive participles show such behaviour with re-spect to the occurrence of schwa, the theme [-i] and the masculine plural402.2. Status of the high vowel [i]inflection [-i]. A class of regular (non-semelfactive) participles acts in pre-cisely the same way. These are participles that are formed from roots witha √CVCr- structure, i.e. such roots contain two consonants that follow thevowel, of which the second one is a sonorant (represented by ‘r’). Let usterm this group √CVCr-class participles. Stress in this group may also bevariable: it may occur on the root or on the theme, and the expected con-sequences witnessed in ∅ ∼ i-class semelfactive participles in Tables (2.42)and (2.43) follow. Consider √pown- ‘fill’:sg du plmasc "pòwn-∅-ú-∅ "pòw@n-∅-l-á "pòw@n-∅-l-ífem "pòw@n-∅-l-á "pòw@n-∅-l-é "pòw@n-∅-l-éneut "pòw@n-∅-l-ú "pòw@n-∅-l-á "pòw@n-∅-l-áTable 2.45: √CVCr-class participle (stress: √"σ − σ)sg du plmasc "pòwn-∅-ú-∅ pow"n-ì-l-’a pow"n-ì-l-∅fem pow"n-ì-l-á pow"n-ì-l-é pow"n-ì-l-éneut pow"n-ì-l-ú pow"n-ì-l-á pow"n-ì-l-áTable 2.46: √CVCr-class participle (stress: √σ−"σ)If stress is on the root syllable, which is the case in (2.45), a schwa occursbefore the final root-consonant [n], the theme is [-∅], and the masculineplural exponent is [-i]. If the stress, however, is on the theme, as in (2.46),no schwa occurs before the root-final [-n], the theme is [-i], and the masculineplural exponent is [-∅]. The masculine singular is, expectedly, exempt fromany alternation, shows root-stress and has the theme [-∅].An explanation is order. Why should we not simply treat the √CVCr-class participles as semelfactive participles? Firstly, the roots √pown- ‘fill’,√d@Rgn- ‘rub’, √pRazn- ‘empty’ listed in (2.48) never occur without the final[n]. This is not true of the semelfactive participles where regular verbs can beformed without the [-n], cf. the semelfactive participle based on √pix- ‘blow’,["pìx-@n-∅-l-í] ‘having given a quick blow (’ vs. the regular participle["pìx-á-l-∅] ‘having blown (’. Secondly and more importantly, none ofthe roots in (2.48) actually display semelfactive semantics. Be it the verb["pòwn-é-∅] ‘fill (’ or the participle ["pòw@n-∅-l-í] ‘fill (’, noneof them denote an instantaneous and, at the same time, perfective action.In fact, the semantics of √pown- is best described as denoting a ‘continuousaction of filling (something)’, which means that its denotation is not even412.2. Status of the high vowel [i]perfective. This is true of all the roots in in (2.48), and this is how they aresharply contrasted with semelfactives: for instance, the regular participle["pìx-á-l-∅] ‘blow (’ denotes an imperfective action of ‘blowing’, whilethe semelfactive participle ["pìx-@n-∅-l-í] ‘blow (’ denotes the actionof ‘a quick, short blow’.It should also be noted that √pown- and √pRazn- also exist as adjectives["pow@n] ‘full (’ and ["pRaz@n] ‘empty (’, and the [n] here could beinterpreted as the adjectivizing suffix /-n/ (["pOt] ‘sweat (’ → ["pOt-@n]‘sweaty (’, etc.), but this is not so likely because constructions withthe adjectiziving /-n/ do not usually form verbal structures. Notice that√d@Rgn- ‘rub’, on the other hand, cannot exist as an adjective *["d@Rg@n],and it forms a uniform class with √pown- and √pRazn-. For this reason,we will rather assume that the adjectives ["pow@n] and ["pRaz@n] are rootssuffixed with a zero adjectivizing morpheme in this thesis.12The class of √CVCr-class does not consist exclusively of [CVCn]-typeroots. It seems that other roots that contain a sonorant as the second codaconsonant of the root also pattern with this group. Consider √misl- ‘think’:sg du plmasc "mìsl-∅-ú-∅ "mìs@l-∅-l-á "mìs@l-∅-l-ífem "mìs@l-∅-l-á "mìs@l-∅-l-é "mìs@l-∅-l-éneut "mìs@l-∅-l-ú "mìs@l-∅-l-á "mìs@l-∅-l-áTable 2.47: Participle built on √misl- (stress: √"σ − σ)Notice that this root forms participles in exactly the same way as the otherroots in the √CVCr-class (viz. those with the structure [CVCn-]) whenthey occur with a stressed root syllable. √misl- cannot undergo variablestress assignment, though: the stress must remain fixed on the root syllable.Much like the other members of this participial class, √misl- conveys noperfective or semelfactive semantics.Some of the roots that follow this pattern are the following:12A future study that compares√CVC-n adjectives with the√CVCn-∅ adjectives wouldbe of interest for morphological inquiry. Examples (39) and (40) in section 4.3 introducea diagnostic that confirms the assumptions on √pown-, and potentially other such roots.422.2. Status of the high vowel [i]root "pòw@n-∅-l-í / pow"n-ì-l-∅ ‘fill’d@Rgn- "d@`Rg@n-∅-l-í / d@Rg"n-ì-l-∅ ‘rub’pRazn- "pRàz@n-∅-l-í / pRaz"n-ì-l-∅ ‘empty’misl- "mìs@l-∅-l-í ‘think’k@Rm- "k@`r@m-∅-l-í ‘feed (cattle)’Table 2.48: Roots that form √CVCr-classWe have established that the √CVCr-class of participles cannot have thesame morphological make-up as the semelfactives. And yet, it is interestingthat the group of √CVCr-class participles displays exponence identical tothat of the semelfactives. The theme must always be either [-i] or [-∅] andthe masculine plural is either exponed by [-i] or [-∅]. But what is relevantat this point is that this group of participles shows alternation between [-i]and [-∅] and [@] that respects the same conditions of stress assignment asthe same alternations in ∅ ∼ i-class participles and ∅ ∼ i-class semelfac-tive participles do. In general, the observation about the complementarydistribution of [i], zero and schwa under different stress conditions carriesover from the regular (non-semelfactive) participles. This complementaryrelation, which appears phonological in nature, is perfectly obeyed by thetheme suffixes, but not so perfectly by the suffix [-i]. This suffix can ap-pear unstressed as well, but crucially even this can only happen under veryspecific phonotactic conditions: in the data that we have seen so far, [-i] only appears when the root has the ‘√CVCr’ phonotactic shape,or when the root is followed by the semelfactive suffix [-n], again creatingthe structure where the root is followed by a sonorant.Our discussion on NM Slovenian participles draws to an end, but twomore participles need to be discussed, which are quite different from anydescribed so far. These are the participles based on the roots √b- ‘be’ and√S- ‘go’, both of which consist of a single consonant. Let us first discuss√S-, which forms participles with the [-∅] theme, the participial suffix [-l]and the usual inflections. Here is its paradigm:sg du plmasc "S-∅-ù-∅ "S-∅-l-à "S-∅-l-ìfem "S-∅-l-à "S-∅-l-E` "S-∅-l-E`neut "S-∅-l-ù "S-∅-l-à "S-∅-l-èTable 2.49: Single-consonant roots: √S- ‘to go’432.2. Status of the high vowel [i]The theme is [-∅] throughout the paradigm, and the masculine plural inflec-tion is always [-i]. No specific stress patterns can be observed because allthe forms are monosyllabic. Now, consider the paradigm of √b- ‘be’:sg du plmasc "b-ì-w-∅ "b-∅-l-à "b-∅-l-ìfem "b-∅-l-à "b-∅-l-E` "b-∅-l-E`neut "b-∅-l-ù "b-∅-l-à "b-∅-l-èTable 2.50: Single-consonant roots: √b- ‘to be’The theme of this participle is always [-∅], as well, except in the masculinesingular form, where it occurs as [-i]. The masculine plural exponent is [-i],as in (2.49). Notice, also, that stress is always on the inflection in all theforms but the masculine singular form, where it occurs on the theme [-i]. Itwould appear that stress and the theme [-i]∼[-∅] alternation are correlatedin this participle, as well. More crucially, however, we have now discoveredthat the [-i] may also occur in the absence of a ‘√CVCr’ environment;it seems that the [-i] simply appears in these two paradigms becauseit can be stressed due to it being the only vowel in the form. This stillimplies that the [-i] is correlated with phonotactic environment: itmust appear either in the ‘√CVCr’/‘√CVC-r’ environment, or if it bearsstress. More on this alternation shall be said in section Interim overview: Verbs and participlesAs we have now reached the end of our detailed discussion on participialmorphology in NM Slovenian, now is the appropriate time to emphasize themost relevant observations that shed light on the [i]∼[∅]∼[@] alternation inverbs and participles in NM Slovenian. In the verbal system, as discussedin section 2.2.2, we discussed the a-class, i-class, ∅-class and e-class ofverbs. Throughout the entire verbal system, the theme vowel only occurs ifit is stressed, but never as unstressed.In the system of participles, the theme vowel has the same distribution:it never surfaces unstressed. In addition, participles exhibit variable stressassignment, where the stress varies between the root and the theme. Withi−∅-class participles, i−∅-class semelfactive participles and also √CVCr-participles the theme is always as zero when the root is stressed but when thetheme is stressed is is realized as [-i]. This is a active synchronic alternationthat seems indicative of a vowel-deletion process affecting the theme vowel442.2. Status of the high vowel [i]/-i/, as [i] and zero are in perfect complementary distribution with respectto stress assignmentFurthermore, we have made a crucial observation with respect to suffix [-i]. This suffix will never be realized as [-i] when it is unstressed,except when it is preceded by a ["CVC] configuration that is followed bya sonorant: all semelfactives, making up ["√CVC-n], as well as √CVCrshaped roots trigger the realization of the as [-i]. It is also realized as[-i] when it can be stressed, as expected, which occurs with the two single-consonant roots. This alternation seems to be connected to the alternationof the theme [-i] with zero, but with an extra phonotactic condition.All the verbal and participial data that we have examined so far suggestthat the alternation between the theme [-i] and zero, and the [-i] andzero, are phonological in nature: they are correlated with purely phonologi-cal factors and cut across different morphological classes of participles. Thestatus of these two alternations is further elaborated on in section 2.2.6 andit receives a formal analysis in Roots and prefixesWe have now determined that the high vowel [i], when realizing the themevowel [-i] and the masculine plural exponent [-i], alternates with zero andschwa within two different stress-based realizations of the same participialparadigm. This alternation seems to correlate with the variable stress as-signment: recall from sections 2.2.3 and 2.2.4 that the theme [-i] has to occuras zero if it is not stressed, while the masculine plural inflection occurs as[-i] only when it is stressed, or when it is unstressed but this happens onlyin ‘√CVCr’/‘√CVC-r’ phonotactic environments.Since the only cases of alternation between [i], zero and schwa that wereobserved were in the suffixes, it is important to also survey roots and prefixesto determine if any [i] segments alternate with zero or schwa under the samestress conditions. Let us begin with roots. There do not appear to be anyroots in NM Slovenian that would have an [i] in a position where it directlyfollows a stressed syllable. However, [i] may occur in a position that directlyprecedes the stressed syllable:root i"gR-à-l-á ‘play’pis- pi"s-à-l-á ‘write’isk- is"k-à-l-á ‘search’Table 2.51: Roots with an unstressed [i] in ptc452.2. Status of the high vowel [i]In the cases listed above and others not given here, [i] occurs without anyalternations being induced. In any derived or underived form, the rootalways retains its [i] vowel segment.A more complex situation can be found in prefixes. In order to discussprefixes and the possible alternations of [i], we must first determine whichprefixes in NM Slovenian contain [i]: two possible prefixes exist that contain[i], viz. [pRi-] with the meaning ‘close to, nearby’ (or it may denote semanticperfectivity), and the other prefix is [iz-], which has the meaning ‘from, outof’ (and it may also denote semantic perfectivity). It is important to mentionthat prefixes in general in NM Slovenian cannot be stressed,13 which meansthat it will not be possible to find any positions in which [pRi-] and [iz-]would occur in a position that directly follows a stressed syllable. However,[pRi-] does seem to have an alternant, viz. [p@R-] with quite similar semanticsand a schwa instead of an [i] vowel. Let us review some data to determineif this could be a possible"√tìsk-á-m p@R-"√tìsk-a-l-á ‘press’p@R-√di"S-ì-m p@R-√di"Sà-l-á ‘smell’p@R-√"stòp-@´m p@R-√"stòp-l-á ‘stand close’p@R-√lo"Z-ì-m p@R-√lo"Z-ì-l-á ‘attach, add’pRi-"√zn-à-m pRi-"√zn-à-l-á ‘confess’Table 2.52: Prefixes [pRi-] and [p@R-] (verbs and participles)nounnompRi-"√tìsk ( ‘(air) pressure’pRi-"√dìx ( ‘feel, sense’pRi-√"stO`p ( ‘approach’pRi-√"lóg-a ( ‘side dish’pRi-"√zn-à-n-j-é ( ‘praise’Table 2.53: Prefixes [pRi-] and [p@R-] (nouns)Notice that while [pRi-] is almost always found with nouns, the prefix [p@R-] isalmost always found with verb and participles. However, this is not a clear-cut division, as [pRi-] may occur with participles and nouns, cf. [pRi-"√zn-à-l-á] ‘confess (’ and [pRi-"√zn-à-n-j-é] ‘praise (’. However,13There are, of course, cases that constitute a somewhat grey area. Consider ["pR-√id-@m] ‘I come (’ vs. [o"d-√id-@m] ‘I leave (’ vs. [i"z-√id-@m] ‘I am published(’, where it is not clear if the prefix in the first case is [pRi-] or just [pR-], but weshall set such cases aside for the purpose of this thesis.462.2. Status of the high vowel [i]in most of the forms above, there is an important semantic difference inthe use of [pRi-] and the use of [p@R-]: when [p@R-] is used with a verb orparticiple, the meaning is completely compositional (transparent), but thisis not the case with the nouns that use [pRi-]. For instance, while [p@R-"√tìsk-a-l-á] ‘press (’ denotes an act of ‘pressing’, the corresponding noun[pRi-"√tìsk] ‘pressure (’ denotes ‘pressure’, but as a technical term inthe sense of ‘blood pressure’ or ‘air pressure’. Consider also [p@R-√di"Sà-l-á]‘to emanate smell (’ as compared to the noun (built on the sameroot, viz. √dix-) [pRi-"√dìx] ‘a sense (’, which would only be used whendescribing, for instance, the mood set by a certain poem or a song, but notto describe the act of ‘emanating a smell’. In all the cases above, the prefix[p@R-] conveys the transparent, compositional meaning, whereas the prefix[pRi-] conveys a more abstract, presumably non-compositional meaning.14The observation that [p@R-] correlates with compositional semantics canbe even better illustrated as the prefix [p@R-] also occurs with nouns. Con-sider the following"√klòp-@´m p@R-"√klòp-l-á ‘connect’p@R-"√klùtS-@´m p@R-"√klùtS-l-á ‘plug in’p@R-"√tìsk-á-m p@R-"√tìsk-a-l-á ‘press’Table 2.54: Prefix [p@R-]: no semantic difference (verbs and participles)nounnomp@R-"√klO`p ( ‘connecting’p@R-"√klùtS-@´k ( ‘plug-in’p@R-"√tìsk ( ‘a press’Table 2.55: Prefix [p@R-]: no semantic difference (nouns)The meaning of the verbs, participles as well as nouns seems to be completelytransparent with the prefix [p@R-]. Notice that the participle [p@R-"√tìsk-a-l-á] ‘press (’ that was already listed in Tables (2.52) and (2.53) isagain listed in (2.54) and (2.55), crucially because its corresponding nounmay also be formed with [p@R-], i.e. [p@R-"√tìsk] ‘a press’ (, which hasa completely transparent meaning, denoting the act of ‘pressing something’.14Notice that [pRi-] in [pRi-"√zn-à-l-á] ‘confess’ (’ and the noun [pRi-"√zn-à-n-j-é] ‘praise ( seems to yield a non-compositional meaning in both cases. It is likelythat the combination of [pRi-] and the root √zn- ‘to know’ yields a non-compositionalinterpretation.472.2. Status of the high vowel [i]It seems that [pRi-] and [p@R-] are correlated with different semantics.While it is possible that these are two lexically distinct prefixes, becausethey exhibit different semantics, it is also possible that they are two mor-phologically determined allomorphs of the same prefix. We leave this issuefor a semantic treatment of NM Slovenian. What is crucial, however, is thatno alternation of [i] with zero and schwa that would be correlated with dif-ferent stress-based realizations of the same paradigm can be found. It seemsthat in the prefixes, [i] shows no stress-conditioned alternations at all.For completeness, let us also consider the prefix [iz-], for which, as wewill see, there exists a similar prefix that is perhaps its"√bRùx-n-@´m iz-"√bRùx-@n-l-á ‘burst out’iz-"√d-à-m iz-"√d-à-l-á ‘betray’iz-"√dèl-á-m iz-"√dèl-a-l-á ‘create’iz-"√klòp-@´m iz-"√klòp-l-á ‘turn off’Table 2.56: Prefix [iz-] (verbs and participles)nounnomiz-"√bRùx ( ‘outburst’iz-"√d-á-j-à ( ‘betrayal’iz-"√dèl-k ( ‘product’iz-"√klO`p ( ‘turning off’Table 2.57: Prefix [iz-] (nouns)As can be observed [iz-] may occur with verbs, participles and nouns, and itreveals no alternations of any kind. It does, however, seem that [iz-] has alexically/morphologically conditioned allomorph. Let us first mention that a[z-]/[s-] prefix exists in NM Slovenian, also denoting perfectivity of some sort.However, [z-]/[s-] is a separate prefix, as it generally is in Slovenian, thoughsome of its occurrences overlap with the meaning typically associated with[iz-]. Historically speaking, such cases are the result of high vowel deletion([iz-] > [z-]/[s-]) that have become lexicalized, as noted by Žaucer (2002: 33),blurring the division between [iz-] and [z-]/[s-].15 However, we need not beconcerned with this distribution, as [iz-] may occur in verbs, participles andnouns, and does not seem to correlate with any stress-related factors: when15The two prefixes [pRi-] and [p@R-] must also be the historical result of high vowel dele-tion and subsequent schwa-epenthesis, but, as discussed above, they show no synchronicphonological connection.482.2. Status of the high vowel [i][iz-] is, however, realized as [z-]/[s-], this only correlates with the occurrenceof specific roots; in other words, the distribution of [iz-] and it’s allomorph[z-]/[s-] truly is lexicalized, as Žaucer observes.The important observation made in this section is that no alternationof [i] with schwa and zero of the type found in the participles can be foundwith [i] vowels in the prefixes or in roots. In fact, the high vowel [i] does notseem to be restricted phonologically in any way in roots and prefixes.2.2.6 Stress and alternationsThe present section discusses the alternations that occur between the twopossible stress-based realizations of the same participial paradigm of theregular and semelfactive i ∼ ∅-class participles, and also √CVCr-classparticiples, in NM Slovenian. The theme vowel always occurs as zero orschwa in the realizations where stress is assigned to the root, and it alwaysoccurs as /-i/ when stress is assigned to the right of the root syllable. Inaddition, the masculine plural exponent /-i/ may occur in the semelfactivei ∼ ∅-class participles and √CVCr-class participles: this occurs in therealization of the paradigm with stress assignment to the root syllable, butit must be replaced by zero in realizations of the same paradigm with stressassignment to the theme vowel.The factor that underlies these alternations of theme vowel and mas-culine plural exponents is the two possible stress-based realizations of thesame paradigm. This is why this section is dedicated to discussing this phe-nomenon, while chapter 3 will discuss the status of schwa and the high vowel/i/ in NM Slovenian.Variable stress assignmentNM Slovenian has two possible ways of realizing its most prominent particip-ial paradigms, as described in detail in section 2.2.3. Recall the example/√jòk-a-l-a/ ‘having cried (’, which is an a-classIII participle, andthat it has two possible realizations: either as ["jòkalá] or [jo"kàlá]. Stressmay either occur on the root syllable or on the theme-vowel: however, thephonological properties of the formed word have no clear correlation withstress assignment, which implies that these different positions of stress as-signment do not seem to be triggered by any phonological factor. Rather,different stress patterns correlate either with specific roots, or with specificmorphosyntactic constructions. This situates the regulation of stress as-signment in NM Slovenian in the domain of morphosyntactic (i.e. lexical)492.2. Status of the high vowel [i]specification. This is precisely what happens with i ∼ ∅-class participles,i ∼ ∅-class semelfactive participles and √CVCr-class participles. Suchstress-realization essentially seems to be a case of morphological class vari-ability: assignment of stress on the root syllable is the only option for someparticiples that employ the theme /-a/ (viz. a-classI), and assignment ofstress to the theme vowel is the only option available for some other par-ticiples that employ the theme /-a/ (viz a-classII). It seems that a thirdgroup of participles, like /√jòk-a-l-a/, may belong to one class or the other.Such morphological class variability, however, can be explained in twodistinct ways in a generative theory of morphosyntax. Embick (2008) dis-cusses both these approaches and defends the one that subscribes to thefollowing principle:(12) Single Output (Embick 2008: 65)An input N to a derivation yields a single output.The Single Output hypothesis essentially refers to morphosyntactic inputs.Notice that in the NM Slovenian data, two distinct stress patterns may re-alize the same morphosyntactic input form, which is in conflict with theSingle Output hypothesis (provided that we dismiss the option of lexicalrepresentations that are accidentally homophonous on the segmental level).The Single Output approach to variable realization of the same morpholog-ical paradigm, which is upheld by Embick (2008), can only explain the NMSlovenian data in one way: a NM Slovenian speaker must possess two dis-tinct grammars, G1 and G2. In G1, /√jòk-a-l-a/ surfaces as ["jòkalá], while inG2 it surfaces as [jo"kàlá]. This implies that G1 specifies root-stress for thisparticiple, while G2 specifies theme-stress. Such an approach to variabilityis compatible with existing work on morphosyntactic variation and change(Kroch 1989; Pintzuk 1991), and also language learning (Yang 2002).The other approach discussed by Embick (2008) is that which incorpo-rates a ‘probabilistic’ component in the grammar. Under this approach, theNM Slovenian variable stress assignment can be captured in the followingway: the rule R that regulates stress assignment (a morphosyntactic rulein NM Slovenian, as noted above) is regulated by a ‘probabilistic opera-tor’ P, which means that P essentially determines whether R applies in agiven derivation or not, and its application is probabilistic. Embick (2008:68) claims that the probabilistic approach to variation may in some casesweaken the ‘Modularity Assumption’, viz. that ‘Grammar and language useare modularly distinct’, which implies that we no longer have a theory ofpure linguistic competence. Embick claims that this happens with cases502.2. Status of the high vowel [i]where ‘external’ factors, e.g. sociolinguistic factors, play a role in determin-ing which ‘variant’ is used, which introduces the notion of ‘use’ to a theory ofgrammar, viz. ‘socilinguistic contexts [would need to be] built into the prob-ability calculation’ (Embick 2008: 68) in the grammar. However, this is notan assumption that is accepted by everybody: for instance, Nevins & Parrot(2010) disagree with Embick (2008) on this issue and claim that a proba-bilistic component that explains some variable aspects of grammar need notweaken the Modularity Assumption, which means that the use of probabilis-tic operators need not push a theory from the domain of linguistic compe-tence to the domain of use. While we cannot engage in further discussionon this topic here, we will ultimately side with Nevins & Parrot (2010) andmay assume a probabilistic operator in our analysis (though nothing crucialwill hinge on this), especially since the stress variability in NM Sloveniandoes not seem to be correlated with any apparent sociolinguistic factors.Probabilistic approach: P(stress)At this point it becomes important to assess if the two approaches to vari-ability can make different predictions for the analysis of the alternationsbetween /-i/, zero and schwa. To give a concrete example, let us take themasculine plural of a √CVCr-class participle, namely ["√pòw@n-∅-l-i] /[√pow"n-ì-l-∅] ‘having filled’. The alternation of [-i], zero and schwa be-tween the two realizations of the morphological paradigm of the participlebuilt on √pòwn- is correlated with stress assignment, as already discussedabove and in section 2.2.3. If we are dealing with one grammar with proba-bilistic stress assignment, then it is clearly the case that the alternations inquestion are correlated with the variable stress patterns:(13) "√pòw@n-∅-l-i / √pow"n-ì-l-∅ under probabilistic stress assignmenta. "√σ-σ["pòw@nlí] ; thm [∅], [-i]b. √σ-"σ[pow"nìl] ; thm [-i], [∅]If this is parsed by a single grammar, then the alternation in the thm must be phonological because it is conditioned by a phonological factor,viz. the location of stress, and the alternations are manifested under thesame morphosyntactic conditions (i.e. those that make up a participle with√pown-). According to this, it is tempting posit a process of vowel deletion,512.2. Status of the high vowel [i]so that the theme [-i] can be deleted when the root is stressed, and that thefinal masculine plural exponent [-i] is deleted when the theme is stressed.16Such an analysis could then assume that all the alternating participleshave /-i/ as the exponent of thm specified in the UR, and we can evenextend this to the ∅-classI and i-classI/II because they reveal the exactsame distribution of thm as [-i] and zero under different stress assignments,as discussed in 2.2.3 and 2.2.4. These alternations of [i] with zero and schwa,under different, probabilistic stress conditions, would be rendered as a directresult of vowel deletion that targets [i] in specific positions. Additionally, wecould also claim that all of the participles (even the ones that employ othertheme vowels) have exponed by /-i/ in the UR, and that this /-i/ isdeleted by the same process that we have posited to derive the deletion ofthe theme vowel. If such a process of vowel deletion has any grounding inthe phonology (in 3.2 we will show that it indeed does17), then this is a veryeconomic way of deriving the alternations in question: all that the grammarneeds to possess is a (morphosyntactic) operator that induces different stressassignment probabilistically (which is also needed on independent groundsfor a-class participles) and a vowel deletion process in the phonology. Nospecial morphologically predetermined allomorphy need be stipulated.However, one could alternatively speculate that the two surface formsgiven in (13) are the result of pure morphological manipulation: it would, inprinciple, be possible to claim that morphology is sensitive to the locationof stress, and that it assigns /-∅/ for the thm when the stress is on theroot syllable and /-i/ for the under these same conditions. This wouldderive the form in (13a). To derive the form in (13b), we would have toclaim that another morphological rule that is also sensitive to the presenceof stress assigns /-i/ as thm when the stress occurs to the right of the rootsyllable and /-∅/ to when the stress directly precedes it.However, there are two issues with such a purely morphological analysis.Firstly, it is very stipulative since it is not grounded in any way, whilethe process of vowel deletion can be grounded phonologically (see section3.2), as it correlates with purely phonological factors, making it predictable:ignoring this would imply a lost phonological generalization. Secondly, amorphological solution is much less economical than the solution of vowel16The schwa occurring in (13a) can be explained as the result of general phonotacticrequirements of NM Slovenian – see section 3.1 on this.17Specifically, the exponent /-i/ surfaces only under specific phonotactic condi-tions: recall that it only occurs with roots with a ‘CVC+sonorant’ structure, cuttingacross different morphological classes, and the constraint driving its deletion can be ap-propriately grounded in terms of markedness, see 3.2522.2. Status of the high vowel [i]deletion: to derive the forms in (13) morphologically, it is necessary toaccommodate three distinct lexically conditioned factors: (i) two differentstress assignments, (ii) two allomorphs of thm for the same root, (iii) twoallomorphs of for the same root. The phonological solution of voweldeletion, however, must assume that the grammar contains a process ofvowel deletion, operating under certain factors, and that two stress patternsare possible for the given underlying form. This is, essentially, an Occam’sRazor argument against the purely morphological analysis of (13).The most economical and grounded (as we are yet to show) analysis istherefore the one which, firstly, assumes that the system of morphology setsspecific stress patterns for specific classes of participles, and that these stress-rules may be probabilistic (regulated by P) for some classes; and secondly, itassumes that the phonological grammar contains a process of vowel-deletiontargeting the high vowel [i], as schematized below:Morphology P(stress) → /"√pòwn-i-l-i/ /√pow"n-ì-l-i/Phonology Vowel deletion "pòwnlí pow"nìl@-epenthesis "pòw@nlí vac.["pòw@nlí] [pow"nìl]Table 2.58: Probabilistic stress and vowel-deletionMultiple grammars approachNow that we have explained why vowel-deletion should be employed undera probabilistic theory of stress assignment, let us turn to the alternativeapproach, viz. the multiple grammars solution, which respects the SingleOutput hypothesis. Let us again consider the masculine plural of the par-ticiple ["√pòw@n-∅-l-i] / [√pow"n-ì-l-∅]:(14) "√pòw@n-∅-l-i / √pow"n-ì-l-∅: multiple-grammars (first version)a. G1 ; root stress, thm /-∅/, /-i//"√pòwn-∅-l-i/ → ["pòw@nlí]b. G2 ; thm stress, thm /-i/, /-∅//√pòw"n-i-l-∅/ → [pow"nìl]Much like the probabilistic approach to variability, the multiple-grammaranalysis can also attempt a purely morphological analysis of the two out-puts ["pòw@nlí] and [pow"nìl]. It is possible to claim that G1, which places532.2. Status of the high vowel [i]stress on the root syllable, contains an input form /"√pòwn-∅-l-i/, whichproduces the output ["pòw@nlí]. G2, on the other hand, assigns stress on thetheme vowel, and contains the input /√pòw"n-i-l-∅/, the output of which is[pow"nìl]. However, even in the multiple-grammars approach it seems thatsuch a morphological solution should be disfavoured. Assume the alterna-tive: G1 has stress assignment on root syllables, and G2 has stress assign-ment on theme vowels. If we simply assume that the phonological grammarof both G1 and G2 contains a vowel deletion process, the outputs ["pòw@nlí]and [pow"nìl] follow from that assumption automatically.