PHONOTACTICS of CZECH Phd Thesis

Faculty of Arts, Masaryk University Department of Linguistics and Baltic Studies

General Linguistics

Aleš Bičan

PHONOTACTICS OF CZECH PhD Thesis

Supervisor: prof. PhDr. Marie Krčmová, CSc.

Brno 2011

I hereby confirm that the present thesis was worked out independently. All sources are acknowledged in the accompanying list of references.

Brno, July 20, 2011 Aleš Bičan

ii CONTENTS

Symbols vi Acknowledgments vii

Chapter One: Introduction 1 1.1 Outline of the work 1 1.2 Theory and description 3 1.3 Modern Standard Czech 6 1.4 Previous phonotactic descriptions of Czech 8

Chapter Two: Phonematics of Czech 11 2.1 Introduction 11 2.2 Consonants 13 2.3 Interpretation of the affricates 16 2.4 Vowels 21 2.5 Semiconsonants 24 2.6 Neutralization of voicing 25 2.7 Neutralization of place of articulation of nasals 31

Chapter Three: Prolegomena to a phonotactic analysis 32 3.1 Phonotactic entities 32 3.2 Accent and diaereme 34 3.3 Signs and their types 41 3.4 Accidental gaps vs. structural restrictions 45 3.5 “Syllabification” 49

Chapter Four: Distributional unit: Theoretical background 55 4.1 Distributional unit as a domain of distribution 55 4.2 Positions 57 4.3 Functional dependency: nuclear and peripheral entities 62 4.4 Occurrence dependency 64 4.5 Semi-distributional unit 66 4.6 Phonotagms and syllables 67 4.7 Constructing a distributional unit 71

iii Chapter Five: Distributional unit: Application to Czech 77 5.1 Nuclear position 77 5.2 Peripheral positions 79 5.3 Archi-positions 85 5.4 Phonotactic properties and collocational restrictions 90 5.5 Postscript 94

Chapter Six: Phonotagms 95 6.1 Major-type phonotagms 95 6.2 Minor-type phonotagms 99

Chapter Seven: Distribution in the nuclear context 109 7.1 Position class ‘n’ 109 7.2 Nuclearity of /r/, /l/ 112 7.3 Position class ‘N’ 117

Chapter Eight: Distribution in the pre-nuclear context 121 8.1 Position class ‘e1’ 121 8.2 Position class ‘e2’ 127 8.3 Position class ‘e3’ 130 8.4 Position class ‘E2’ 132 8.5 Position class ‘E3’ 134 8.6 Position class ‘pre1’ 134 8.7 Position class ‘pre2’ 138 8.8 Accidental appendices 146

Chapter Nine: Distribution in the post-nuclear context 152 9.1 Position class ‘i1’ 152 9.2 Position class ‘i2’ 153 9.3 Position class ‘i3’ 159 9.4 Position class ‘I’ 160 9.5 Position class ‘im’ 161

Chapter Ten: Properties of peripheral combinations 162 10.1 Peripheral combinations and their length 162 10.2 Phonematic constituency of combinations of length 2 167 10.3 Reducibility and resolvability 171 10.4 Phonematic constituency of combinations of length 3 175 10.5 Phonematic constituency of combinations of length 4 and 5 179 10.6 Further criteria: expandability, pairedness and mirror effect 182

iv Chapter Eleven: Pre-nuclear combinations 187 11.1 Distributional types DE0 and DE1 187 11.2 Distributional type DE2 188 11.3 Distributional type DE3 195 11.4 Distributional type DE4 203 11.5 Distributional type DE5 207

Chapter Twelve: Post-nuclear combinations 209 11.1 Distributional types DI0 and DI1 209 11.2 Distributional type DI2 210 11.3 Distributional type DI3 213

Chapter Thirteen: Combinations of nuclear and peripheral phonemes 215 13.1 ‘CV’ combinations 215 13.2 ‘VC’ combinations 223 13.3 Combinability of nuclear phonemes with peripheral combinations 226 13.4 ‘CVC’ combinations 230

Summary and conclusion 236

Appendices A: Collocational restrictions 245 B: Peripheral combinations and their properties 252 C: Freight-yard schemes 266 D: Distribution of nuclear phonemes 268 E: Alternative analysis of major-type phonotagms 271

References 273

v SYMBOLS

/…/ phonological representation […] phonetic (allophonic) representation → is interpreted as, corresponds to /T/, /F/, /S/ etc. archiphonemes ‘C’ peripheral (non-nuclear) phoneme

‘Cn’ a certain number of peripheral phonemes ‘V’ nuclear phoneme

‘VR’ nuclear semiconsonant ‘O’ occlusive ‘F’ fricative ‘N’ nasal (also archi-position ‘N’) ‘R’ sonant ‘L’ labial ‘P’ palatal ‘A’ alveolar ‘K’ velar ‘I’ isolated consonant or semiconsonant, i.e. not specified as to the place of articulation (also archi-position ‘I’) ‘e1’, ‘pre1’, ‘i3’ etc. positions ‘E2’, ‘E3’ etc. archi-positions pos position class (e.g. pos ‘e1’) ∀ all phonemes of a particular class ∈ includes, contains ~ complement, except for {…} a set/class of items ∩ intersection of classes vi ACKNOWLEDGMENTS

I would like to thank to my supervisor Marie Krčmová for giving me what I needed most: freedom to do it my way and criticism when I went too much astray. All errors in the final form of this work are of course my own only.

Most of all I want like to Lenka and my parents for the support and encouragement.

The work was written as part of the project Výzkumné centrum vývoje staré a střední češtiny (od praslovanských kořenů po současný stav) [Center for the Research of Old Czech and Middle Czech (its development from the Proto-Slavonic roots to the present day)] (MŠMT ČR LC546).

vii

Chapter One INTRODUCTION

This work analyzes the distribution and combinations of phonemes of Modern Standard Czech, in short, the phonotactics of Czech. An enterprise of this kind usually requires setting up some frame upon which phoneme occurrences and combinability could be stated. For us, the frame is a phonotagm, a unit in its phonetic manifestation largely coextensive, but not necessarily identical with a syllable. The phonotactic structure is accounted for by means of a model called distributional unit. It was originally applied to Pekingese Chinese (Mulder 1968), and later successfully tested on other languages: San Martín Quechua (Howkins 1972), Yulu (Gabjanda 1976), English (El-Shakfeh 1987), Sudanese Arabic (Dickins 2007) and partly on French and Russian (Rastall 1993). De- fined as a network of positions, the distributional unit is an underlying template behind all attested phonotagms allowing for their simple and exhaustive description. We will offer a detailed account of the phonotagm constituency by means a simple formal device (i.e. distributional unit), an extensive analysis of phonotagm-initial combinations, of which there are as many as 431 tokens in our database, of phonotagm-final combinations, of which there are as many as 101 tokens, and of combinations of nuclear phonemes (vowels and nuclear /r/, /l/) with non-nuclear phonemes (consonants and non- nuclear /r/, /l/) within a single phonotagm. In this respect it is the most comprehensive phonotactic analysis of Czech.

1.1 Outline of the work

The work is organized as follows. Chapters 1–3 deal with necessary preliminaries. Be- sides this introduction, the inventory of phonemes in Czech is discussed in Chapter 2. They are sorted to three basic classes, to consonants, vowels and semiconsonants. The

1 chapter also considers under what circumstances differences between phonemes are valid and when they are not. Chapter 3 defines phonotactic entities and considers several problems connected with a phonotactic analysis such as the status of phonological forms of words or syllabification. It also explains why phonological forms built of several phonotagms are not analyzed here because they belong to so-called para- phonotactics and must be considered together with features like accent. Chapters 4–5 discuss the key notion distributional unit. Chapter 4 explains the motivation behind this model and its role as a domain of distribution of phonemes. It is as a self-contained bundle of positions. The positions are crucial in understanding what a distributional unit; they are not viewed as relative placements of phonemes, but as divisions within phonotactic constructions corresponding to immediate constituents as relata of phonotactic relations. Chapter 4 also defines nuclear and peripheral entities, and relationships between phonotagms and phonic syllables. The model is applied to Czech in Chapter 5. In this language the distributional unit has five pre-nuclear, one nuclear and three post-nuclear positions. To make the analysis more precise, several archi-positions are introduced; they are a particular kind of positions accounting for special distribution of certain phonemes. After the construction of the distributional unit, Chapter 6 deals with its instances, phonotagms. Two types of phonotagms are recognized: major-type phonotagms corresponding to all self-contained phonotactic entities, and minor-type phonotagms functioning as special attachements to major-type phonotagms. Chapters 7–9 deal with the distribution of phonemes of Czech, that is, with their occurrences within the positions of the distributional unit. A position can be viewed as a sum of phonotactic properties (= distributional and combinational potentials of phonemes) which are unique for each position; phonemes occurring in a certain position share these properties. Chapter 7 discusses phonotactic properties of the nuclear position, which is occupied by vowels, and of the nuclear archi-position, which is occupied by /r/ and /l/. The ability of the latter to function as nuclear entities is reviewed here, too. Chapter 8 examines properties of the five pre-nuclear positions and of the two pre- nuclear archi-positions. Finally, post-nuclear positions and archi-positions, and their phonotactic properties are discussed in Chapter 9.

2 The main topic of Chapters 10–13 is combinations of phonemes in Czech. Chapter 10 discusses various properties of pre-nuclear and post-nuclear combinations such as their length, phonematic constituency, reducibility and resolvability. Chapter 11 divides pre-nuclear combinations according to their length, and sorts them to a series of distributional patterns. Chapter 12 does the same for post-nuclear combinations. Chapter 13 considers the ability of nuclear phonemes to combine with non-nuclear ones and the ability of pre-nuclear phonemes to co-occur with post-nuclear ones. No other previous analysis of Czech has considered this problem. Finally, Chapter 14 gives a summary and conclusion to this work. It is supplemented by several appendices. Appendix A summarizes restrictions introduced during our analysis limiting the occurrence and combinability of phonemes. Appendix B provides a complete listing of peripheral combinations found to exist in Modern Standard Czech. Appendix C reproduces so-called freight-yard schemes of the pre-nuclear and post- nuclear sections of the distributional unit. Appendix D illustrates the distribution of nuclear phonemes. Finally, Appendix E offers an alternative analysis of major-type phonotagms inspired by but superseding the one in Kučera (1961).