(15) /√pòwn-i-l-i/: multiple-grammars (final version)a. G1 ; root stress, Vowel deletion/√pòwn-i-l-i/ → ["pòw@nlí]b. G2 ; thm stress, Vowel deletion/√pòwn-i-l-i/ → [pow"nìl]In other words, selecting a less complex analysis is a matter of derivationaleconomy: the morphological analysis requires the two grammars to posit(i) different stress assignments, (ii) different theme vowels, and (iii) differ-ent masculine plural exponents, all for the same root, viz. √pown-. Thephonological alternative, however, only requires the two grammars to positdifferent stress assignments, while the input forms (i.e. /√pòwn-i-l-i/) canbe precisely the same, and the phonological grammars both contain thevowel deletion process. This makes it possible to maintain minimal differ-ence between the two grammars (surely a desirable result), as the phonolog-ical approach only requires them to differ in terms of stress assignment, noother morphologically specified allomorphs need to be posited. It seems thateven under the multiple-grammars analysis, the solution to the two outputs,viz. ["pòw@nlí] and [pow"nìl], should be derived by a phonological process ofvowel deletion that targets the high vowel [i].Predictions for UR’sIn this section, we have established that the variable stress-based alterna-tions between [i], zero and schwa need to be recognized as a phonologicalreaction on the different (variable) stress patterns. This implies that thethm [-i] is specified in the UR for the participial classes which participiatein the aforementioned alternation, and the exponent should be ana-lyzed as /-i/ in the UR of every participle (full phonological argumentation542.2. Status of the high vowel [i]will be given in 3.2).18class thm thm m.pla. a-class /-a/ /-i/ [-a] [-∅]b. i-class /-i/ /-i/ [-i] [-∅]c. i ∼ ∅-class /-i/ /-i/ [-i]∼[-∅]∼[@] [-∅]d. ∅-classI /-i/ /-i/ [-∅] [-∅]e. √CVCr-class /-i/ /-i/ [-i]∼[-∅] [-i]∼[-∅]f. ∅-classsemelf /-i/ /-i/ [-∅] [-i]g. i ∼ ∅-classsemelf /-i/ /-i/ [-i]∼[-∅] [-i]∼[-∅]h. ∅-classII /-∅/ /-i/ [-∅] [-∅]i. √single-C /-∅/ /-i/ [-∅] [-i]Table 2.59: Participial UR’s and SR’sNow all the participles in NM Slovenian have /-i/ as the exponent; wewill be able to show (in 3.2) that the absence of /-i/ on the surfacein (a–d) and (h) above is due to the same phonological process of deletionthat deletes the thm /-i/ under different stress conditions in the remainingparticipial classes. This is what makes specifying /-i/ in these classes a muchmore economical solution than to specify an additional morphologically de-termined allomorph (viz. /-∅/) for their exponent,19 i.e. the phonologycan derive the absence of in (a– d) and (h) ‘for free’, since it alreadycontains the appropriate process that is needed to derive the rest of the par-ticipial classes. However, the crucial piece of evidence will come from thefollowing observation: the /-i/ happens to show up as [-i] whenevera √CVCr (sonorant) cluster occurs, which cuts across the morphologicalclasses (semelfactive and √CVCr-class participles) and so requires the ex-planation of why /-i/ surfaces to be in the domain of phonology. Thiswill be an important factor that we will consider when we discuss the detailsof the vowel deletion process that targets [i] in section 3.2. The argumentsfor positing the /-i/ as the only available exponent are, therefore, thatof derivational economy and that of capturing a phonological generalization.18The case of the√single-C ["blì] ‘to be (’ and the supposed [i]∼∅ alternationsit exhibits (given in (2.50)) will be explained as a result of morphologically determinedallomorphy in section3.2 and chapter 4. For now, let us just assume it has a zero themesuffix.19In fact, some theories of morphological exponence, such as the version of DistributedMorphology advocated by Embick (2010), predict that it is impossible to specify two mor-phologically determined allomorphs of a morpheme that follows the same overt morpheme;in our case, this is the participial morpheme (overt /-l/), which is followed by the m.plinflection.552.2. Status of the high vowel [i]Notice that an analysis of vowel deletion, where the vowel [i] is targetedin specific position, also allows us to unify the classes in (b– g) in terms oftheir theme vowel specification. If we supply all these classes with the theme/-i/, the phonology should output the correct surface forms, as [i] and zero(and schwa) are in perfect complementary distribution with stress in theseclasses. The ∅-classII might also seem like a likely candidate for this atfirst glance, but looking back section 2.2.3, the roots of this class seem tohave no root vowel specified in the UR. In addition, their stress specificationis such that they would stress any theme vowel that would be assigned tothem in the UR: recall that √mR- ‘die’ builds a participle with two schwasand no [i], viz. [u-"√m@´R-∅-@`l-∅] ‘having died (’. The single consonantroots in (i) are also inherently different from any other participles in NMSlovenian, which is why they cannot be assigned /-i/ as the exponent oftheir theme suffix; we shall not elaborate on them here, but see sections 3.2and 4 for discussion.Observe that in Table (2.59) the three stress-based instantiations of thea-class participle have been subsumed under one rubric, as the differentstress realizations trigger no relevant alternations (alternations across theirdifferent paradigms, or the same paradigm with different stress assignmentfor the case of variable stress). However, looking at the underlying forms,it now seems that we have unified a great number of the other seeminglydifferent classes morphologically: the ∅-classI, i-class and i ∼ ∅-class,√CVCr-class as well as the semelfactives have the same underlying forms,in terms of the theme vowel and the inflections, just with different stressrealization. The semelfactives are somewhat different still, because of theirspecial semantics (and morphology, recall the semelfactive morpheme /-n/),but crucially even they now have the same theme vowels and the same ex-ponents of inflection as the other mentioned classes. Such a unification ofdifferent morphological participle ‘classes’ is a benefit of positing a phono-logical vowel deletion process that targets [i], which is surely a desirableresult.56Chapter 3Phonological analysisThis chapter offers the phonological analyses of schwa and the high-vowel/i/ in NM Slovenian. In 3.1, schwa is analyzed as an epenthetic vowel whichis the result of a repair strategy for illicit phonotactic consonant clusters. In3.2, the unstressed high vowel /i/ is analyzed as undergoing deletion in word-final position. 3.3 examines the status of the alternations that occur in themasculine singular of participles, while 3.4 examines whether /i/-deletionalso applies in nouns and adjectives in NM Slovenian.3.1 SchwaUp to now, we have encountered the vowel schwa in two specific positionsin our discussion of participles in NM Slovenian: one was its occurrencebetween the root and the participial exponent /-l/ in i ∼ ∅-class participles,cf. ["√xRàn-∅-@´l-∅] ‘having fed (’, and the other between the root andthe semelfactive exponent /-n/ in semelfactive participles, cf. ["√pìx-@n-l-i] ‘having blown (’. This section, however, presents a more generalpicture of schwa in NM Slovenian in order to discuss its distribution, anddetermine precisely what its role is in the system of participles. In whatfollows, the standard version of Optimality Theory (Prince & Smolensky2004) is assumed in the discussion of the phonological properties of schwain NM Slovenian.3.1.1 Phonotactics and schwaThe only pre-existing treatment of Slovenian schwa in a theory of generativephonology is that given in Jurgec (2007a,b), which concentrates on StandardSlovenian as spoken in the capital city, Ljubljana. We will show that thecrucial aspects of schwa distribution in that version of Slovenian carry overto NM Slovenian. Jurgec treats most instances of schwa in Slovenian ascases of epenthesis motivated by purely phonotactic reasons, viz. sonoritysequencing through the constraint SonSeq (Prince & Smolensky 2004) andthe avoidance of consonantal nuclei through *Nuc/C.573.1. SchwaMuch like Standard Slovenian, NM Slovenian contains noun paradigmswith nouns whose roots consist of only two or more consonants. Two suchexamples are given below, viz. √ps- ‘dog’ and √sn- ‘dream’:√p(@)s- ‘dog’ √s(@)n- ‘dream’ "p@`s-∅ "s@`n-∅ "ps-à "sn-à "ps-ù "sn-ùTable 3.1: Schwa in nounsOnly the first three cases are given above, i.e. the accusative, locative andinstrumental are omitted for ease of exposition. Notice that schwa onlysurfaces in the nominative case in which no overt suffix is available. However,in the other cases with overt suffixes the schwa is not present. The sameobservation can be extended to adjectival paradigms:√"pO`t-(@)n- ‘sweaty’ √dèl-(@)n- ‘partial’ "pO`t-@´n-∅ "dèl-@´n-∅ "pO`t-n-á "dèl-n-á "pO`t-n-ú "dèl-n-úTable 3.2: Schwa in adjectivesIt is in principle possible to analyze these alternations between zero andschwa in two ways: we may assume that schwa is specified in the UR andthat it deletes wherever it would occur in an open syllable on the surface,or we may assume that schwa is not specified underlyingly and that it isepenthesized in the contexts with zero suffixes. For Standard LjubljanaSlovenian, Jurgec (2007a,b) analyzes such instances of schwa as cases ofphonotactically triggered epenthesis; that analysis is based on two observa-tions: one is that syllabic consonants appear to be non-existent in LjubljanaSlovenian, which offers the generalization that such phonotactic configura-tions are avoided. Jurgec analyzes this as an effect of a high-ranked *Nuc/Cconstraint:(16) *Nuc/C (Prince & Smolensky 2004)Assign a violation mark for every consonant in the nucleus.Jurgec interprets the occurrence of schwa as the result of epenthesis inducedby the phonotactic requirements of the language; for instance, *[ps"] is not alicit phonotactic configuration in Ljubljana Slovenian, which is why schwa583.1. Schwais epethesized as a repair strategy. NM Slovenian does not seem to haveany instances of consonantal nuclei either, which is why this same reasoningmay be adopted for our analysis. Notice that specifying schwa underlyinglyand assuming that the grammar deletes it in certain positions would renderthe occurrence of schwa much more accidental: a deletion analysis wouldneed to specify the schwa in the UR of every root and suffix where schwaalternation occurs on the surface. Since such schwa alternation seems tooccur across different morphological contexts, an epenthesis analysis seemsless stipulative, explaining schwa as a phonotactic repair strategy.The second observation is that Ljubljana Slovenian never shows se-quences of coda consonants that constitute a rise in sonority, nor does itshow sequences of coda consonants that are insufficiently dissimilar to con-stitute a clear drop in sonority. This can be couched in the tendency of codaclusters to exhibit a fall in sonority, following the assumptions of a sonorityscale like the following:(17) Sonority scale (Clements 1990; Smolensky 1995; Parker 2011)vowels > glides > liquids > nasals > fricative > stopThis scale is valid for Slovenian (Jurgec 2007a: 7), and Jurgec takes theobservation at hand to imply a generalization, encoded by a high-rankedSonoritySequencing constraint (Clements 1990), prohibiting rising (orrelatively ‘level’) sonority in complex codas. We may use the constraintSonSeq as defined by Kager (1999):(18) SonSeq (Kager 1999: 267)Complex onsets rise in sonority, and complex codas fall in sonority.Jurgec again explains the occurrence of the alternating schwa as the re-sult of phonotactically motivated epenthesis; in other words, it occurs asstrategy that repairs illicit sonority relations in coda clusters. The samegeneralization is again found in NM Slovenian, which we discuss below.The core data that we will examine will crucially involve the lateral [l]and the nasal [n]. Liquids and nasals, in general, are unable to form the sec-ond consonant of a coda cluster in NM Slovenian (and Standard Slovenian)(Jurgec 2007a,b), which is what the high-ranked SonSeq correctly predicts.If the first consonant in the cluster is an obstruent, then this yields an {ob-struent+sonorant} coda cluster, constituting an illicit rise in sonority in thecoda. Recall the adjective ["pO`t-@´n-∅] ( ∼ ["pO`t-n-á] (‘sweaty’, where the nominative singular masculine form contains a schwathat breaks up the potential ‘[t]+[n]’ coda cluster. If, on the other hand,593.1. Schwawe have {sonorant+sonorant} coda clusters, the two consonants do not ex-hibit a sufficient fall in sonority, even if the second consonant is a nasal andthe first one a lateral liquid. Consider the adjective ["dèl-@´n-∅] (∼ ["dèl-n-á] ( ‘partial’, where the nominative singular contains aschwa that breaks up the potential ‘[l]+[n]’ coda cluster. This seems tobe particularly strong generalization because we may find such instances ofschwa with relatively recent loanwords:√"fil(@)m- ‘film’ √sa"tùR(@)n- ‘Saturn’ "fìl@´m-∅ sa"tùR@´n-∅ "fìlm-á sa"tùRn-á "fìlm-ú sa"tùRn-úTable 3.3: Schwa in loanwordsThe sonority based restriction in NM Slovenian seems to yield a strong gen-eralization and so renders such alternating cases of schwa fully predictable(given the existence of the lexical items discussed above, of course). In sec-tion 3.1.4, we will see that ‘obstruent+obstruent’ coda clusters need not bephonotactically illicit in NM Slovenian, as they may not trigger epenthe-sis (e.g. []σ , or []σ are fully licit), but ‘consonant+sonorant’ codaclusters are always illicit, triggering epenthesis. The latter is true acrossthe grammar. This implies that NM Slovenian motivates a specific sonorityscale that determines which clusters count as expressing ‘sufficiently falling’sonority, but this is not surprising given the fact that different languagesmay motivate different sonority scales (Steriade 1982; Rice 1992).Arguing for an analysis of schwa that posits its presence underlyinglyand then has the grammar perform schwa-deletion in open syllables seemsmore stipulative, as all the schwa vowels discussed up to now can be ex-plained as the result of a phonotactic repair strategy. Throughout the restof this section, we will show that the relevant instances of schwa (those inthe participial system) can be directly predicted by the two phonotactic con-straints that we have discussed. We will thus assume an analysis of schwathat explains its presence through phonotactically motivated epenthesis.2020While the analysis of schwa, as epenthesis, presented in this thesis captures its overallpattern in NM Slovenian, potential counter-examples may be identified. I have comeacross only one such counter-example, viz. ["b@z@k-∅] ‘elderberry (’ ∼ ["b@zg-a]( – thanks to Joseph Stemberger for pointing it out – where the ‘[s/z]+[k]’ codacluster does not usually trigger epenthesis in NM Slovenian; see section 3.1.4 and footnote29 in particular for more on this. However, cases of ‘consonant+sonorant’ coda clustersseem to be generally absent in NM Slovenian, which still offers a relevant, though morespecific phonotactic generalization. Future work should examine this in greater detail.603.1. SchwaHowever, in the final subsection (3.1.5), we will briefly contrast this analy-sis with a schwa-deletion account of the NM Slovenian data, and show thatsuch an alternative analysis misses some basic observations in NM Slovenian.Since the schwa-epenthesis analysis manages to capture a plausible expla-nation through the two phonotactic restrictions that we have discussed, wewill demonstrate that it is ultimately more appropriate to choose such ananalysis over the deletion one.We may, therefore, proceed by assuming that alternating schwa vowelsare due to phonotactically motivated epenthesis, but we will proceed care-fully and always check if such phonotactic motivation can be found. The twophonotactic generalizations presented above were formalized by positing ahigh-ranked SonSeq constraint and a high-ranked *Nuc/C. To constructan analysis we must also assume that the grammar of NM Slovenian has thefollowing two constraints that will have to be low-ranked:(19) a. *@Assign a violation mark for any schwa in the output.b. DepAssign a violation mark for any segment in the output that hasno correspondent in the input.These constraints must be ranked in a way to promote schwa-epenthesis asa repair strategy for configurations with consonantal nuclei.21/ps-∅/ SonSeq *Nuc/C Dep *@a. "ps"*!b. ☞ "p@s * *Table 3.4: Repairing consonantal nuclei: /ps-∅/ → ["p@`s]/sn-∅/ SonSeq *Nuc/C Dep *@a. "sn"*!b. ☞ "s@n * *Table 3.5: Repairing consonantal nuclei: /sn-∅/ → ["s@`n]21Note that from now on I omit tones from the representation in the candidates of OTtableaux primarily for better readability and because they play no role in the topics underdiscussion.613.1. SchwaIn this way, schwa is epenthesized in order to avoid having [s] or [n] as thesyllable nucleus in the examples above. This same ranking also promotesschwa-epenthesis as a repair for illicit sonority relations in coda clusters:/"film-∅/ SonSeq *Nuc/C Dep *@a. "film *!b. ☞ "fil@m * *Table 3.6: Repairing illicit sonority: /"fìlm-∅/ → ["fìl@´m]/"pO`t-n-∅/ SonSeq *Nuc/C Dep *@a. "pOtn *!b. "pOtn"*!c. ☞ "pOt@n * *Table 3.7: Repairing illicit sonority: /"pO`t-n-∅/ → ["pO`t@´n]These tableaux illustrate that very often it is SonSeq and *Nuc/C togetherthat promote schwa epenthesis as a phonotactic repair strategy. However,for ease of exposition, I will only represent *Nuc/C in the tableaux fromnow on, unless specific reference to SonSeq will be necessary.22The constraint set used above does explain why schwa must occur, how-ever, it says nothing about the site of epenthesis. Notice that these con-straints alone would not be able to exclude candidates that would epenthe-size schwa in absolute word-final position, yielding *["ps@], *["sn@], *["pO`tn@´]and *["fìlm@´]. For now, we shall assume that NM Slovenian must also con-tain a high-ranked constraint that prohibits schwa from occurring in opensyllables, which explains the ungrammaticality of *["ps@], *["sn@], and otherexamples.23 We, therefore, require a markedness constraint along the fol-lowing lines:22Note that there are unpredictable cases of schwa in NM Slovenian – see footnote 20and section 3.1.4. This implies that schwa may also be specified underlyingly, which meansthat the ranking of Max(@) with respect to *@ will need to be ‘Max(@) ≫ *@’, where Dep(@)is ranked lower than the Dep-constraints on the other vowels. This predicts a system withunderlying schwa, as well as schwa-epenthesis, which is precisely what we need in orderto adequately model NM Slovenian.23A candidate such as *["@ps] is also not an option – it would violate SonSeq, but alsofor another important reason: in most cases, schwa in initial position is not availablebecause it violates constraints that prohibit deletion or insertion in initial position in theProsodic Word – see footnote 33.623.1. Schwa(20) *@]σ (MacBride 2000: 8)Schwa is not in an open syllable.Macbride (1996) explains the failure of schwa in Berber to occur in opensyllables (Guerssel 1976; Chtatou 1982; Dell & Tangi 1992) as the resultof the high-ranked constraint *@]σ , and similar phenomena are attested inSalish and Wakashan languages (Shaw 1996; Blake 2000). For now, we shallmake use of *@]σ . Because we will analyze the data in a step-by-step fashion,we will ultimately be forced to abandon *@]σ for an Alignment constraint(see section 3.1.3), which will unify the range of data that we will consider.Observe *@]σ at work in NM Slovenian:/ps-∅/ *Nuc/C *@]σ Dep *@a. "ps"*!b. "ps@ *! * *c. ☞ "p@s * *Table 3.8: Schwa epenthesis and *@]σ : /ps-∅/ → ["p@`s]/sn-∅/ *Nuc/C *@]σ Dep *@a. "sn"*!b. "sn@ *! * *c. ☞ "s@n * *Table 3.9: Schwa epenthesis and *@]σ : /sn-∅/ → ["s@`n]In what follows, we will see that *@]σ explains most epenthesis sites in thedata of verbs and participles.It is now time to discuss the status of schwa in verbs and participles.In the remainder of this section, we will consider the interaction of schwaand the high vowel /i/; specifically, we will discuss the interaction of schwaepenthesis and vowel deletion. Recall from the previous section (2.2.6) thatthe verbal theme vowel /-i/ as well as the masculine plural exponent /-i/,occurring in participles, both delete on some occasions. What the precisedetails of this deletion process are is left for the section that will deal ex-clusively with /i/-deletion (section 3.2). For now, we can just assume thata constraint such as *˘i is found in the phonology of NM Slovenian, andthat it prohibits any unstressed [i] vowels. First, consider the simple case ofi ∼ ∅-class participles:633.1. Schwaur of stem /-i/ /-a/ /-e//√xRàn-i-l-/ ‘feed’ "xRàn@´l "xRànlá "xRànlé/√kùR-i-l-/ ‘burn’ "kùR@´l "kùRlá "kùRléTable 3.10: i ∼ ∅-class participles: Root stressur of stem /-i/ /-a/ /-e//√xRàn-i-l-/ ‘feed’ xRa"nìl xRà"nilá xRà"nilé/√kùR-i-l-/ ‘burn’ kù"Ríl kù"Rilá kù"RiléTable 3.11: i ∼ ∅-class participles: Theme stressThe two tables above show two different stress realizations of the same forms,with which the occurrence or absence of the theme vowel [-i] correlates. Thiscorrelation implies that the theme vowel /-i/ is deleting, as established insection 2.2.6. However, the final masculine plural exponent is also presentunderlyingly (as also established in 2.2.6) and it deletes as well. Noticethat the schwa, in turn, only shows up in the masculine plural forms withroot stress and nowhere else. Such a distribution of schwa can be easilymotivated by the constraints that we have been assuming so far. In additionto those constraints, we assume the constraint *˘i for now, which prohibits[i] in unstressed syllables, and also Max(i), which prohibits /i/-deletion:/xRan-i-l-a/ *Nuc/C*i˘ *@] σMax(i)Dep *@a. "xRanila *!b. ☞ "xRanla *c. "xRan@la *! * * *Table 3.12: Schwa in i ∼ ∅-class participles: /"xRan-i-l-a/ → ["xRànlá]643.1. Schwa/xRan-i-l-i/ *Nuc/C*i˘ *@] σMax(i)Dep *@a. "xRanili **!b. "xRanil *! *c. "xRanli *! *d. "xRanl"*! **e. ☞ "xRan@l ** * *f. "xRanl@ *! ** * *g. "xRan@l@ **! ** ** **Table 3.13: Schwa in i ∼ ∅-class participles: /"xRan-i-l-i/ → ["xRàn@´l]]In (3.12), the theme vowel /-i/ deletes, but no schwa occurs, i.e. no schwa isepenthesized because no illicit phonotactic configuration is created through/i/-deletion. The schwa in candidate c. is ruled out both by *@]σ and thegeneral violation of the low-ranked constraints that regulate the presence ofschwa. Let us move on to (3.13): in this form, there are two /i/-vowels todelete, which could potentially result in the participial /-l/ syllabifying asthe syllables nucleus (candidate d.); of the forms with schwa, only candidatee. is available, as candidates f. and g. violate *@]σ , as well as incur doubleviolations of the lower-ranked constraints. This tableau shows that *@]σmust be active in the system to rule out candidates such as f., which wouldotherwise fare equally well as candidate e. under the rest of the constraints.What is crucial about these participial data that we have discussed sofar is that a constraint such as *@]σ does seem to be necessary to account forthe distribution of schwa. Specifically, it prohibits schwa in open syllables,just like in the system of nouns discussed before, which is exactly what weseem to require to derive the system of participles.Let us now turn to the √CVCr-class. In the previous section we showedthat √CVCr-class participles are essentially a case of i ∼ ∅-class partici-ples in that they have the same UR specification: /-i/ for thm and, of course,/-i/ for of stem /-i/ /-a/ /-e//√pòwn-i-l-/ ‘feed’ "pòw@nlí "pòw@nlá "pòw@nlé/√pRàzn-i-l-/ ‘slap’ "pRàz@nlí "pRàz@nlá "pRàz@nléTable 3.14: √CVCr-class participles: Root stress653.1. Schwaur of stem /-i/ /-a/ /-e//√pòwn-i-l-/ ‘feed’ pow"nìl pow"nìlá pow"nìlé/√pRàzn-i-l-/ ‘slap’ pRaz"nìl pRaz"nìlá pRaz"nìléTable 3.15: √CVCr-class participles: Theme stressSchwa in this class occurs only in the root stress version of the paradigm.The interesting fact is that schwa is not epenthesized in the same site wherethe theme would be deleted, but it rather occurs between the two finalconsonants of the root. The constraint ranking we have set up previouslycan easily derive this situation:/"pown-i-l-i/ *Nuc/C*i˘ *@] σMax(i)Dep *@a. "pownili **!b. "pown"li *! * *c. "pown@li * *! * * *d. ☞ "pow@nli * * * *Table 3.16: Schwa in √CVCr-class ptc: /"pown-i-l-i/ → ["pòw@nlí]/pow"n-i-l-i/ *Nuc/C*i˘ *@] σMax(i)Dep *@a. pow"nili *!b. ☞ pow"nil *c. pow@"nil *! * * *Table 3.17: Schwa in √CVCr-class ptc: /pow"n-i-l-i/ → [pow"nìl]In Table (3.16), no candidates in which the final /-i/ would be deleted,such as *["pown@l], are given as we are yet to discuss why this /-i/ surfaces(see section 3.2 on this); for now let us just consider the candidates thatretain the final /-i/. Notice that the schwa actually epenthesizes in a sitethat is different from that of the deleted vowel: it occurs between the finaltwo consonants of the root. The constraints in Table (3.16) automaticallyexplain this as the result of the constraint *@]σ (i.e. a ban on schwa inopen syllables), which unifies the distribution of schwa in √CVCr-class663.1. Schwaparticiples with the rest of the participial system and, in fact, the nounsystem: recall that schwa is epenthesized in /sn-∅/ ‘dream (’ toyield ["s@n], and a candidate such as *["sn@] must be ruled out somehow – hereby *@]σ . This represents further evidence in favour of the constraint *@]σ .In Table (3.17) the epenthesized schwa is simply redundant (/i/-deletioncreates no illicit phonotactic structure for it to repair), which stems fromthe general violation of Dep and *@.We can now consider the system of semelfactive participles. The distri-bution of schwa in both these classes is precisely as in the √CVCr-classof participles, which means that it can be unified with the rest of the datathat we have discussed so far. Consider the following examples:ur of stem /-i/ /-a/ /-e//u-√gàs-n-i-l-/ ‘turn off’ u"gàs@nlí u"gàs@nlá u"gàs@nlé/√màx-n-i-l-/ ‘slap’ "màx@nlí "màx@nlá "màx@nléTable 3.18: Schwa in semelfactive participles: Root stressur of stem /-i/ /-a/ /-e//u-√gàs-n-i-l-/ ‘turn off’ ugas"nìl ugas"nìlá ugas"nìlé/√màx-n-i-l-/ ‘slap’ max"nìl max"nìlé max"nìléTable 3.19: Schwa in semelfactive participles: Theme stressIn the semelfactives, schwa is epenthesized between the root and the semelfac-tive exponent /-n/, rather than between /-n/ and the participial exponent/-l/, in order to avoid a violation of *@]σ : surface forms such as *[u"gàsn" lí]and *[u"gàsn@lí] are clearly ungrammatical. The only difference betweenthe √CVCr-class participles and the semelfactives is one of morphologicalstructure: the √CVCr-class participles contain a root with a final con-sonant cluster where the second consonant is a sonorant (cf. √pown-, ‘fill’,√misl- ‘think’, etc.), while in the semelfactives the root is followed by thesemelfactive morpheme (cf. √gas-n- ‘turn off’, √max-n- ‘slap’, etc.). Butfor phonotactic purposes, they both show a ‘consonant + sonorant’ cluster,so it is not surprising that they show identical behaviour with respect to thephonotactically triggered process of schwa epenthesis.Notice that it is the ‘consonant + sonorant’ cluster after the stressedvowel that creates an illicit phonotactic structure: for instance, /"CVCr-l-i/can only map to *["CV.Cr".li] (where ‘r’ represents any sonorant, as before) ifno epenthesis is available, which is ungrammatical in NM Slovenian. It seemsto be an issue of syllabification, as the sonorant cannot form a part of the673.1. Schwacoda nor a part of the following onset, for independent phonotactic reasons.24That this is the case is confirmed by roots that are of the ‘CVCC’ phonotacticshape where, crucially, the final consonant is an obstruent. Consider anexample that fits in the ∅-class of participles, viz. √ust- ‘mouth:ur of stem /-i/ /-a/ /-e//iz-√ùst-i-l-/ ‘to mouth’ i"zùst@´l i"zùstlá i"zùstléTable 3.20: Participle built on √ust- ‘mouth’As can be observed above, there is no need to epenthesize schwa between [s]and [t], as [st] can be a well-formed coda as well as an onset cluster in NMSlovenian (e.g. [√kòst] ‘bone (’, [√stòw] ‘chair (’, etc.).Schwa epenthesis between the two consonants that precede the participial/-l/ is, therefore, needed only when the second consonant is a sonorant. No-tice that the presence of the final /-i/ on the surface is also conditionedby this: whenever a ‘consonant + sonorant’ cluster occurs after the stressedvowel, the exponent must surface as unstressed [-i]. This is an inter-esting correlation and one that will be crucial for our analysis of /i/-deletionin section 3.2. The discussion of this correlation is postponed until then.So far, all the cases of schwa that we have examined occur in phonotacticcontexts where we would expect epenthesis to occur as a repair strategy inNM Slovenian. We have also only witnessed evidence that corroboratesthe use of the constraint *@]σ : schwa epenthesis seems to avoid creatingconfigurations with schwa in open syllables.3.1.2 Cyclic applicationWe have by now presented several pieces of evidence which indicate thatschwa epenthesis in NM Slovenian occurs as a phonotactic repair strategy,and that it obeys the constraint *@]σ , which prohibits schwa in open sylla-bles. However, some curious properties with respect to the distribution ofschwa still need to be discussed. Some participles exhibit schwa in a positionwhere it is not expected. These are participles that belong to the ∅-classof participles and their roots only seem to consist of a few consonants un-derlyingly. Consider the following examples:24For completeness, *["] and *["CVC.rli] are equally ungrammatical. This stemsfrom the insufficient sonority fall in the coda cluster and insufficient rise in the onset, asdiscussed before, but also from the general phonotactic prohibition on σ[nl...] onsets,which are non-existent in NM Slovenian and probably in any dialect of Slovenian.683.1. Schwaur of root√"mR-/ ‘die’ u"mRèm u"m@´R@`l/po-√dR-/ ‘knock over’ po"dRèm po"d@´R@`l/u-√"pR-/ ‘resist’ u"pRèm u"p@´R@`l/√tsvR-/ ‘fry’ "tsvRèm "tsv@´R@`lTable 3.21: Schwa in an open syllable (in ∅-class participles)While these roots form participles with a zero verbal theme, they employthe /-e/ theme vowel when they form verbs. A first person singular verbalform and a masculine plural participial form is given for each root above.Notice that two schwa-vowels occur in the participial forms: a schwa in anopen syllable and a schwa in a closed syllable. The first occurrence of theschwa violates the ban on schwa on open syllables (viz. *@]σ). It seems thatboth these schwa-vowels must be epenthetic: compare the verb [u-"mR-è-m] ‘Idie’ with the participle [u-"m@´R-∅-@`l] ‘having died’. No schwa surfaces in theverbal forms, which is expected; the theme vowel is the only vowel availableand so is stressed.25 However, since the participial form is built with a zerotheme, no vowel is available, which is why schwa epenthesis occurs withinthe root to avoid creating syllabic consonants. But even so, it is unclearwhy two schwa-vowels occur in the participle: it is clear that *[u"m@´Rl] isinsufficient, [rl] cannot form a licit coda due to sonority restrictions in NMSlovenian, nor can one of the consonants become syllabic. *[u"mR@`l], onthe other hand, is completely fine in terms of phonotactic restrictions thatgovern NM Slovenian. It is, in fact, equivalent to the verbal form, exceptthat the stressed vowel is a schwa instead of the mid vowel [e]. The schwaeven occurs in a closed syllable, which satisfies *@]σ . And yet, only [u"m@´R@`l]is the grammatical form here.In chapter 4, where we discuss the morphology-phonology interface, wewill assume the framework of Distributed Morphology (Halle & Marantz1993; Embick 2010), which will allow us to treat phonological computationin cycles, implying that phonology can compute outputs of previous ap-plications of phonology. That phonological application is cyclic was firstconceived in Chomsky & Halle (1968), and this is also one of the basicclaims of Lexical Phonology (together with Stratal OT) (Kiparsky 1982a,b;Mohanan 1986; Kiparsky 2000; Bermúdez-Otero 2011), Cophonology The-ory (Inkelas & Zoll 2007; Inkelas 2008, 2011) and other instantiations ofOptimality Theory, which we do not discuss here. What is relevant for the25The initial [u] in u-√mR- is a prefix, and it is never stressed, as expected for Slovenian.The stress always falls on the syllable that follows [u], whichever syllable that is.693.1. Schwacurrent discussion is that it is the notion of a cycle that greatly disam-biguates the data with the redundant schwa-vowels in Table (3.21). Withthe schwa-related constraints that we have proposed so far, the data in Ta-ble (3.21) make perfect sense if we assume that we are dealing with cyclicapplication of phonology: recall that we are dealing with verbal and par-ticipial forms, and that participles are inherently built on verbal stems. If weassume that the verbal stems are computed in one phonological cycle, andthat the rest of the structure, be it the verbal inflection, or the participialsuffix and inflection, is computed in another, the data in Table (3.21) beginto make sense:Input – verbal stem /u-"mR-e/i-deletion vac.@-epenthesis vac.SR [u"mRe]Input – full verb /u"mRe-m/i-deletion vac.@-epenthesis vac.SR [u"mRem]Table 3.22: Cyclic application exemplified with √"mR- ‘die’: verbsInput – verbal stem /u-"mR-∅-/i-deletion vac.