1.2 Theory and description

If you want to plow up a field, you need a plow or at least a spade. It is extremely difficult to plow up a field without it. But even the possession of some equipment is not enough: you also need to know how to use it. The same thing is with languages. If you want to describe a language, you need a tool, some linguistic theory with which the description could be achieved, but you also need to have a certain methodology specifying how such a description is to proceed. The theoretical and methodological framework adopted in this work is based on a semiotic paradigm introduced by Jan Mulder and Sándor Hervey (Mulder 1968, 1989, 1998; Mulder – Hervey 1972, 1980; Hervey 1996; also Dickins 1998 and Mulder – Rastall 2005) known as Axiomatic Functionalism. It is a structuralist and functionalist approach stemming from the teachings of Ferdinand de Saussure (1972) and of the Pra-

3 gue School, especially of Nikolai Trubetzkoy (1939). Its phonological theory is particularly based on that of André Martinet (Martinet 1956, 1991; see also Akamatsu 1988, 1992), being, however, more formalized. In the manner of thinking about linguistic theories it is very similar to Louis Hjelmslev’s Glossematics; both approaches insist that a theory “is in itself independent of any experience; [i]n itself, it says nothing at all about the possibility of its application and relation to empirical data” (Hjelmslev 1961: 14; Mulder 1996: 18). On the philosophical level Mulder and Hervey have been much influenced by the philosophy of science of Karl Popper (cf. Popper 2002, 2003). They have adopted his falsificationism and modified it to create their own version (Mulder 1993). This modification concerns, among others, the ontological difference between theory and description (Mulder – Hervey 1980: 15–28, Mulder 1996). They represent two different sets of statements. A theory is an axiomatically and deductively built set of notions and models whose purpose is provide a means for producing descriptions of desired phenomena. Theories are primarily evaluated in terms of deductive-logical consistency, and can only be indirectly tested in light of their ability to engender consistent, adequate and simple descriptions (Hervey 1996). In other words, a theory must be internally consistent and must provide an appropriate framework for accounting for certain phenomena. It is also a virtue of a theory if it is simple, but one can have a relatively complicated theory provided that it allows for simple descriptions. A description is an act of applying a theory to a particular range of phenomena in order to produce a meaningful account of it. Hypothe- ses about speech phenomena are tested against relevant data in an attempt to refute them. A hypothesis, in order to be testable at all, must be formulated in such terms so that its refutation is, at least in principle, possible (Popper ibid.). The sets of hypotheses that have withstood refutation (i.e. which are corroborated) are adopted as statements about the described phenomena. Just like a theory a description should be internally consistent: it is inadmissible for it to contain statements contradicting each other. Also, a description should be externally consistent with the theory: it must not be an arbitrary mixture of statements and ad hoc concepts proposed to reach a desired goal at any cost. It is often the case that linguistic descriptions use notions without defining them, but rather assuming their universal va-

4 lidity. Such descriptions, though seemingly plausible and appealing, are useless because they are in fact devoid of any meaning. A notorious example of this is the concept of phoneme. It is used in many phonological descriptions, but few of them bother to specify its definition, taking it for grated. However, there is no universal definition of phoneme; rather, this notion is defined by every theory. Consequently, a phoneme stands for something else in one description and something else in another description, which may and does lead to confusion. Aware of these dangers, Mulder and Hervey have always, like Hjelmslev, insisted on precisely and explicitly formulated definitions of notions they employ; it was one reason they have produced so-called Postulates for Axio- matic Functionalism (the latest version in Mulder – Hervey 2009) which lay out the fundamental assumptions of the approach (the axioms), and provide a network of definitions of all theoretical notions of the theory. Having adopted Mulder and Hervey’s approach, we constantly make recourse to the Postulates without always citing them. Besides consistency, which is an essential requisite of any description (and theory), it is also required of descriptions that they be adequate and simple.1 While consistency can be tested, adequacy and simplicity may rather be relative terms. A description is adequate if it fulfills its goal, and provides an exhaustive and understandable account of a selected range of phenomena. A successful description should in addition be able to account for potential phenomena, i.e. it should be able to state the possibilities of the system. Our work will therefore not only describe the structure of existing combinations of phonemes in Czech, but will also offer a principled means for predicting structurally possible but unattested combinations. A description may be regarded simple if it does not contain redundant elements. One should not introduce new components unless necessary. Sometimes, however, it makes more sense to sacrifice simplicity in order to make a description more adequate. Or we can afford to have some parts of a description less simple in order to make other parts more simple and/or adequate. The degree of adequacy and simplicity and the balance between them are always a matter of judgment of the describer (cf. Dickins 1998: 42–6). So, while we have strived for making our de-

1 Cf. also Hjelmslev’s so-called empirical principle: “The description shall be free of contradiction (self-consistent), exhaustive, and as simple as possible” (Hjelmslev 1961: 11).

5 scription as consistent as possible, it is not the only way it could have been done, but it is a way we believe to be simple, adequate and novel.

1.3 Modern Standard Czech

The language described here is Modern Standard Czech. Modern Czech is a West Slavic language spoken by the majority of inhabitants of the Czech Republic at the beginning of the 21st century and during the 20th century; our description is thus a synchronic one. By Standard Czech we mean a variety of Modern Czech officially codified and having a special status among all forms of Czech. In the Czech tradition it is usually known as spisovná čeština (“literary Czech”), but we prefer using the other term without making any distinctions. Therefore, we describe one particular variety of Modern Czech, not Modern Czech as a whole. We believe the latter may not in fact be possible because Czech is not one homogenous structure but a conglomeration of several varieties with several intermixed systems. Standard Czech is one of such systems, one which is more or less clearly delimitated, even though the precise boundary between Czech which is standard and Czech which is not standard cannot always be set. Yet the distinction must be made because the phonology of Standard Czech is different to that of non-standard Czech. Differences are numerous; they concern not only different phoneme inventories but also different phonotactics. For example, Standard Czech does not allow phonotagm-final combinations of a fricative and /ř/ while non-standard Czech does: cf. otevř, a variant of otevři “to open (imper. sg.)”. Similarly, Standard Czech does not allow phonotagm-final combinations where the velar occlusive is preceded by another occlusive, but non-standard Czech does: cf. nedotk’, a variant of nedotkl “he did not touch”. Standard Czech exists primarily but not exclusively as a written language because Czechs mostly use dialects or so-called Common Czech (obecná čeština) in their casual speech. The pronunciation of Standard Czech has been codified in the official handbook Výslovnost spisovné češtiny I (“Pronunciation of literary Czech I” = VSČ, 2nd edition

6 1967) and its second volume Výslovnost spisovné češtiny (= Romportl et al. 1978).2 It has also been described elsewhere, notably in Hůrková (1995), Palková (1997), Krčmová (2008), Zeman (2008) and in the grammar books mentioned below. Several detailed phonetic descriptions of this variety of Czech exist, the most comprehensive ones being Hála (1962), Palková (op. cit.) and Krčmová (op. cit.). Phonetic and orthoepic descriptions of Standard Czech are not by themselves sufficient if we aim to account for its phonotactics. Phonotagms occur within phonological forms of words. There are several dictionaries of Standard Czech the most important of which are Slovník spisovného jazyka českého (“Dictionary of the literary Czech language”, 1960–1971, = SSJČ), and more recent Slovník spisovné češtiny (“Dictionary of literary Czech”, 3rd edition 2008, = SSČ). The most comprehensive dictionary remains to be Příruční slovník jazyka českého (“Concise dictionary of the Czech language”, 1935–1957, = PSJČ)3. However, this dictionary does not contain only words recognized as standard, “literary”, but also words whose status may be dubious because they are rather archaic or created ad hoc by literary writers. If a word occurs only in PSJČ, this fact is explicitly mentioned. Yet this dictionary is important because it lists words containing phoneme combinations not found elsewhere (e.g. /PStř/ in pstřeň). Quite often, we also cite place names and proper names because they also contain phoneme combinations not attested elsewhere, for example /jř/ in the surname Kejř or /břv/ in the place name Břve. We mention such combinations, but they do not represent the basic data for our analysis. Also, we make no systematic analysis of onomatopoeic expressions if their phonotactic structure is markedly deviant in comparison with the structure of “normal” vocabulary. These include expressions such as bzz, kšc, pf or pšt. As a rule, dictionaries list words in their uninflected form. However, many phoneme combinations are found only in inflected forms (such as /jSK/ in vojsk, gen. pl. of vojsko). The grammar of Standard Czech and its inflections are also sufficiently described in various reference books such as Česká mluvnice (“Czech grammar”, 6th edition, = Havránek – Jedlička 1986), Mluvnice češtiny (“Grammar of Czech”, particularly the

2 The book has the same title like the 1967 one except for the number I. The former deals with pronunciation of foreign words while the latter with pronunciation of Czech as a whole. 3 PSJČ and its database, SSJČ and the glossary of PSJČ, SSJČ and SSČ are available online; see references.

7 second volume, 1986, abbreviated MČ2), Příruční mluvnice češtiny (“Concise grammar of Czech”, 2nd edition 1996, = PMČ), and Mluvnice současné češtiny 1 (“Grammar of contemporary Czech 1”, 2010, = MSČ). Another valuable resource is Internetová jazyk- ová příručka (“Internet language reference book”) available online and maintained by the Institute of the Czech Language of the Academy of Sciences of the Czech Republic. We have also used Český národní korpus (“Czech National Corpus”), the synchronic base SYN, but this source does not contain only texts from Standard Czech and should be cautiously treated.4

1.4 Previous phonotactic descriptions of Czech

This work is not the first or the only phonotactic description of Czech. If phonotactics is broadly understood as an account of the distribution and combinations of phonemes within certain distributional frame, and an account of the structure of such a frame, it is evident that linguists could not have avoided this topic, if only they were to say that a phoneme x can or cannot combine with a phoneme y. One could trace back statements of this kind e.g. in De Orthographia bohemica from the 15th century (usually ascribed to Jan Hus; a Czech translation in Nechutová et al. 1982: 57–69). However, we will con- fine ourselves to the most recent works. Probably the earliest comprehensive treatise on phoneme combinations in Czech is Hattala (1870). Hattala provides an impressive list of word-initial consonantal combinations and sorts them to several classes. Some of the combinations mentioned are dubious or occurrent in words probably no longer used in the Czech of Hattala’s time, and some are missing (such as /vj/ in věk which is no consonantal combination for him). Other descriptions of phoneme combinations stemmed from attempts to create the short- hand for Czech. Summarizing these attempts, Trnka (1937, in particular pp. 31–41) discusses several combinational restrictions on phonemes, and the ratio of occurrences of

4 We will provide English glosses of the analyzed words only when the meaning is somehow important to the topic under discussion.

8 consonants with vowels and with other consonants; he also provides lists of attested consonantal combinations in a single morpheme and across morphological boundaries. Phoneme combinations are also discussed in some early articles by Mathesius (see below), but a properly phonological analysis of word-initial and word-final consonantal combinations was published as late as 1972 by Novotná (1972). Although her primary focus was spoken Czech, and although she excludes combinations occurring in allegedly foreign-origin words, her work was at that time the most comprehensive description of this kind analyzing as many as 308 word-initial and 20 word-final combinations. A partial list of consonantal combinations is also given in Trnka (1972, reprinted in Trnka 1982) together with the frequency of their occurrence in words of different syllable length (see also Mazlová 1946). The quantitative aspect of phoneme combinations is also investigated in Ludvíková – Kraus (1966), Ludvíková (1968, 1972a), and Těšite- lová et al. (1985). The frequency of Czech sounds (not phonemes!) and their combinations based on the Czech National Corpus has been examined in Bartoň et al. (2009).5 Recently, the probability of the occurrence of consonantal combinations at the beginning, at the end of and in the middle of words, stress groups and prosodic phrases has been discussed in Volín – Churaňová (2010). Combinability of Czech phonemes is also discussed in Mathesius (1929, 1931a, 1931b, 1931c; see also Mathesius 1947), Vachek (1932; see also Vachek 1968), Trnka (1966), Novotná-Hůrková (1980), MČ1 (1986), and Grygarová-Rechzieglová (1993). Phoneme combinability has also been examined for various Slavic languages, Czech included. Sawicka (1974) is one such attempt including lists and a partial classification of word-initial and word-final combinations. Tolstaja (1968, 1974) discusses consonantal combinations in Slavic languages examining their properties (like resolvability, cf. Chapter 10 here). According to Sawicka (2009), which is a general account of phoneme combinations in Slavic languages, combinations of vowels in Slavic languages are discussed in the PhD. thesis Grupy samogłoskowe w językach słowiańskich (Toruń, 2002; unavailable to us) by Anna Korytowska. Analyzing several languages, not only the Slavic ones, Appel (1957–8) provides a list of initial and final consonantal combinations

5 The work, however, presents only rough data without interpretation, and is methodologically very dubious. See also the next note.

9 in Czech, some of which being rather doubtful. Properties of word-initial and word-final consonantal combinations derived from an analysis of 104 languages, Czech included, are discussed in Greenberg (1978). Another line of phonotactic research, not necessarily exclusive to the previous one, has focused on the phonological structure of higher-level units such as syllables or words. Mathesius (1929; reprinted and translated in Mathesius 1947) describes the constituency of words whose phonological forms are built of one, two, three or four phonemes. With the basic criterion being whether a phoneme is a vowel or a consonant, there are 20 structural types of words (e.g. ‘V’ for a, ‘CV’ for to, ‘CVVC’ for neuč, ‘CCCV’ for sklo etc.). Vachek (1940) supplemented Mathesius’ analysis by examining words built of five phonemes; 24 more structural types were added. The structure of more complex words is described in Trnka (1972), though with a primary emphasis on consonantal combinations occurring in words of different syllable length. One can also get some impression of the phonological structure of Czech words from Bartoň et al. (2009) based on the analysis of the Czech National Corpus.6 Probably the only analysis of the syllable in Czech was done by Kučera (1961). He establishes its structure in terms of a grid of several positions7. He then specifies membership of each position, and introduces a number of constraints on the combinability of phonemes. The model reappears in a reduced and slightly modified form in Kučera – Monroe (1968). It is also reproduced in Palková (1997: 276) as the model of the syllable structure in Czech, but curiously enough, Palková has adopted the reduced variant from Kučera – Monroe (1968: 55), not from Kučera (1961: 74, 79–80). The Czech syllable has also been analyzed from the quantitative perspective in Ludvíková (1972a, 1972b, 1976, 1978) and in Těšitelová et al. (1985). The syllable structure of Slavic languages is discussed in Lekomceva (1968) and also in Sawicka (1985).

6 The data presented here are, as noted by the authors, distorted by a fact that abbreviations have been included. For example, words of the structure ‘CC’ are said to belong among the 100 most common types. Czech has no such word, only abbreviations like př. for příklad “example”. 7 His use of the term position differs to ours; see Chapter 4 and also Appendix E.

10 Chapter Two PHONEMATICS OF CZECH

2.1 Introduction

This chapter presents the inventory of phonemes in Czech. It is intended to be an alternative to the existing phonological analyses (Kučera 1961, Vachek 1968, Horálek in MČ1), bringing some new and hopefully more adequate solutions. In describing the phonematics of a language we must start with the definition of phoneme. Like any other notion in the theory, it is a model accounting for and applicable to certain speech phenomena. It acquires two types of identity: paradigmatic and syntagmatic (Mulder – Hervey 2009). From the paradigmatic perspective, which is the domain of phonematics, the phoneme is a self-contained unordered bundle of distinctive features.1 The distinctive features express the phoneme’s distinctive function which is the sum of oppositions the phoneme, via its allophones, partakes in. It is these oppositions that determine the paradigmatic identity of the phoneme. From the syntagmatic perspective, which is the domain of phonotactics, the phoneme is the minimum phonotactic entity; it is the smallest phonological entity capable of entering into ordering relations. Phonemes are grouped into phonotagms which are self-contained phonotactic entities (see Section 3.1). Unless a language has one-phoneme words like Czech a, phonotagms correspond to self-contained combinations of phonemes. Within these combinations, every phoneme has a certain phonotactic function. Those functioning as nuclear entities only are vowels, and those functioning as peripheral, i.e. non-nuclear entities are consonants. Czech in addition has a class of so-called semiconsonants capable of functioning both

1 The terminology used here does not necessary coincide with that used by other linguists.

11 as nuclear and peripheral entities.2 Moreover, within a self-contained combination of phonemes every phoneme is assumed to occupy a certain position of the distributional unit (see Chapter 4). A position is an intension of distributional and combinational potentials of a certain class of phonemes, and the membership of a phoneme in a position is that which determines its syntagmatic identity. The traditional method by which the paradigmatic identity of phonemes of a given language is set is known as commutation test (Akamatsu 1988 and 1992). It calculates the distinctive function of particular phonic segments in specific contexts (Mulder 1968: 127–8). Phonemes are abstractions of phonic segments with the same distinctive function. By virtue of their mutual commutability (opposability), it is possible to analyze phonemes into unordered sets of distinctive features. In phonology commutation is a choice between phonological entities (or between the presence and absence of a phonological entity) which are in functional opposition in a given context (Mulder – Hervey 2009). As the commutation test setting up the phonemes of Czech has been performed in detail elsewhere (Bičan 2008a), we will only present its results. Before doing that, we must mention that the definition of commutation implies the existence of an oppositional relation between some phonological entities in a certain context. The context is here of crucial importance. A relation taking place in one context need not necessarily obtain in another. The inoperability or suspension of an opposition in a given well-defined context is called neutralization. As a concept, neutralization has been operated with in many linguistic approaches and variously defined (cf. Davidsen-Nielsen 1978, Akamatsu 1988). The notable difference between these and the present approach is the fact that we hold that the concept neutralization necessarily implies another concept: archiphoneme. It is an entity resulting from neutralization and occurring in a certain phonotactic context which has triggered the neutralization. It is formally defined as the intersection of the sets of distinctive features qua sets characterizing two or more phonemes in the other contexts. Being an intersection, an archiphoneme is logically included in the phonemes from which it results and is functionally equivalent (but not identical!) to them. It means that an archiphoneme is “a phoneme in

2 See Section 4.3 on more detailed definitions of these classes. Note that we will use the terms vocoid and contoid for phonetically defined types of sounds (after Pike 1947).

12 a sub-system which, when projected into the over-all system, is represented there by two or more phonemes” (Mulder 1968: 114). It is a logical consequence of neutralization, which is itself implied from the concept of opposition (commutation). If there is an oppositional relation between two or more distinctive features characterizing certain phonemes in one context (context A), but the relation cannot be postulated in another context (context B) because the characterization is redundant there, it follows that the entity occurring in context B cannot be equated with any of the phonemes of context A. It implies from the definition of phoneme as a bundle of distinctive features.

2.2 Consonants

Consonants are phonemes that occur in non-nuclear positions only, which means they can never constitute nuclei of phonotagms and are dependent on them. Figure I visualizes the decomposition of the Czech consonants into distinctive features and indicates their mutual oppositional relations. The notion distinctive feature as employed here little to do with binaristic and universalist distinctive feature theories such as the one envi- sioned by Jakobson (in Jakobson – Halle 1956)3. The distinctive features given in the figure are not chosen from a pre-established set of universal distinctive features, and there is no need for them to be binary (cf. also Akamatsu 1992: 35ff.). They are first of all functional entities and their function is distinctive.4 If something is distinctive, it must distinguish something, and so a feature can be regarded distinctive if and only if it distinguishes the phoneme it is part of from another one. /p/ is voiceless in Czech because it is not voiced (i.e. it is not /b/), it is labial because it is not alveolar, palatal, or velar (i.e. not /t/, /ť/ or /k/), and it is occlusive because it is not fricative or nasal (i.e. not /f/ or /m/). Thus, /p/ is ‘voiceless labial occlusive’—in Czech. The latter stipulation is

3 Jakobson’s theory was applied to Czech by Kučera (1961) and Vachek (1968). 4 Moreover, they are not necessarily identical with phonetic properties of sounds. So, if phonemes are described as containing the distinctive feature ‘palatal’, it does not mean they are all realized as palatal sounds (see below). Distinctive features, just like phonemes, are phonological, not phonetic entities. We will always make clear when we speak about realizations and when about distinctive features of phonemes. For detailed phonetic descriptions of the sounds of Czech, see Hála (1962), Palková (1997) and Krčmová (2008) with references.

13 essential because the features are not universal but unique for every language because every language embodies a different set of oppositions between phonemes. On the contrary, /m/ is neither voiceless nor voiced in Czech because the features ‘voiceless’ and ‘voiced’ do not distinguish anything in the case of the nasals, and cannot be distinctive for them. /m/ is therefore ‘labial nasal’.

Occlusive Fricative Nasal Voiceless Voiced Voiceless Voiced Labial /p/ /b/ /f/ /v/ /m/ Alveolar /t/ /d/ /s/ /z/ /n/ Palatal /ť/ /ď/ /š/ /ž/ /ň/ Velar /k/ /g/ /x/ /h/

Figure I: Consonants of Czech. The affricates are analyzed as two-phoneme combinations. Outside the system of proportion stand /j/ ‘approximant’ and /ř/ ‘spirant’.