@-epenthesis u"m@RSR [u"m@R]Input – participle /u"m@R-l-i/i-deletion u"m@Rl@-epenthesis u"m@R@lSR [u"m@R@l]Table 3.23: Cyclic application exemplified with √"mR- ‘die’: participlesThe assumption is that the verbal stem is always built first, and then aphonological cycle is run on it. After that, the grammar may build either averb or a participle out of it – depending on the stem. When the grammaris constructing a verb with the theme /-e/, this yields no special effects.However, when constructing a verbal stem with a zero theme, the grammarmust first build a verbal stem, which triggers a pass through the phonol-ogy with the effect that we want: the UR at that point is the verbal stem703.1. Schwa/u"mR-∅-/, and when it is processed by the phonology of NM Slovenian,schwa is epenthesized for the expected phonotactic reasons. The output ofthis cycle is [u"m@R], which is used as the input for the following particip-ial cycle. In this cycle, the input, therefore, is /u"m@R-l-i/, which is thenprocessed by the phonology. The final /-i/ is deleted and schwa epenthesisoccurs for the expected phonotactic reasons. The first seemingly redundantschwa in [u"m@R@l], which occurs in an open syllable, is preserved from theprevious, verbal cycle.Notice that, with such an analysis in place, it is absolutely crucial toassume that the first cycle computes the root and the theme together. Ifthis were not the case, then we would predict that schwa epenthesis occursin the verbal forms of the examples above (e.g. *[u"m@Rem] vs. [u"mRem]),as that would mean that only the root is processed in the first cycle and,given its phonotactic form (√mR-), it would require a repair through schwa-epenthesis. The data that confirm this can also be found in other participialclasses. For instance, take the root √pown- ‘fill’: it will have to be processedtogether with the theme vowel when the stress is on the theme, so that noepenthesis occurs to repair the illicit ‘[wn]’ coda, yielding [pow"ni] as thecorrect output of the first cycle, and not *[pow@"ni]. The same can be saidabout roots like √jamR- ‘moan’ that take the /-a/ theme, which the firstcycle needs to output as ["jamRa] and not as *["jam@Ra]. In short, there isample evidence for stating that the first cycle has to encompass the roottogether with the theme, but to the exclusion of the following suffixes.Let us return to the illustration in Tables (3.22) and (3.23). That schemamakes use of rule-based mechanics, which explains the preservation of theschwa from the previous verbal cycle automatically. However, in a system ofcyclic phonology, where phonological computation is driven by Optimality-Theoretic principles, we must explain why the schwa is retained in the par-ticipial cycle. We can do this by positing a high-ranked faithfulness con-straint that prohibits the deletion of any schwa in the output of the previouscycle. We term this output the ‘base’:(21) Max-BaseAssign a violation mark for any segment x that is in the output ofthe previous phonological cycle iff there is no correspondent of x inthe output of the present cycle.Let us now use this constraint to illustrate the derivation of the masculineparticiple [u"m@´R@`l], which must pass through two phonological cycles:713.1. Schwa/u"mR-∅/ *Nuc/CMax-Base*˘i *@] σMax(i)Dep *@a. u"mR"*!b. u"mR@ *! * *c. ☞ u"m@R * *Table 3.24: Cyclic derivation of [u"m@´R@`l]: 1st cycle/u"m@R-l-i/ *Nuc/CMax-Base*˘i *@] σMax(i)Dep *@a. u"m@Rli *!b. u"m@Rl"*! * *c. u"mR@l *! * * *d. ☞ u"m@R@l * * * **Table 3.25: Cyclic derivation of [u"m@´R@`l]: 2nd cycleIn Table (3.24) above, the verbal stem with no overt theme undergoesepenthesis under the expected constraints (epenthesis avoids creating opensyllables with schwa). This is then taken as input in the next, participial cy-cle in Table (3.25).26 Since the schwa epenthesized in the verbal cycle is nowpart of the input, it can be referred to by Max-Base independently of anyother schwa-vowels that are epenthesized in the participial cycle. The rank-ing prohibits the deletion of the input-specified schwa, but another schwa isneeded to prevent the creation of a syllabic consonant. In this way, the firstschwa becomes grounded phonologically as a remnant from the ‘previous’cycle, where it was required phonotactically. But in the second cycle, itonly needs to surface for reasons of high-ranked faithfulness, which capturesthe observation that it is phonotactically not needed. This approach, then,implies that the participial system of NM Slovenian needs to pass throughat least two stages of phonological evaluation, viz. the verbal cycle and theparticipial cycle. The exact details of such a cyclic analysis, specifically whywe should even call the first cycle a ‘verbal’ one and the second a ‘participial’26No candidate such as *[u"m@Rl@] is given here. This is discussed further down.723.1. Schwaone, together with a formal definition of a ‘phonological cycle’, will be givenin section 4.3, where we discuss the phonology-morphology interface.A general question that arises with a cyclic analysis of the participialsystem is whether such an analysis makes any unwelcome predictions for therest of the data in NM Slovenian. Section 3.2 discusses this question anddemonstrates that a cyclic analysis actually explains an important aspectof the process of /i/-deletion, which means that the distribution of schwaand /i/-deletion can be adequately unified under a cyclic analysis. Theremaining part of this section, however, is dedicated to the evaluation of theconstraint *@]σ , which we have been using, to see how it fares under the newpossibilities that are opened under a cyclic analysis.3.1.3 Re-evaluating the ban on schwa in open syllablesUp to now, we have been constructing an analysis in which *@]σ , prohibitingschwa in open syllables, has played an important role. Now that we haveintroduced a new variable into our analysis, viz. the cyclic application ofphonology in the participial system, the adequacy of this constraint must bereconsidered. Let us return to the cyclic derivation of [u"m@´R@`l], in particularthe participial cycle, which is given in Table (3.25) above. Footnote 26 onpage 72 highlights the fact that no candidate such as *[u"m@Rl@] is given inthe tableau in Table (3.25). This is the case because the set of constraintsthat is used in that tableau seems to be insufficient to rule out *[u"m@Rl@],where the final schwa is epenthesized in the absolute word-final position.The participial cycle from Table (3.25) is here repeated with the addedcandidate *[u"m@Rl@]:/u"m@R-l-i/ *Nuc/CMax-Base*˘i *@] σMax(i)Dep *@a. u"m@Rli *!b. u"m@Rl"*! * *c. u"mR@l *! * * *d. ☞ u"m@R@l * * * **e. ☞ u"m@Rl@ * * * **Table 3.26: Participial cycle of [u"m@R@l]733.1. SchwaCandidate d. (the grammatical candidate) cannot be evaluated as more orless harmonic than candidate e. This stems from the way syllabificationinteracts with schwa: in candidate d., the input-specified schwa ends up inan open syllable because the epenthesis of the second schwa pushes the [R]into its onset, while in candidate e., the epenthesis of schwa in absolute finalposition pushes the [R] into the coda position of the input-specified schwa.In this way, both candidates have precisely one schwa-vowel contained in anopen syllable, and they also incur exactly the same number of violations ofthe other constraints. Our analysis must, however, explain why candidatee. is ruled out.Is it perhaps time to re-think the constraint *@]σ? It is indeed the casethat there are other ways of dealing with schwa that fails to occur in open syl-lables: van Oostendorp (2000: 141) discusses the status of schwa-epenthesisin Dutch, where schwa has a similar distribution to that in NM Slovenian,viz. it can never surface in the absolute final position of the word. While itis a fact that schwa would end up in an open syllable in the absolute finalposition of the prosodic word, it is not clear that its distribution is regulatedby a constraint such as *@]σ . In van Oostendorp (2000: 142), the prohibitionof schwa in absolute word-final position is encoded through a GeneralizedAlignment constraint (McCarthy & Prince 1993) that requires the right edgeof the morphological word to be aligned with the right edge of the prosodicword. In this way, van Oostendorp explains that the avoidance of epenthesisin edge positions is grounded in the need to keep the morphological edgesof words intact. We too can formulate such a constraint:(22) Align(MWd, R; PWd, R)∀x∃y such that x is a Morphological Word and y is a Prosodic Word,where the rightmost edge of x coincides with y.While our definition does not define the relation ‘coincide’, we take it tobe a primitive in (22) above and assume that ‘coincide’ demands the finalsegments that have morphological affiliation overlap with the final segmentdominated by the Prosodic Word. This implies that epenthesis will incura violation of Align(MWd, R; PWd, R), as the epenthetic segment will‘intervene’ between the edge of what constitutes the Morphological Wordand the edge of the Prosodic Word. This constraint, in practical terms, isthe same as Edge-Integrity (Kang 2004). For our data, Align(MWd, R;PWd, R) would prohibit schwa-epenthesis in final position, but not in anyword-internal positions. Looking back at the data discussed in this section,*@]σ was actually mostly used to rule out epenthesis in final position.743.1. SchwaBut what about cases like ["pòw@nlí] ‘having filled (’? Schwa iscrucially not epenthesized in an open syllable to avoid creating *["pòwn@lí].However, we now know, for independent reasons, that participles need topass through the verbal cycle of phonology before the rest of the participleis built. This implies that the input to the verbal cycle is /pown-i/. Could aconstraint such as Align(MWd, R; PWd, R) be used to rule out *["pòwn@´]in the verbal cycle, and so render ["pòw@´n] as the most harmonic candi-date? Align(MWd, R; PWd, R) would definitely be violated, and it wouldbe violated by *["pòwn@´], but not ["pòw@´n], as the latter does not containan epenthetic vowel between the edge of the Morphological Word and theProsodic Word. Let us see if Align(MWd, R; PWd, R) can also solve thepuzzle of evaluating [u"m@´R@`l] vs. *[u"m@´Rl@`]:/u"m@R-l-i/ *Nuc/CMax-Base*˘i Al-RMwd,PwdMax(i)Dep *@a. u"m@Rli *!b. u"m@Rl"*! * *c. u"mR@l *! * * *d. ☞ u"m@R@l * * **e. u"m@Rl@ *! * * **Table 3.27: Participial cycle of [u"m@R@l] with Align(MWd, R; PWd, R)Align(MWd, R; PWd, R) correctly predicts that candidate d. is the winner.Even though the relevant candidates have deleted the final /-i/, Align(MWd,R; PWd, R) is only violated by candidate e., but not by candidate d. Thisis because Alignment constraints do not typically encode an input-outputcorrespondence relation (which is typically done by faithfulness constraints),but only refer to whatever string can be identified in the output that hasmorphological affiliation. Since Align(MWd, R; PWd, R) manages to solvethe puzzle of evaluating [u"m@´R@`l] vs. *[u"m@´Rl@`], and since it can also derivethe epenthesis site of schwa in the rest of the data (in a cyclic analysis),we have no further use for *@]σ in our analysis. It seems that Align(MWd,R; PWd, R) is the constraint that governs the site of epenthesis in NMSlovenian, and not a ban on schwa in open syllables.753.1. Schwa3.1.4 Residual issuesIn the present section, we have provided a description of schwa in NM Slove-nian and have accounted for its distribution: in the relevant cases, schwa inNM Slovenian is epenthesized to repair a phonotactically illicit form, andthe site of epenthesis is regulated by Align(MWd, R; PWd, R). What isrelevant for the present thesis is the fact that schwa in NM Slovenian hasphonotactic motivation and acts in predictable ways. There are some occur-rences of schwa that cannot be accounted for with certainty at this point,but these do not seem to be immediately relevant for the data discussed inthis thesis. Such is the occurrence of schwa in the verbal system of ∅-classparticiples, which were discussed in Table (2.12) in section 2.2.2 and arerepeated below:indic. du pl1st.p "xRàn-∅-@´m "xRàn-∅-vá "xRàn-∅-mó2nd.p "xRàn-∅-@´S "xRàn-∅-tá "xRàn-∅-té "xRàn-∅-t3rd.p "xRàn-∅-∅ "xRàn-∅-tá "xRàn-∅-jóTable 3.28: Verbs formed with [-∅] (∅-class) du pl1st.p "xRán-∅-và "xRán-∅-mò2nd.p "xRán-∅-∅ "xRán-∅-tà "xRán-∅-tè3rd.pTable 3.29: Verbs formed with [-∅] (∅-class)Schwa in the paradigms of this class of verbs occurs in the first and secondperson singular forms of the indicative. However, let us first concentrate onthe infinitival form. Recall, from section 2.2.2, that schwa can also occurin the infinitive in this class of verbs: crucially, no schwa occurs after rootsending in a sonorant, but schwa does occur in roots ending in an obstruent,as shown in Tables (2.14) and (2.15) in section 2.2.2, respectively. This iswhy no schwa occurs in the inifinitival form of √xRan- above. Such a bi-furcation of infinitival forms could perhaps be explained through phonotac-tically motivated schwa-epenthesis. We mentioned at the beginning of thissection that sonority sequencing is active in NM Slovenian. This is some-thing that can explain the well-formedness of ‘sonorant+[t]’ coda sequences763.1. Schwaas in ["xRànt] above, but ‘obstruent+[t]’ coda-sequences could perhaps beargued to constitute a cluster where the two consonants are insufficientlydifferent in terms of sonority, and would so require schwa-epenthesis. Inother words, ‘obstruent+obstruent’ could be a sufficient coda environmentto warrant epenthesis. There is, however, insufficient evidence to makesuch a claim: infinitives that belong other verbal classes show no schwaepenthesis in ‘obstruent+obstruent’ clusters; cf. the e-class ["√nE`s-∅-t] ‘tocarry’ and ["√pE`tS-∅-t] ‘to bake’. Also, ‘obstruent+obstruent’ need not re-quire epenthesis in the noun system, consider cases of ‘[s]+[t]’ sequences:["√kòst-∅] ‘bone (’, ["√mòst-∅] ‘bridge (’, ["√pàst-∅]‘trap (’, etc. It seems that schwa in the infinitive of ∅-classverbs is regulated morphologically in some way.The schwa in the first and second person singular forms in the indicativeis equally puzzling. The schwa in the first person singular forms could still beargued to be epenthetic: any ‘consonant+[m]’ coda sequence would requireepenthesis in NM Slovenian, either due to the sonority restrictions (since[m] would never sufficiently different in terms of sonority than the precedingconsonant), or in avoidance of syllabifying the [m] as a nucleus. However,the schwa in the second person singular cannot be explained through phono-tactically motivated epenthesis so readily. Roots such as √xRan- ‘feed’, aswell as √kuR- ‘burn’ show a schwa before the inflection [-S]. It would bepossible to claim that the fall in sonority between any sonorant and [S] issimply not sufficient in NM Slovenian, but the issue lies in the fact thatsuch clusters are permitted in the system of nouns: consider, for instance,["stORS-∅] ‘pine cone (’. Several possibilities are open for this: ei-ther sonority sequencing is slightly different in the noun system than in theverbal system (which is a stipulative solution), or the schwa in these twoforms (or at least the second person singular one) is not epenthetic at all. Alikely possibility is that this schwa is a morphologically/lexically determinedallomorph of the second person singular indicative category, viz. /-@S/, onlyoccurring with roots that form ∅-class verbs. The third option is that allthe instances of the first and second person singular indicative suffixes are/-@m/ and /-@S/, respectively, and that the schwas usually delete when theycome in contact with a theme vowel in order to avoid creating a hiatus.These are stipulations that need to be explored in the future. Notice thatthis unexpected occurrence of schwa would also make it difficult to analyzethe ∅-class verbs in a way that would parallel the participles: in the par-ticiples, we were able to claim that in most cases, the surface zero theme wasthe result of an underlyingly specified /-i/ theme vowel. The evidence forsuch a claim in the system of participles is abundant, but this is not a very773.1. Schwalikely option for the verbal system. Further study of the verbal system inthe future is needed to determine the status of the aforementioned schwas.Another curious phenomenon can be observed in the paradigms of nounswith a diminutive suffix. Nouns that are formed with bare roots show noschwa epenthesis in ‘obstruent+obstruent’ codas, but if the second obstruentbelongs to the diminutive suffix, then schwa may occur in that position.Consider [√o"bísk-∅] ‘visit (, ["√písk-∅] ‘whistle (’,["√blísk-∅] ‘lightning (, and compare them to the diminutives[is-"√pís-@k-∅] ( ∼ [is-"√pís-k-à] ‘note (’, ["√líst-@k-∅]( ∼ ["√líst-k-à] ‘small leaf (’, ["√tsùk@R-tS@k-∅] ( ∼ ["√tsùk@R-tSk-á] ‘little sugar (’.The problem is that the two diminutive suffixes trigger schwa/zero alter-nations in contexts where one obstruent follows another, which are contextsthat usually require no epenthesis repair (‘obstruent+obstruent’ codas aregenerally phonotactically well-formed in NM Slovenian). Historically speak-ing, the occurrence of such vowel/zero alternations with Slavic diminutivesuffixes is not unexpected at all since the schwa must be the historical re-sult of yer realization.27 Yer-vowels typically trigger synchronic vowel/zeroalternations in the diminutives of most Slavic languages with the followingcrucial characteristics: the alternating vowel shows up in cases where thediminutive suffix is followed by a zero inflection, and it disappears whenthe diminutive is followed by an overt vowel inflection; also, the alternationin question does not seem to be phonotactically triggered, though its al-ternation is conditioned phonologically; see Gussman (2007), Jarosz (2008)and Chociej (2009) on Polish, Kenstowicz & Rubach (1987) on Slovak andGouskova (2012) on Russian. For instance, in Polish the masculine nomina-tive singular and feminine genitive plural case-forms have a zero inflection,where the alternating vowel in the diminutive suffix surfaces.In the cases that we looked at, viz. nominative and genitive singular mas-culine forms, this general Slavic observation is also true of NM Sloveniandiminutives, but the observation that the schwa in the diminutive wouldsurface with every zero inflection does not hold; to explain this we needto look at more data. The schwa in the suffix [-tS@k] does occur in themasculine ["tsùk@R-tS@´k-∅] and not in the genitive form ["tsùk@R-tSk-á], as predicted, as the latter has an overt inflection. The same istrue of the [-@k] suffix: observe ["√líst-@k-∅] ( ∼ ["√líst-k-à] ‘leaf(’. Let us now observe the feminine forms. With the suffix [-k],the schwa fails to surface in the because feminine forms have an27Thanks to Gunnar Ólafur Hansson for reminding me of this.783.1. Schwaovert inflection, as predicted: ["√lòpat-k-á] ‘little shovel (’. How-ever, the form rather surprisingly fails to reveal the schwa in thediminutive suffix as well, even though the inflection is zero: ["√lòpát-k-∅]( The same obtains for feminines constructed with the diminutive[-tS@k]: ["√òw-tSk-á] ‘little sheep (’ ∼ ["√òw-tSk-∅] ( that the form of feminines is not inert to phonotactic re-pairs through schwa epenthesis in any way. Consider the following exam-ples that illustrate schwa epenthesis for phonotactic reasons: ["√féRm-à]‘company (’ ∼ ["√féR@`m-∅] ( and ["√StèRn-á] ‘water-well(’ ∼ ["√StèR@´n-∅] (, the schwa that occurs in the diminutive forms in the nominativesingular of masculines is not phonotactically conditioned in any way, nor isit predictable from the phonological context in which the diminutive occurs.It is likely that we are dealing with morphologically specified allomorphs,viz. /-tS@k/ ∼ /-tSk/ and /-@k/ ∼ /-k/, which are the historical remnant ofSlavic yer realization in the diminutive suffixes.Another curious instance of schwa occurs in the verbal system. A processof ‘e∼@’ alternation can be found, where the theme vowel [e] alternates withschwa (these data were initially presented in section 2.2.2):sg du pl1st "pìx-n-@´-m "pìx-n-e-vá "pìx-n-e-mó2nd "pìx-n-@´-S "pìx-n-é-ta "pìx-n-e-té3rd "pìx-n-é-∅ "pìx-n-é-ta "pìx-n-e-jóTable 3.30: e∼@-alternationIt is curious that schwa should alternate with [e], which is a theme vowelavailable in the verbal system. However, no such alternation is ever foundif the theme vowel [e] is stressed: cf. [pod"√dR-e-m] ‘knock over (’,[po"√dR-e-S] (, [po"√dR-e-∅] (, etc. The alternation in Ta-ble (3.30) could perhaps be analyzed if we assumed that a markedness con-straint is prohibiting the mid vowel [e] in unstressed syllables, such as *e˘,but that the restriction on schwa in the final position in the prosodic words,or the restriction on schwa in open syllables somehow blocks this in all theforms but the 1st and 2nd singular. However, it is unclear how such arestriction could be achieved, as the system that we have developed so farhas no way of ruling out candidates such as *["pìx@n] ( or *["pìx@nvá](1p.du), etc. In addition, the postulation of two phonological cycles in theverbs and participles further complicates matters as this implies that only793.1. Schwathe stem /"pìx-n-e-/ is processed first without any additional inflectionalsuffixes. The ‘e∼@’ alternation is perhaps not even an alternation in thephonological sense, but may be an elaborate case of morphologically fixedallomorphy or the result of some other interaction of phonology and mor-phology. We will refrain from discussing the e∼@-alternation from now on;an analysis of this phenomenon is best left for a future study of schwa inNM Slovenian.The data presented in this subsection are, however, not of immediateconcern to the present thesis, as we are mainly concerned with the particip-ial system. While the distribution of schwa is an important aspect of ouranalysis, we have already provided a sufficient breadth of evidence whichindicates that schwa often acts in predictable ways and is an epenthetic seg-ment. The cases of schwa that are important for this thesis are those thatoccur between a consonant and the participial suffix /-l/ or the semelfactive/-n/ (or simply root-final /l/ or /n/): in such environments, schwa is pre-dicted to occur in any form in NM Slovenian – as discussed at the beginningof this section, because an ‘obstruent+{l, n}’ sequence would always yield anunfavourable rise in the sonority in the coda position, and a ‘sonorant+{l,n}’ sequence would yield an insufficient sonority drop in the coda position,all as expected in central Slovenian generally (Jurgec 2007b). The questionof the second person singular schwa in the verbal system, as well as thequestion of the schwa in the diminutive, are therefore independent topics,which we have only explained briefly. They should, however, be investigatedin the future to create a clearer picture of the verbal and noun systems ofNM Slovenian.3.1.5 Epenthesis vs. deletion accountIn this final section of the discussion on schwa, we will contrast the twopossible analyses of the schwa in NM Slovenian, viz. the epenthesis accountthat we have been following versus the deletion account, which we have notdiscussed. A brief overview of the schwa epenthesis account is given in (23):(23) Schwa-epenthesisa. Nouns: √sn- ‘dream’/sn-∅/ → ["s@n] ( → ["sna] ( Participles: √xRan- ‘feed’Cycle 1: /"xRan-i/ → ["xRan]Cycle 2: /"xRan-l-i/ → ["xRan@l] ( Schwac. Participles: √pown- ‘fill’Cycle 1: /"pown-i/ → ["pow@n]Cycle 2: /"pow@n-l-i/ → ["pow@nli] ( Participles: u-√mR- ‘die’Cycle 1: /u-"mR-∅/ → [u"m@R]Cycle 2: /u"m@R-l-i/ → [u"m@R@l] ( of the schwa-vowels here are specified underlyingly, but are explainedas the result of a phonotactic repair, viz. schwa epenthesis. This repairoccurs for reasons discussed at the beginning of the discussion on schwa,viz. to avoid faulty sonority sequencing in coda clusters and syllabic conso-nants. The crucially residual schwas that seem to have different motivationand cannot be adequately explained by this approach are the following (seeprevious subsection for discussion):(24) Residual schwas under epenthesis analysisa. Schwa in infinitives of ∅-class verbsb. Schwa in and indicative of ∅-class verbsc. e∼@-alternation in e-class verbsThe alternative analysis involves schwa deletion, as already mentioned.This means that the presence of schwa is posited underlyingly, as in thenoun /s@n-a/, and that the schwa undergoes deletion on the surface becauseit occurs in an open syllable, yielding ["sna] (cf. *["]). Notice that herewe cannot appeal to the use of an alignment constraint that would delete theschwa because it would intervene between the edges of the MorphologicalWord and the Prosodic Word, as the schwa in *["] is word-internal. Bypursuing such a strategy, we arrive at the following analysis:(25) Schwa-deletiona. Nouns: √sn- ‘dream’/s@n-∅/ → ["s@n] ( → ["sna] ( Participles: √xRan- ‘feed’Cycle 1: /"xRan-i/ → ["xRan]Cycle 2: /"xRan-@l-i/ → ["xRan@l] ( Participles: √pow@n- ‘fill’Cycle 1: /"pow@n-i/ → ["pow@n]Cycle 2: /"pow@n-@l-i/ → ["pow@nli] ( Schwad. Participles: u-√m@R- ‘die’Cycle 1: /u-"m@R-∅/ → [u"m@R]Cycle 2: /u"m@R-@l-i/ → [u"m@R@l] ( that schwa has to be posited within many roots, and to derive thepresence of schwa between the root and the participial /-l/ we must saythat the UR of the participial suffix is in fact /-@l/. This creates a problemwith participial forms where the stress is realized on the theme vowel, asin [pow"nil] ‘having filled (’, where we must assume that there aresome higher-ranked constraints that prohibit the hiatus between [i] and [@](cf. *[pow"ni@l]). The same has to be assumed with all participles thatemploy other theme vowels than /-i/: for instance, consider *["joka@l] ‘havingcried (’ vs. ["jokal]. However, NM Slovenian employs glide-insertion asthe hiatus resolution strategy, where the relevant examples may be foundin nouns, even in borrowings such as /ni"vo-a/ ‘level (’ with thesurface form [ni"voja].28 This would predict the surface form of /"jok-a-@l-i/to be *["jokaj@l] and not the correct form ["jokal]. This presents a potentialproblem for the deletion account of schwa.Another crucial aspect of this analysis that we must consider is the twocycles that we have posited in the computation of the verbal and participialforms. Are these cycles still necessary in the schwa-deletion analysis? Con-sider [u"m@R@l] ‘having died (’: can the direct mapping from /u-"m@R-@l-i/ to [u"m@R@l] be achieved? This does not seem possible as the candidate*[u"mR@l] would win – it has a schwa in a closed syllable and violates noneof the phonotactic constraints that are active in NM Slovenian, whereas[u"m@R@l] violates the ban on schwa in open syllables, and *[u"m@Rl] violatessonority sequencing in the final coda cluster, while *[u"m@Rl"] creates an illicitconsonantal nucleus. This is again impossible to derive without positing twophonological cycles: if /u-"m@R-∅/ is processed in the first cycle, outputting[u"m@R], then Max-Base can protect the schwa processed in the first cyclefrom deletion in the second cycle, which is very similar to our analysis withepenthesis. It seems that the schwa-deletion analysis also requires two cyclesto compute verbs and participles.So far, the schwa-deletion account does not seem to be in any way sim-pler than the schwa-epenthesis account. It is even less plausible given thetype of hiatus resolution it would employ. But let us also consider the listof ‘residual’ schwas that we could not explain through phonotactically mo-tivated schwa-epenthesis:28Thanks to Joseph Stemberger and Douglas Pulleyblank for reminding me of this.823.1. Schwa(26) Residual schwas under deletion analysisa. Schwa in infinitives of ∅-class verbsb. Schwa in and indicative of ∅-class verbsc. e∼@-alternation in e-class verbsRecall from the previous section that the schwas found in the infinitivalforms of ∅-class verbs never occur in an open syllable, which means thatthe deletion account cannot explain the presence and absence of the schwasin these forms through purely phonotactic motivation either. The same canbe said about the schwas in the first and second person singular forms ofthe indicative – they never occur in an open syllable. But what about thee∼@-alternation in e-class verbs? Here we could posit a hiatus, such as [e@],which would be resolved by deleting the [e]. However, all other instancesof hiatus are resolved by glide insertion, as noted above, implying that nopurely phonological solution is apparent under a deletion account either.Overall, it seems that the schwa-deletion account explains no more thanthe schwa-epenthesis account. To a large extent, these two approaches arenotational equivalents, in the sense that they produce the same outputsfor the data that we examined, with the exception of the hiatus resolutionwhere the deletion account fares worse than the epenthesis one. But, ingeneral, there is an an important exception between these accounts: schwa-epenthesis may be explained as a phonotactically motivated repair strategy,which makes it less stipulative than an analysis of lexically specified schwain certain positions, as the latter is just the result of accidental lexical spec-ification, whereas the former provides explanations about the phonologicalgrammar of NM Slovenian. For this reason, we will maintain the analysis of(the relevant cases of) schwa as the result of epenthesis in this thesis, thoughnothing particularly crucial seems to hinge on this.2929The ultimate analysis that would unify the distribution of all the schwas in NM Slove-nian will probably require a combination of (a) phonotactically triggered epenthesis and(b) some other mechanism that will explain the schwa in the diminutives, and also incases like ["b@z@k-∅] ‘elderberry (’ ∼ ["b@zg-a] (, as given in footnote 20,where the ‘[s]+[k]’ coda cluster is perfectly licit phonotactically. That (a) is needed isargued for by the general absence of ‘consonant+sononrant’ codas, which seem to containa schwa – these reflect the general phonotactic requirements of NM Slovenian. (b) cases,on the other hand, are purely lexical; they must be the result of historical yer-realization,which seem to play no role in the phonotactic make-up of NM Slovenian. Parallel ob-servations are found in Czech, Slovak (Mellander 1999) and Polish (Chociej 2009), andspecific mechanisms are employed in the literature to explain their patterning, consistingof a form of epenthesis for the phonotactically-driven alternations and other mechanismsfor the lexical ones (Gussman 2007); see also Scheer (2012).833.2. On the deletion of the high vowel [i]3.2 On the deletion of the high vowel [i]In sections 2.2.6 and 3.1, which dealt with topics of stress variability andschwa epenthesis, respectively, we assumed that the phonological componentin NM Slovenian contains a process that deletes the theme vowel /-i/ andthe masculine plural inflection /-i/, in some but not all unstressed positions.In fact, in 2.2.6 we argued that positing a process of /i/-deletion shouldbe preferred to a purely morphological/lexical analysis of the alternationsbetween schwa, [i] and zero because a phonological analysis does not miss aphonological generalization, and it constitutes a more economical analysis.These arguments presuppose that the phonological grammar can derive thedeletion in a plausible way. This section demonstrates that this is the case.In section 2.2.6, we discussed two phonological generalizations with re-spect to the alternations of the theme vowel and the masculine plural ex-ponent. First, in i ∼ ∅-class participles, the absence of any overt themevowel or masculine plural exponent, and the occurrence of schwa, is cruciallycorrelated with the type of stress-realization the paradigm undergoes: withroot stress realization, the theme and the exponent surface as zero(schwa may occur where necessary), while with theme stress realization, thetheme surfaces as [i], but surfaces as zero. The second phonologicalgeneralization pertains to the surfacing of the as [i] even though it isunstressed; this is crucially correlated with roots that have a ‘CVCr’ shapeand semelfactives, which yield a comparable phonological string of segments.3.2.1 First generalization: stress and i ∼ ∅Let us now discuss the first generalization in greater detail. Observe someexamples of the alternation of [i] and [∅] under different realizations of stress:ur of stem /-i/ /-a/ /-e//√xRàn-i-l-/ ‘feed’ "xRàn@l "xRànlá "xRànlé/√kùR-i-l-/ ‘burn’ "kùR@´l "kùRlá "kùRléTable 3.31: i ∼ ∅-class participles: Root stressur of stem /-i/ /-a/ /-e//√xRàn-i-l-/ ‘feed’ xRa"nìl xRa"nìlá xRa"nìlé/√kùR-i-l-/ ‘burn’ ku"Rìl ku"Rìlá ku"RìléTable 3.32: i ∼ ∅-class participles: Theme stressWith the roots specified here, the cannot surface as [i] at all. Thetheme, however, can only surface as [i] when stressed. This is reminiscent843.2. On the deletion of the high vowel [i]of a process of ‘high vowel deletion’ where /i/ deletes if it is unstressed.In terms of phonetic plausibility, high vowel deletion can indeed be suf-ficiently grounded: high vowels ([i], [u]) require ‘extreme’ articulation in-volving a ‘close jaw position’, which ‘corresponds to low inherent duration’(Crosswhite 1999: 66), and short duration decreases the perceptibility ofvowels, according to Steriade (1994), thus inviting reduction or deletion.The high vowel /i/ deleting in unstressed positions is, therefore, a wellgrounded phonological process. The crucial question that arises now, iswhat the precise scope of such a process is. The theme vowel /-i/ as wellas the /-i/ inflection delete; the former occurs word-medially and thelatter word-finally. However, there are instances of unstressed [i] that do notundergo deletion in the participles of NM Slovenian: such examples includeroots with an unstressed [i], as in [√pi"s-a-l-a] ‘having written ( [√i"gR-a-l-a] ‘having played (, and also prefixes in which anunstressed [i] never deletes – recall the prefixes [iz-] and [pRi-], which werediscussed in section 2.2.5. A