As we can see, the backbone of the consonantal phonematic system is formed by two proportional oppositions: ‘labial’ ~ ‘alveolar’ ~ ‘palatal’ ~ ‘velar’ on the one hand, and ‘occlusive’ ~ ‘fricative’ ~ ‘nasal’ on the other. The first may be conveniently referred to as manners of articulation, and the second as places of articulation. However, these are just useful labels, not functional categories; it is not phonemes that are articulated, but their realizations. In addition, there is a proportional opposition ‘voiceless’ ~ ‘voiced’ which is not applicable to nasals, though. We can conveniently call it voicing. The phonemes /p/, /b/, /f/, /v/, /m/ are all grouped together as labials, even though /p/, /b/ and /m/ are bilabial in realization (oral stops [p], [b] and nasal stop [m], respec- 5 tively), whereas /f/, /v/ are realized as labio-dental fricatives [f] and [v] , respectively. The place of articulation is interconnected with the manner of articulation here, and hence there is no need to postulate separate distinctive features ‘bilabial’ for /p/, /b/, /m/ and ‘labio-dental’ for /f/, /v/. They are all ‘labial’, i.e. phonetically characterized by the activity of lips. The labial nasal /m/ is also realized as labio-dental nasal stop [ɱ] before /f/ and /v/ (in e.g. tramvaj or nymfa).

5 The sound [v] has some peculiar phonetic properties; see below, Section 2.6.

14 The phonemes /s/ and /z/ are grouped together with /t/, /d/, /n/ under alveolars;6 they are realized as alveolar fricatives/sibilants [s], [z], alveolar oral stops [t], [d] and as alveolar nasal stop [n], respectively. Before /ň/, the phonemes /t/, /d/ may also—there is free variation—be realized as palatal stops [c], [ɟ] in words like matně and hodně; similarly, /n/ may be realized before /ť/ and /ď/ as palatal nasal stop [ɲ] in words like kontě (cf. VSČ: 61–2).7 The phonemes /š/, /ž/ are grouped together with /ť/, /ď/8 and /ň/ under palatals; they are realized, respectively, as post-alveolar fricatives/sibilants [ʃ], [ʒ], and palatal stops, oral and nasal: [c], [ɟ] and [ɲ]. The relation between /t/, /d/ and /s/, /z/ is the same as (i.e. is proportional to) the relation between /ť/, /ď/ and /š/, /ž/ as well as that between /p/, /b/ and /f/, /v/, and between /k/, /g/ and /x/, /h/. The phonemes /k/, /g/, /x/, /h/ are all interpreted as velars. In realization they are characterized by articulation in the back of the month and/or throat. The phonemes /k/ and /g/ are realized as velar stops, voiceless [k] and voiced [ɡ], respectively; /x/ is realized as voiceless velar fricative [x]. However, /h/ is realized as voiced laryngeal fricative [ɦ]. Despite this realizational discrepancy, /h/ functions in the system as the voiced counterpart of /x/. The opposition /x/ ~ /h/ is proportional to all other voicing oppositions and is liable to neutralization. The phonetic counter-domain of this neutralization is assimilation of voicing; in such processes [ɦ] is changed/assimilated to [x], see the examples under (1), in particular the fourth one (the capital letters are archiphonemes, see Section 2.6).

(1) [xata] chata “cottage” × [xat] chat “cottage (gen. pl.)” → /xata/ × /xaT/ [brada] brada “chin” × [brat] brad “chin (gen. pl.)” → /brada/ × /braT/ [soxa] socha “statue” × [sox] soch “statue (gen. pl.)” → /soxa/ × /soX/ [noɦa] noha “leg” × [nox] noh “leg (gen. pl.)” → /noha/ × /noX/

6 We argued in Bičan (2008a and 2010a) that /n/ should be viewed rather as a so-called hyper- phoneme containing a hyper-feature ‘alveolar/velar’ (on the notions see Mulder – Hervey 1980: 104–21). However, since this is of little importance to the phonotactics, we refrain from this here and place /n/ under alveolars. 7 According to Novotná (1962), the palatal realizations are more common. 8 See Machač – Skarnitzl (2004) for a recent phonetic analysis of the Czech palatal stops.

15 Viewed from a different perspective, /p/, /b/, /t/, /d/, /ť/, /ď/, /k/, /g/ are all interpreted as occlusives; /f/, /v/, /s/, /z/, /š/, /ž/, /x/, /h/ as fricatives; and /m/, /n/, /ň/ as nasals. The classification into occlusives, fricatives and nasals corresponds to the phonetic scope of these phonemes: all occlusives are realized as oral stops, all fricatives as fricatives, and all nasals as nasal stops. Moreover, all occlusives and fricatives are distinguished as to being either voiceless (/p/, /t/, /ť/, /k/, /f/, /s/, /š/, /x/) or voiced (/b/, /d/, /ď/, /g/, /v/, /z/, /ž/, /h/). The nasals /m/, /n/, /ň/ are indifferent to this distinction, even though they are in realization always voiced. A phoneme cannot be distinctively (phonologically) voiceless unless it is opposed to its voiced counterpart and vice versa. Outside the main system of proportions stand two phonemes: spirant /ř/ and approximant /j/, each characterized by one distinctive feature only. The label spirant is an arbitrary one (it was chosen for the lack of a better one-word term), and has nothing to do with the phonetic realization of /ř/ which is realized as a fricative vibrant, either voiced [r̝] or voiceless [r̝̊]. The voiced one occurs at the beginning of a word before a vocoid, between two vocoids, and in the vicinity of a phonetically voiced consonant unless there is a phonetically voiceless consonant next to it (cf. Petřvald, a place name realized as [pɛtrvalt̝̊ ]). The voiceless one occurs before and after a voiceless sound and/or at the end of a word before a pause (in e.g. keř [kɛr]̝̊ or tři [tr̝̊ɪ]). The sounds are very unique in the world’s languages, and there may be no other language but Czech where they would be realizations of a separate phoneme, not just facultative variants. The second phoneme standing outside the system of proportions is /j/ defined as ‘approximant’ and realized as voiced palatal approximant [j] in all instances of its occurrence. We will use the term sonant for the class of phonemes which includes /j/, /ř/ and non-nuclear /r/, /l/ (for which see below). In other words, sonants are those peripheral phonemes which are not occlusives, fricatives or nasals.

2.3 Interpretation of the affricates

Traditional analyses of Czech usually postulate two other consonantal phonemes: /c/ and /č/, realized, respectively, as voiceless alveolar affricate [ʦ] and voiceless post-

16 alveolar affricate [ʧ]. Strictly speaking, we should add two more affricates, their voiced counterparts: [ʣ] and [ʤ]. However, in our analysis all are interpreted as corresponding to combinations of two phonemes, namely to /Ts/, /Tš/, /Tz/ and /Tž/9, respectively. As it is not immediately relevant to this work, we will not discuss here theoretical and methodological problems concerning when and why a certain sound or a combination of sounds correspond to one or two phonemes. It has been much discussed in literature.10 The present section will only demonstrate the adequacy and simplicity of such an analysis. Its simplicity can be instantly shown: If the affricates were interpreted as single phonemes, the Czech phoneme inventory would have additional members. So, instead of positing four additional phonemes behind [ʦ], [ʧ], [ʣ], [ʤ], we account for them with combinations of phonemes already part of the system, namely with /Ts/, /Tš/, /Tz/, /Tž/. However, our analysis has other advantages, too. To begin with, let us review why most analysts of Czech prefer interpreting the affricate [ʦ] and [ʧ] as separate phonemes /c/ and /č/. The reasons are various (see Palková 1997: 240–1 and Novotná-Hůrková 1980 for an overview). First, phonetic reasons are adduced. The affricates behave like single sounds, albeit with two phases, but the two phases cannot be separated by explosion. This is unlike real combinations /ts/, /tš/ (see below) where the occlusive and constrictive phases are clearly distinguished. However, phonetic criteria need not be necessarily decisive in a phonological analysis. The fact that the affricates [ʦ], [ʧ] have similar characteristics like single sounds does not mean that they must correspond to single phonemes. There is no a priori correlation “one sound = one phoneme” because phonemes are entities of a quite different ontological nature than sounds, that is to say, a phoneme in functional phonology is not a general- ized family of sounds but an abstract model accounting for certain speech phenomena. A combination of two phonemes /ts/ (i.e. /Ts/) may be a model for a single speech sound [ʦ] if the model is appropriate and can be applied to the given speech event.

9 This analysis is the most adequate given the phonetic nature of the affricates. To keep things simple, we will discuss only [ʦ] and [ʧ] because their voiced counterparts [ʣ] and [ʤ] are rather rare, yet they belong to the phonological system of Czech. [ʣ] occurs word-initially probably only in dzinkat (SSJČ); [ʤ] is found in many words like džus or džungle. 10 See e.g. Mulder (1968: 28–30), Hervey (1972: 355–9), Martinet (1965: 109–23) and Trubetzkoy (1939: 50–9). In Bičan (2008a) we discussed the problem in relation to Czech.

17 What is more important is the fact that the affricates [ʦ], [ʧ] can be confronted with genuine combinations of sounds [t.s], [t.ʃ] (and also with [t.ts], [t.tʃ]). If so, we cannot maintain that [ʦ], [ʧ] correspond to /Ts/, /Tš/ because we would fail to account for a 11 functional difference between [ʦ], [ʧ] and [t.s], [t.ʃ]. The following examples show the difference could indeed be functional:12

(2) (a) [klaʦku] klacku (dat. sg. of klacek) × [klat.sku] Kladsku (loc. sg. of Kladsko) [poʧiːt] počít × [pot.ʃiːt] podšít rd [ɦrʧiː̩ ] hrčí (3 person sg. of hrčet) × [ɦrt.ʃiː̩ ] hrdší (comparative of hrdý) (b) [praːʦɛ] práce × [praːt.sɛ] prát se [vjɛʦɛm] věcem (dat. pl. of věc) × [vjɛt.sɛm] vjet sem [raːʦɛ] ráce × [raːt.sɛ] rád se

These pairs might be strong evidence for the communicational difference between the affricates [ʦ], [ʧ] and the combinations [t.s], [t.ʃ]. However, the examples must be carefully examined before any conclusion is reached. First of all, the difference between the forms under (2a) is somewhat artificial or rather potential. Although the words like Kladsku or hrdší may be pronounced with a sequence stop + fricative, research has shown that the pronunciation with the affricate [ʦ], [ʧ] is much commoner (Novotná 1962, Novotná-Hůrková 1974); this pronunciation is also recognized as orthoepic (VSČ: 56, Hůrková 1995: 32–3, Palková 1997: 334, Krčmová 2008: 188, 205, Zeman 2008: 112). On the other hand, the pronunciation with [t.s], [t.ʃ] occurs only in careful speech, and apparently when a speaker wants to indicate a morphological boundary (cf. hrdý “proud” vs. hrdší “prouder”). The situation with the examples under (2b) is slightly different, though. In this case the pronunciation with an affricate is also common but not considered to be correct in standard Czech. The codification requires pro-

11 Cf. the situation of Polish where the affricates are single phonemes because they can be compared in the same context to stop–fricative sequences as in [tʃɨ] trzy “three” × [ʧɨ] czy “if”. 12 The examples and pronunciation were taken from Kučera (1961: 32), Vachek (1968: 71) and Palková (1997: 240). We use a dot to indicate the difference between affricates [ʦ], [ʧ] and sequences of two sounds [t] + [s], [t] + [ʃ], respectively.

18 nunciation with stop + fricative so that the grammatical boundary between the words is not obliterated and the words not confused. Be it as it may, there is one important feature in common: the stop–fricative pronunciation in contrast to the affricate pronunciation is used to highlight boundaries between forms. If this is the function of the stop + fricative sequences, it is reasonable that they be interpreted as instances of so-called boundary signals (Grenzsignale; Trubetzkoy 1939). In the present theoretical framework this situation can be accounted for by means of the notion diaereme, roughly corresponding to what others have called (dis)juncture (see Section 3.2). The diaereme (transcribed as ‘#’) will be a phonological representation of phonetically salient boundaries between forms13. Consequently, the opposition between [ʦ], [ʧ] and [t.s], [t.ʃ] does not obtain within one and the same syllable but across two syllables because the latter combination occurs only across syllable boundaries. It is irrelevant whether the boundary coincides with a word boundary in prát se or with a morpheme boundary in hrdší as long as the boundary is phonetically salient and functional. Accordingly, the forms under (2a) are interpreted in the following manner:

(3) [praːʦɛ] práce × [praːt.sɛ] prát se → /#prāTse#/ × /#prāT#se#/ [vjɛʦɛm] věcem × [vjɛt.sɛm] vjet sem → /#vjeTsem#/ × /#vjeT#sem#/ [raːʦɛ] ráce × [raːt.sɛ] rád se → /#rāTse#/ × /#rāT#se#/

One of the advantages of this analysis is such that the rule expressing the merge of práce with prát se is quite simple in contrast with a similar rule in analyses operating with the separate phonemes /c/ and /č/. For us, /#prāTse#/ práce and /#prāT#se#/ prát se coincide when the medial diaereme is deleted (i.e. /T#s/ becomes /Ts/), which happens when the boundary is neglected. The alternative analyses have to explain why two phonemes /t/ and /s/ turn to the phoneme /c/ if they stand next to each other.

13 See also Bičan (2008b). In that article, however, the affricates are interpreted as simple phonemes /c/ and /č/. We did not want to burden it with a discussion why they should rather be interpreted as combinations of two phonemes as the topic of the article was something else.

19 Another advantage of our analysis manifests itself on the phonotactic level, which is of particular importance for the present work.14 The combinability of phonemes at the beginning of a phonotagm becomes more transparent once /Ts/, /Tš/ and /Tz/, /Tž/ are accepted as valid combinations of an occlusive and an alveolar or a palatal fricative because they parallel other combinations of the same type. See Figure II where the question marks can be replaced by /Ts/, /Tš/, /Tz/ and /Tž/, respectively.

/Sp/ /Sb/ /St/ /Sd/ /Sk/ /Sg/ [sp] [zb] [st] [zd] [sk] [zɡ] /Ps/ /Pz/ /Ks/ ? ? N.A. [ps] [bz] [ks]

/Šp/ /Šb/ /Št/ /Šd/ /Šk/ N.A. [ʃp] [ʒb] [ʃt] [ʒd] [ʃk] /Pš/ /Pž/ /Kš/ ? ? N.A. [pʃ] [bʒ] [kʃ] Figure II: Combinations of occlusives and fricatives at the beginning of a phonotagm. N.A. stands for Not Attested. The questions marks can be replaced by /Ts/, /Tš/, /Tz/ and /Tž/.

/ST/ /SK/ /ŠT/ N.A. N.A. N.A. [st] [sk] [ʃt] /PS/ /KS/ /PŠ/ /KŠ/ ? ? [ps] [ks] [pʃ] [kʃ] Figure III: Combinations of occlusives and fricatives at the end of a phonotagm. N.A. stands for Not Attested. The questions marks can be replaced by /TS/ and /TŠ/.

The situation is similar in the post-nuclear context where the combinations /TS/, /TŠ/ have the same constituency as /PS/, /PŠ/ and /KS/, /KŠ/. See Figure III where the question marks stand for /TS/ and /TŠ/, respectively. And there are in addition two post- nuclear three-consonant combinations /PST/ and /KST/. If the affricate in [pɛːʦ.t] péct is analyzed as /TS/, the resulting combination /TST/ parallels the structure of /PST/ and /KST/ (similar combinations with /Š/ are not attested). We can see that if the affricates

14 Admittedly, once the affricates are analyzed as two-phoneme combinations, we get unusual combinations /TTs/, /TTš/, /TSTs/ and /TŠTš/ across phonotagm boundaries as in [vjɛt.ʦɛm] vědcem → /#vjeTTsem#/ (cf. vjet sem and věcem in (3)), [raːt.ʦɛ] rádce → /#rāTTse#/ (cf. ráce and rád se in (2) and (3)), [rats.ʦɪ] racci → /#raTSTsi#/, and [krɛʧ.ʧiː̝̊ ] křeččí → /#křeTŠTší#/. As far as we can see, this is the only disadvantage of our analysis.

20 are analyzed as two-phoneme combinations, these combinations follow structural patterns attested for other pre-nuclear and post-nuclear combinations, which in turn simpli- fies the overall phonotactic analysis. If we analyzed the affricates as separate phonemes /c/ and /č/, we would have to set a restriction on the occurrence of /t/ and /d/ (better: on their archiphoneme /T/); we would have to say that they cannot be followed by /s/, /z/, /š/, /ž/, though the other occlusives can.

2.4 Vowels

Vowels are phonemes that occur in nuclear position only; they are not dependent on any other phonemes. Their inventory is given in Figure IV.

Front Back Central High Mid High Mid Short /i/ /e/ /a/ /u/ /o/ Long /ī/ /ē/ /ā/ /ū/ /ō/ Diphthongal /ë/ /ä/ /ö/

Figure IV: Vowels of Czech.

The Czech vowel system is traditionally sorted to high (/i/, /u/), mid (/e/, /o/) and low (/a/). The high and mid vowels are distinguished as to being either front (/i/, /e/) or back (/u/, /o/), while /a/ is central. Furthermore, a distinction between short and long is made for all vowels, though sometimes the phonemic status is denied for /ō/ due its prevalent occurrence in phonological forms of words of foreign origin. However, the existence of minimal pairs such as [loʒɛ] lože ~ [loːʒɛ] lóže proves the phonological value of /ō/ as opposed to /o/. Our analysis of the vocalic system is not in fact much different to the traditional one. The crucial difference lies in the interpretation of diphthongs [au],̯ [ɛu̯] and [ou̯] as corresponding to single phonemes, not to combinations of two. From the 15 historical perspective, the diphthong [ou̯] is a “native” sound in Czech whereas [au̯] and [ɛu̯] were introduced with the adoption of foreign words. The latter two do not oc-

15 See Studenovský – Trpák (2004) and Studenovský (2008, 2010) for recent phonetic analyses of the Czech diphthongs.

21 cur in many words, but should still be regarded as part of the sound system of Czech. The diphthong [au]̯ is attested, for example, in auto “car”, which is one of the most used words, and, moreover, there is no other word for car in Czech. The diphthong [ɛu̯] occurs, for example, in euro, which is a fairly common word, too. The problem of diphthongs was discussed in phonological literature (see Palková 1997: 196–7, cf. also Vachek 1968, Chapter III). Standpoints were various just as were the arguments for and against the monophonemic evaluation. However, it should be borne in mind that whether something is a phoneme or not must be decided on the grounds of the definition of the phoneme in the theory adopted for the description, a point already mentioned in the previous section. Arguments about the origin of the particular diphthongs, about their phonic properties and/or about morphonological alterna- tions between alleged diphthongs and simple vowels do not simply hold because the definitions provided by the theory are prior to any of these stipulations. If the Czech diphthongs are interpreted as combinations of two phonemes, they are usually viewed as combinations of /a/, /e/, /o/ with /u/, i.e. as /au/, /eu/ and /ou/. We do not hold this view. If [au],̯ [ɛu̯] and [ou̯] were to correspond to two-phoneme combinations /au/, /eu/ and /ou/, we should be able to show each element in these combinations complies with the definition of phoneme. Phonemes, being minimum syntagmatic entities, should be functionally orderable, but the alleged phoneme /u/ in /au/, /eu/ and /ou/ does not have this capacity: within the domain of one phonotagm /u/ always follows either of /a/, /e/, /o/, and hence the order of /au/, /eu/, /ou/ is predictable and not functional. The condition “within one phonotagm” is necessary because the phonotagm is the domain upon which the distribution of phonemes and phonotactic relations between them are described. This is important because the reverse combinations /ua/, /ue/, /uo/ do occur in Czech but not within one phonotagm—they occur across two phonotagms. Moreover, there are genuine combinations of /a/, /e/, /o/ and /u/ which also occur across two phonotagms; compare the examples in (4).