process that targets any unstressed [i] seg-ment would therefore make incorrect predictions, as it would also delete the[i]-segments in all roots and prefixes. The lack of deletion in roots couldbe accounted for by a high-ranked positional faithfulness constraint thatprotects roots (e.g. Max-Root), following Beckman (1998), but no suchconstraint can account for the lack of deletion in the prefixes, as prefixescannot be referred to in a way that would exclude suffixes, because only‘root’ and ‘affix’ are normally treated as theoretical primitives. The scopeof this process that targets unstressed [i] segments for deletion must thus bedefined some other way.Let us assume that unstressed [i] is only targeted for deletion in abso-lute word-final position. Can such a process be grounded appropriately?In other words, are there sufficent factors of markedness that would triggerthe deletion of [i] in that specific prosodic position? While the high vowel[i] may be prone to deletion due to the phonetic factors discussed above(Crosswhite 1999; Steriade 1994), it is even more likely that it would delete inthe word-final position. The absolute final position in the prosodic word is avery marked position: it constitutes an intrinsically ‘weak prosodic position’(Coetzee 2004: 128), which is due to the more general ‘prosodic weakness offinal syllables, which are liable to de-stressing, de-voicing, shortening, trun-cation, and so on, under purely phonological conditions’ (Prince & Smolensky2004: 137). Coetzee (2004) proposes the following markedness constraintthat prohibits vowels in word-final position:(27) *v˘]σ ]Pwd (Coetzee 2004: 128)853.2. On the deletion of the high vowel [i]Do not allow a vowel in an unstressed prosodic word-final syllable.This constraint is akin to Free-V, a constraint prohibiting word-final un-stressed vowels, which was proposed by Prince & Smolensky (2004), and itessentially achieves the same effect. For our purposes, we can define it away that will only prohibit the vowel [i] in word-final position:30(28) *˘i]σ ]PwdDo not allow [i] in an unstressed prosodic word-final syllable.Deleting unstressed [i] only in absolute word-final position can, therefore,be appropriately grounded as a phonological process.While using such a constraint explains why unstressed [i] vowels mayoccur in roots and prefixes, it does not independently explain the deletionof the theme vowel /-i/. The theme vowel is never in absolute word-finalposition in the participles of NM Slovenian. At this point, it becomes im-portant to look back at the section on schwa (3.1), where the behaviour ofschwa with certain roots required us to posit cyclic application of phonology;specifically, we spoke of a verbal cycle (through which the verb/verbal stempasses) and the participial cycle (through which the participle passes). Thisdistinction must now be respected in all analyses of participles in NM Slove-nian. As it happens, the distinction between a verbal and a participial cycleactually automatically does away with any problems that we had in connec-tion with *˘i]σ ]Pwd being able to delete the theme vowel /-i/: the theme vowelis always an inherent part of the verbal stem, which means that the theme/-i/ will always be in absolute word-final position when deriving participles,as /√root-thm/, constituting the verbal stem, will have to pass throughthe verbal cycle of phonology before it can derive the participle. *˘i]σ]Pwdcan, therefore, adequately account for the deletion of both the theme /-i/and the inflection /-i/.To reiterate, *˘i]σ ]Pwd triggers the deletion of any unstressed [i] vowel inthe verbal as well as the participial cycle: in the verbal cycle it deletes thetheme /-i/, which occurs in word-final position there, and in the participialcycle it deletes the /-i/, which is also in word-final position. Let usillustrate this with the participle built on √xRàn- ‘feed’, by giving bothderivational cycles below:30It might very well be that only *v˘]σ]Pwd is high-ranked in NM Slovenian, and that thefaithfulness constraints which protect all the vowels from deletion, except [i], are rankedabove it. However, in what follows, we will proceed to use *˘i]σ ]Pwd.863.2. On the deletion of the high vowel [i]/"xRan-i/ *Nuc/CMax-Base*˘i]σ] PwdAl-RMwd,PwdMax(i)Dep *@a. "xRani *!b. ☞ "xRan *Table 3.33: Deriving /"xRàn-i-l-i/ → ["xRàn@´l] ( verbal stem/"xRan-l-i/ *Nuc/CMax-Base*˘i]σ] PwdAl-RMwd,PwdMax(i)Dep *@a. "xRanli *!b. "xRanl"*! *c. ☞ "xRan@l * * *d. "xRanl@ *! * * *e. "xRan@l@ *! * ** **Table 3.34: Deriving /"xRàn-i-l-i/ → ["xRàn@´l] ( participleThe two tableaux above represent the verbal and the participial cycles. Inthe first cycle, the verbal theme vowel /-i/ occurs in the absolute word-finalposition and, because of this, must be deleted. In the participial cycle, is also in word-final position and must delete for the same reason.It should be noted that the tableaux presented above should also containa high-ranked Ident-constraint (McCarthy & Prince 1995) on the featuresthat make up [i], so that no reduction of [i] is possible and that deletion ispromoted instead. Since nothing relevant interacts with this assumption,Ident-constraints will not be included in any of the tableaux below.3.2.2 Second generalization: CVCR sequencesLet us now turn to the second generalization, which concerns ‘√CVCr’ and‘√CVC-r’ sequences of segments (recall that ‘r’ stands for any sonorant).As discussed in the previous sections, the inflection does not always873.2. On the deletion of the high vowel [i]delete, but must surface on some occasions. It surfaces in the semelfactiveparticiples:ur of stem /-i/ /-a/ /-e//u-√gàs-n-i-l-/ ‘feed’ u"gàs@nlí u"gàs@nlá u"gàs@nlé/√màx-n-i-l-/ ‘slap’ "màx@nlí "màx@nlá "màx@nléTable 3.35: in semelfactive participles: Root stressur of stem /-i/ /-a/ /-e//u-√gàs-n-i-l-/ ‘feed’ ugas"nìl ugas"nìlá ugas"nìlé/√màx-n-i-l-/ ‘slap’ max"nìl max"nìlé max"nìléTable 3.36: in semelfactive participles: Theme stressThe semelfactive participles respect the generalization of i ∼ ∅-alternation asfar as theme vowels are concerned: when the theme vowel cannot be stressed(i.e. with root stress) it deletes. But for the /-i/, this generalization,which also holds with the i ∼ ∅-class of participles, is not quite surface truefor semelfactive participles. When stress occurs on the theme vowel, as inTable (3.36), the /-i/ does actually delete, but it surfaces with rootstress in Table (3.35). This is clearly not predicted by the constraints thatwe set up in the previous section: the final unstressed /-i/ should delete.The /-i/ also surfaces in participles whose root has a ‘√CVCr’phonotactic shape, as demonstrated below:ur of stem /-i/ /-a/ /-e//√pòwn-i-l-/ ‘feed’ "pòw@nlí "pòw@nlá "pòw@nlé/√pRàzn-i-l-/ ‘slap’ "pRàz@nlí "pRàz@nlá "pRàz@nlé/√mìsl-i-l-/ ‘think’ "mìs@llí "mìs@llá "mìs@lléTable 3.37: in √CVCr-class participles: Root stressur of stem /-i/ /-a/ /-e//√pòwn-i-l-/ ‘feed’ "pow"nìl pow"nìlá pow"nìlé/√pRàzn-i-l-/ ‘slap’ "pRaz"nìl pRaz"nìlá pRaz"nìléTable 3.38: in √CVCr-class participles: Theme stressThe behaviour of these participles is identical to that of the semelfactves:the theme vowel respects the generalization about deletion, but the m.plsurfaces with root stress. The roots shown above for this class are √pown-,√pRazn-, and also √misl- (the latter only has a root stress realization).883.2. On the deletion of the high vowel [i]These participles have a different morphological and semantic structure thanthe semelfactives, but crucially, the phonotactic structures of their rootsseem to form a uniform phonological class with the ‘root+/-n/’ semelfactivestructure. The surfacing of the is crucially correlated with roots thathave a ‘√CVC’ structure and are followed by a sonorant ([n] or [l] in thedata presented here).31 While the participles in which the /-i/ surfacesbelong to different morphological classes, they are identical in terms of theirphonological structure.Finally, the /-i/ also surfaces with participles that are built onsingle-consonant roots. Below is an example, viz. √S- ‘to go’:sg du plmasc "S-∅-ù-∅ "S-∅-l-à "S-∅-l-ìfem "S-∅-l-à "S-∅-l-E` "S-∅-l-E`neut "S-∅-l-ù "S-∅-l-à "S-∅-l-èTable 3.39: in single-consonant roots: √S- ‘to go’With roots that consist of only a single consonant underlyingly, the surfaces, but here it is stressed and not unstressed, which makes itspresence on the surface expected. The fact that no other vowel except theinflectional suffixes is present in such participles offers an explanation forwhy the /-i/ surfaces: it is the only vowel available in the masculine pluraland it is therefore stressed. The only other root that consists of a singleconsonant is √b- ‘to be’. We will set √S- and √b- aside for now (theyreceive further discussion at the end of this subsection); what is crucial fornow is that the surfaces with these two roots because it can be stressed.The data in which the surfaces are of utmost importance for theunderlying status of /-i/ in the entire participial system of NM Slovenian.Recall that this /-i/ does not surface at all in the a-class, i-class,and i ∼ ∅-class participles. This is why it might be tempting to analyzethis /-i/ as a special exponent of that is specific to some morphologicalclass – and, indeed, the semelfactive participles, which are morphologicallyand semantically different, seem a likely candidate for this. However, assoon as we acknowledge that this /-i/ also surfaces with completely regularroots that happen to have the same phonological structure as semelfactives,but are not semelfactives at all (viz. √CVCr-class), such an analysis be-comes less plausible. The √CVCr-class of participles is identical to the31Not many other roots that would be followed by a sonorant other than [n] or [l] andwould take the theme /-i/ exist in NM Slovenian, but see section 2.2.3 for an overview.893.2. On the deletion of the high vowel [i]i ∼ ∅-class morphologically and semantically, the only difference being thephonological shape of their root (‘CVCr’): this means that these two seem-ing morphological classes need to be collapsed into one, as already proposedin section 2.2.6 in Table (2.59); ‘i ∼ ∅-class’ and ‘√CVCr-class’ are noth-ing more than descriptive labels, as both types of participles must have theunderlying structure /√root-i-l-/, and, when forming masculine plurals,/√root-i-l-i/. This observation is very important because it means thatthe claim made above, viz. that the systematically fails to surface ini ∼ ∅-class participles, is not true: the actually has a very systematicoccurrence on the surface, as it only correlates with ‘√CVCr’ root shapes.This systematic correlation of the phonotactic shape of the root andthe occurrence of /-i/ is, therefore, phonological in nature, and itcrucially cuts across different morphological classes: the /-i/ surfaces withregular i ∼ ∅-class (√CVCr) participles and semelfactives, and also withsingle-consonant roots, in which it receives stress. Such a distribution thatreveals a clear correlation of purely phonological factors speaks in favourof a universal /-i/ exponent of the in the participial system, deletingwith ‘√CVC’ root shapes, and surfacing with ‘√CVCr’ root shapes (and‘√CVC-r’ semelfactives, of course). This reaffirms the hypothesis that /-i/is the universal exponent for in all participles, which was first advancedin section 2.2.6. This is particularly plausible given the fact that the m.plexponent in nouns an adjectives is also typically /-i/ (see 3.4 on this).However, before we can finally confirm the hypothesis of the universal/-i/ for participles in NM Slovenian, we must check if the surfacing of this/-i/ with the phonological structure specified above can be grounded in thephonology. Let us first observe what the current set of constraints that wehave been assuming predicts for these forms. Recall that when derivingparticiples, the verbal stem first passes through one cycle of phonology andthe participle passes through another. Let us construct a participle on theroot √pown- ‘to fill’:903.2. On the deletion of the high vowel [i]/"pown-i/ *Nuc/CMax-BaseAl-RMwd,Pwd*˘i]σ] PwdMax(i)Dep *@a. "powni *!b. "pown"*! *c. "pown@ *! * * *d. ☞ "pow@n * * *Table 3.40: Verbal cycle: /"pòwn-i/ → ["pòw@´n]In the verbal cycle, the theme vowel /-i/ deletes because it is unstressed andoccurs in the very final position of the prosodic word, as expected. However,we run into problems when attempting to derive the correct surface form ofthe participle:/"pow@n-l-i/ *Nuc/CMax-BaseAl-RMwd,Pwd*˘i]σ] PwdMax(i)Dep *@a. / "pow@nli *!b. "pow@nl"*! *c. ☞ "pow@n@l * ** **d. "pown@l *! * * *e. "pow@nl@ *! * ** **Table 3.41: Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí]Our current set of constraints predicts that the final /-i/ must delete inany circumstances. This is why the tableau above cannot give adequatepredictions: the surface form must be candidate a., viz. ["pow@nli], and notcandidate c., *["pow@n@l]. It is therefore a fact that we must change theconstraints in some way. The current ranking, as it stands, cannot predictthe ungrammaticality of *["pow@n@l]. What is curious about this candidateis that it contains two adjacent unstressed syllables in which schwa is in thenucleus. We have not encountered any such forms before; we have encoun-tered sequences of a stressed schwa, followed by an unstressed schwa (recall913.2. On the deletion of the high vowel [i]/u-√"mR-∅-l-i/ → [u"m@R@l] ‘having died (’), but no sequence oftwo unstressed schwa vowels. This could very well be the factor that pro-hibits the surfacing of candidate c.Avoidance of sequences of two unstressed syllables that contain schwacan be explained as a plausible phonological process. In Dutch, a simi-lar restriction obtains: under some additional morphological restrictions, asequence of two syllables whose nuclei are realized by schwas is prohibited(Booij 1995; van Oostendorp 2000, 2010). Van Oostendorp uses a constraintsuch as *@@, which simply prohibits any adjacent schwa-containing syllables,to derive this effect. There is also considerable evidence from the historicaldevelopment of Germanic languages for the dispreference of two adjacentunstressed schwa-hosting syllables (Shannon 1986, 1991): Shannon derivesthis from the more common tendency of avoiding the repetition of adjacentunstressed syllables, which he terms the ‘Syllable Sequence Law’. However,it is also possible such an effect actually stems from a more basic princi-ple of metrical organization of phonology in NM Slovenian. Coetzee (2002:50) explains the avoidance of two unstressed schwa-bearing syllables in He-brew through the constraint *Lapse (Elenbaas & Kager 1999), which pro-hibits any two adjacent unstressed syllables and prefers a steady sequenceof ‘stressed+unstressed’ syllables. It is possible that the avoidance of twoadjacent unstressed schwas in NM Slovenian stems from such a constraint: ifthe faithfulness constraints for most vowels are high-ranked (above *Lapse),there should be no concern for *Lapse affecting such vowels, but in princi-ple only epenthetic vowels (schwa) and vowels whose faithfulness constraintsare ranked below *Lapse. This means that the avoidance of sequences ofunstressed schwas can be grounded phonologically. For the purposes of thisanalysis, we may use the following constraint:(29) *@˘@˘ (*Lapse-based)Assign a violation for any sequence of adjacent unstressed syllablesthat contain schwa in the nucleus.However, the constraint *@˘@˘ is does not seem to be generally surface-true in NM Slovenian. Counter-examples may be found in diminutive con-structions in the noun system. Recall the (noun) diminutives [-k]/[-@k] and[-tSk]/[-tS@k], which were discussed in the section on schwa epenthesis (3.1).With roots that end in a consonantal cluster, where the second consonant isa sonorant, such as √tsukR- ‘sugar’, it is possible to attach the diminutive[-tSk]/[-tS@k] to the noun built on this root, and this creates a sequence oftwo unstressed schwas: consider the noun ["tsùk@´R-∅] ( ∼ ["tsùkR-á] ( with expected schwa epenthesis, and then the diminutive923.2. On the deletion of the high vowel [i]["tsùk@R-tS@´k-∅] ( The nominative singular form with the diminu-tive yields the otherwise prohibited sequence of two unstressed schwas, asstated by *@˘@˘ (*Lapse-based). However, recall that the schwa that occurs inthe diminutive forms in the nominative singular of masculines is not phono-tactically conditioned in any way. As discussed towards the end of section3.1, in such diminutive forms, the schwa seems to be specified as part ofthe suffix in the input. But regardless of that, the fact that the schwa mayoccur in the nominative forms such as ["tsùk@R-tS@´k-∅] is problematic, and itis not clear if there could be a higher ranked constraint that would protectit. Everything indicates that *@˘@˘ is not a *Lapse constraint after all.A different constraint is needed. In the following few paragraphs we willexplain that the ungrammaticality of sequences of two unstressed schwasactually stems from a more rudimentary restriction on the metrical orga-nization of phonological segments. To do so, we must first introduce someassumptions which are necessary to arrive at that conclusion. A crucialquestion about metrical constituency in NM Slovenian is whether the lan-guage parses segments into trochaic or iambic feet. Based on the data thatwe have observed so far, stress is not predictable from any phonological fac-tors, and it may only be manipulated by specific morphological constructions(especially in derived environments). This means that stress must be theproperty of lexical and morphological specification in NM Slovenian. Themost typical and, therefore, default type of stress in non-derived environ-ments would be word-initial stress, when the word begins with the exponentof the root, of course (prefixes are generally unstressable). This is also verytypically reflected in the system of nouns and adjectives (for some typicalparadigms see section 3.4). This points to the conclusion that NM Slove-nian parses feet into trochees, precisely like Standard Ljubljana Slovenian(Jurgec 2007a). This seems to be a particularly well founded assumptionfor verbal stems, as cases of [σ"σ]-type stress are only induced by specificmorphological constructions there, as already mentioned, or when the wordhas a prefix.32 Generally speaking, what is claimed for Standard LjubljanaSlovenian in Jurgec (2007a) in terms of stress patterning, holds true for NMSlovenian. It, therefore, seems appropriate to interpret NM Slovenian as alanguage that parses metric structure into trochaic feet.We will also assume that foot construction can be maximally binary,along the lines of Prince (1985), Hyde (2001) and Hyde (2002). In particular,Hyde claims that the grammar can only generate maximally binary feet, by32Note that some bisyllabic roots do show [σ"σ]-stress in non-derived environments, butthis again correlates with no phonological factors.933.2. On the deletion of the high vowel [i]specifying this as a condition on the component GEN , and this is often atacit assumption of much work on foot structure, as in Elenbaas & Kager(1999); see Rice (2007) for discussion. This implies that our grammar willonly produce feet built on up to two constituents; for cases such as ["σσσ], wewill assume that the final syllable is left unfooted, as in [("σσ)σ]. This seemsappropriate given the fact that such syllables seem to receive no secondarystress in NM Slovenian, at least not in verbs and participles, but our analysisbelow is still compatible with other approaches to such ‘redundant’ syllables.In terms of what constituents actually build feet in NM Slovenian, thereseems to be no independent evidence to say whether NM Slovenian buildsfeet on moras or just simply syllables: since NM Slovenian seems to show nocontrastive length, such evidence is not all that easy to come by. Other vari-eties of Slovenian are construed to only count syllables, and not moras, whenconstructing feet (Jurgec 2007a), and we may adopt this same assumptionfor NM Slovenian as well for the time being.Above, we only considered the example ["pow@nli], along with the prob-lematic ungrammatical candidate *["pow@n@l]. However, an explanatorilyadequate analysis must explain several pieces of data at the same time,which is why we will consider a broader range of examples:input to 2nd cycle"m@R-l-{∅, -i}/ ‘die’ u"m@Ru u"m@R@l/"pow@n-l-{∅, -i}/ ‘fill’ "pownu "pow@nli/"xRan-l-{∅, -i}/ ‘feed’ "xRanu "xRan@lTable 3.42: Range of data to considerThese data show two forms from the participial paradigms for the threegiven roots, viz. the masculine singular form, where the participial suffix/-l/ is realized as [u], and the masculine plural form, where the /-i/ under-goes deletion (or fails to do so). Why the /-l/ is realized as [u] is a tangentialissue (see the end of this section, and also section 3.3 on this), but the wayschwa behaves in these forms is curious: we learned in the previous sec-tion on schwa that schwa specified in the first cycle of phonology cannotbe deleted in the second since it is protected by Max-Base – however, theschwa specified in the previous cycle does delete in the masculine singularforms, but just not with roots like √mR-. These data offer three points thatneed to be explained: (i) why the /-i/ surfaces with /√pown-/, and/CVCr/ forms in general, but not the other forms, (ii) why the schwa maydelete in the, even though it cannot delete in any of the other forms,and (iii) why schwa does not delete in the forms with roots like √mR-.943.2. On the deletion of the high vowel [i]Observe the masculine plural forms first. The three forms given in Table(3.42) could be explained by a constraint that does not allow the final con-sonant of the ‘base’ (i.e. the output of the previous cycle) to be re-syllabifiedoutside the foot: with [("po.w@n)li] the base is [("po.w@n)], and in the prob-lematic candidates such as *[("po.w@).n@l], the final segment of the base isre-syllabified outside the foot. This could be captured by an Alignmentconstraint (McCarthy & Prince 1993) of the following type:(30) Align(Base, R; Ft, R)∀x∃y such that x is a Base and y is a Foot, where the rightmost edgeof x coincides with y.This constraint needs to be ranked immediately above Max-Base to achievethe desired effect. Consider the second cycle of [("po.w@n)li]:/("po.w@n)-l-i/ *Nuc/CAl-RMwd,PwdAl-RBase,FtMax-Base*˘i]σ] PwdMax(i)Dep *@a. ☞ ("po.w@n).li * *b. ("po.w@).nl"*! * * *c. ("po.w@).n@l *! * * **d. ("pow.n@l) *! * * * *e. ("po.w@n).l@ *! * * **Table 3.43: Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí]The new Alignment constraint successfully rules out the problematic candi-date (i.e. candidate d.).33 Let us check whether this constraint predicts therest of the data in Table (3.42). Consider the second cycle of [u.("m@.R@l)]:33Notice that candidates such as *[("po.w@n).@l] are not included in the tableau above.Such a candidate would win under just the constraints given above. However, we needto assume that NM Slovenian always syllabifies coda consonants as onsets if a vowelfollows. The standard treatment for this is to assume that Onset is high-ranked. How-ever, since NM Slovenian does contain onsetless syllables in the initial position of thePWd, we need to assume that something along the lines of an Anchor-Left constraint(McCarthy & Prince 1995, 1999), or Edge-Integrity (Kang 2004), dominates Onset.This would demand correspondence between the segments of the left edge in the PWdand rule out any unwanted vowel deletion, protecting just the initial onsetless syllables.953.2. On the deletion of the high vowel [i]/u.("m@R)-l-i/ *Nuc/CAl-RMwd,PwdAl-RBase,FtMax-Base*˘i]σ] PwdMax(i)Dep *@a. u.(" * *! *b. ☞ u.("m@.R@l) * * * **c. u.("mR@l) * *! * * *d. u.("m@R.l@) *! * * * **Table 3.44: Participial cycle: /u.("m@R)-l-i/ → [u.("m@.R@l)]The new Alignment constraint is always violated because the system hasno choice but to parse two syllables into a foot where the input only con-tains a monosyllabic foot (which must be driven by higher-ranked constraintwhich we do not discuss here). Within this subset of candidates, candidateb. is correctly predicted to emerge as the winner. Let us now move on todiscussing the second cycle of [("xRa.n@l)]:/("xRan)-l-i/ *Nuc/CAl-RMwd,PwdAl-RBase,FtMax-Base*˘i]σ] PwdMax(i)Dep *@a. (" * *!b. ☞ ("xRa.n@l) * * * *c. ("xRan.l@) *! * * * *Table 3.45: Participial cycle: /("xRan)-l-i/ → [("xRa.n@l)]For the same reasons as with [u.("m@.R@l)], [("xRa.n@l)] must surface as thewinning candidate. It seems that the new Alignment constraint (Al-RBase,F t)that we have added to our constraint hierarchy might be the appropriate wayto proceed. However, recall that we should attempt to construct an analysiswhere we also explain the curious distribution of schwa in the masculinesingular forms in Table (3.42). With these forms, a critical complicationarises. Consider the second cycle of [("]:963.2. On the deletion of the high vowel [i]/("po.w@n)-l-∅/ *Nuc/CAl-RMwd,PwdAl-RBase,FtMax-Base*˘i]σ] PwdMax(i)Dep *@a. ☞ ("po.w@).nu * *b. / (" * *!Table 3.46: Participial cycle: /("po.w@n)-l-∅/ → [("]With the masculine singular forms, the schwa must delete, but the Align-ment constraint that we have proposed gives wrong predictions here: sincethe alignment constraint is violated in the masculine singular form anyway,because of the inclusion of the suffixal [u] in the foot, the candidate thatdoes not violate Max-Base is selected, which in this case is candidate a. –an ungrammatical candidate. We could, in principle, attempt to argue thatthe schwa deletes because this is demanded by an even higher-ranked con-straint, but it is not certain what that constraint could be. The schwas thatare remnants from the previous cycle must always be preserved in the formswith overt vowel inflections, and this seems to be interacting with the con-straint that blocks the deletion of the /-i/. Because the same schwadeletes with zero inflections ( forms), this looks very suspiciously likethe same interaction effect. One could, in principle, argue that the schwadeletes because the forms trigger a separate ‘cophonology’ which li-censes schwa deletion of this kind, but no independent evidence that wouldsupport this can be mustered.In light of this, I would like to propose that the pattern witnessed in Table(3.42) is not due to an Alignment constraint, but is in fact due to a morefundamental faithfulness effect on the metrical level of representation; it willbe shown that we seem to be dealing with a type of mora preservation fromthe previous cycle. To pursue this analysis, let us assume that NM Sloveniancontains moraic constituents within syllables and also that it assigns morasto coda consonants – i.e. that it is a ‘Weight-by-Position’ language (Hayes1989, 1995).34 These assumptions will allow us to tease apart the range ofdata presented in Table (3.42), and pinpoint the generalization that connects34Note that no aspect of Moraic Theory (Hayes 1995) precludes the possibility ofsyllable-counting languages also specifying moras as part of their metrical organization,even if foot-construction never ‘looks below’ the syllable level; in fact Kager (1995) ex-plicitly states that moras are still present in syllable-counting languages.973.2. On the deletion of the high vowel [i]them. The basic metrical representations are laid out below.These assumptions on foot structure in NM Slovenian predict that theparticipial form ["xRàn-l-á] ‘having fed (’ has the following structure:PWdFσµa n l aµσµx RFigure 3.1: Structure of ["xRànlá]The two syllables form a trochaic foot, which means that they are bothdominated by the foot node F . The first syllable contains two moras, andthe second one. With the masculine plural form of this participle we predictmuch the same. Consider ["xRan@l] (σµa nµ@ lµσx RFigure 3.2: Structure of ["xRàn@´l]The first syllable dominates one mora and the second dominates two. Wealso predict the following structure participial forms such as ["pòw@nlí]‘having filled (’:983.2. On the deletion of the high vowel [i]PWdFσµop wµ@σµnσl iµFigure 3.3: Structure of ["pòw@nlí]As before, the foot encompasses two syllables, where the first syllable ismonomoraic and the second syllable bimoraic. Now recall that participlesare constructed in two cycles of phonological application. In the first cy-cle, the output is the verbal stem ["pòw@´n] and the output of the second(participial) cycle is ["pòw@nlí]. This is shown below.PWdFσµop wµ@σµnPWdFσµop wµ@σµnσl iµFigure 3.4: Cyclic derivation of ["pòw@nlí]The crucial question of this section of course is, why candidates such as*["pòw@n@´l] must be rendered ungrammatical. The foot structure of such afailed derivational outcome is given below.993.2. On the deletion of the high vowel [i]PWdFσµop wµ@σµnPWdFσµo*p wµ@σnµ@ lσµFigure 3.5: Cyclic derivation of *["pòw@n@´l]Notice the apparent difference between [("pò.w@n).lí] and *[("pò.w@).n@´l]:while the former completely preserves the foot-internal mora count fromthe verbal cycle, the latter must inevitably lose a mora by re-syllabifyingthe coda consonant [n]. It therefore seems that the constraint that we re-quire is actually a faithfulness constraint, crucially one that requires thenumber of moras specified in the foot in the input to be maintained on thesurface. Such a constraint could, of course, be satisfied by producing a can-didate such as *[("pòw.n@´l)], which would have four moras in the foot, but*[("pòw.n@´l)] is impossible for independent reasons (viz. because it incurs aviolation of Max-Base by deleting the schwa in the base).Such a constraint would make no unexpected predictions for the rest ofthe cyclic /i/-deletion and schwa epenthesis analysis that we have developedso far. Consider the derivation of [u"m@´R@`l] ‘having died (’:PWdFσµµ@m RPWdFσ@µm Rµ@ lσµFigure 3.6: Cyclic derivation of [u-"m@´R@`l]Note that the prefix [u-] is excluded from the representation above. Theoutput of the verbal cycle is [u.("m@´R)], which yields exactly two moras in1003.2. On the deletion of the high vowel [i]the foot. The output of the participial cycle, on the other hand, containsthree moras in the foot because the participial /-l/ is parsed within the footas the final coda consonant. It seems that the constraint that we requireis satisfied by there being the same (or greater) amount of moras in theinput and output foot, which means that the mora that supports the footcan come from different segments – in this case, these are the second schwaand participial [-l] suffix. Notice that *[("pò.w@).n@´l] in Table (3.5) cannotescape the violation of such a metrical constraint, as the final coda [l] occursoutside the foot.The generalization that we are dealing with is thus quite fundamental:the phonology of NM Slovenian requires the preservation of the amount ofmoras specified in the input foot. More specifically, it requires that there bea mora specified in the output foot for every mora that occurs in the foot inthe input. But crucially, it seems to demand no input-output correspondencebetween the moras themselves. The task of ‘maintaining’ mora count canbe performed by other moras. Constraints that achieve faithfulness effectswithin feet have been proposed in the literature: based on the work byBeckman (1998), a Positional Faithfulness constraint that refers to feet, suchas Max-Ft, is often used (Itô et al. 1996; Hall 2000; van Oostendorp 2004).However, Positional Faithfulness only refers to positions in the output, butwe require a constraint that refers to the foot in the input and output. Wecan capture the generalization discussed above in a Max-type constraint,but with the following definition:1013.2. On the deletion of the high vowel [i](31) Max-F(µ)For every input mora µI that is associated with a foot FI , there hasto exist an output mora µO which is associated with FO, where FIR FO.As indicated above, this constraint is not a fully typical Max-constraint: itneeds to refer to the foot-internal position of a mora in the input and out-put, which means that it is not a typical Positional Faithfulness constraint(Beckman 1998), as Positional Faithfulness only refers to output positions.Also, Max-type constraints usually force an input-output correspondencerelation between the input and output units that they are evaluating. Thisdoes apply to the foot here (hence, the dictum FI R FO), but not to themoras. In other words, this constraint does not require that the same in-put mora match the output mora in the foot, but merely that some moracorresponds to it in the output foot; in that sense, this constraint is moregeneral than a typical Max-constraint. The side effect of the constraintMax-F(µ) will therefore be the preservation of ‘mora count’ within the foot(where the mora count may be increased), which is not necessarily achievedby preserving specific input moras. What the best way of formalizing therequired correspondence relation here is I leave for a future study, and fornow, resort to using Max-F(µ), as it will be sufficient for the purposes ofthis analysis.At this point, we may finally return to the computation of the second(participial) cycle in (3.41), where the constraints that we used produced theungrammatical output *["pow@n@l]. We can now use the constraint Max-F(µ) to derive the correct surface form:/("poµ.w@µnµ)-l-i/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ☞ ("poµ.w@µnµ).li *b. ("poµ.w@µ).nl"*! * *c. ("poµ.w@µ).n@l *! * **d. ("poµwµ.n@µlµ) *! * *e. ("poµ.w@µnµ).l@ *! * **Table 3.47: Participial cycle: /"pòw@´n-l-i/ → ["pòw@nlí]1023.2. On the deletion of the high vowel [i]The tableau above illustrates that the phonological grammar is forced toprefer the surface realization of the /-i/, even if this violates *˘i]σ ]Pwd,because deleting a foot-internal mora that corresponds to some mora the in-put is even less preferred. The /-i/ could still undergo deletion, but then theschwa from the previous cycle would have to be deleted to keep at least threemoras in the foot (cf. candidate d.), which is, however, not an option, forwhich we have independent evidence from the distribution of schwa: schwaspecified in the previous cycle must be preserved on the surface. We haveonly discussed the example with √pown-, but the phonological computationin the two cycles should be precisely the same for the semelfactives, as well.The example that we discussed above was a case of root stress. Letus also briefly present the computation of the two cycles with stress oc-curring on the theme vowel. In such cases, the computation is much morestraightfoward. Recall that stress is a matter of lexical specification, whichis why it is reflected in the UR in the tableaux below (we will discuss stressassignment in greater detail in section 4)./pown-"i/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ☞ pow("niµ)b. po("w@µnµ) *! *Table 3.48: 1st cycle: /pown-"i/ → [pow"ni]The stress is realized on the theme vowel and the structure that emergesviolates none of the constraints that we have specified in the tableau. Moreconcretely, though the theme [-i] is word-final, it is stressed and thereforecannot undergo the deletion otherwise demanded by *˘i]σ ]Pwd.1033.2. On the deletion of the high vowel [i]/pow("niµ)-l-i/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. pow("niµ.liµ) *!b. ☞ pow("niµlµ) *Table 3.49: 2nd cycle: /pow("niµ)-l-i/ → [pow("niµlµ)]In the participial cycle, the occurs in the word-final position and itis unstressed, which is why it must undergo deletion. This shows that theconstraint ranking we have provided is able to generate the forms with bothstress patterns in the two cycles of phonology.Let us now discuss the rest of the data presented in Table (3.42) todemonstrate the predictions of Max-F(µ). We start by giving the tableauxfor the second cycles of [("xRa.n@l)] ‘having fed (’ and [u.("m@.R@l)] ‘hav-ing died ("xRaµnµ)-l-i/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ("xRaµnµ.liµ) *!b. ☞ ("xRaµ.n@µlµ) * *c. ("xRaµnµ.l@µ) *! * *Table 3.50: 2nd cycle: /("xRan)-l-i/ → [("xRa.n@l)]In the case with [("xRa.n@l)], the final /-i/ may delete and schwa may epenthe-size, as is reflected in candidate b. This is possible because these modifica-tions do not cause the foot in candidate b. to contain fewer moras than theinput foot. A similar situation can be observed with [u.("m@.R@l)]:1043.2. On the deletion of the high vowel [i]/u.("m@µRµ)-l-i/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. u.("m@µRµ.liµ) *!b. ☞ u.("m@µ.R@µlµ) * *c. u.("mR@µlµ) *! * *d. u.("m@µRµ.l@µ) *! * *Table 3.51: 2nd cycle: /u.("m@R)-l-i/ → [u.("m@.R@l)]With [u.("m@.R@l)], preserving two moras in the output foot is enough becausethe input foot only contains two moras, as well. This allows the lower-rankedconstraints to decide the winner of the derivation: the winner is candidateb., which does not violate *˘i]σ ]Pwd, Max-Base, nor Al-RMwd,Pwd.All the cases that we have presented up to now can also be analyzed byassuming a high-ranked Alignment constraint that requires the right edgeof the base and the right edge of the foot to coincide. The crucial pieceof data comes from the masculine singular participle form, viz. [("]‘having filled’, which could not be successfully analyzed with the previously-mentioned Alignment constraint:/("poµ.w@µnµ)-l-∅/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ("poµ.w@µ).nu *!b. ☞ ("poµwµ.nuµ) *Table 3.52: 2nd cycle: /("po.w@n)-l-∅/ → [("]The constraint Max-F(µ) is successful in predicting candidate b. as themost harmonic option, which is precisely where the Alignment constraint(discussed above) failed. Max-F(µ) unifies the entire range of data specifiedin Table (3.42). In particular, it predicts the following: when the systemencounters a zero exponent of inflection in the participial paradigm, the1053.2. On the deletion of the high vowel [i]schwa will be forced to delete to preserve at least three moras specified inthe foot because the input foot also contains three moras. To repeat whatwas already stated before in this section, the alternation that forces theparticipial /-l/ to be realized as [-u] in the is a completely tangentialissue. If the system did not contain this alternation, our analysis would stillpredict that the schwa from the previous cycle would undergo deletion:/("poµ.w@µnµ)-l-∅/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ("poµ.w@µ).nl"*! *b. ("poµ.w@µ).n@l *! *c. ("poµ.w@µnµ).l@ *! *d. ☞ ("poµwµ.n@µlµ) * *Table 3.53: 2nd cycle: /("po.w@n)-l-∅/ → [("pow.n@l)] (no l ∼ u)This tableau illustrates that whether the schwa specified in the first cycledeletes or not is crucially tied to whether the infection has an overt or azero exponent: with zero exponents, as in the masculine singular forms, theschwa deletes so that a sufficient amount of moras remain specified in theoutput foot. A word-final epenthetic schwa, as in candidate c., does retainthe same amount of moras in the foot, but it incurs a crucial violation ofAl-RMwd,Pwd, because this epenthesis causes a misalignment between theProsodic Word and the Morphological Word. The system has no choicebut to select candidate d. as the winner. This piece of evidence may beunified with the rest of the data under an analysis that evokes a form ofmoraic faithfulness, but it cannot under any of the other approaches thatwere discussed in this section. For this reason, we shall accept this latestanalysis, which claims that for any input foot-internal mora, there must bean output foot-internal mora.With the analysis in this section, we have now explained why the m.plsurfaces with ‘√CVCr’ and ‘√CVC-r’ structures: it surfaces to avoid creat-ing feet that have fewer moras than the foot produced by the previous cycleof phonology, and also to avoid deleting the schwa which was also epenthe-sized in that previous cycle. With simple ‘√CVC’ roots, on the other hand,there is no danger of failing to maintain the needed mora count: consider1063.2. On the deletion of the high vowel [i][("xRàµnµ)] the output of the verbal cycle of /"xRàn-i/ ‘feed’, which for theparticipial input /("xRàµnµ)-l-i/ ‘having fed (’ outputs [("xRàµ.n@´µlµ)],maintaining minimally the amount of moras specified in the input foot. Insuch contexts, the /-i/ may delete. This is a plausible account ofthe deletion and surface retention of the /-i/. This confirms that thephonological component can indeed easily predict the distribution of /-i/ if it is universally specified as the underlying exponent of mascu-line plural in the participles of NM Slovenian.3.2.3 Single consonant rootsFinally, let us turn out attention to the single-consonant roots that are usedto form participles. In NM Slovenian, only two such roots exist, viz. √S- ‘togo’ and √b- ‘to be’, whose paradigms are given below:sg du plmasc "S-∅-ù-∅ "S-∅-l-à "S-∅-l-ìfem "S-∅-l-à "S-∅-l-E` "S-∅-l-E`neut "S-∅-l-ù "S-∅-l-à "S-∅-l-èTable 3.54: Single-consonant roots: √S- ‘to go’sg du plmasc "b-ì-w-∅ "b-∅-l-à "b-∅-l-ìfem "b-∅-l-à "b-∅-l-E` "b-∅-l-E`neut "b-∅-l-ù "b-∅-l-à "b-∅-l-èTable 3.55: Single-consonant roots: √b- ‘to be’As was already noted before, these two participles are important becausethe /-i/ is allowed to surface, but crucially for different reasons thanthe ‘√CVCr/√CVC-r’ cases, which we discussed above. The /-i/ seemsto be able to surface simply because it bears stress in these two participles:because it is stressed, it is not subject to deletion even though it occurs inthe word-final position of the prosodic word.However, in every other respect, these two participial formations seemto be very unusual when compared to all other participles in NM Slovenian.Firstly, it is curious that stress lands on the inflections in all (but the Stress is never allowed to move to the inflection in any of the otherparticipial classes; however, this could probably be connected to the factthat these two participles contain roots which consist of a single consonant.1073.2. On the deletion of the high vowel [i]Secondly, the √b- ‘to be’ formation seems to show an [i] vowel following theroot in the masculine singular form. This is notated as a theme vowel in thetable above, viz. ["b-ì-w-∅]. Since this root is not used in the correspondingverbal paradigm of ‘to be’ (√b- is a suppletive root), it is difficult to assessif the [i] vowel in the masculine singular is truly a theme vowel or perhapsjust a part of the suppletive root. Thirdly, the occurrence of the [i] in themasculine singular could be taken as evidence that this [i] is also presentunderlyingly in the remaining forms of the paradigm. However, this wouldimply a deletion process which could only be motivated by assuming thatstress first occurs on the [i], but then shifts away to the inflections, renderingthe [i] unstressed and subject to deletion. But recall that no such processof deletion is to be recovered in any other data in NM Slovenian; in fact,unstressed [i] vowels (in pre-stress position) contained in roots and prefixescan never undergo deletion, as shown in section 2.2.5. Positing /i/ uniformlyin the paradigm is, therefore, a very stipulative solution.Fourthly, since positing /i/ in this participial class and assuming its dele-tion is not a good solution, we would expect schwa epenthesis to occur in thefirst (verbal) cycle of phonology to render the root phonotactically licit, butno schwa occurs anywhere on the surface. In other words, if the suppletiveroot here is really √b-, then we expect the following cyclic analysis: /b-∅/is the verbal stem, constructed with a zero theme suffix, and the first cycleis run on this stem. The output of such a cycle would have to be *["b@], orperhaps *["@b], to render it phonotactically licit. The input to the secondcycle would then be /("b@)-l-i/, and our current system of constraints pre-dicts the output for such an input to be *[("b@l)], or *[b@.("li)] if we assumethat the stress shifts somehow. The schwa has to remain in the output,unless there is some higher-ranked constraint that demands its deletion, butno such constraint seems to offer itself. This problem could be solved if wecould assume that these two roots are not computed in two cycles for somereason.The two roots discussed here seem very different with respect to all otherparticipial formations in NM Slovenian. However, this is perhaps not tooodd given the cross-linguistic exceptionality of verbs/participles such as ‘togo’ and ‘to be’. In section 4.3, where we discuss the morphology-phonologyinterface, we will illustrate how these two verbs can avoid phonological pro-cessing in two cycles.This section concludes the discussion of the purely phonological aspectsof /i/-deletion. We have demonstrated that unstressed /i/ must delete inabsolute word-final position, unless this would yield a sequence of two un-stressed syllables that both contain schwa, in which case /i/ must surface1083.3. Masculine singularunstressed. The surface distribution of the theme vowel /-i/ and the m.plinflection /-i/ is thus fully predictable by the phonology, on the assumptionthat verbs and participles are processed in two phonological cycles, for whichwe saw independent motivation in the previous section on schwa. This con-firms the initial hypothesis, which stated that any alternations between [i],zero and schwa are due to the process of /i/-deletion and its interaction withthe general phonotactic requirements of NM Slovenian.3.3 Masculine singularThroughout the description of the participial system which we gave in sec-tion 2.2.3, we constantly had to omit any discussion of the masculine singularform. This section is dedicated to this form of the participial paradigm inNM Slovenian, most notably to the following alternation it exhibits: theparticipial suffix /-l/ needs to be realized as either [w] or [u] in the this form of the participial paradigm offers another observation: often,stress does not appear on the theme in the, while it does in the restof the paradigm. In short, the form realizes the participial /-l/ in aspecific way, and it often disallows theme stress. We start by discussingthe latter, which is followed by a discussion of the former. The goal of thissection is to determine whether the curious behaviour of the form hasany implications for the analysis of /i/-deletion that we have proposed.3.3.1 Special stress patternThe masculine singular form of the participial paradigm may often only dis-play root stress: while stress will occur on the theme vowel in the remainingforms of the paradigm, this will not happen in the form. Consider theparticiples formed with the theme /-a/ (only masculine forms are given):sg du plmasc "dìx-ó-w-∅ "dìx-a-l-á "dìx-á-l-∅sg du plmasc "pE`l-ó-w-∅ pe"l-à-l-á pe"l-à-l-∅Table 3.56: a-class root and theme stress: √dìx- ‘breathe’ and √pèl- ‘drive’1093.3. Masculine singularsg du plmasc "jòk-ó-w-∅ "jòk-a-l-á "jòk-á-l-∅masc "jòk-ó-w-∅ jo"k-à-l-á jo"k-à-l-∅Table 3.57: a-class variable stress: √jòk- ‘cry’The a-class participles have a three-way split with regard to stress: withroots, such as √dìx-, stress must occurs on the root, while with others, suchas √pèl-, the stress must occur on the theme vowel. With yet others, suchas √jòk-, the stress realization is variable, as discussed in 2.2.6. However,notice that in none of these paradigms is the form affected by this.Even with √pèl-, where the stress must obligatorily occur on the themevowel in the paradigm, the has stress on the root. These stress ‘shifts’seem to be controlled by morphology, as there is no phonological factor thatthey could be attributed to; it, therefore, seems that root stress is simplyan inherent property of the category of participles in NM Slovenian.All the participles that have the theme vowel /-i/ specified underlyinglyand undergo /i/-deletion with root stress, but not with theme stress, sharethis property with the a-class: the can never show theme stress (theseare i ∼ ∅-class and √CVCr-class participles; see section 2.2.3 for a layoutof all the paradigms). However, some roots that also form participles withthe theme vowel /-i/ and have the theme vowel stressed obligatorily do infact allow the theme to be stressed in the form. But this only appliesto some roots:sg du plmasc ka"d-ì-w-∅ ka"d-ì-l-á ka"d-ì-l-∅sg du plmasc "dùb-∅-ú-∅ du"b-ì-l-á du"b-ì-l-∅Table 3.58: i-class participles: √kàd- ‘smoke’ and √dùb- ‘get’Both the roots above form i-class participles and realizing the stress on thetheme is obligatory. However, while √kàd- permits the stress to be realizedon the theme vowel in the as well, √dùb- does not permit this as thestress must occur on the root in the form. In our description of NMSlovenian participles in section 2.2.3, we called participles, like those formedwith √kàd-, i-classI participles, and those that are built on roots like√dùb- we called i-classII participles. These examples show that whether1103.3. Masculine singularthe form of the participle will allow the stress to occur on the themesimply depends on what root the participle is built on. This then impliesthat NM Slovenian must possess a special rule of morphology that controlssuch stress assignments, one specific to the form with some roots.Roots that do allow stress modification to the form can be construedas the result of the ‘elsewhere’ effect, i.e. they are subject to the stress ruleslike any other form in the paradigm. While this morphological rule needsto be assumed to derive the paradigms correctly, it has no negative impacton our analysis of /i/-deletion. In fact, some notion of /i/-deletion in theverbal stem cycle is necessary to be able to derive the paradigms above. Forinstance, take the ["dùbú] in Table (3.58) above: the input for its verbalcycle will be /"dùb-i/ with an unstressed [i] in final position, and our accountof /i/-deletion precisely predicts that it will output ["dùb], as is necessaryto construct the participial form ["dùbú]. This provides additional evidencefor our analysis of /i/-deletion and so further strengthens our claim.The [l] ∼ [w] ∼ [u] alternationAs was pointed out previously, the forms of participles reveal a curiousalternation: the participial /-l/ is realized as either [w] or [u]. Looking backat the a-class participles, last given in Tables (3.56) and (3.57), we canquickly notice this alternation at work: cf. ["jòk-ó-w] ( ∼ ["jòk-a-l-á](m.du) ∼ ["jòk-á-l] ( However, [l] does not alternate with [w] only inthe system of participles, but also in the system of nouns and adjectives:√stòl- ‘chair’ √vòl- ‘ox’ "stòw-∅ "vòw-∅ "stòl-á "vòl-á "stòl-ú "vòl-úTable 3.59: l ∼ w in nouns: presence of alternation√pre"dàl ‘drawer’ √Sàl- ‘scarf’ pre"dàl-∅ "Sàl-∅ pre"dàl-á "Sàl-á pre"dàl-ú "Sàl-úTable 3.60: l ∼ w in nouns: absence of alternationThe alternation can be found only in the nominative singular in masculinenouns. Furthermore, only some roots undergo this alternation, while somedo not, as illustrated by the distinction between √stòl- and √Sàl-.1113.3. Masculine singular√bèl- ‘white’ √tsèl- ‘whole’ "bèw-∅ "tsèw-∅ "bèw-gá "tsèw-gá "bèw-mú "tsèw-mú "bèl-ím "tsèl-ímTable 3.61: l ∼ w in adjectivesIn the system of adjectives, on the other hand, this alternation is much moregeneral; there do not seem to be any systematic exceptions to it. Notice thatthe alternation in the adjectives is not limited to any particular grammaticalcategory, but crucially has a phonological correlate: /l/ seems to be realizedas [w] if it occurs in coda position, and this is generally how this rule isconstrued for Slovenian (p.c. Peter Jurgec). We could test the generality ofthis rule in the system of nouns above if any of the cases (other than thenominative) have a zero exponent, or an exponent that contains no vowel,and so renders /l/ in coda position. But this is unfortunately not the case– all the non-nominative cases of masculine nouns have overt exponents,beginning in vowels. However, feminine and neuter nouns do realize thegenitive plural with a zero exponent, which means we could use feminineand neuter roots ending in /l/ to test the phonological status of the l ∼ walternation. Consider √o"bàl- ‘beach’ and √vE`sl- ‘oar’:√o"bàl- ‘beach’ √vE`sl- ‘oar’ o"bàl-á "vE`sl-ú o"bàl-é "vE`sl-á o"bàl-í "vE`sl-ú o"bàl-∅ "vE`s@´l-∅Table 3.62: l ∼ w alternation in feminine and neuter nounsNotice that the final /l/ is not realized as [w] in the genitive plural of feminineand neuter nouns, which means that the l ∼ w alternations is correlated withmore than just phonological factors. It would seem that it is necessarilycorrelated with morphological factors, as well. This could be dealt within a number of ways: perhaps, in the noun system, the masculine singularexponent is not zero after all, but is prespecified in some way that triggersthe l ∼ w alternation, for instance through a system of floating features(Wolf 2005, 2007). And perhaps this is true of the adjectival system aswell, in that the adjectivizing (derivational) morpheme35 is not zero, but is35The derivational morphemes are not indicated in the tables above, but nouns andadjectives are assumed to contain a zero derivational morpheme between the root and the1123.3. Masculine singularprespecified in this way, which would explain why the l ∼ w alternation canbe manifested throughout the entire adjectival paradigm. Another way todeal with this would be to assume a system of cophonologies (Inkelas & Zoll2007; Inkelas 2008, 2011), where only the in the nouns would triggera special cophonology that endorses the l ∼ w alternation, and the entireadjectival paradigm would do the same. How precisely this is dealt with isnot too relevant for the present thesis; we leave the exact details of such anaccount to future studies.How this could connect to the participial system is quite obvious. Inthe participial paradigm, we have a problem similar to that found in nouns:while the form does realize the /-l/ as [w] in coda position, the m.plalso often realizes the participial /-l/ in coda position, but there it mustalways surface as [l] and never as [w], which is much like comparing with above. If it were possible to analyze this al-ternation in the participial form as the consequence of the same m.sginflection as that found in nouns, this would be a very welcome result (infact, in section 3.4 we will show that the participial exponents of inflectionare shared by noun forms in the nominative case). In that way, the mappingof /-l/ to [w] in the form of participles could also be either due to somesort of prespecification of the inflection, or due to a pass through thecophonology that would trigger the alternation in question.The segment /l/, however, does not only alternate with [w], but also with[u]. It seems that a coda /l/ (in the right morphological contexts) is realizedas [w] if it is immediately preceded by a vowel on the surface, but as [u] if it isimmediately preceded by a consonant on the surface. For instance, comparethe ["jòkow] ‘having cried’ from (3.57) with ["dùbú] ‘having got’from Table (3.58). Independent evidence is also available from nouns, whichshows that this observation holds true. Consider the nominal paradigms ofconstructed on the roots √kO`tl- ‘vat’ and √O`sl- ‘donkey’:√kO`tl- ‘vat’ √O`sl- ‘donkey’ "kO`tú-∅ "`Osú-∅ "kO`tl-á "`Osl-á "kO`tl-ú "`Osl-úTable 3.63: l ∼ u alternation in nounsinflection. Most versions of Distributed Morphology (Halle & Marantz 1993) would makethis assumption, since roots are treated as ‘acategorial’ morphemes that always need tobe categorized by a categorizing morpheme – see section 1.1.1133.3. Masculine singularThese data from the system of nouns confirm that /l/ which would occurin a coda position after a consonant is realized as [u]. If we incorporatethis observation into our analysis of cyclic /i/-deletion, the state of affairsis directly predicted. As mentioned before, the realization of the participial/-l/ as [u] in cases like ["dùbú] ‘having got’ follows naturally from ourprevious analysis because the verbal stem /dùb-i/ must first pass throughthe verbal cycle of phonology, where the theme vowel /-i/ is deleted, whichimplies that the input to the participial cycle for the form will be/"dùb-l-∅/; this way, the participial /-l/ will occur directly after a consonant.This unifies the distribution of [w] and [u], which are derived from /l/, inthe system of nouns and participles.What remains to be discussed is the compatibility of the alternation inquestion with the cyclic analysis of schwa-epenthesis and /i/-deletion thatwe advanced in the previous section. In connection with that analysis, wewill indicate a possible way of treating the alternation in question, but nodetailed analysis will be given, as that would take us too far outside the scopeof the present thesis. Recall examples such as ["pòw@n-l-í] ‘havingfilled’ with its corresponding UR /"pòwn-i-l-i/, which is computed in twocycles: the verbal cycle inputs /"pòwn-i/ and outputs [("pò.w@´n)], while theinput to the participial cycle is /("pò.w@´n)-l-i/, with the output [("pò.w@n).lí](see sections 3.1 and 3.2 for details). The form of this paradigm is[("pò] with the UR /"pòwn-i-l-∅/. According to our analysis of cyclic/i/-deletion and epenthesis, this form should be computed in two steps, aswell: the verbal cycle would receive /"pòwn-i/ as input and would output[("pò.w@´n)]. The input for the participial cycle then has to be /("pò.w@´n)-l-∅/. Since the final surface form is [("pòw.nú)], the schwa inserted in thefirst cycle needs to delete. We explained the deletion of this schwa as theexpected consequence of our analysis of mora preservation within the foot.We also indicate that the reason why this schwa needs to delete is completelydivorced from the l−w−u-alternation. The two tableaux that illustrate thisare repeated here:1143.3. Masculine singular/("poµ.w@µnµ)-l-∅/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ("poµ.w@µ).nu *!b. ☞ ("poµwµ.nuµ) *Table 3.64: 2nd cycle: /("po.w@n)-l-∅/ → [("]/("poµ.w@µnµ)-l-∅/ *Nuc/CAl-RMwd,PwdMax-F(µ)Max-Base*˘i]σ] PwdMax(i)Depa. ("poµ.w@µ).nl"*! *b. ("poµ.w@µ).n@l *! *c. ("poµ.w@µnµ).l@ *! *d. ☞ ("poµwµ.n@µlµ) * *Table 3.65: 2nd cycle: /("po.w@n)-l-∅/ → [("pow.n@l)] (no l ∼ u)The deletion of the schwa is crucially correlated with the presence of the zeroinflection: if the exponent of the inflection is not overt, which is the casewith the, then the grammar has no choice but to select a candidatewith a deleted schwa to satisfy Max-F(µ), so that the output foot containsenough moras to match those in the input foot.Now recall that there are, in principle, two possible ways of dealingwith the l ∼ w ∼ u-alternation, as discussed a few paragraphs up: we couldresort to presepecification of the exponent to induce the alternation(e.g. a floating feature that docks onto [-l]), or we could claim that the m.sgsubscribes to a cophonology that triggers the alternation. Under such acophonology analysis, we could claim that the demands a pass throughthe cophonology that enforces the l ∼ w ∼ u-alternation, but crucially afterthe participial cycle of phonology has applied. The output of the participialcycle would be ["pòwn@´l] as shown in Table (3.65) and the cophonology trig-gered by the category would require the coda /l/ to turn to [w], whichwould yield *["pòwn@´w]. The question that would then become important iswhy this candidate is not the grammatical outcome. The answer very likely1153.4. Nouns and adjectiveslies in the fact that a ‘[@w]’ sequence, as it occurs in *["pòwn@´w], is not a licitconfiguration in NM Slovenian; this is not an unreasonable assumption be-cause no such sequences seem to exist in any surface form in NM Slovenian.If this led to the deletion of the schwa, in order to satisfy the constraint thatprohibits [@w], then [l] would directly follow a consonant in the coda andwould be replaced by [u], as expected.A very similar scenario can be envisioned under a prespecification anal-ysis: the prespecified material (e.g. a floating feature) would need to dockon the participial /-l/. Again, it is conceivable that a candidate such as*["pòwn@´w] would be produced, which means that a ‘[@w]’ sequence wouldagain be created. If ‘[@w]’ sequences are illicit, as suggested above, thenthe schwa would delete and /-l/ would again occur in a position directlyfollowing a consonant. This would mean that it would have to surface as[u], yielding [("pòµwµ.núµ)]. The details of any of the two analysis sketchedout here will need to be investigated in a future study, but we can at leastdemonstrate that they are unrelated to the crucial aspects of our analysisof @-epenthesis and /i/-deletion.Whatever the most appropriate analysis is, we have demonstrated thatour analysis of cyclic /i/-deletion and schwa-epenthesis is compatible withseveral different approaches to realizing /l/ as [w] or [u] under specific mor-phological conditions. Choosing the most appropriate approach is a rela-tively independent topic that should be taken up in future work.3.4 Nouns and adjectivesIn sections 2.2.2, 2.2.3 and 3.2, we established that the phonological gram-mar of NM Slovenian promotes a process of vowel deletion. Specifically,any unstressed high vowel /i/ that occurs in the word-final position of theprosodic word undergoes deletion. There are several alternations that at-test to this process, all discussed in the sections specified above. However,those sections only examine the status of the high vowel /i/ in the verbaland participial paradigms of NM Slovenian. Because of this, the goal of thepresent section is to outline the most prominent paradigms of nouns andadjectives in NM Slovenian, to determine whether /i/ has the same distri-bution as in the paradigms of verbs and participles. A crucial discovery ofthis section will be that the high vowel /i/ is not restricted in the same wayin nouns or adjectives: in other words, we will show that /i/ may alwayssurface unstressed in the word-final position of the prosodic word.1163.4. Nouns and adjectivesLet us begin with nouns. The nominal system of NM Slovenian is verysimilar to that of Standard Slovenian (Toporišič 2000; Herrity 2000): nounscode gender, number and case. The following masculine, feminine and neuterparadigms represent the biggest and most productive nominal paradigms fortheir gender in NM Slovenian:sg du plnom "vlàk-∅ "vlàk-á "vlàk-ígen "vlàk-á "vlàk-ów "vlàk-ówdat "vlàk-ú "vlàk-omá "vlàk-ómacc "vlàk-∅ "vlàk-á "vlàk-éloc "vlàk-ú "vlàk-íx "vlàk-íxinst "vlàk-óm "vlàk-omá "vlàk-íRoots: √stòl- ‘chair’, √kòl- ‘pole’, √kRùx- ‘bread’, etc.Table 3.66: Noun system: masculine nounssg du plnom "mèst-ú "mèst-á "mèst-ágen "mèst-á "mèst-∅ "mèst-∅dat "mèst-ú "mèst-omá "mèst-ómacc "mèst-ú "mèst-á "mèst-áloc "mèst-ú "mèst-íx "mèst-íxinst "mèst-óm "mèst-omá "mèst-íRoots: √màsl- ‘butter’, √vE`sl- ‘oar’, √O`kn- ‘window’, etc.Table 3.67: Noun system: neuter nounssg du plnom "xìS-á "xìS-é "xìS-égen "xìS-é "xìS-∅ "xìS-∅dat "xìS-í "xìS-amá "xìS-amáacc "xìS-ó "xìS-é "xìS-éloc "xìS-í "xìS-áx "xìS-áxinst "xìS-ó "xìS-amá "xìS-amíRoots: √Zènsk- ‘woman’, √tsE`st- ‘road’, √mòk- ‘flour’, etc.Table 3.68: Noun system: feminine nounsLet us first determine the distribution of /i/ in these paradigms.36 In themasculine paradigm, /i/ may surface as a stressed vowel, but it may also36Some cases in the paradigms might require a different tonal pattern than others, but I1173.4. Nouns and adjectivessurface as unstressed, and this may occur in the word-final position of theprosodic word. Consider the masculine plural and the instrumental pluralexponents, both of which are realized as unstressed [i] in such a position.The same distributional properties of /i/ also carry over to the neuter andfeminine paradigms: in the neuter paradigm the instrumental plural is alsorealized as unstressed [i], while in the feminine paradigm this is the realiza-tion of the dative singular suffix. What is curious is that these are positions(unstressed word-final positions) in which any /i/ would delete in verbs andparticiples. The only time the /-i/ fails to delete in word-final un-stressed position in the participles is when this would create phonotacticallyillicit configurations that could not be repaired by schwa-epenthesis. But wehave no such cases here: the /-i/ in the ["vlàk-í] could delete withoutfear of violating any phonotactic constraints.These observations suggest that the distribution of /i/ is very different inthe system of nouns when compared to the system of verbs and participles.37What is even more striking is that the suffixal exponents of the number,gender and case in the nominative forms (written in bold font) are exactlythe same as in the system of participles. Consider the participial paradigmagain, and below it the nominal paradigm in the nominative case:sg du plmasc "pòwn-∅-ú-∅ "pòw@n-∅-l-á "pòw@n-∅-l-ífem "pòw@n-∅-l-á "pòw@n-∅-l-é "pòw@n-∅-l-éneut "pòw@n-∅-l-ú "pòw@n-∅-l-á "pòw@n-∅-l-áTable 3.69: Participle √pòwn-í-l- ‘to fill’: gender-number exponentsuniformly give paradigms with a single tonal pattern here, which are also possible. Eitherway, tonal specification plays no role in our analysis, so it need not be a matter of concern.37It should be noted that with more complex suffixal morphology in nouns, some suf-fixes display a possible alternation between zero and [i], however, this is highly variable.Consider the diminutive [-(i)ts]: ["slik-a] ‘picture ( → ["slik-ts-a] ‘small picture (’vs. ["gred-a] ‘garden patch (’ → [gre"d-its-a] ‘small garden patch (’, though clearcases with a stressed [-"its] as this one are quite rare. There are also productive excep-tions which are very numerous and common: ["rok-a] ‘hand (’ → ["rok-its-a] ‘smallhand (’, ["torb-a] ‘bag (’ → ["torb-its-a] ‘small bag (’, etc. Furthermore,no surrounding [i] vowel participates in this seeming