(4) [ʔaktu.alɪtɪ] aktuality × [na.utʃɪt] naučit [du.ɛl] duel × [nɛ.umɲɛt] neumět [vaku.oviː] vakuový × [do.utʃɪt] doučit

22 As the transcriptions hint, all of these combinations are breakable to two syllables and correspond to two phonotagms. On the other hand, the diphthongs [au],̯ [ɛu̯], [ou̯] belong to single syllables and correspond to single phonotagms. The alleged combinations /au/, /eu/, /ou/ cannot thus be confronted with the reverse combinations /ua/, /ue/, /uo/ within one phonotagm, and their order cannot be regarded functional. This is not true for other combinations with vowels, though. Let us compare the situation with the sequences [aj], [ɛj], [ɪj], [oj], [uj] (see also Vachek 1968, Chapter 3 and references therein). In this case it is not possible to analyze them as single phonemes because the order of the segments is functional. This is to say: if we assume that they correspond to combinations /aj/, /ej/, /ij/, /oj/, /uj/, we can immediately prove these are genuine combinations because we find, within one phonotagm, their permutations /ja/, /je/, /ji/, /jo/, /ju/. Furthermore, we can find within a single phonotagm the combinations /āj/, /īj/, /ōj/, /ūj/ and /jā/, /jī/, /jō/, /jū/16, but not /āu/, /ēu/ or /ōu/.17 Accordingly, the appropriate solution is to analyze [au],̯ [ɛu̯], [ou̯] as realizations of single phonemes and not of combinations of two phonemes. Note that the diphthongal vowels are here transcribed with a dieresis as /ë/, /ä/ and /ö/ rather than /eu/, /au/ and /ou/ because they are, by functional necessities, single phonemes, and there is a functional difference between the diphthongal vowels /ë/, /ä/, /ö/ and two-phoneme combinations /eu/, /au/, /ou/. Cf. /pöTšeK/, [pou̯tʃɛk] pouček “little puck” and /pouTšeK/, [po.utʃɛk] pouček “theorem (gen. pl.)”. Once the diphthongs are accepted as realizations of single phonemes, we can see there are three types of vocalic quality: [ɪ], [ɛ], [a], [u], [o] × [iː], [ɛː], [aː], [uː], [oː] × 18 [ɛu̯], [au],̯ [ou̯] . To account for these differences, we have posited a proportional opposition ‘short’ ~ ‘long’ ~ ‘diphthongal’, and interpreted the vowels as short /i/, /e/, /a/, /u/, /o/, long /ī/, /ē/, /ā/, /ū/, /ō/, and diphthongal /ë/, /ä/, /ö/. The difference between the particular vowels within these sets is accounted for by two proportional oppositions: ‘front’ ~ ‘central’ ~ ‘back’ on the one hand, and ‘high’ ~ ‘mid’ on the other. The first

16 The combinations /ēj/, /jē/ are very rare if occurrent at all. 17 /āu/ is found, but only across two phonotagms as in /nāušňiTse/ náušnice. 18 The IPA symbols do not do justice to the realizations of the Czech vowels because they are neither open nor close but rather neutral to this phonetic distinction. It is only long [iː] that is more close than short [ɪ] (see Podlipský et al. 2009). Both are high front unrounded vocoids. The vocoids [u], [uː] are high back and rounded. The vocoids [ɛ], [ɛː] are mid front unrounded, and [o], [o] are mid back rounded. The vocoids [a], [aː] are unrounded low central.

23 accounts for the difference between front /i/, /e/, /ī/, /ē/, /ë/, central /a/, /ā/, /ä/, and back /u/, /o/, /ū/, /ō/, /ö/. Fínally, to account for the difference between /i/, /ī/ and /e/, /ē/, and for /u/, /ū/ and /o/, /ō/, we have posited the second proportional opposition ‘high’ ~ ‘mid’. Accordingly, /i/, /ī/, /u/, /ū/ are high, and /e/, /ē/, /o/, /ō/ are mid. This distinction is not relevant for the diphthongal and central vowels. In conclusion, let it be noted that the names of the distinctive features are completely arbitrary. The terms high and mid could have been replaced by high and low or by close and open, but we stick to mid for /e/ and /o/ because this is the traditional label for them. However, the terms high and mid do not imply there are any low vowels. From the phonetic point of view, [a] as a realization of /a/ is low, but we cannot define it as ‘short central low’ because central vowels are always low in Czech; consequently, low cannot be a distinctive feature of /a/ as it does not distinguish anything. Likewise, the diphthongal vowels cannot be defined as either high or mid because these features are not distinctive; once again, they do not distinguish anything.

2.5 Semiconsonants

Semiconsonants are phonemes that can stand both in nuclear position and non-nuclear position. Their occurrence is dependent on the presence of at least one other phoneme. Czech has two such phonemes. The first is /l/ characterized by only one distinctive feature ‘lateral’ and realized as voiced alveolar lateral approximant. The second is /r/ characterized by a distinctive feature ‘vibrant’ and realized as voiced alveolar vibrant.19 Two types of realizations of these phonemes have been recognized, syllabic and non- syllabic. They are traditionally said to be syllabic between two contoids and at the end of a word after a contoid, and non-syllabic when occurring in the vicinity of a vocoid and at the end of a word before a contoid (Frinta 1909: 107, 115). Examples are given in Figure V. However, there have been disputes whether the variants differ in their sound characteristics. It has been argued that they do not, and that the impression of syllabicity

19 It is interesting to note that in spoken Czech /r/ is often realized as a flap (Machač 2009), and /l/ is often velarized (Volín 2002). Cf. also Machač – Skarnitzl (2009).

24 is determined by the context (Hůrková – Hlaváč 1981, Palková 1997: 232–3, cf. also Bičan 2010b). Phonologically, however, it is of little importance whether there is or there is not any phonetic difference between the two variants because conditions under which /r/, /l/ function as nuclear entities are clearly definable (see Section 7.2).

Non-syllabic Syllabic [raːno] ráno [srna̩ ] srna [lɛs] les [vl̩k] vlk [rtuc] rtuť [viːtr]̩ vítr [lscɪviː] lstivý [nɛsl]̩ nesl Figure V: The distribution of non-syllabic and syllabic realizations of /r/, /l/

Finally, let us note that the phoneme /m/ is not treated here as a semiconsonant, even though it may be realized as syllabic [m̩ ] in sedm, osm, its derivates and in a handful of place names. All of these words can be alternatively pronounced with [um], i.e. [sɛdum] instead of [sɛdm̩ ], so the syllabic pronunciation is regarded as accidental. Likewise, the occasional occurrence of syllabic [n̩] instead of [ɛn] in names like Trautenberk is viewed as accidental.

2.6 Neutralization of voicing

It was mentioned in Section 2.1 that oppositions operative in some contexts may be inoperative in others. The latter being the case, we may, under some clearly defined cir- cumstances20, speak about neutralization of an opposition. It is the commutation test upon which we decide whether and which oppositions are neutralized. The commutation test performed on the Czech data has revealed that there is a functional difference in the voicing of obstruents in the context before a vocoid and between two vocoids. To account for, we have postulated a proportional opposition ‘voiceless’ ~ ‘voiced’ pertaining to all occlusives and fricatives. However, the situation is different at

20 Certain oppositions may be actually invalid due to defective distribution of phonemes in certain contexts. The difference between neutralization and defective distribution is, however, clearly delimited in functional phonology; see Mulder (1968: 112–3, 1989: 238).

25 the end of words. Here the voicing of the word-final obstruents is predictable from the context. First, they are always voiceless if the word is terminated with a pause (cf. led “ice” and let “flight”, both realized as [lɛt]). Second, they are always voiceless if there is no pause, and if the next word begins with a phonetically voiceless sound (cf. led taje “the ice melts”, realized as [lɛt‿tajɛ]). Third, they are always voiceless if there is no pause and if the next word begins with vibrant [r] or with lateral [l] or with a nasal or with [v] (cf. led může “the ice can”, realized as [lɛt‿muːʒɛ]). Finally, they are always voiced if there is no pause, and the next word begins with a phonetically voiced stop or fricative (cf. let bude “the flight will be”, realized as [lɛd‿budɛ]).

Occlusive Fricative Nasal Labial /P/ /F/ /m/ Alveolar /T/ /S/ /n/ Palatal /Ť/ /Š/ /ň/ Velar /K/ /X/

Figure VI: Consonantal system in contexts of neutralization of voicing. Outside the system of main proportions again stand /j/ ‘approximant’ and /ř/ ‘spirant’.

Phonologically, this situation is interpreted as neutralization of voicing. This term refers to the fact that the opposition voiceless ~ voiced, which has been postulated for the pairs /p/ ~ /b/, /f/ ~ /v/, /t/ ~ /d/, /s/ ~ /z/, /ť/ ~ /ď/, /š/ ~ /ž/, /k/ ~ /g/, and /x/ ~ /h/, is not valid at the end of words. The word-final obstruents cannot be realizations of none of these phonemes due to the predictability of voicing. As fully predictable features cannot be part of phonological representation of words, the consonants occurring at the end words are phonologically neither voiceless nor voiced. Consequently, the consonantal sub-system at the end of words looks like that in Figure VI. The capital letters represent so-called archiphonemes. As explained in Section 2.1, an archiphoneme is a phoneme in a sub-system which corresponds to two or more phonemes, defined by the intersection of the distinctive features common to these phonemes. Thus, /P/ is an archiphoneme which is defined as ‘labial occlusive’ and which at the end of words represents /p/ ‘voiceless labial occlusive’ and /b/ ‘voiced labial occlusive’. In realization, it has phonetic features of these phonemes, i.e. it is a bilabial oral stop, with the voicing appropriate for a given context. The same is true for the other archiphonemes. For the sake

26 of convenience, the archiphonemes arising from neutralization of voicing will be called voicing archiphonemes. Simply said, voicing archiphonemes are occlusives and fricatives whose voicing is predictable from the context they occur in. The contexts can be summarizes as follows:

Contexts of neutralization of voicing (α) The end of a phonological form of a word.21 (β) Before any phonologically voiceless consonant. (γ) Before any phonologically voiced consonant except for /v/. (δ) Before any voicing archiphoneme. (ε) Before /ř/ if the latter is followed by a voiceless or voiced consonant with the exception of /v/. (ζ) Before /ř/ if the latter is the last consonant of a word.

The first context (α) has already been mentioned: it is the end of a phonological form of a word coinciding with the end of a phonotagm. The forms given under (5) are interpreted accordingly.22

(5) [lɛt] led → /leT/ [ʔaʃ] až → /aŠ/ [potok] potok → /potoK/

The second context (β) is the one before a voiceless consonant, and the third (γ) is the one before a voiced consonant with the exception of /v/ (see below). In Czech it holds that only phonetically voiceless obstruents can precede voiceless obstruents, and only phonetically voiced obstruents can precede voiced ones (i.e. combinations like [sp] or [zb] exist, but not [sb] or [zp]). There are several ways to interpret this situation, but we find most adequate to state that neutralization of voicing takes place before a voiceless or voiced phoneme, and so the mentioned phonetic combinations [sp] and [zb] are

21 See Section 3.3 for the definition of phonological form and word. 22 It should be noted that neutralization of voicing does not take place at the end of what we call minor-type phonotagms such as /ob/ from /obvaS/ obvaz. We will return to this in Section 6.2.

27 interpreted as /Sp/ and /Sb/, respectively. It is in accord with the fact that most of assimilatory processes are regressive (anticipatory) in Czech: when an obstruent is combined with another, the voicing of the first is assimilated to the voicing of the second. Compare the examples in (6).

(6) [ʒaːba] žába “frog” × [ʒapka] žabka “little frog” → /žāba/ × /žaPka/ [prosiː] prosí “(he) pleads” × [prozba] prosba “plea” → /prosī/ × /proSba/ [tvuːj] tvůj “your” × [dva] dva “two” → /tvūj/ × /dva/

The context “before /v/” is the only exception where neutralization of voicing does not take place. That is, the opposition between the voiceless and voiced consonants is operative before /v/. It is also operative before the nasals /m/, /n/, /ň/, and sonants /r/, /l/, /j/ and /ř/ (with two exceptions dealt with immediately). Neither of the latter phonemes is, however, phonologically voiced, but /v/ is. Although there could be alternative analyses, we think it more adequate to interpret /v/ as the voiced counterpart of /f/, both forming a pair proportional to those like /s/ ~ /z/, /x/ ~ /h/. The reasons the oppositions between the voiceless and voiced consonants is valid before /v/ seem to be partly dia- chronic, partly perhaps due to the phonetic nature of [v]. The sound [v] is usually characterized as voiced labio-dental fricative, being a counterpart of [f], voiceless labio- dental fricative. However, the friction of [v] is acoustically very small compared to other fricatives (Volín – Skarnitzl 2005).23 From the historical point of view, this peculiarity is usually explained by assuming that the v-sound was originally a labial-velar approximant [w] (Frinta 1916, Romportl 1973: 105–17, Volín – Skarnitzl 2006a, 2006b). If true, it was a sound with similar characteristics like [r], [l] and [j], before which both voiceless and voiced sounds can generally stand. However, in Modern Czech the origin of the sound is blurred because [v] loses, in morphological processes, its voicing at the end of a word and before a voiceless sound. Therefore, we find best to say that the opposition between voiceless and voiced consonants is neutralized before

23 Czech is not the only language where neutralization of voicing is operative before voiceless and voiced consonants except for /v/ before which both types of consonants can occur. In Russian (Padgett 2002), Hungarian and Slovak (Bárkányi – Kiss 2010) it behaves similarly.

28 any and all voiceless consonants and before any and all voiced consonants with an exception of /v/ before which this opposition is valid. See the examples in (7).

(7) [paːvɪ] pávi “peacocks” × [paːf] páv “peacock” → /pāvi/ × /pāF/ [sliːva] slíva “plum” × [sliːfka] slívka “little plum” → /slīva/ × /slīFka/

The fourth context of neutralization (δ) is the one before any voicing archiphoneme. From the phonetic perspective, this again derives from the fact that combinations of obstruents are either voiceless or voiced in Czech. They need not be only of two but also of three or four obstruents. Neutralization of voicing is transitive; it applies to both phonotagm-initial and phonotagm-final context, and even across phonotagm boundaries:

(8) [zaːpst] zábst → /zāPST/ [fskvjɛt] vzkvět → /FSkvjeT/ [ʔɛvropskiː] evropský → /evroPSkī/

The fifth context (ε) is the one before /ř/ if the latter is immediately followed by a consonant that is either voiceless or voiced; /v/ is an exception again. Examples follow.

(9) (a) [xrtaːn̝̊ ] chřtán → /Xřtān/ [krcɪnɪ̝̊ ] křtiny → /Křťini/ [trpɪt̝̊ ] třpyt → /TřpiT/ (b) [ɦrbɛt̝ ] hřbet → /XřbeT/

To say it in other words, in the context ‘_řO’ where ‘O’ is either a phonologically voiceless or voiced occlusive, the opposition between voiceless and voiced consonants is not valid. If the occlusive ‘O’ is voiceless, /ř/ is realized as voiceless [r̝̊], and the consonant before it is also realized as voiceless due to assimilatory processes as in (9a). For example, in /Xřtān/ the /X/ is realized as voiceless [x] because it takes over, via /ř/, the voicing of the voiceless /t/. On the other hand, if the occlusive ‘O’ is voiced, /ř/ is realized as voiced [r̝], and the preceding consonant is also realized as voiced as in (9b). For

29 example, in /XřbeT/ the /X/ is realized as voiced [ɦ] because it takes over, again via /ř/, the voicing of the voiced /b/. In any case, the phonological difference between voiceless and voiced consonants is not valid before /ř/ + occlusive. However, neutralization of voicing does not take place before /ř/ if the latter is followed by a consonant indifferent to the distinction between voiceless and voiced conso- nants24 or by /v/, which is again an exception. The examples are under (10). Thus, in /hřMňeT/ the /h/ is not an archiphoneme because it could be, at least potentially, confronted with the voiceless /x/. That the voiceless consonants can occur before /řM/ is proven by /třMen/.

(10) (a) [ɦrmɲɛt̝ ] hřmět → /hřMňeT/ [ɦrmiː̝ ] hřmí → /hřMī/ [trmɛn̝̊ ] třmen → /třMen/ (b) [brvɛ̝ ] Břve (place name) → /břve/ [pɛtrvalt̝̊ ] Petřvald (place name) → /petřvalT/

While discussing /ř/, we should return to the end of phonological forms. We have already explained that the opposition between voiceless and voiced consonants is inoperative here. The opposition is also inoperative at the end of a phonological form before word-final /ř/. This is the sixth context (ζ) for neutralization of voicing. If standing at the end of a phonological form of a word, the phoneme /ř/ is realized as voiceless [r̝̊] and for that reason only phonetically voiceless obstruents can precede it:

(11) [pɛpr]̝̊ pepř → /pePř/ [vɛpr]̝̊ vepř → /vePř/ [dovɲɪtr]̝̊ dovnitř → /dovňiTř/

24 In effect, it is only one such consonant, the archiphoneme /M/; see the next section.

30 2.7 Neutralization of place of articulation of nasals

There is one other and last neutralization process affecting the system of consonants. It involves the opposition between /m/, /n/ and /ň/. This opposition is neutralized before any consonant or semiconsonant in the phonotagm-initial context. It is a result of the fact that only [m] is found before any contoid at the beginning of a syllable. No other nasal occurs here, which means there can be no phonological difference between the individual places of articulation for the nasals because it is wholly predictable. Conse- quently, the opposition between /m/ ‘labial nasal’, /n/ ‘alveolar nasal’ and /ň/ ‘palatal nasal’ is not valid here resulting in the archiphoneme /M/ whoch is defined simply as ‘nasal’ (and always realized as [m]). Examples are given in (12). Another context where the opposition is neutralized is after /ř/: again, it is only [m] that occurs here, which makes the place of articulation again wholly predictable. Examples can be found in under (10a) above. In other contexts the opposition /m/ ~ /n/ ~ /ň/ is valid.

(12) [mlɛːko] mléko → /Mlēko/ [mstniː] mstný → /MStnī/ [smlouva̯ ] smlouva → /sMlöva/ [smrt] smrt → /sMrT/

31 Chapter Three PROLEGOMENA TO A PHONOTACTIC ANALYSIS

3.1 Phonotactic entities

On the theoretical level phonotactics is a system of phonotactic entities or, which amounts to the same, a system of syntagmatic entities in a phonological system (Mulder – Hervey 2009).1 On the descriptive level it is an analytic account of such a system. A syntagmatic entity is an entity capable of functional ordering or an entity with constituents commutable with orderable entities. In other words, there are two kinds of phonotactic entities, the minimal and maximal one. The minimal phonotactic entity is a phoneme. Phonemes are orderable, but their constituents, i.e. distinctive features, are not. The maximal phonotactic entity is a phonotactic construction. It is a self-contained combination of phonemes which are its immediate as well as ultimate constituents. The phonotactics describes both types of entity. On the one hand, it accounts for the structure and constituency of phonotactic constructions (i.e. it determines of which phonemes or phoneme types phonotactic constructions are built of), and on the other, it accounts for the distribution and combinations of phonemes within such constructions (i.e. it determines the limits of actual occurrences and mutual combinability of phonemes). By distribution of a phoneme we mean the set of its occurrences within a phonotactic construction in a given language (Mulder – Hervey op. cit.). The total set of occurrences of a phoneme cannot be in fact determined because there are potentially infi- nitely many phonotactic constructions. Yet, we can usually generalize them and talk about, for instance, the occurrence of phonemes in nuclear position or the occurrence of phonemes after /t/. To be more precise, there are two types of occurrence of a phoneme:

1 On the theory of phonotactics in general, see e.g. Sigurd (1968) and Goldsmith (2009) with references. Cf. also Basbøll (1994) who distinguishes five different foundations of phonotactics.

32 its occurrence in a position within the distributional unit and its occurrence (in fact, co- occurrence) with other phonemes of the same phonotactic construction (Mulder 1968: 223). Thus, phonemes not only occur in certain context, they also combine with each other. Formally, a combination of phonemes (or a phoneme combination) will be for us any complex of two or more phonemes. If such a combination is self-contained, we will call it a phonotactic construction.2 The difference between combinations of phonemes and phonotactic constructions rests on the latter having the property of self- containedness. Something is self-contained if it represents all relative dependencies of all of its constituents or of all of its members, i.e. of phonemes in the case of phoneme combinations (Mulder 1989: 84–7). In short, it means a phonotactic construction is a complete, autonomous and well-formed combination of phonemes. It does not require, at a given level of analysis, anything else, and can function on its own. As a rule, directly attested phonological forms of sentences are self-contained (see Section 3.4). Although the self-containedness of a combination implies its well-formedness, the reverse does not hold. A well-formed combination is not necessarily self-contained in phonotactics. A pre-nuclear combination /Tb/ is well-formed in Czech (but not in Eng- lish, for instance), but it is not self-contained because it is not an autonomous phonotactic entity. By definition, consonants are dependent on some other entity whose presence they require. In Czech this entity is either a vowel or a semiconsonant. However, this dependency is not represented in the combination in question. On the other hand, a combination /TbāT/ (cf. dbát) is well-formed and self-contained in Czech because all dependencies of its constituents, i.e. of /T/, /b/, /ā/ and /T/ are represented there: the consonants /T/ and /b/ are dependent on the vowel /a/, and so is the final /T/; the complex as a whole does not, at the level of phonotactics, require anything else. To put it simply, a combination is self-contained if it does not require any other element outside the combination in question (El-Shakfeh 1987: 94). In the next chapter it will argued that the phonotactic structure of a language is describable upon the so-called distributional unit defined as a self-contained bundle of positions. Every phoneme of a phonotactic construction is assumed to occupy a certain

2 That is, we distinguish between a construction and a combination. For Mulder – Hervey (op. cit.), these terms are synonymous, but we find useful to make this terminological difference.