alternation: ["slik-ts-i] (,[gre"d-its-i] (, ["torb-its-i] (, etc. It is not so convincing that this is a phono-logical process ‘proper’ – it is likely a case of morphologically fixed allomorphy, which isexpected as these cases were historically derived by a process of high vowel deletion.1183.4. Nouns and adjectivessg du plmasc "vlàk-∅ "vlàk-á "vlàk-ífem "xìS-á "xìS-é "xìS-éneut "mèst-ú "mèst-á "mèst-áTable 3.70: Nouns in nominative case: gender-number exponentsNotice that exactly the same exponents are used so that even the syncretismsbetween the different forms in the paradigm are the same. Recall thatparticiples do not code case, but only gender and number, while nouns inaddition also code case. It would seem that the exponents presented in boldabove are the default exponents for gender and number and are used in thenominative case forms of nouns as such, viz. as default exponents.38 Onthe other hand, one could claim that the exponents of gender and numberin the participial paradigm are just ‘accidentally’ homophonous exponents,but this would be a grave stipulation, one that is avoided by assuming thatthey constitute the default set of exponents.The assumption that we are dealing with a set of default number andgender exponents is further confirmed by the paradigms of adjectives. Ad-jectives in NM Slovenian, much like nouns, code gender, number and case.They follow the following paradigm, for which only the nominative caseforms are given:sg du plm "lèp-∅ "lèp-á "lèp-íf "lèp-á "lèp-é "lèp-én "lèp-ú "lèp-á "lèp-áTable 3.71: Adjectives in NM Slovenian (nominative case)Notice that what was said about nouns above can also be applied to adjec-tives: they seem to use the same exponents as participles and nouns, whichconfirms that these exponents must be the defaults used for expressing gen-der and number in NM Slovenian. Now that we have also introduced theadjectival paradigms, observe that /i/ in the masculine plural form maysurface unstressed without any consequence.38This idea follows naturally from frameworks of morphology that permit exponent un-derspecification for morphosyntactic features. In short, the framework Distributed Mor-phology, which is taken up in this thesis (see sections 1.1 and chapter 4), would analyzesuch ‘default’ exponents by positing a rule that inserts them for any specific combinationof gender and number features, without making reference to case features.1193.4. Nouns and adjectivesSince it seems that we are dealing with a constant set of ‘default’ expo-nents expressing gender and number, we face quite a curious situation: exponent /-i/ undergoes deletion in the participles of NM Slovenian,but this same exponent fails to delete in nouns and adjectives, in phonolog-ical positions where it is predicted to delete in the participles, viz. in theword-final position of the prosodic word.120Chapter 4At the PF-InterfaceThis chapter discusses the data from NM Slovenian from the standpoint ofthe PF-interface. Section 4.1 is concerned with the morphosyntactic aspectsof NM Slovenian. In particular, it examines the morphosyntactic structureof verbs and participles; special emphasis is paid to the spell-outs of theaspectual head Asp0. Section 4.2 offers a discussion on the spell-out ofphonological domains at PF. Several approaches to the spell-out of phonol-ogy are discussed with emphasis on how phonological cyclicity is formallyimplemented. In section 4.3, we discuss how the phonological cycle thatencompasses the verbal stem (the root and theme vowel, or aspectual suffix)can be derived in NM Slovenian. Section 4.4 discusses the process of /i/-deletion that occurs in NM Slovenian verbs and participles but not nouns oradjectives, revealing that it cannot be modelled by the existing approachesto ‘construction-specific’ phonology. Section 4.5 discusses a tentative pro-posal that attempts to appropriately limit the construction-specific processof /i/-deletion in NM Slovenian.4.1 Morphosyntax of NM SlovenianThe goal of this section is to present a brief analysis of the morphosyntaxof NM Slovenian, to the extent that is relevant for the data which we havebeen discussing so far. As was already explained in chapter 1, the generaloutlook on morphosyntax taken up here is that of Distributed Morphology(Halle & Marantz 1993; Embick 2010), coupled with Minimalist assumptionson syntactic computation (Chomsky 1993, 1995, 2001); see chapter 1 for abrief introduction to the crucial concepts. Below, we first start by analyzingthe structure of verbs and participles, specifically the syntactic position ofthe theme vowels and aspectual suffixes.Our analysis of the morphosyntactic formation of verbs and participleswill be largely based on that proposed by Marvin (2002) for Standard Slove-nian, which is appropriate, given that the relevant morphosyntactic detailsspecified below are identical for NM Slovenian. According to Marvin (2002:83-90), verb formation is derived by successive cyclic head-movement of the1214.1. Morphosyntax of NM Slovenianroot to the tense head T0. We represent this with the following structuresfor verbs in NM Slovenian:TPT0 AspPAsp0 vPv0 √rootÐ→ TPT0Asp0v0√root v0Asp0T0AsPAsp0 vPv0√rootFigure 4.1: Verb formation: [[[[√root] v0] Asp0] T0]The structure illustrated here fits the categories that we have found in theverbs in NM Slovenian so far: they must encode different aspectual prop-erties and also tense. Verbs alone express present tense in NM Slovenian,precisely as in Standard Slovenian – see Marvin (2002) for details. Partici-ples, however, cannot occur independently, but require a slightly differentsyntactic configuration: they must always be accompanied by an auxiliaryverb. Marvin (2002: 83-90) proposes that participles in Slovenian do notundergo head-movement to T0, and that T0 has an independent spell-out,viz. the auxiliary verb. Following Marvin’s analysis, we may represent thestructure of participles in NM Slovenian in the following way:TPT0 PtcPPtc0 AspPAsp0 vPv0 √rootÐ→ TPT0 PtcPPtc0Asp0v0√root v0Asp0Ptc0AsPAsp0 vPv0 √rootFigure 4.2: Participial formation: [[[[√root] v0] Asp0] Ptc0]1224.1. Morphosyntax of NM SlovenianThe root undergoes successive cyclic head-movement up to Ptc0, but not toT0. Notice that this reflects the observation that (NM) Slovenian partici-ples code no tense on their own, whereas the verbs do. Notice, also, thatthere is no syntactic head in the verbal and participial structures above thatwould correspond to an agreement suffix: we will follow the proposal byHalle & Marantz (1994), which assumes that the agreement suffix on verbsand participles is the spell-out of the agreement head Agr0 adjoined to thepre-existing verbal/participial complex at PF.An important question that we must ask is which head spells out thetheme vowels.39 This is relatively straightforward for the aspectual suffixes:we may assume that those are spell-outs of Asp0. To determine the syntacticposition of these verbal theme vowels, we must consider their distribution,particularly in relation to the aspectual suffixes. For this reason, let us firstspend a few paragraphs discussing aspectual suffixes.As is typical of Slavic in general, imperfective aspect is the default as-pect encoded by ‘bare’ verbal stems (Slabakova 2001), but different aspec-tual readings may be triggered by prefixation and suffixation. This is trueof Standard Slovenian, as discussed in Toporišič (2000: 348) and Žaucer(2002: 1), and it is also true of NM Slovenian.The only aspectual suffix thatwe have discussed so far is the semelfactive, /-n-i/ for participles and /-n-e/for verbs. In the previous two chapters, we treated these two allomorphsof the semelfactive as bi-morphemic, viz. as ‘semelfactive+theme’, but inthis section we will actually treat them as monomorphemic, viz. as /-ni/and /-ne/, which is how Dickey (2003) analyzes the two semelfactive allo-morphs in Slovenian, and Svenonius (2004a: 183) offers a similar analysisof the Russian semelfactive. Below we will briefly sketch the alternative(bi-morphemic) analysis and explain that it makes no crucially differentpredictions for our analysis. Recall that the semelfactive reading reflectsa perfective and ‘punctual’ event. However, Slovenian makes use of otheraspectual suffixes, as well: these would be /-eva/ and /-uje∼ova/ (Dickey2003), which are usually termed secondary imperfectives in most literatureon Slavic languages and they trigger an ‘iterative reading’ (Slabakova 2001:84). In NM Slovenian, the two secondary imperfectives are /-eva/ and /-uje∼(u)va/. While /-eva/ may be used in verbs and participles, /-uje/ is theallomorph used in verbs, while /-(u)va/ the one used in participles. The [u]in /-(u)va/ occurs in this exponent variably. A typical property of secondary39Marvin (2002) offers a different treatment of theme vowels than that proposed here,but she does not discuss the relation between theme vowels and aspectual suffixes at all,which is crucial for our proposal above.1234.1. Morphosyntax of NM Slovenianimperfectives in Slavic is that they may only attach to verbal stems that aretelic, which means that they imply a semantic event ‘end-point’ (Slabakova2001: 84). Since bare verbal stems in Slavic are typically imperfective, asmentioned above, they are rendered telic by the addition of a prefix. Ob-serve the following table which illustrates that this is precisely the case inNM Slovenian, as well:imperfinf telicinf scnd.impinf√ris-a-t iz-Ris-a-t iz-Ris-u"va-t *Ris-uva-t ‘draw’√pix-a-t na-pix-a-t na-pix-u"va-t *pix-uva-t ‘blow’√jok-a-t ob-jok-a-t ob-jok-u"va-t *jok-uva-t ‘cry’√rez-a-t iz-Rez-a-t iz-Rez-u"va-t *Rez-uva-t ‘cut’√goR-e-t do-goR-e-t do-go"R-eva-t *goR-eva-t ‘burn’√xlap-e-t iz-xlap-e-t iz-xla"p-eva-t *xlap-eva-t ‘evaporate’Table 4.1: -uje∼(u)va and -eva as scnd.imperf in NM SlovenianThe prefixes /iz-/, /na-/ and /do-/ are telic because they imply an end-point of the (verbal) event that they scope over. Notice that the secondaryimperfectives may only attach to telic verbal stems, and never to bare, im-perfective verbal stems, hence the ungrammaticality of all the forms in thefourth column. Their semantic contribution is also that of ‘iteration’: [√ris-a-t] denotes an imperfective event of ‘drawing’, while [iz-√ris-a-t] denotesa complete event of ‘drawing’, but [iz-√ris-uva-t] denotes an iteration ofseveral completed events of ‘drawing’. These data also enable us to pin-point the syntactic position of the secondary imperfectives, specifically, wecan now claim that they are generated above the telic prefixes because theymust scope over any such prefix. Since these secondary imperfectives maynever stack with the semelfactive suffixes, we may assume that they are bothspell-outs of Asp0, the aspectual head above the v0. This is quite a typicalassumption on Slavic: Gribanova (2015) assumes this too for Russian.40Now that we have introduced the crucial distributional facts of the as-pectual suffixes (the semelfactives and the secondary imperfectives), we mayturn to theme vowels. It is interesting that the theme vowels never seemto coincide with any of the aspectual suffixes. Consider the following table,which shows that it is impossible to stack the theme vowel /-a/ and thesemelfactive suffixes:40Slabakova (2001: 86) positions the Slavic telic prefixes in a head below Asp0 andanalyzes the secondary imperfectives as spell-outs of an Event head, E0, directly belowAsp0 (Travis 1994, 2010), which is also an option for Slovenian, but one that shall not beexplored here.1244.1. Morphosyntax of NM Slovenianimperf[a] semelf√but-a- ✕ "but-a-∅ "but-a-l-a√pix-a ✕ "pix-a-∅ "pix-a-l-a✕√but-ne/ni "but-ne-∅ "but-ni-l-a✕√pix-ne/ni "pix-ne-∅ "pix-ni-l-aTable 4.2: Imperfective (default: -a) vs. semelfactive -ni∼neThe theme vowel [-a] can never co-occur with a semelfactive, which suggeststhat theme vowels occupy the same position as the semelfactive, viz. Asp0.41We can gather more support for this in the distribution of theme vowels inrelation to the secondary imperfectives:imperf[a,e] scd.imp√pix-a ✕ "pix-a-∅ "pix-a-l-a√Ris-e ✕ "RiS-e-∅ "ris-a-l-a✕√pix-uje/-uva na-pi"x-uje-∅ na-pix-u"va-l-a✕√Ris-uje/-uva iz-Ri"s-uje-∅ iz-Ris-u"va-l-aTable 4.3: Imperfective (default: -a,-e) vs. scnd.imperf -uje∼(u)vaNotice that regardless of whether the theme vowel in the imperfective verbalstem is [-a] or [-e], the stem will simply contain [-uje] in verbs and [-uva] inparticiples when a secondary imperfective is constructed. A similar situationoccurs with the /-eva/ secondary imperfective:42imperf[i,e] scd.imp√goR-i ✕ go"R-i-∅ go"R-e-l-a√xlap-i ✕ xla"p-i-∅ xla"p-e-l-a✕ do-√goR-eva- do-go"R-eva-∅ do-go"R-eva-la✕ do-√xlap-eva- iz-xla"p-eva-∅ iz-xla"p-eva-l-aTable 4.4: Imperfective (default: -i,-e) vs. scnd.imperf −evaWith the /-eva/ suffix, many of the imperfective verbal stems to which itattaches have the [-i] theme vowel in verbs, and [-e] in participles. When asecondary imperfective is constructed, the [-i] and [-e] are replaced by [-eva].41Note that the forms listed here are all underlying phonological representations: /"but-ni-l-a/ must, of course, surface as ["but-@n-l-a].42With the secondary imperfective exponent /-eva/, the allomorph /-uje/ may alsooccur in the verbal forms with some roots.1254.1. Morphosyntax of NM SlovenianIt, again, seems to be the case that the suffix /-eva/ cuts across the themevowel specification in the imperfective verbal stem.It seems that theme vowels are generally in complementary distributionwith the aspectual suffixes; this is especially clear with the semelfactivesand the /-uje∼(u)va/ secondary imperfective. For this reason, we will as-sume that theme vowels are also spell-outs of Asp0. This does not seem tobe unique to Slovenian, as Gribanova (2015) and Gribanova & Harizanov(2015) also note that theme vowels are tied to Asp0 in Russian.We must also mention two alternatives to this analysis. It may bepossible to treat the secondary imperfectives and semelfactives as /-uv-a/,/-uj-e/, /-ev-a/ and /-n-i/, /-n-e/, where the additional vowel is actually atheme suffix. If we assume, together with Oltra-Massuet (1999) and Embick(2010), that theme vowels are actual thm0 heads adjoined to the relevanthead at PF, we could say that /-uv/, /-uj/ and /-ej/ are the exponents ofAsp0 and that the following vowels are the exponents of thm0 adjoined toAsp0 – this is what Gribanova (2015) proposes for Russian. The second al-ternative is to say that the ‘extra’ vowels are actually the exponents of Asp0themselves, but that the secondary imperfectives (viz. /-uv/, /-uj/ and /-ev/) are the exponents of an Event head (E0), hosted in an Event Phrase(Travis 1994, 2010), located directly below Asp0, which is what Slabakova(2001: 86) proposes for the secondary imperfectives. Notice that what thesetwo analyses share with the one suggested above is the generalization thatthe ‘theme vowels’ must occur in a projection above the verbalizer v0, eitherin Asp0 or directly above Asp0, and that they are tied to aspect in someway. However, our original analysis will be sufficient for our purposes. Theremaining two analyses described here could very well be adopted, thoughwe would not benefit from this here in any way, which is why the simpleroption is selected.An additional example can also be given: ‘bare’ verbal stems indeedtypically encode imperfective aspect, but this is not necessarily always thecase. With some verbal roots, a difference in the selection of the themevowel is sufficient to express a a contrast between perfect and imperfect as-pect: Žaucer (2002: 1) cites an example from Toporišič (2000: 348-350),who shows that, in Standard Slovenian, [√pik-a-ti] ‘stab (inf)’, with animperfective reading, is contrasted by [√pitS-i-ti] ‘stab (inf)’,43 which hasa perfective reading. A very similar scenario obtains in NM Slovenian: con-sider ["pik-a-t] ‘stab (inf)’, which has an imperfective reading, and ["pitS-@t](‘stab inf)’, which has a perfective reading. This actually provides further43It is assumed that the root-final /k/ palatalizes to [tS].1264.1. Morphosyntax of NM Slovenianevidence for the idea that theme vowels are in fact exponents of Asp0, asthey seem to be closely connected with the interpretation of aspect.According to the assumptions on morphosyntax that we have specified sofar, a participle would have the following structure prior to any linearizationprocedure at PF, but after Agr0 is adjoined to Ptc0:Ptc0Ptc0Asp0v0√feedxRanv0Asp0iPtc0lAgr0iFigure 4.3: Participle /√xRan-i-l-i/ ‘having fed (’The exponents specified in the tree above are, of course, not present priorto the linearization of this structure, and are only given for reasons of clearexposition. We can now list some of the basic Vocabulary Insertion rulesthat insert exponents for the given syntactic heads. We will not attempt toaccount for any of specific exceptions with this brief analysis, but rather tryto account for the most general patterns of verbal and participial exponencein NM Slovenian since the focus of this thesis is not on segmental exponenceas such. Starting with the ‘theme’ vowels, these are inserted as the defaultexponents of Asp0 for specifics lists of roots:(32) Theme vowels inserted for Asp0 in verbs[Asp0] ↔ a / ⟨{√jok-, √pix-, ...}, v0 ⟩[Asp0] ↔ i / ⟨{√xRan-, √del-, ...}, v0 ⟩[Asp0] ↔ e / ⟨{√pel-, √mR-, ...}, v0 ⟩(33) Theme vowels inserted for Asp0 in participles[Asp0] ↔ a / ⟨{√pel-, ...}, v0 ⟩ ⟨ Ptc0 ⟩[Asp0] ↔ ∅ / ⟨{√mR, ...}, v0 ⟩ ⟨ Ptc0 ⟩The list of Vocabulary Insertion (VI) rules in (32) shows that the threebasic theme vowels are inserted for Asp0 in the presence of different rootswhen constructing verbs. Some of the VI-rules in (32) also apply when1274.2. On the nature of Spell-Out at PFconstructing a verbal stem that is used to build a participle: this is the casewith the rule that inserts /-a/ and /-i/. But for most roots that are assigned/-e/ in verbs, we need a special rule that inserts /-a/ when building a verbalstem which is used to construct a participle, which is given in (33). Sincethe roots that take the theme /-e/ actually further bifurcate in participles,we also need a special rule for roots such as √mR-, which are assigned zerothemes in participles; this is captured in the second rule in (33). Thatthe rules in (33) will block the third rule in (32) is ensured by the SubsetPrinciple (see section 1.1), since the rules in (33) require a more specificcontext, viz. the presence of the Ptc0 head.(34) Semelfactives and secondary imperfectives in verbs[Asp0[semelf]] ↔ ne / ⟨ v0 ⟩[Asp0[scd.imp]] ↔ uje / ⟨{√pix-, √Ris-, ...}, v0 ⟩[Asp0[scd.imp]] ↔ eva / ⟨{√goR-, √xlap-, ...}, v0 ⟩(35) Semelfactives and secondary imperfectives in participles[Asp0[semelf]] ↔ ni / ⟨ v0 ⟩ ⟨ Ptc0 ⟩[Asp0[scd.imp]] ↔ uva / ⟨{√pix-, √Ris-, ...}, v0 ⟩ ⟨ Ptc0 ⟩The rules in (34) and (35) account for the exponence of the semelfactive andsecondary imperfective suffixes. As before, all the suffixes have allomorphsspecific to the participle, except for /-eva/, which is accounted for by positingthe third rule in (34) alone.Notice that the contexts of all the rules above specify actual exponentsof roots; all the rules are assumed to actually specify no exponents, so thatinstead of √mR- ‘die’ the list in reality specifies √die. The exponents areonly used for reasons of clearer exposition.4.2 On the nature of Spell-Out at PFUp to now, we have only considered the spell-out of syntactic structure – thiswas done in section 1.1. There, we explained that in a phasal approach tosyntax (Chomsky 2000, 2008), the syntactic component produces syntacticcycles by spelling out syntactic structure to the interfaces ‘phase by phase’.This section, on the other hand, is dedicated to discussing how the spell-outof phonological structure proceeds at the PF-interface: this level of spell-outrefers to structure made up by exponents, which is the result of linearizationand Vocabulary Insertion (see section 1.1 for details). We will particularlybe concerned with how the domains of phonological spell-out are defined.1284.2. On the nature of Spell-Out at PFSome theories of phonology, most prominently many instantiations ofOptimality Theory (Prince & Smolensky 2004), advocate a purely globalapproach to phonological spell-out: entire words must be spelled out simul-taneously, under such a view, without any mechanism of cyclic application.On the other hand, frameworks such as that of Cophonology Theory (Orgun1996; Inkelas & Zoll 2007; Inkelas 2008, 2011) make use of cyclic applicationof phonology that is determined by the morphological structure of the word.Consider the word [[[[√root] x] y] z]:π3(rootxy, z)π2(rootx, y)π1(root, x)root xpi1ypi2zpi3Figure 4.4: Cyclic application in Cophonology TheoryCophonology Theory allows the grammar to contain several ‘cophonologies’:if one part of the grammar exhibits productive phonology that is not com-patible with productive phonology in another part of grammar, then wemay say that this grammar contains two cophonologies. In CophonologyTheory, cophonologies are tied to specific morphological constructions: amorpheme may bear a diacritic that triggers a phonological cycle, in otherwords, it triggers a pass through the phonology. The diacritics that triggerthese cophonologies are represented with π above in Figure (4.4). Everycophonology π triggered by a morpheme is construed as a function thatconcatenates two (exponents of) morphemes and produces a phonologicaloutput. Note that each of the morphemes may be specified with a differentcophonology, essentially π1, π2 and π3: Inkelas (2008: 8) explains that theapplication of the phonological cycles represented in Figure (4.4) proceedsfrom the bottom up: π1(root, x) must take place first, whereby π1 processes/root-x/, then π2(rootx, y) applies, with π2 processing /rootx-y/, and fi-nally π3(rootxy, z) applies, with π3 processing /rootxy-z/. Inkelas & Zoll(2007) term this rigid course of application the Stem Scope property:(36) Stem Scope (Inkelas & Zoll 2007: 144)The scope of morphologically conditioned phonology is the stemformed by the word-formation construction in question.1294.2. On the nature of Spell-Out at PFThe Stem Scope property predicts that no morpheme in a structurally higherposition can be influenced by a cophonology specified in a structurally lowerposition: in Figure (4.4), π2 can only affect the morphological constituentsthat it scopes over, but crucially not morphemes ‘up the tree’.In Lexical Phonology (Kiparsky 1982a,b; Mohanan 1986) and Stratal OT(Kiparsky 2000; Bermúdez-Otero 2011), phonology is spelled out in severalstrata: the stem once, it is built, passes through the Stem-level phonologyand when the entire word is constructed, it passes through the Word-levelphonology. Several words together then pass through the Postlexical, orPhrasal-level of phonology. These levels may all endorse different phonolo-gies, much akin to different cophonologies in Cophonology Theory, just thatthese are fixed for the entire language and cannot vary from one morpho-logical construction to another. However, at the Stem-level phonology (andthe Word-level, in some versions of these theories), morphemes making upthe stem may be specified to trigger a pass through this level’s phonologybefore the structure is sent to the next level. These are essentially phonolog-ical cycles, again like in Cophonology Theory, but with a difference: thesecycles may only represent a reiteration of the Stem-level’s phonology andcannot trigger different, ‘construction-specific’ phonological effects.The system of phonology that Embick (2010, 2013) assumes is cruciallyrooted in the framework proposed by Halle & Vergnaud (1987). Along theirlines, Embick assumes the following three components:Cyclicphonz→ M-Wordphonz→ PhrasalphonFigure 4.5: Embick’s phonological grammarThis division of labour between the different phonological components di-rectly reflects that in Lexical Phonology and Stratal OT, where the ‘Cyclicphonology’ corresponds to the Stem-level, the ‘M-word phonology’ to theWord-level, and the ‘Phrasal phonology’ to the Postlexical level. For com-pleteness, we should mention that in conjunction with Halle & Vergnaud(1987), this system is driven by phonological re-write rules and not OT-style computation. Consider the morphological construction [[[[√root] x]y] z] again: after the linearization procedure has taken place with the out-put [√root⊕x, x⊕ y, y⊕ z], Vocabulary Insertion takes place. Imaginethat Vocabulary Insertion inserts the exponent /β/ for y. This exponentmay bear a diacritic which triggers phonological effects that are specific toformations that include this morpheme. Notice that this precisely mirrors1304.2. On the nature of Spell-Out at PFthe phonological cycles triggered by diacritic specification in CophonologyTheory. Let us construct a sample derivation in Embick’s system, wherethe exponent of y, /β/, bears a diacritic (let us call it ‘π’) which triggers aphonological cycle (‘π-cycle’):(37) Spelling out [√root⊕x, x⊕ y, y⊕z] in the phonologya. Step # 1: VI inserts /α/ for xb. Step # 2: VI inserts /β/pi for y→ Phonological cycle π triggered on /√root-α-β/→ Output of π-cycle: [√rootαβ]c. Step # 3: VI inserts /γ/ for zd. Step # 4: Ship off /√rootαβ-γ/ to the M-Word phonologyNote that all the exponents could potentially be specified for a diacritic,which would imply three phonological cycles throughout the word beforethe derivation reached the M-Word phonology. These cycles directly reflectthe cyclic application of cophonologies in Cophonology Theory. The onlydifference lies in the fact that Embick’s π-cycles trigger phonological re-writerules of the SPE fashion (Chomsky & Halle 1968), whereas cophonologiesinvolve whole rankings of OT-style constraints. It is equally possible that noexponent would bear a diacritic in (37), which would mean that the string ofexponents would be directly shipped off to the M-Word phonology, withouttriggering any phonological cycles.In (37), we stated that a π-cycle is triggered by a diacritic π on theexponents of morphemes and not on the morphemes (i.e. linearized syntacticheads) themselves. We adopt this rationale by following Embick (2013) whooffers a brief discussion on this topic. Embick (2013: 12) points to anobservation made by Halle & Vergnaud (1987): Halle & Vergnaud make theobservation that the English suffix -ity triggers a phonological (π) cycle, butthe suffix -ness does not.44 Embick explains that both these suffixes seem tobe exponents of the nominalizing head n0, which implies that two exponentsof the same morpheme may or may not trigger a cycle. This implies thatπ-cycles are triggered on the exponents of morphemes and are, in that sense,a ‘by-product’ of the PF-interface. We will require this assumption in ouranalysis of phonological cyclicity, as discussed in the following section, 4.3.44Halle & Vergnaud (1987) analyze this as a property that follows from the distinctionbetween ‘Level 1’ and ‘Level 2’ affixes, which, in turn, reflects the Stem vs. Word leveldistinction in Lexical Phonology and stems from the earlier work on this by Kiparsky(1982a,b).1314.3. Phonological cyclicity in NM SlovenianIn section 1.1, we subscribed to the idea that syntactic computation pro-ceeds in phases (Chomsky 2001, 2008), which implies that the PF-interfacereceives ‘chunks’ of structure in a manner that is fully regulated by thesyntax itself. The inevitable consequence of this is that a phase-cycle isnot only a syntactic cycle, but also a phonological cycle. Works such asMarvin (2002), Marantz (2007), Samuels (2009), Piggott & Travis (2013)and Newell & Piggott (2014) argue that phase-cycles indeed form phono-logical cycles. This is the over-reaching proposal that we will ultimatelysubscribe to in section 4.5. However, Embick (2013) notes that π-cycles,which we have explained as a by-product of the PF-interface, cannot over-lap with phase-cycles: Embick shows that π-cycles are sometimes triggeredby exponents of non-phase heads and that phase-heads themselves need nottrigger a π-cycle at all. This implies that two types of cyclicity exist:(38) Two types of cycles45a. Phase-cycles: they are determined by syntax, and are realizedas a ‘piece of structure’ that the PF-interface receivesb. π-cycles: within the piece of structure (a phase) sent to the PF-interface, exponents of heads may trigger phonological cycles –these are π-cyclesIn this sense, a phase-cycle is a ‘hard’ cycle that cannot be manipulated atPF, but π-cycles are an optional by-product of the interface. This is why π-cycles are used to encode ‘construction-specific’ and exceptional phonologicaleffects in the way that we discussed at the beginning of this section.4.3 Phonological cyclicity in NM SlovenianNow that the relevant basics of NM Slovenian morphosyntax were discussedin section 4.1 and the assumptions on cyclicity in 4.2, it is crucial that wediscuss how the phonological cyclicity in the phonological computation ofverbs and participles is formally implemented at the interface. In particular,we will explain how the morphosyntactic observations on theme vowels andaspectual suffixes tie in with the issue of cyclicity.45Embick (2013) discusses this same distinction and he calls pi-cycles ‘φ-cycles’, butthat term is avoided in the present thesis so as not to be confused with morphosyntactic‘φ-features’, even though the symbol pi does not correspond to a letter representing thesegment [f] in Greek.1324.3. Phonological cyclicity in NM Slovenian4.3.1 Phase status of verbal stemsRecall from section 4.2 that we made the distinction between syntactic cycleson the one hand, and phonological cycles on the other. Syntactic cycles, alsoknown as phases (Chomsky 2001, 2008), send a ‘chunk’ of syntactic struc-ture to the PF-interface, where it is linearized, replaced by phonologicalexponents (through Vocabulary Insertion), and also processed phonologi-cally. A phase-cycle, therefore, inevitably forms a domain for phonologicalcomputation – essentially a phonological cycle. In 4.2, we subscribed tothe definition of a phase domain where a phase head triggers spell-out ofdomains that already contain a phase-cyclic head, following Embick (2010,2013) and Marantz (2013). This means that the root will always be pro-cessed in the same phase-cycle as the first categorial head that categorizesit. For our verbal and participial structures discussed above, this meansthat entire verbs and participles will be computed in the same phase-cycle,as they only contain one phase head, the verbalizer v0.Notice that it is actually crucial, for our phonological analysis, that theroot be computed together with the exponents of Asp0. For instance, if theroot √pown- ‘fill’ were first processed in a phase, separated from the theme/-i/, then we would expect the input to the verbal cycle to be /("po.w@n)-i/with an epenthesized schwa in the root. This is unproblematic if the stressis on the root throughout the derivation, as the final /-i/ is predicted todelete and the output would be the predicted [("po.w@n)]. But if the stresshas to occur on the theme /-i/, we predict that /("po.w@n)-i/ would surfaceas *[po.w@.("ni)], and not as the correct form [pow.("ni)] because the schwaepenthesized in a previous cycle would be retained (see sections 3.1 and 3.2for details). Our analysis, so far, seems compatible with a phasal approachto syntax, since a phase will include both the root and the Asp0 head in thesame phase-cycle.Before we continue discussing phonological cyclicity, let us turn to a po-tential morphosyntactic issue that roots such as √pown- present. As stated,it is crucial that the roots and the theme vowels are computed in the samephonological cycle and, therefore, also in the same phase-cycle. This is es-pecially crucial for roots such as √pown- ‘fill’ and √d@Rgn- ‘rub’. In section2.2.3 on page 42 we noted that √pown-, √d@Rgn- and other such formationsmust be roots and not cases of roots suffixed with some [-n] suffix. We havealso been assuming that these roots form verbal stems directly, with no inter-vening categorial heads between √root and v0. This assumption is trivialfor √d@Rgn- because it cannot exist as a noun or adjective as *["d@Rg@n],but this is not trivial at all for √pown-, which does exist independently as1334.3. Phonological cyclicity in NM Slovenianan adjective, viz. ["pow@n] ‘full (’. At first glance, it seems that wemay be forced to claim that √pown- must first build an adjectival stemwhich is then used to construct the verbal stem, ultimately constituting ade-adjectival verb or participle. This would imply the structure [[[[√pown]a0] v0] ...], which would mean that [[√pown] a0] would constitute the firstphase-cycle, and this would inevitably trigger schwa-epenthesis and causegrave problems for our account of cyclic deletion and epenthesis, as indi-cated above. However, we will maintain that √pown- does actually build averbal stem directly, viz. [[[√pown] v0] ...], as there is independent semanticevidence to support this claim.Adjectives such as ["pow@n] ‘full (’, ["pRaz@n] ‘empty (’ and[(@)R"detS] ‘red (’ share a common property: they may be construedas gradable adjectives (Kennedy 1997). A typical property of gradable ad-jectives is that they can be modified by degree adverbials such as ‘very’ or‘quite’ Kennedy (1997: 1). This is true of ["pow@n-∅] and [(@)R"detS-∅]:(39) Adjectival modification by degree adverbial zelua. Telethese:m.plsodibarrels:m.plsobe:3p.plzeluverypownifull:adj.m.plThese barrels are very full.b. Talethis:f.sgslikaphoto:f.sgjebe:3p.sgzeluvery(@)RdetSared:adj.f.sgThis photo is very red.In (39) above, the two adjectives are modified by the degree adverbial zelu‘very’, which shows that the two adjectives are gradable. Kennedy & Levin(2008) demonstrate that de-adjectival verbs, if constructed on the base ofa gradable adjective, reflect the semantic properties of gradable adjectives.This implies that de-adjectival verbs (and participles) constructed on anadjectival base should exhibit the same gradability properties as their ad-jectival counterparts. However, the de-adjectival participles that √pown-and √RdetS- supposedly construct are very different:(40) Adjectival modification by degree adverbial zelu in verbal stemsa. *JanezJohnjeaux:3p.sgzeluveryna-pownufill:ptc.m.sgtelethese:m.plsodebarrels:m.plJohn filled these barrels very much.b. JanezJohnjeaux:3p.sgzeluverypo-RdetSiwredden:ptc.m.sgtolethis:f.sgslikophoto:f.sgJohn reddened this photo very much.1344.3. Phonological cyclicity in NM SlovenianIn (40a), we can see that the participle built on √pown- cannot be modifiedby the degree adverbial zelu, but this is in fact possible for the participlebuilt on √RdetS-. These data imply that √pown- truly does not have thestructure [[[[√pown] a0] v0] ...], but rather [[[√pown] v0] ...], as no degreemodification is possible in the verbs and participles it builds – there is noadjectival head for the adverbial to scope over. The root √RdetS-, on theother hand, does seem to first construct an adjective, so that its verbalstem has the structure [[[[√RdetS-] a0] v0] ...], precisely because it allowsdegree modification. This is a direct confirmation of the morphosyntacticassumptions that we have been making so far.464.3.2 pi-cycles in the verbal stemLet us now turn back to the topic of phonological cyclicity. Recall that aphase-cycle is always also a phonological cycle, but, as was discussed in sec-tion 4.2, ‘smaller cycles’ triggered by specific morphemes may occur withinthe larger phase-cycle. We termed these cycles π-cycles – ‘π’ implying thatthey are a by-product of the PF-interface and are not correlated with thesyntactic component in any way. It is important to ask if a theory of the PF-interface truly requires such cycles, as a more parsimonious solution wouldbe one that could derive all cyclicity effects in phonology through phase-cycles, which have independent syntactic motivation. But as Embick (2013)points out, π-cycles seem necessary for empirical reasons. This is also thecase with our data. In our analysis of the verbal and participial system,we had to posit two cycles of phonological computation in sections 3.1 and3.2. For our analysis, it was crucial that the root is computed together withthe exponents of Asp0 (theme vowels, aspectual suffixes), but crucially notwith the participial exponent or the exponents of the agreement head. Itis not possible for there to be two phase-cycles in these structures becauseonly one categorial head is present (viz. v0), which means that we mustposit a π-cycle somewhere in the verbal stem. The appropriate candidateseems to be the Asp0 head, as all the theme vowels and aspectual suffixesare its exponents. If Asp0 triggers a π-cycle, this means that [[[√root ]v0] Asp0] are computed first, and the output of this cycle is then computedwith [[Ptc0 ] Agr0 ] for participles and [[T0 ] Agr0 ] for verbs, all of which iscontained in one phase. And this is precisely what we require.Recall, from section 4.2, that the π-cycles are indicated on morphemes by46Please note that the phase-cyclic status and properties of adjective formations willnot be examined in any other way in this thesis. Such research should be conducted infuture studies.1354.3. Phonological cyclicity in NM Sloveniandiacritic indices, and that these diacritics are specified on the exponents ofmorphemes (i.e. exponents of syntactic heads) rather than on the morphemesthemselves, as discussed in Embick (2013: 12). In this sense, whether a spe-cific morphosyntactic construction undergoes a π-cycle or not is a completelyarbitrary matter of lexical specification. This is an assumption that is verymuch needed for our data. Recall from section 3.1 that the two cases of sin-gle consonant roots √b-/√bi- ‘be’ and √S- ‘go’ do not undergo two cycles,whereas all roots such as √mR- ‘die’ must undergo two cycles, and this is alsothe case with all constructions that take the theme /-i/ and all semelfactivesbuilt with /-ni/.For the √mR-type cases, we found the evidence for a cycle in the over-application of schwa-epenthesis on the surface, as in [u"m@R@l] ‘having died(’ – the exponent of Asp0 here is a zero theme. The cases that involvethe theme /-i/ and the semelfactive /-ni/ require two cycles so that the un-stressed [i]-vowels in these two suffixes can be computed in the word-finalposition and, therefore, can undergo word-final vowel deletion. Anotherpiece of evidence is also found with √CVCr roots (r represents any sono-rant), where an epenthesized schwa must remain in a position before thesonorant and not in a different position on the surface, as in [√pow@n-l-i]‘having filled (’. This suggests that the exponents of Asp0, /-∅/, /-i/and /-ni/, are the triggers of π-cycles. They must, therefore, bear a diacritic,viz. /-∅/pi, /-i/pi and /-ni/pi, which triggers a π-cycle as soon as VocabularyInsertion inserts one of these two exponents for Asp0.47The cases with √b-/√bi- and √S-, on the other hand, cannot be com-puted in two cycles: recall that the participles constructed from these tworoots also involve a zero theme; the √bi- form occurring in the masculinesingular only must be a suppletive form of the root, as we discussed in sec-tion 3.1 on page 108. A pass through the first π-cycle would inevitably leadto schwa epenthesis, yielding *["b@] and *["S@], and the second cycle wouldthen have to output *["b@l] and *["S@l]:input 1st cycle Input 2nd cycle/√b-∅-/ *("b@) /("b@)-l-i/ *("b@l)/√S-∅-/ *("S@) /("S@)-l-i/ *("S@l)Table 4.5: √b- ‘be’ and √S- ‘go’ under two cycles ( the case with the theme /-∅/, nothing seems to preclude zero exponents alsointroducing pi-diacritics into the computation. In terms of Distributed Morphology, thismeans that no segmental exponent is inserted, but only the diacritic is.1364.3. Phonological cyclicity in NM SlovenianThe correct outputs here are ["bli] and ["Sli] with no schwa-epenthesis. Noticethat the final /-i/ is stressed, which in itself is problematic for a cyclictreatment: the final inflections, such as the /-i/ above, are never stressedin any of the other participial forms. The roots √b- and √S- are trulyexceptional in this respect, as well. In all the other participles and verbsin NM Slovenian, the stress is either on the root, which is probably thedefault option, or it may also occur on the theme, but never beyond it.This observation suggests that stress is typically assigned in the first π-cycle, where it either occurs on the root or the theme, but in the second,participle cycle, the stress remains fixed on the theme. However, if no cycleis triggered by the exponent of Asp0 for the roots √b-/√bi- and √S-, it isdirectly predicted that the final inflections will receive stress and that noschwa-epenthesis will occur, simply because the roots (in all but the no vowel. Since the Asp0 spells out a zero theme with these tworoots, this cannot be the same zero theme that is spelt out with roots suchas √mR-. Let us call this new zero theme /-∅2/, and the one occurring with√mR- /-∅1/. /-∅2/ must not be specified for a π-cycle, but /∅1/ must be.The crucial list of exponents that must or must not be specified to un-dergo a π-cycle is the following:(41) π-specification of Asp0 exponentsπ-specified/-i/pi , /-ni/pi, /-∅1/piπ-unspecified/-∅2/This formalizes the observation that /-i/, /-ni/ and /-∅1/ will trigger a π-cycle, but /-∅2/, which occurs only with √bi-/√b- ‘be’ and √S- ‘go’, willnot. Other exponents of Asp0 may or may not be specified for a π-cycle; it isirrelevant for our phonological analysis if they are or are not specified. How-ever, it is likely that all the other theme vowels, /-a/ and /-e/, and aspectualsuffixes which attract stress are also specified for a π-cycle: this may be thecase because the theme and aspectual suffixes are in the word-final positionin the first cycle, where an Alignment constraint (McCarthy & Prince 1993)could trigger alignment of stress with that edge. This would explain whystress may never occur to the right of Asp0-exponents, but we leave a formalinvestigation of this for a future study.4848Because all Asp0-exponents seem to trigger a pi-cycle, it seem fruitful to explore, inthe future, if the lack of cyclicity with √b- and √S- could be derived from independentfactors – for instance, a different position in the syntactic tree.1374.4. /i/-deletion at the interfaceThe analysis of π-cyclicity that is advanced above may appear somewhatstipulative. However, positing π-cycles is a mechanism used to formalize ex-ceptional patterns which must be specified lexically and are, therefore, un-predictable. This is the same stance that Cophonology Theory (Orgun 1996;Inkelas et al. 1997; Inkelas & Zoll 2007; Inkelas 2008, 2011) and Stratal OT(Kiparsky 2000; Bermúdez-Otero 2011) would need to take to explain thepresence or absence of phonological cycles in different morphological con-structions. The √b-/√bi- and √S- roots form a specific morphologicalconstruction that behaves differently in terms of cyclicity than the otherparticiples in NM Slovenian, and positing a zero theme specific to these tworoots is a way of formalizing a ‘construction-specific’ effect – or, in this case,rather a lack of it.4.4 /i/-deletion at the interfaceIn the previous sections of this chapter, we provided a basic description ofNM Slovenian morphosyntax and we also gave a brief analysis of the phono-logical cyclicity that is found in the verbs and participles. All our analyses,phonological and morphosyntactic, have been successfully captured by theapproach of Distributed Morphology (Halle & Marantz 1993) coupled witha cyclic OT-style grammar (Prince & Smolensky 2004). However, the goalof this section is to address a problem that cannot be so readily dealt withby the existing theories of the PF-interface. In the paragraphs below, we willexplain that the /i/-deletion process that we found in the verbs and partici-ples of NM Slovenian is a ‘construction-specific’ phonological effect whosedomain of application requires a very specific view on spell-out domains.In section 3.2, we explained that the verbs and participles in NM Slove-nian are processed by a phonological grammar that productively licenses/i/-deletion in the word-final position of the Prosodic Word. Some typicalexamples include the following:input 1st cycle Input 2nd cycle/√xRan-i/ "xRan /"xRan-l-i/ "xRa.n@l‘feed’/√pix-ni/ "pi.x@n /"pi.x@n-l-i/ "‘blow (semelf)’/√pown-i/ "po.w@n /"po.w@n-l-i/ "‘fill’Table 4.6: /i/-deletion in word-final position ( participles)1384.4. /i/-deletion at the interfaceThe table above represents three typical participial formations: a simple√CVC root (√xRan-), a semelfactive (√pix-ni-) and a √CVCr root(√pown-). In the first cycle, only the verbal stem is processed: here theword-final theme /-i/, or the /i/ vowel in the semelfactive suffix /-ni/ aredeleted. In the second cycle, the whole participle is processed, and here theword-final /-i/ deletes with simple √CVC roots, but not with semelfactivesor √ CVCr roots, for independent phonotactic reasons (see section 3.2 fora formal phonological analysis). The crucial observation is that verbs andparticiples are processed by phonology that promotes /i/-deletion in word-final position. The constraint ranking that this phonology subscribes to isthe following (again, see section 3.2 for details):⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩*Nuc/CSonSeqAl-RMwd,PwdMax-F(µ)⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭≫ { Max-Base } ≫ { *˘i]Pwd } ≫⎧⎪⎪⎪⎨⎪⎪⎪⎩Max(i)Dep*@⎫⎪⎪⎪⎬⎪⎪⎪⎭Figure 4.6: Constraint ranking (/i/-deleting phonology)The crucial aspect of this constraint hierarchy that licenses /i/-deletion inverbs and participles is the *˘i]Pwd ≫ Max(i) ranking.However, in section 3.4, we showed that this process of /i/-deletion mustin fact be limited to verbs and participles, as no /i/-deletion takes place inthe systems of nouns and adjectives. Consider the following examples:input/√kol-i/ " ‘poles (n)’/√vlak-i/ " ‘trains (n)’/√gOvoR-i/ "gO.vo.Ri ‘speeches (n)’/√lep-i/ "le.pi ‘nice (adj)’/√bel-i/ " ‘white (adj)’Table 4.7: No /i/-deletion ( nouns/adjectives)In the nouns and adjectives of NM Slovenian, no /i/-deletion takes place.Notice that with simple √CVC roots, the word-final /-i/ is allowed to sur-face, which is the exact position where it has to delete in verbal and par-ticipial formations.49 Recall from section 3.4 that the /-i/ in the nouns49Nothing like a word ‘minimality effect’, in the sense of Orie & Pulleyblank (2002),that would require the word to be bisyllabic is blocking /i/-deletion, because /i/-deletionstill fails to occur with polysyllabic roots (cf. √gOvoR- above) where minimality effectswould not apply, as the structure resulting from deletion would still be bisyllabic.1394.4. /i/-deletion at the interfaceand adjectives is the same exponent that codes in the participles.We are dealing with a case of ‘construction-specific’ phonology: verbsand participles subscribe to an /i/-deleting phonology, but nouns and ad-jectives do not. In the system of phonology advocated by Embick (2013),which was discussed in section 4.2, this can be formalized by claiming thata morpheme specific to verbal and participial constructions triggers a setof re-write rules that include an /i/-deleting rule. In Cophonology Theory(Inkelas & Zoll 2007; Inkelas 2008, 2011), which was also discussed in sec-tion 4.2, this would be formalized in a very similar way: we would needto say that verbs and participles subscribe to an /i/-deleting cophonology,whereas nouns and adjectives subscribe to an /i/-preserving cophonology.The /i/-preserving cophonology must thus involve a re-ranking of Max(i)so that it comes to dominate *˘i]Pwd:⎧⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎩*Nuc/CSonSeqMax(i)Al-RMwd,PwdMax-F(µ)⎫⎪⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎪⎭≫ { Max-Base } ≫ { *˘i]Pwd } ≫ { Dep*@ }Figure 4.7: Constraint ranking (/i/-preserving phonology)Max(i) needs to be high-ranked in the /i/-preserving cophonology, wherethe Max-constraints on other segments in NM Slovenian are presumablypositioned (since no other deletion seems apparent). Other re-rankings ofthe lower-ranked constraints might be necessary in this cophonology, but, ifthis is the case, it stems from independent reasons that we will not pursuehere. The important thing to observe is that a minimal re-ranking derivesthe difference between the cophonology in which verbs and participles areprocessed and the one which derives nouns and adjectives.Let us now discuss how the phonological grammar assumed by Embick(2010, 2013) would formally implement this construction-specific phonolog-ical effect of /i/-deletion. Recall from 4.2, that Embick’s phonology consistsof a Cyclic level, which feeds the ‘M-Word’ level, and the latter feeds aPhrase level phonology. Any construction-specific phonological effects haveto be triggered by diacritics specified on the exponents of affixes. Thismeans construction-specific effects will only be possible on the Cyclic levelof phonology, as the M-Word level phonology is fixed for the entire grammar.Given our discussion of participles in 4.1, a linearized string of morphosyn-tactic heads that make up a participle would be the following:1404.4. /i/-deletion at the interface(42) Linearized participles in NM Slovenian[√root⊕v0, v0⊕Asp0, Asp0⊕Ptc0, Ptc0⊕Agr0]One of the exponents of the heads in this string will need to bear a diacriticthat will bring about an /i/-deleting phonological effect. In section 4.3, weexplained that some exponents of Asp0 are the ones responsible for trigger-ing a π-cycle in the verbal stem; recall that, in this cycle, any word-finalexponent of Asp0 that ends in /i/ undergoes /i/-deletion, which suggeststhat the diacritic ‘π’ specified on the exponents of Asp0 is responsible forbringing about /i/-deletion as well as triggering a cycle.Before we proceed with formalizing this construction-specific /i/-deletionprocess, a brief caveat is needed. As mentioned before, Embick’s systememploys phonological re-write rules. Since we have been using an OT-stylegrammar, as discussed in section 1.1, we need to assumed that π-diacriticsdo not trigger the application of phonological rules, but rather trigger apass through a specific cophonology, in the spirit of Cophonology Theory(Inkelas & Zoll 2007; Inkelas 2008, 2011). This assumption is a necessaryconsequence of assuming an OT-based phonological grammar with cyclic ap-plication. However, the discussion of the observations in this section shouldbe just as relevant for rule-based approaches to phonology.Since it is the exponents of Asp0 that trigger a cophonology π thathas the constraint ranking that promotes /i/-deletion, we can construct thefollowing derivation:(43) Spelling out [√feed⊕v0, v0⊕Asp0, Asp0⊕Ptc0, Ptc0⊕Agr0[]]a. Step # 1: VI inserts /xRan/ for √feedb. Step # 2: VI inserts /∅/ for v0c. Step # 3: VI inserts /i/pi for Asp0→ Phonological cycle π triggered on /xRan-∅-i/→ Output of π-cycle: ["xRan]d. Step # 4: VI inserts /l/ for Ptc0e. Step # 5: VI inserts /i/ for Agr0[]→ Ship off /"xRan-l-i/ to the M-Word phonologyIn (43) above, it is the theme vowel /-i/ that triggers the π-cycle as anexponent of Asp0. This appropriately triggers deletion of the theme /-i/,which is in word-final position in the first cycle. However, no additionalcycles are triggered after that and /"xRan-l-i/ is sent off to the M-Word levelphonology. Since the M-Word level phonology should be fixed for the entire1414.4. /i/-deletion at the interfacegrammar, it is not clear how the derivation in (43) could delete the /-i/. We could assume that the M-Word level phonology also triggers/i/-deletion, which would then produce the correct output ["xRa.n@l]. How-ever, this would give entirely incorrect predictions for nouns and adjectives,predicting /i/-deletion where /i/ must in fact be preserved on the surface.If the M-Word level phonology does not trigger /i/-deletion, the derivationin (43) predicts that word-final unstressed /i/ would be preserved in nounsand adjectives, but also in verbs and participles, outputting *["].We could also try to argue that the /-i/ triggers an additional π-cycle before the whole word is sent to the M-Word level phonology. Thiswould output the correct participial form in (43), viz. ["xRa.n@l]. However,this assumption also causes a critical complication: recall, from our dis-cussion in section 3.4, that the /-i/ used in the participles should betreated as the same exponent that also codes masculine plural in nouns andadjectives (in the nominative case). The evidence that we supplied for this in3.4 was the fact that the whole paradigm of gender and numbers exponentsin participles, as compared to nouns and adjectives, seems to be preciselythe same. If the /-i/ in (43) triggers a π-cycle, then all masculineplural nouns and adjectives should undergo the same cycle of /i/-deletion,but this does not happen. One could stipulate that the used in theparticiples is a different exponent that is accidentally homophonous to theone used in nouns and adjectives, but such a stipulation would be entirelyunconvincing given the paradigm of number-gender exponents mentionedabove and discussed in 3.4.We have encountered a serious problem: the Cyclic level of phonol-ogy cannot derive the deletion of the /-i/, and the M-Word level ofphonology fails to do the same thing. It seems that the distinction betweenthe Cyclic level and M-Word level phonologies over-generates and under-generates at the same time: it either predicts over-application or under-application of /i/-deletion. However, this problem is not unique to thephonological grammar used by Embick (2010, 2013). Cophonology Theory(Inkelas & Zoll 2007; Inkelas 2008, 2011) suffers from a similar problem.Observe the derivation from (43) transformed into a representation such aswould be used in Cophonology Theory:1424.5. Proposal: Domains of phonological Spell-Outπ(√xRan,-∅,-i)√xRan -∅ -ipi-l-iFigure 4.8: Deriving /"xRan-i-l-i/→["xRa.n@l] in Cophonology TheoryRecall, from section 4.2, that Cophonology Theory subscribes to the StemScope principle, which dictates that the cophonology triggered by a suffix canonly affect whatever the suffix scopes over. In Figure (4.8) above, the theme/-i/, which is responsible for triggering the /i/-deleting π-cycle, only scopesover the root and the zero verbalizing suffix, but not over the final /-i/.We could attempt to endow the /-i/ with another π-diacritic, but thiswould run into the same problems as Embick’s system discussed above: thissystem also makes /i/-deletion either under-apply or over-apply.The problem here is that the exponent /-i/ can attach to differentderivational bases and it seems to be influenced by the phonology to whichthe respective derivational base subscribes to: if /-i/ attaches to a participialbase (i.e. [ptc]+/-i/), it will be processed by the /i/-deleting cophonology,but if it attaches to an nominal or adjectival base (i.e. [adj]+/-i/), it willbe processed by the /i/-preserving cophonology. It is not clear how thiscould be derived in the phonological grammar that Embick (2010, 2013)subscribes to: the major problem with an approach like this is the overlyrigid way in which phonological domains are defined. Embick’s system dele-gates construction-specific phonological effects to the Cyclic level of phonol-ogy, while this feeds the M-Word level of phonology. As we discussed above,this rigid distinction between the fixed Cyclic and M-Word levels of phonol-ogy either causes under-application of over-application of /i/-deletion. Inthe following section, we will propose that phonological domains should bedefined on a purely ‘construction-specific’ basis, specifically in a way thatdirectly follows syntactic phases.4.5 Proposal: Domains of phonological Spell-OutIn the previous section, we explained that the phonological grammar that isused by Embick (2010, 2013) cannot derive the construction-specific effectof /i/-deletion in NM Slovenian. We also explained that the problem chieflystems from the rigid definition of phonological domains which that view ofphonology assumes: the rigid distinction between the fixed levels of Cyclic1434.5. Proposal: Domains of phonological Spell-Outphonology and M-Word phonology is problematic. In this section, we willfirst explain that the domains of phonological spell-out should be set on aconstruction-specific basis, specifically in a way that directly follows fromphasal spell-out of syntax. In addition, we will also discuss a brief andtentative formal solution that will successfully derive the application of /i/-deletion in NM Slovenian.To explain why the exponent /-i/ undergoes deletion in the par-ticiples but not in nouns in adjectives we will subscribe to the general re-search program set out by Marvin (2002), Marantz (2007), Samuels (2009),Piggott & Travis (2013) and Newell & Piggott (2014), who argue that a syn-tactic phase-cycle also forms a domain for phonological computation. Sincethe fixed distinction between Cyclic and M-Word phonology levels, as advo-cated by Embick (2010, 2013), cannot predict the domain of /i/-deletion inNM Slovenian, we should investigate whether a phase cycle is the domain inwhich construction-specific phonological effects are processed. Recall that aparticiple contains only one phase head, the verbalizer v0, which means thatverbs and participles are computed in a single phase-cycle. Simple nounsand adjectives also contain a single phase-head, the nominalizer n0 and ad-jectivizer a0, respectively, also forming a single phase. The generalizationabout the domain of /i/-deletion in NM Slovenian can be captured verysimply by referring to phase-cycles as follows:(44) i-deletion generalization/i/-deletion occurs in verbal phases, but not in nominal or adjectivalphases.A phase domain encompasses the entire word in NM Slovenian participles,which means that the word-final /-i/ is also a part of this phase. If/i/-deletion targets the phase as a phonological domain, then we directlypredict that the /-i/ will delete in such phases. Since /i/-deletiontakes place in verbs and participles in NM Slovenian, this means that itis contained to the verbal phase – no such deletion occurs in nominal oradjectival phases. Notice that by assuming a phase-cycle as the domainfor /i/-deletion, we have also explained why /i/-deletion seems to operate inverbs and participles: participles are constructions that are derived with thesyntactic head Ptc0 attaching to the verbal stem; Ptc0 is a functional andnot a categorial head, which also means that Ptc0 cannot be a phase-headand so must still be contained in the verbal phase.The generalization in (44) is supported by the nouns and adjectives thatwe discussed previously, as they all show the attachment of the /-i/ and1444.5. Proposal: Domains of phonological Spell-Outit does not undergo deletion. However, even better data can be found in NMSlovenian: participial stems may be used to form de-participial adjectiveswith the adjectivizer /-n/. Consider the following datum with √xRan ‘feed’:input output/√xRan-i-l-i/ "xRan@l (√xRan-i-l-n-i/ xRa"nilni ( 4.8: De-participial adjectiveAbove, the /-i/ deletes when a participle is constructed, but if theparticipial stem is used to construct a de-participial adjective with /-n/, /-i/ never deletes. This is directly expected given the generalization in(44), since the /-i/ is in a different phase in the de-participial adjective:with a de-participial adjective, such as [[[[[[√feed] v0] Asp0] Ptc0] a0] #0],it is in the adjectival phase, but with pure participles it is in the verbalphase. If the constructions above were processed by the same phonologythat processes verbal and participial constructions, then we would expectthe output to be *[xRa"nil@n], and not [xRa"nilni].The work on phase-cycles as phonological domains (Marvin 2002; Marantz2007; Samuels 2009; Piggott & Travis 2013; Newell & Piggott 2014) looksfor instances of cyclicity in phonology and attempts to determine if these in-stances can correspond to phase boundaries. The contribution of the presentthesis is in the evidence, presented above, which indicates that phase-cyclesseem to play a crucial role in determining the domains for construction-specific phonological processes. This seems to be the case with /i/-deletionin NM Slovenian which applies in verbal phases only.If a phase-cycle is the domain in which construction-specific phonologyis processed, this creates some friction with the fixed levels of phonology as-sumed by Embick (2010, 2013) who follows Halle & Vergnaud (1987): underthat view, any construction-specific phonology is tied to the Cyclic level and,after that, whole words are processed on the M-Word level. Since the ver-bal phase is a complete word in NM Slovenian, undergoing a construction-specific phonological effect, it seems that the distinction between Cyclicphonology and M-Word phonology should be done away with. It seems moreappropriate to think of the phase, and not the word, as a ‘central phono-logical domain’, within which π-cycles may be triggered. In what follows,we will subsume this idea that the phase is a central phonological domainin order to present a possible, tentative solution to the issue of the domainof /i/-deletion. However, it should be noted we will not explore or discussthe predictions that such an assumption may have for the wider organiza-tion within the phonological grammar in terms of its levels (e.g. Phrasal1454.5. Proposal: Domains of phonological Spell-Outlevel vs. other levels). While this issue is of substantial importance, lengthlimitations on this thesis prohibit us from exploring it further.If the phase is the ‘central phonological domain’, we still need to explainhow a π-cycle that is triggered by an exponent of Asp0 is able to deter-mine the cophonology within which the /-i/ is processed. Recall thelinearized participle and its exponents:(45) a. Step # 1: VI inserts /xRan/ for √feedb. Step # 2: VI inserts /∅/ for v0c. Step # 3: VI inserts /i/pi for Asp0d. Step # 4: VI inserts /l/ for Ptc0e. Step # 5: VI inserts /i/ for Agr0[]As soon as /i/pi is inserted for Asp0, a π-cycle is triggered on /√xRan-∅-i/. After that, Vocabulary Insertion proceeds to insert /-l/ and /-i/. Sincewe are assuming that there is no M-Word phonology, we can claim thatthe phonology simply ‘cycles out’ on these remaining exponents, using thecophonology π previously set in the phonological grammar by the diacriticπ. Nothing is shipped off to a separate M-Word level phonology as in Em-bick’s system, but the computation is simply continued to the end of thephase. This stems from the idea that the phase sets the crucial domain forphonological computation and not the morphological word.To formally implement the idea that the phonological grammar G may‘cycle out’ on the exponents /-l/ and /-i/, we need to be very explicit abouthow a cophonology is set in G. Let us assume that G has a buffer that storesthe relevant π, which is essentially a list of ranked constraints:(46) Buffer of Ga. Phonological grammar G has a buffer which stores instructionsfor ranking constraints π (a cophonology): G{pi}b. If an exponent specifies π1, then G{pi1}c. If no exponent specifies π, then G{pi0} (use default cophonology)The crucial idea behind this formalization is that a cophonology needs to becontained in the buffer of the grammar, but if no cophonology is specified,the grammar just uses the cophonology that is treated as default. In NMSlovenian, this would be the /i/-preserving cophonology because it is morewide-spread through the grammar than the /i/-deleting one.It now becomes crucial to set a locality constraint on the buffer. We willassume that the buffer retains the cophonology until it either encounters a1464.5. Proposal: Domains of phonological Spell-Outnew π-diacritic, or when it encounters the end of the phase-cycle. We canassume that the end of the phase-cycle the buffer needs to ‘reset’:(47) Buffer localityThe buffer will retain πn introduced by an exponent until:a. an exponent specifies πmb. the end of the phase-cycle (where the buffer resets)That the buffer can be overridden by a new cophonology is a standard as-sumption on how cyclicity works either on the Cyclic level of phonology inthe system used by Embick (2010, 2013), or in Cophonology Theory (Orgun1996; Inkelas & Zoll 2007; Inkelas 2008, 2011); here it is just stated for-mally. The other locality constraint on the buffer is essentially that of aphase-boundary: the idea that the buffer has to reset with each phase is aformal implementation of phases interrupting construction-specific phono-logical processes. The locality constraint in (47), therefore, predicts that thebuffer can only be tampered with in one of the two cases just discussed (a.or b.). In the NM Slovenian participles, as shown in (45), the buffer will beset for an /i/-deleting cophonology, which is triggered by πi on the Asp0 ex-ponent /-i/: after the πi-cycle has been run on /√xRan-∅-i/, the /i/-deletingcophonology πi should be retained in the buffer, as no new π is introduced,and the end of the phase has not been reached. This should ensure that the/i/-deleting cophonology will affect the /-i/. The locality constraintin (47) becomes a necessity as soon as we assume that there is no M-Wordlevel phonology which the whole participle could be shipped off to – thegrammar G needs to literally ‘cycle out’ on the yet unprocessed exponentswith whatever it has in the buffer. The locality constraint in (47) basicallyonly requires the buffer to be tampered with when absolutely necessary –only when one of the two locality boundaries is met.50Let us now derive the process of /i/-deletion or the lack of it in NM Slove-nian, by assuming that the cophonologies are stored in the proposed buffer.We need to assume that NM Slovenian has two cophonologies: the defaultcophonology π0 is the /i/-preserving one, whereas π1 is the construction-specific cophonology that triggers /i/-deletion. Let us first construct therelevant derivations with place-holders standing in for the real syntacticheads and exponents, before we move to NM Slovenian, to explain how thebuffer we defined above works as a model:50Future inquiry should determine the validity of this analysis; if it proves to be on theright track, this requirement could be derived from principles of computational efficiency,which natural human language follows in some way or another (Chomsky 2001, 2005).1474.5. Proposal: Domains of phonological Spell-Out⟨√root0, x0, y0, z0, w0 ⟩/√α β γ δ ǫ /G{pi0} output: [αβγδǫ]Figure 4.9: Single phase cycle with no π-cyclesIn Figure (4.9) above, the concatenated heads [√root0⊕x0, x0⊕y0, y0⊕z0,z0⊕w0] are shown, which are represented as a list of n-tuples. These headsare all contained within one phase-cycle. When Vocabulary Insertion insertsthe exponents /α/, /β/, /γ/, /δ/ and /ǫ/ for the list of syntactic heads, allthese exponents are processed in one phonological cycle. Notice that no π-diacritic is specified on any of the exponents, which is why the buffer of thegrammar G sets π0, the default cophonology for phonological computation.Now consider Figure (4.10):⟨√root0, x0, y0, z0, w0 ⟩/√α β γpi1 /G{pi1}1st cycle[αβγ]/αβγ – δ ǫ /G{pi1}2nd cycle output: [αβγδǫ]Figure 4.10: Single phase cycle with a π-cycle (π1)1484.5. Proposal: Domains of phonological Spell-OutHere, the list of syntactic heads again constitutes a phase-cycle. However,when Vocabulary Insertion inserts the exponent for y0, the exponent comesspecified for a π-cycle. As soon as the exponent /γ/pi1 is inserted, a cycleis run on /√αβγ/, which is processed by G{pi1}, with cophonology π1 set inthe buffer. After this cycle has produced an output, Vocabulary Insertioncontinues with insertion, producing the exponents /δ/ and /ǫ/. After that,these two exponents need to be processed phonologically, which is why thegrammar G{pi1} simply cycles out on these remaining exponents, retaining π1in the buffer because no other π is specified on these exponents and becauseno intermediate phase-boundary is encountered.Let us now finally turn to NM Slovenian. The following example illus-trates the derivation of the participle ["√xRan-@l] ‘having fed (’ from/√xRan-i-l-i/. All the syntactic heads are contained in the verbal phase here,as noted before, and the Asp0 theme exponent /-i/ will bear π1:⟨√feed0, v0, Asp0, Ptc0, Agr0⟩/√xRan ∅ ipi1 /G{pi1}1st cycle["xRan]/"xRan – l i /G{pi1}2nd cycle output: ["xRan@l]Figure 4.11: Deriving the participle ["√xRan-@l] ‘having fed’The diacritic π1 specified on the Asp0 exponent /-i/ triggers a π-cycle andsets π1 in the buffer; this means that the π-cycle is processed by the /i/-deleting cophonology. After the π-cycle has produced an output, the remain-ing exponents, viz. /-l/ and /-i/, are inserted. Now the grammar needs tocycle out on these to process them phonologically. Since these two exponents1494.5. Proposal: Domains of phonological Spell-Outbear no π and they are contained in the same phase-cycle as the previouslyinserted exponents, the grammar retains π1 in the buffer for their phono-logical processing. This is how the word-final /-i/ is deleted. Withnouns and adjectives, we expect G to set π0 in its buffer since no π-diacriticseem to be specified on any of the exponents there; this means that simplenouns and adjectives listed in Table (4.7) will be processed according tothe schema outlined in Figure (4.9): that schema implies that whole nounsand adjectives are processed in a single phonological cycle with the default,/i/-preserving cophonology π0 set in the grammar’s buffer.The approach which we have proposed here introduces the idea that thephonological grammar has a ‘cophonology’ buffer which obeys phasal local-ity. This approach explains the π-specification of exponents as the source ofconstruction-specific phonological effects, and their boundary is delineatedby the end of a phase. This explains construction-specific phonological ef-fects as an arbitrary property stipulated lexically on the exponents, whichis essentially how Embick (2013) and Halle & Vergnaud (1987) treat them,and this is also how Cophonology Theory (Inkelas & Zoll 2007; Inkelas 2008,2011) treats them. Recall that, in section 4.3, we explained that only someexponents of Asp0 need to trigger a π-cycle: these were the theme /-i/, thesemelfactive /-ni/ and the zero theme /-∅1/, whereas the rest could po-tentially trigger them or not (except for the second zero theme /-∅2/ usedsolely with √b- ‘be’ and √S- ‘go’ formations). The analysis involving thegrammar’s buffer and phasal locality actually forces us to assume that allthe other exponents of Asp0 (again, except /-∅2/) trigger a π-cycle. This isbecause /i/-deletion needs to be triggered by introducing π1, the /i/-deletingcophonology, through a diacritic. Crucially, this diacritic needs to ‘persist’into the next cycle of phonological computation so that the /-i/ willdelete when the grammar ‘cycles out’ on it. However, this is again not muchdifferent from approaches such as that advanced by Cophonology Theory,where construction-specific phonological effects must be stipulated lexicallyon some exponents in some manner.This section has discussed a crucial solution that explain how the /i/-deleting cophonology can target the word-final /-i/ in the participlesof NM Slovenian. The proposal that we have developed involves doing awaywith the rigid distinction between Cyclic phonology and M-Word phonol-ogy that Embick (2010, 2013) employs. This is replaced by a more directreference to phases in phonological computation, and we have also intro-duced the notion of a ‘buffer’ that the phonological grammar operates with.The positive theoretical aspect of this approach is that it attempts to de-rive phonological domains as direct reflections of syntactic cycles (phases).1504.5. Proposal: Domains of phonological Spell-OutThis approach is also empirically superior in that it manages to derive theconstruction-specific /i/-deletion process in NM Slovenian where other ap-proaches fail. However, a crucial caveat is in order: this approach wasdeveloped to account for the /i/-deletion process in NM Slovenian in a waythat follows from (some) independent properties of the grammar, but thisis nevertheless a first, tentative attempt at deriving processes like this in aformally tractable fashion. In the future, this approach will certainly need tobe modified for more adequate empirical coverage, if not for theory-internalconsiderations. With that in mind, it should be treated as a platform forfuture discussion of such phenomena and by no means a final proposal.151Chapter 5ConclusionThe present thesis has presented data from Novo mesto Slovenian, a SouthSlavic language, which have revealed a process of unstressed /i/-deletion.The interesting aspect about this process is that it appears to be construction-specific: /i/-deletion is found in verbs and participles, but not in nouns oradjectives, where unstressed /i/ is fully preserved on the surface.In chapter 3, we proposed an analysis of /i/-deletion in a way that inter-twines it with schwa epenthesis, which also appears to be operative in Novomesto Slovenian. We encountered examples where schwa-epenthesis over-applied on the surface, for which we had posit two cycles of phonologicalcomputation. It should be noted that this step proposes a specific accountof schwa epenthesis in verbs and participles in Novo mesto Slovenian, and assuch it will be relevant for the examination of other neighbouring dialects.However, a particularly curious aspect of the cyclic analysis should be em-phasized: in section 3.2 in (31), we were forced to propose the constraintMax-F(µ), which is not a typical Max-constraint, as it refers to input andoutput moras (within the foot) which do not need to be in a correspon-dence relation. This was a necessary step to derive the generalization thatthe masculine plural /-i/ surfaces with a ["CVC]-configuration followed bya sonorant (see 3.2 for details). Since the generalization itself is sound, fu-ture work should consider different formal implementations of it, perhapseven reducing it down to the interaction of several constraints. The variousdifferent formal implementations were not examined further here becausethe over-reaching focus of this thesis is elsewhere, and also due to lengthlimitations.In chapter 4, we turned back to the process of /i/-deletion and studiedit from the perspective of the PF-interface. Recall that the masculine plu-ral agreement suffix /-i/, may attach to a participial stem, as in [ptc]-/i/,and under specific phonotactic conditions undergo deletion. But when thissame suffix attaches to an adjectival or nominal stem, as in [adj]-/i/, it ispreserved under the same phonotactic conditions in which it deletes withparticiples. We demonstrated that this is presents an empirical challengefor frameworks such as Cophonology Theory (Inkelas & Zoll 2007; Inkelas152Chapter 5. Conclusion2008, 2011), where the final attaching suffix, in our case /-i/, cannot ‘inherit’the phonology specified in the derivational history of the stem. For ap-proaches such as that of Embick (2013), who follows the phonological levels(Cyclic/Word/Phrasal level) of Halle & Vergnaud (1987), the phenomenonwas also problematic: such an approach either under-generates by position-ing /i/-deletion in the stem (on the Cyclic level), where it cannot affect thesuffix /-i/, or it over-generates by positioning /i/-deletion on the Word levelphonology, where it affects the entire grammar (i.e. all word classes). In4.4, we came to the conclusion that this problem is rooted in the way suchapproaches define phonological domains. The fixed Cyclic vs. Word leveldistinction is too rigid and too restrictive. In section 4.5, we subscribed tothe general research program set by Marvin (2002), Marantz (2007), Samuels(2009), Piggott & Travis (2013) and Newell & Piggott (2014), which seeksto uncover reflections of syntactic phases (Chomsky 2001, 2008) in phonol-ogy. We noted that an adequate generalization about the /i/-deletion factscan be made if it is assumed that phonological domains are generally set byphases and not by fixed phonological levels. The important contribution ofsuch an analysis is the observation that phases also form a locality boundaryfor construction-specific phonological processes.In the remainder of section 4.5, we formulated a brief and tentativeproposal that formally implements the idea that phonological computationproceeds in phases. We proposed that the ‘instructions’ for phonologicalprocessing are stored in a buffer of the phonological grammar, which obeysphasal locality: more specifically, in Novo mesto Slovenian verbs and partici-ples, an exponent in the stem triggers /i/-deletion, sending the instructionsfor an /i/-deleting phonology to the phonological buffer. This means thatthe stem is computed first, to the exclusion of the participial /-l/ and mascu-line plural /-i/. But after the stem has been computed, the /-i/ is added tothe stem and the grammar ‘cycles out’ on whatever unprocessed exponentsare left in the phase. When doing so, it uses the same phonology in thebuffer that was specified by the stem because no phase boundary intervenesto ‘reset’ the buffer. This is a first attempt at deriving construction-specificprocesses like that of /i/-deletion. Future research should fully flesh out thepredictions it makes, and in particular examine how it could be made towork with multi-phasal spell-out. If this proposal is on the right track, itwould provide an important argument for a modularly distinct, but unifiedtheory of grammar, where the locality constraints on syntactic computationare witnessed ‘all the way down’.As an additional note, the data and analysis presented in this thesisshould be of interest for a diachronic inquiry. It is a well known fact that153Chapter 5. ConclusionSlovenian dialects had undergone a more general historical process of high-vowel reduction and deletion (Lenček 1982; Greenberg 2000), which musthave resulted in the system found in Novo mesto Slovenian, where /i/-deletion survived into the synchronic grammar, but in a limited fashion,reduced to word-final positions (rendering it apocopal) in only verbs andparticiples. The diachronic path of development that led to this situationshould be of substantial interest to theories of phonology and morphology.154BibliographyAnderson, S. (1982). Where’s Morphology? Linguistic Inquiry, 13, 571–621.Aronoff, M. (1976). Word Formation in Generative Grammar. MIT Press.Arregi, K. & Nevins, A. (2012). Morphotactics: Basque Auxiliaries and theStructure of Spellout. Studies in Natrual Language and Linguistic Theory86. New York: Springer.Becker, M. & Jurgec, P. (2015). Interaction of tone and ATR in Slovenian.In Kehrein, W., Köhnlein, B., Boersma, P., & van Oostendorp, M. (Eds.),Segmental structure and tone, Berlin. de Gruyter.Beckman, J. N. (1998). Positional Faithfulness. PhD thesis, University ofMassachusetts Amherst, Amherst, MA.Bermúdez-Otero, R. (2011). Cyclicity. In van Oostendorp, M., Ewen, C.,Hume, E., & Rice, K. (Eds.), The Blackwell companion to phonology, (pp.2019–2048)., Malden, MA. Wiley-Blackwell.Biskup, P. (2012). Slavic prefixes and adjectival participles. In Ziková, M. &Dočekal, M. (Eds.), Slavic Languages in Formal Grammar: Proceedingsof FDSL 8.5, (pp. 271–289)., Frankfurt am Main. Peter Lang.Blake, S. (2000). On the distribution and representation of schwa inSliammon Salish: descriptive and theoretical perspectives. PhD thesis,University of British Columbia, Vancouver, BC.Bonet, E. & Harbour, D. (2012). Contextual allomorphy. In Trommer,J. (Ed.), The morphology and phonology of exponence, (pp. 195–235).,Oxford. Oxford University Press.Booij, G. (1995). The Phonology of Dutch. Oxford: Oxford University Press.Bye, P. & Svenonius, P. (2012). Non-concatenative morphology as epiphe-nomenon. In Trommer, J. (Ed.), The morphology and phonology of expo-nence, (pp. 427–495)., Oxford. Oxford University Press.Chociej, J. (2009). Pattern in Polish Vowel-Zero Alternations: An Opti-mality Theory Analysis. Master’s thesis, University of Toronto, Toronto,ON.Chomsky, N. (1993). A minimalist program for linguistic theory. In TheView from Building 20, MIT Working Papers in Linguistics, (pp. 1–52).,Cambridge, MA. MIT Press.155BibliographyChomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press.Chomsky, N. (2000). Minimalist inquiries: The framework. In Martin, R.,Michaels, D., & Uriagereka, J. (Eds.), Step by Step: Minimalist Essays inHonor of Howard Lasnik, (pp. 89–155)., Cambridge, MA. MIT Press.Chomsky, N. (2001). Derivation by Phase. In Kenstowicz, M. (Ed.), KenHale. A Life in Language, (pp. 1–52)., Cambridge, MA. MIT Press.Chomsky, N. (2005). Three factors in language design. Linguistic Inquiry,36 (1), 1–22.Chomsky, N. (2008). On phases. In Freidin, R., Otero, C. P., & Zubizarreta,M. L. (Eds.), Foundational issues in linguistics theory: essays in honor ofJean-Roger Vergnaud, (pp. 133–166)., Cambridge, MA. MIT Press.Chomsky, N. & Halle, M. (1968). The Sound Pattern of English. New York:Harper and Row.Chtatou, M. (1982). Aspects of the Phonology of a Berber Dialect of the Rif.PhD thesis, SOAS, London.Clements, G. N. (1990). The role of the sonority cycle in core syllabification.In Kingston, J. & Beckman, M. E. (Eds.), Papers in laboratory phonologyI: Between the grammar and physics of speech, (pp. 283–333)., Cambridge.Cambridge University Press.Coetzee, A. W. (2002). OO-Correspondence as Cumulativity. University ofMassachusetts Occasional Papers, 26, 27–57.Coetzee, A. W. (2004). What it means to be a loser: Non optimal candidatesin Optimality Theory. PhD thesis, University of Massachusetts Amherst,Amherst, MA.Comrie, B. (1976). Aspect. Cambridge: Cambridge University Press.Crosswhite, K. M. (1999). Vowel Reduction in Optimality Theory. PhDthesis, University of California Los Angeles, LA, CA.Dell, F. & Tangi, O. (1992). Syllabification in Ath-Sidhar Rifian Berber.Journal of African Languages and Linguistics, 13, 125–162.Dickey, S. M. (2003). Verbal aspect in Slovene. In Stolz, T. (Ed.), STUF -Language Typology and Universals, volume 56, (pp. 182–207).Elenbaas, N. & Kager, R. (1999). Ternary rhythm and the Lapse constraint.Phonology, 16, 273–330.Embick, D. (2003). Locality, Listendess and Morphological Iidentity. StudiaLinguistica, 57 (3), 143–169.Embick, D. (2007). Linearization and local dislocation: Derivational me-chanics and interactions. Linguistic Analysis, 33, 303–336.Embick, D. (2008). Variation and morphosyntactic theory: Competitionfractionated. Language and Linguistic Compass, 2 (1), 59–78.Embick, D. (2010). Localism versus globalism in morphology and phonology.156BibliographyCambridge, MA: MIT Press.Embick, D. (2013). Phase cyclces, φ-cycles, and phono-logical (in)activity. MS. University of Pennsylvania.∼embick/ph-cycle-sept.pdf.Embick, D. & Halle, M. (2005). On the status of stems in morphologicaltheory. In Geerts, T., van Ginneken, I., & Jacobs, H. (Eds.), Romancelanguages and linguistic theory 2003, (pp. 59–88)., Amsterdam. John Ben-jamins.Embick, D. & Noyer, R. (2001). Movement operations after syntax. Lin-guistic Inquiry, 32 (4), 555–595.Gouskova, M. (2012). Unexceptional segments. Natural Language and Lin-guistic Theory, (30), 79–133.Greenberg, M. L. (2000). A Historical Phonology of the Slovene Language.Heidelberg: Carl Winters Universitätsverlag.Greenberg, M. L. (2006). A Short Grammar of Standard Slovene. Universityof Kansas.Gribanova, V. (2015). Exponence and morphosyntacticallytriggered phonological processes in the Russian verbal com-plex. Journal of Linguistics, Available through FirstView., V. & Harizanov, B. (2015). Locality and directionalityin inward-sensitive allomorphy: Russian and Bulgarian. To appearin The Morphosyntax-Phonology Connection, Oxford University Press.∼gribanov/downloads/Gribanova HarizanovMSPI.pdf.Guerssel, M. (1976). Issues in berber phonology. Master’s thesis, Universityof Washington, Seattle, WA.Gussman, E. (2007). The Phonology of Polish. Oxford: Oxford UniversityPress.Hall, N. (2000). Max-Position Derives Iterative Footing. In Megerdoo-mian, K. & Barel, L. A. (Eds.), WCCFL 20 Proceedings, number 101-114,Somerville, MA. Cascadilla Press.Halle, M. (1973). Prolegomena to a Theory of Word Formation. LinguisticInquiry, 4 (3-16).Halle, M. (1997). Impoverishment and fission. PF: Papers at the interface:MIT Working Papers in Linguistics, 425–250.Halle, M. & Marantz, A. (1993). Distributed morphology and the piecesof inflection. In The View from Building 20, MIT Working Papers inLinguistics, (pp. 111–176)., Cambridge, MA. MIT Press.Halle, M. & Marantz, A. (1994). Some key features of distributed morphol-157Bibliographyogy. MIT Working Papers in Linguistics 21: Papers on phonology andmorphology, 21, 257–288.Halle, M. & Vergnaud, J.-R. (1987). An Essay on Stress. Cambridge, MA:MIT Press.Harley, H. (2014). On the identity of roots. Theoretical Linguistics, 40,225–276.Hayes, B. (1989). Compensatory Lengthening in Moraic Phonology. Lin-guistic Inquiry, 20 (2), 253–306.Hayes, B. (1995). Metrical Stress Theory: Principles and Case Studies.Chicago: The University of Chicago Press.Herrity, P. (2000). Slovene: A Comprehensive Grammar. London, NewYork: Routledge.Hyde, B. (2001). Metrical and Prosodic Structure in Optimality Theory. PhDthesis, Rutgers, The State University of New Jersey, New Brunswick, NewJersey.Hyde, B. (2002). A restrictive theory of metrical stress. Phonology, (19),313–359.Inkelas, S. (2008). The morphology-phonology connec-tion. Talk given at the 34th annual meeting of theBerkeley Linguistics Society. Berkeley, CA. Available at:∼inkelas/Papers/BLS2008Inkelas.pdf.Inkelas, S. (2011). The interaction between morphology and phonology. InGoldsmith, J., Riggle, J., & Yu, A. C. L. (Eds.), Handbook of phonologicaltheory: 2nd edition, (pp. 68–102)., Oxford. Wiley-Blackwell.Inkelas, S., Orgun, O. C., & Zoll, C. (1997). The implications of lexicalexceptions for the nature of grammar. In Roca, I. (Ed.), Derivations andconstraints in phonology, (pp. 393–418)., Oxford. Clarendon Press.Inkelas, S. & Zoll, C. (2007). Is grammar dependence real? A comparisonbetween cophonological and indexed constraint approaches to morpholog-ically conitioned phonology. Linguistics, 45, 133–172.Itô, J., Kitagawa, Y., & Mester, A. (1996). Prosodic Faithfulness and Cor-respondence: Evidence from a Japanese Argot. Journal of East AsianLinguistics, 5, 217–294.Itô, J. & Mester, A. (1995). Japanese phonology. In Goldsmith, J. A. (Ed.),The Handbook of phonological theory, (pp. 817–838)., Oxford. Blackwell.Itô, J. & Mester, A. (1999). The phonological lexicon. In Tsujimura, N.(Ed.), The handbook of Japanese linguistics, (pp. 62–100)., Oxford. Black-well.Jarosz, G. (2008). Partial ranking and alternating vowels in Polish. InProceedings of CLS 43, volume 41, (pp. 193–206)., Chicago. Chicago Lin-158Bibliographyguistics Society.Jurgec, P. (2007a). Novejše besedje s stališča fonologije: Primer slovenščine.PhD thesis, University of Ljubljana, Ljubljana.Jurgec, P. (2007b). Schwa in slovenian is epenthetic. Handout from SecondAnnual Meeting of the Slavic Linguistics Society. Berlin, August 24th.Jurgec, P. (2011). Slovenščina ima 9 samoglasnikov. Slavistična revija, (59),149–161.Kager, R. (1995). Metrical Stress Theory: Principles and Case Studies byBruce Hayes. Phonology, 12 (3), 437–464.Kager, R. (1999). Optimality Theory. Cambridge. Cambridge UniversityPress.Kang, E. (2004). Edge-Demarcation in Phonology. PhD thesis, CornellUniversity.Kennedy, C. (1997). Projecting the Adjective. PhD thesis, University ofCalifornia Santa Cruz, Santa Cruz, CA.Kennedy, C. & Levin, B. (2008). Measure of change: The adjectival coreof degree achievements. In McNally, L. & Kennedy, C. (Eds.), Adjectivesand Adverbs: Syntax, Semantics and Discourse, (pp. 156–182)., Oxford.Oxford University Press.Kenstowicz, M. & Rubach, J. (1987). The phonology of syllabic nuclei inSlovak. Language, 63 (3), 463–497.Kiparsky, P. (1982a). From cyclic phonology to lexical phonology. In van derHulst, H. & Smith, N. (Eds.), The structure of phonological representa-tions, volume Vol. 1, (pp. 131–175)., Dordrecht. Foris.Kiparsky, P. (1982b). Lexical morphology and phonology. In Yang, I.-S.(Ed.), Linguistics in the morning calm, (pp. 3–91)., Seoul. Hanshin.Kiparsky, P. (2000). Opacity and cyclicity. The Linguistics Review, 17,351–367.Kroch, A. (1989). Reflexes of grammar in patterns of language change.Language Variation and Change, 1, 199–244.Lenček, R. L. (1982). The Structure and History of the Slovene Language.Columbus, OH: Slavica.Lieber, R. (1980). On the Organization of the Lexicon. PhD thesis, MIT,Cambridge, MA.Lieber, R. (1992). Deconstructing morphology. Chicago: University ofChicago Press.Logar, T. (1993). Slovenska narečja. Ljubljana: Založba Mladinska Knjiga.MacBride, A. (2000). Phonological and morphological issues in the distribu-tion of berber schwa. Master’s thesis, University of California Los Angeles,LA, CA.159BibliographyManova, S. (2011). Understanding Morphological Rules: With Special Em-phasis on Conversion and Subtraction in Bulgarian, Russian and Serbo-Croatian. Dordrecht: Springer.Marantz, A. (1984). On the Nature of Grammatical Relations. Cambridge,MA: MIT Press.Marantz, A. (1988). Clitics, morphological merger, and the mapping tophonological structure. In Hammond, M. & Noonan, M. (Eds.), Theoret-ical Morphology, (pp. 253–270)., San Diego. Academic Press.Marantz, A. (1997). No escape from syntax: Don’t try morphological anal-ysis in the privacy of your own lexicon. In University of PennsylvaniaWorking Papers in Linguistics 4(2), Proceedings of the 21st Annual PennLinguistics Colloquium, (pp. 201–225).Marantz, A. (2001). Phases and Words. In he Choe, S. (Ed.), Phases in theTheory of Grammar, (pp. 196–226)., Seoul. Dong In.Marantz, A. (2007). Phases and words. In Choe, S.-H. (Ed.), Phases in theTheory of Grammar, (pp. 191–222)., Seoul. Dong In.Marantz, A. (2013). Locality domains for contextual allomorphy across theinterfaces. In Matushansky, O. & Marantz, A. (Eds.), Distributed Mor-phology Today: Morphemes for Morris Halle, (pp. 95–116)., Cambridge,MA. MIT Press.Marvin, T. (2002). Topics in the Stress and Syntax of Words. PhD thesis,MIT, Cambridge, MA.McCarthy, J. J. & Prince, A. (1993). Generalized alignment. In Booij,G. & van Marle, J. (Eds.), Yearbook of Morphology 1993, (pp. 43–67).,Dordrecht. Kluwer.McCarthy, J. J. & Prince, A. (1995). Faithfulness and reduplicative identity.In Beckman, J., Dickey, L. W., & Urbanczyk, S. (Eds.), University ofPennsylvania Working Papers in Linguistics 18, (pp. 249–384)., Amherst,MA. GLSA Publications.McCarthy, J. J. & Prince, A. (1999). Faithfulness and identity in ProsodicMorphology. In Kager, R., van der Hulst, H., & Zonneveld, W. (Eds.), TheProsody-Morphology Interface, (pp. 218–309)., Cambridge. CambridgeUniversity Press.Mellander, E. (1999). Teasing apart synchronic yers and epenthetic vowelsin czech and slovak. In Herk, G. V. (Ed.), Proceedings of the 1999 An-nual Conference of the Canadian Linguistics Association, (pp. 213–224).,Ottawa, Canada. University of Ottawa.Merchant, J. (2015). How much context is enough? Two cases of span-conditioned stem allomorphy. Linguistic Inquiry, 46 (2), 273–303.Mohanan, K. P. (1986). The theory of lexical phonology. Reidel: Dordrecht.160BibliographyMontague, R. (1970). Universal grammar. Theoria 36, 373–398.Nevins, A. & Parrot, J. K. (2010). Variable rules meet impoverishmenttheory: Patterns of agreement leveling in english varieties. Lingua, 120 (5),1135–1159.Newell, H. & Piggott, G. (2014). Interactions at the syntax-phonology in-terface: Evidence from Ojibwe. Lingua, 150, 332–362.Oltra-Massuet, M. I. (1999). On the notion of theme vowel: A new approachto catalan verbal morphology. Master’s thesis, MIT, Cambridge, MA.Orgun, C. O. (1996). Sign-Based Morphology and Phonology with special at-tention to Optimality Theory. PhD thesis, University of California, Berke-ley, Berkeley, CA.Orie, O. & Pulleyblank, D. (2002). Yoruba vowel elision: Minimality effects.Natural Language and Linguistic Theory, 20 (1), 101–156.Parker, S. (2011). Sonority. In van Oostendorp, M., Ewen, C. J., Hume, E.,& Rice, K. (Eds.), The Blackwell Companion to Phonology, (pp. 1147–1160)., Oxford. Wiley-Blackwell.Pater, J. (2000). Non-uniformity in english secondary stress: the role ofranked lexically specific constraints. Phonology, 17, 237–274.Pater, J. (2007). The locus of exceptionality: Morpheme-specific phonol-ogy as constraint indexation. In Bateman, L., O’Keefe, M., Reilly, E., &Werle, A. (Eds.), University of Massachusetts Occasional Papers in Lin-guistics 32: Papers in Optimality Theory III, (pp. 259–296)., Amherst,MA. GLSA, University of Massachusetts.Pater, J. (2010). Morpheme-specific phonology: Constraint indexation andincosistency resolution. In Parker, S. (Ed.), Phonological Argumentation:Essays on Evidence and Motivation, London. Equinox.Petek, B., Šuštaršič, R., & Komar, S. (1996). An acoustic analysis of con-temporary vowels of the standard Slovenian language. In ProceedingsICSLP 96: Fourth international conference on spoken language process-ing, October 3–6, 1996, Philadelphia, PA, USA, (pp. 133–136). Universityof Delaware.Piggott, G. & Travis, L. (2013). Adjuncts within words and complex heads.In Folli, R., Sevdali, C., & Truswell, R. (Eds.), Syntax and its Limits, (pp.157–74).Pintzuk, S. (1991). Phrase structure in competition: variation and changein Old English word order. PhD thesis, University of Pennsylvania.Prince, A. (1985). Improving Tree Theory. Berkeley Linguistics Society,(11), 471–490.Prince, A. & Smolensky, P. (1993/2004). Optimality Theory: ConstraintInteraction in Generative Grammar. Oxford: Blackwell Publishing.161BibliographyRamovš, F. (1936/1995). Kratka zgodovina slovenskega jezika. Ljubljana:Znanstvenoraziskovalni center SAZU.Rice, C. (2007). The roles of GEN and CON in modeing ternary rhythm.In Blaho, S., Bye, P., & Krämer, M. (Eds.), Freedom of Analysis?, (pp.233–255)., Berlin. Mouton de Gruyter.Rice, K. D. (1992). On Deriving Sonority: A Structural Account of SonorityRelationships. Phonology, 9 (1), 61–99.Rubach, J. (1984). Studies in Generative Grammar: Cyclic and lexicalphonology: the structure of Polish. Dordrecht: Foris Publications.Samuels, B. D. (2009). The Structure of Phonological Theory. PhD thesis,Harvard University, Cambridge, MA.Scheer, T. (2012). Variation is in the lexicon: yer-based and epentheticvowel-zero alternations in Polish. In Cyran, E., Szymanek, B., & Kardela,H. (Eds.), Sound, Structure and Sense: Studies in Memory of EdmundGussmann, (pp. 631–672)., Lublin. Wydawnictvo KUL.Shannon, T. F. (1986). Preference Laws and Syllable Structure in Dutch.Van Oosten & Snapper 1986, 1–26.Shannon, T. F. (1991). On the syllabic motivation of inflectional suffixes inGermanic. In Antonsen, E. H. & Hock, H. H. (Eds.), Studies In GermanicLinguistics : Selected Papers From The 1st and 2nd Symposium On Ger-manic Linguistics,, (pp. 169–183). John Benjamins Publishing Company.Shaw, P. (1996). Headless and weightless syllables. Paper presented atthe 19th Annual Meeting of the Canadian Linguistic Association, BrockUniversity.Skinner, T. R. (2009). Investigations of Downward Movement. PhD thesis,McGill University, Montréal.Slabakova, R. (2001). Telicity in Second Language. Amsterdam. John Ben-jamins Publishing Company.Smolensky, P. (1995). On the structure of the constraint component Con ofUG. ROA 86.Sproat, R. (1985). On Deriving the Lexicon. PhD thesis, MIT, Cambridge,MA.Srebot Rejec, T. (1988). Word accent and vowel duration in StandardSlovene: An acoustic and linguistic investigation, volume Number 226in Slavistische Beiträge. Otto Sagner.Steriade, D. (1982). Greek prosodies and the nature of syllabification. PhDthesis, MIT, Cambridge, MA.Steriade, D. (1994). Licensing by Cue. University of California Los Angeles,MS.Svenonius, P. (2004a). Slavic prefixes and morphology: An introduction to162Bibliographythe Nordlyd volume. Nordlyd, 32 (2), 177–204.Svenonius, P. (2004b). Slavic prefixes inside and outside VP. Nordlyd, 32 (2),205–253.Svenonius, P. (2012). Spanning. MS, CASTL: University of Tromsø., P. (2013). Look both ways: Outward-looking allomor-phy in Icelandic participles. MS, CASTL: University of Tromsø., P. & Bye, P. (2010). Exponence, phonologyand non-concatenative morphology. CASTL, Trømsø. MS.šič, J. (1968). Liki slovenskih tonemov. Slavistična revija, 16, 315–393.Toporišič, J. (1992). Enciklopedija slovenskega jezika. Ljubljana: Cankarjevazaložba.Toporišič, J. (2000). Slovenska slovnica. Maribor: Obzorja.Travis, L. (1984). Parameters and Effects of Word Order Variation. PhDthesis, MIT, Cambridge, MA.Travis, L. (1994). Event phrase and a theory of functional categories. InKoskinen, P. (Ed.), Proceedings of the Canadian Linguistics Association1994, (pp. 559–570)., Toronto. Toronto Working Papers in Linguistics.Travis, L. (2010). Inner Aspect: The Articulation of VP. Dordrecht:Springer.van Oostendorp, M. (2000). Phonological Projection: A Theory of FeatureContent and Prosodic Structure. Berlin: Mouton de Gruyter.van Oostendorp, M. (2004). Crossing morpheme boundaries in Dutch. Lin-gua, (114), 1367–1400.van Oostendorp, M. (2010). Dutch diminutives and the question mark. InRice, C. & Blaho, S. (Eds.), Advances in Optimality Theory: ModellingUngrammaticality in Optimality Theory, (pp. 67–96)., London. EquinoxPublishing.Šuštaršič, R., Komar, S., & Petek, B. (1995). Slovene. Journal of theInternational Phonetic Association, 25, 86–90.Šuštaršič, R., Komar, S., & Petek, B. (1999). Slovene. In Handbook of the In-ternational Phonetic Association, (pp. 135–139)., Cambridge. CambridgeUniversity Press.Žaucer, R. (2002). The Role of Verbal Prefixes in Slavic: Evidence fromSlovenian Locative Denominal Verbs. Master’s thesis, University of Ot-tawa, Ottawa.Wolf, M. (2005). An Autosegmental Theory of Quirky Mutations. In163BibliographyAlderete, J., hye Han, C., & Kochetov, A. (Eds.), Proceedings of the 24thWest Coast Conference on Formal Linguistics, (pp. 370–378)., Somerville,MA. Cascadilla Proceedings Project.Wolf, M. (2007). For an Autosegmental Theory of Mutation. In Bateman, L.,O’Keefe, M., Reilly, E., & Werle, A. (Eds.), University of MassachusettsOccasional Papers in Linguistics 32: Papers in Optimality Theory III,(pp. 315–404)., Amherst, MA. GLSA.Yang, C. (2002). Knowledge and learning in natural language. Oxford, UK:Oxford University Press.164


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items