33 position. A distributional unit is an abstract network of these positions. An instance of a distributional unit where all of its positions are filled with a phoneme or left empty is phonotagm (Mulder – Hervey op. cit.). From the perspective of its constituency, a phonotagm may be either a single phoneme or a phonotactic construction. The first case is encountered in languages where single phonemes can function as phonological forms of words; Czech (cf. a) or French (cf. eau “water”) are such languages. On the other hand, in some languages phonological forms always correspond to complexes of phonemes. An example is San Martín Quechua (Gabjanda 1976) where words are built of at least two phonemes. Whatever the case, phonotagms are self-contained phonotactic entities. In the majority of cases, phonotagms correspond to phonotactic constructions, but since the latter is by definition a complex of two or more phonemes, a phonotagm like /a/ (cf. Czech a) cannot be called a phonotactic construction. It is still self-contained, though.

3.2 Accent and diaereme

Since a phonotagm is a self-contained phonotactic entity, it is follows that there is no more complex phonotactic entity. If there were, the phonotagm would be dependent on such an entity and could not be self-contained. Yet phonological forms of words are seldom built of single phonotagms, at least in Czech. The words like pondělí, automo- bil, nevykonstruovali have phonological forms containing several phonotagms. How- ever, they are not only built of groups of phonotagms, but the phonotagms are in addition characterized by features tying them up, as it were, so that they constitute such and such groups and no other. The phonotagms /o/ and /na/ can, in Czech, be gathered into two groupings /o/ and /na/ (cf. o and na) or to one group /ona/ (cf. ona). Therefore, there must be some additional features involved that determine the groupment, and also the order of the grouped entities (cf. sama × masa in footnote 4 below). These features do not belong to phonotactics, but to another phonological compartment called by Mulder – Hervey (op. cit.) para-phonotactics. The function of para-phonotactics3 is to account

3 For more on para-phonotactics, see El-Shakfeh (1987) and Gardner (1985). On accent and diaereme in Czech, see also Bičan (2008b).

34 for structures and relations that cannot be adequately accounted for in phonotactics. As the phonotagm is the most complex and the maximum entity of phonotactics, para- phonotactics must necessarily account for groups of phonotagms; they consequently assume a new identity on the level of para-phonotactics. The higher-level entities are simply called para-phonotactic entities; they may be either simple or complex. Any para-phonotactic entity consists of base and para- phonotactic features. The base of a complex para-phonotactic entity corresponds to two or more para-phonotactic entities. To put it otherwise, a complex para-phonotactic entity is a group of two or more para-phonotactic entities upon which additional para- phonotactic features are superimposed. An example may be the phonological form of the sentence Na stole leželo jablko “on the table lay an apple” which corresponds to a complex para-phonotactic entity. It consists of several smaller para-phonotactic entities united together by intonation (namely of accent groups corresponding to na stole, leželo, jablko). On the other hand, the base of a simple para-phonotactic entity corresponds either to a single phonotagm (e.g. /noS/ nos) or to a combination of two or more phonotagms (e.g. /oko/ oko). As speech is linearly sequenced, it is a necessity that groups of phonotagms with para-phonotactic bases be manifested in a sequence, although they are merely juxtaposed from the phonotactic perspective. Para-phonotactic features are defined as features that accompany, but not determine, the identity of the respective base (Mulder – Hervey op. cit.). Two types of such features are recognized according to the function they fulfill: distinctive para-phonotactic features and contrastive para-phonotactic features. Within these we can differentiate between several kinds according to their realization or particular function. Distinctive para-phonotactic features are features in a relation of direct opposition with one or more para-phonotactic features. The two most common features of this kind are tones in tone languages such as Mandarin Chinese and phonological forms of intonations.4 However,

4 Other examples are features determining the placement of accentual prominence in languages with a free stress (cf. English (an) import × (to) import), and features determining the sequence of phonotagms in bases, cf. Czech /masa/ masa “mass” × /sama/ sama “alone”, which proves the sequence is distinctive. Accordingly, para-phonotactics as defined by Mulder – Hervey (op. cit.) is not necessarily the same as prosody or suprasegmental level.

35 since Czech is not a tone language, and intonation has no bearing upon a phonotactic analysis, we will not deal with distinctive para-phonotactic features any further. The function of contrastive para-phonotactic features is groupment over and above a phonotactic or para-phonotactic groupment. A phonotactic groupment is a group of two phonotactic entities, and a para-phonotactic groupment is a group of two or more para- phonotactic entities. If the base is simple, the function is trivial because the base has already its unity by being simple, but if the base is constituted by several entities, the function of contrastive para-phonotactic features is to group them into one unit. De- pending on the situation in a particular language, we can speak about two kinds of contrastive para-phonotactic features: accent and diaereme (or juncture, Bičan 2008b). The function of accent is to gather entities into higher-level groups which may be conveniently called accent groups. The function of diaereme is very similar to that of accent: it also gathers entities into a higher-level unitm but this time by virtue of indicating boundaries of the unit, which function is not inherent in accent. This higher-level entity may be called diaereme group. Each language need not differentiate these types, though it is very likely it will have some kind of contrastive para-phonotactic features. The difference between accent and diaereme is convenient particularly for Czech because accent groups do not necessarily overlap with diaereme groups in this language. As a way of illustration let us take the phonotagms /na/, /ho/ and /ře/. They may be gathered into several diaereme groups, say, to /#na#/, /#hoře#/ or /#nahoře#/.5 The function of diaereme is trivial in the case of /#na#/ as its base is a single phonotagm. The function of diaereme is, however, not trivial in the case of /#nahoře#/: first of all, its base corresponds to a conglomeration of three phonotagms grouped together to one diaereme group. More importantly, /#nahoře#/ is something else than /#na#/ plus /#hoře#/ where the same phonotagms are grouped into two diaereme groups. If the latter two are in addition united into another, a higher-level group by the features of accent, they produce the accent group /#na#hoře#/, which is a phonological form of na hoře “on the hill”, whereas /#nahoře#/ is a form of nahoře “above, up on”. In the latter the accent group and the diaereme group coincide, and it may thus be redundant to speak about accent and diaereme here, but we transcribe both for the sake of clarity. To say in a

5 Diaereme groups are put in between a pair of double crosses. Accent groups are underlined.

36 more casual manner, in Czech, accent groups may consist of several diaereme groups because certain monosyllabic words such clitics and prepositions are attached to preceding or following words with which they form accent groups. Considering the difference between /#na#hoře#/ na hoře and /#nahoře#/ nahoře, one may rightly ask how the presence of the middle diaereme in /#na#hoře#/ is justified if both na hoře and nahoře are pronounced identically as [ˈnaɦor̝ɛ]. More generally, we may query how diaeremes and accents are phonetically signaled or whether they are not just notational devices to reflect the grammatical structure. In literature there have been much debate on this topic, but we do not want to engage in it here. Suffice to say that both accent and diaereme (in fact any phonological entity or feature) must have phonetic correlates and be functional. It is because every phonological form is a class of allophones which are phonetic forms with distinctive function (see the next section). The phonetic correlates of accent groups in Czech have for a long time been a subject of research. Although the language is traditionally described as having a stress fixed on the first syllable, it has not been possible to find any stable phonetic properties of this first syllable making it different to the others. In fact, stress groups (in the phonetic sense) are in Czech rather marked by internal cohesion characterized by a certain melodic contour differentiating one stress group from another. This is in perfect agreement with our conception of accent groups (in the phonological sense): they are groups of entities gathered together into one higher-level unit. The phonetic cohesion has been investigated on the following pairs to which we have added phonological representations. The research (Palková – Volín 2003, Palková 2004) has showed that the pairs are phonetically different, and that the hearers can tell them out (cf. also Bičan 2008c).

(1) (a) světlo vnímají “they perceive light” × světlo v ní mají “they have light in it” → /#svjetlo#vňīmajī#/ × /#svjetlo#v#ňī#majī#/ (b) proti vnějším “against internal” × protivnějším “more bothering” → /#proťi#vňejšīm#/ × /#proťivňejšīm#/ (c) ještě je tele “it is still a calf” × ještě jetele “more shamrock” → /#jeŠťe#je#tele#/ × /#jeŠťe#jetele#/

37 (d) o kolovrátek “for a spinning-wheel” × okolo vrátek “around the turnstile” → /#o#kolovrāteK#/ × /#okolo#vrāteK#/

Although the melodic contour can help determine boundaries between the individual accent groups, it may not signal sharp boundaries precisely as the contour is simply continuous. Sharp boundaries may be signaled by diaereme. We thus get to the phonetic correlates of diaereme, and the difference between pairs like na hoře and nahoře which are generally pronounced alike. However, unlike the latter, na hoře may also be realized with a short pause in between na and hoře (i.e. there is a potential pause in-between). Pauses, unless they are accidental interruptions of speech, are the most obvious signals of boundaries. Quite often, too, phonemes across boundaries of phonologically forms are realized in a markedly different way than in contexts with no boundaries. This fact has been known from many languages, for instance, from English where there is a difference between pairs like a name × an aim, see them eat × see the meat etc.6 In Czech boundary-initial vowels may be realized with a glottal stop or with a period of irregular or breathy phonation as in the examples (2).7 The words in (2a–e) differ only in the presence of a glottal stop. The example (2f) is taken from Lehiste (1965) whose meas- urements proved phonetic difference here. Finally, the examples (2h–m) show that the glottal stop need not only occur at what we would characterize as word boundaries. However, in these cases the use of the glottal stop is optional (VSČ: 34–40, Zeman 2008: 74–6, Krčmová 2008: 202–3) and not much occurring (Novotná-Hůrková 1974)8.

(2) (a) [ˈsʔok] z ok “from eyes” × [ˈsok] sok “rival” → /#S#ok#/ × /#sok#/ (b) [ˈsʔuxɛm] s uchem “with the ear” × [ˈsuxɛm] suchem “dryness (instr. sg.)” → /#S#uxem#/ × /#suxem#/

6 See Jones (1931, 1956), Lehiste (1960) and Hoard (1966). Malmberg (1964) mentions similar examples from other languages. 7 Non-modal phonation in Czech has recently been investigated by Skarnitzl (2004). – On the glottal stop and its function in Czech, see also Romportl (1984). 8 Novotná-Hůrková (op. cit.: 118) has mentioned one interesting observation: if occurring between two vocoids as (2i–l), the glottal stop is less common provided that the two vocoids are articulatorily and acoustically close as in (2i) and (2j). To put it otherwise, it is more probable the glottal stop will be used in words like (2k) and (2l) than in the words like (2i) and (2j).

38 (c) [ˈkʔosaːm] k osám “to the axes” × [ˈkosaːm] kosám “scythe (dat. pl.)” → /#K#osām#/ × /#kosām#/ (d) [ˈfʔaktɛx] v aktech “in acta” × [ˈfaktɛx] faktech “fact (loc. pl.)” → /#F#aKteX#/ × /#faKteX#/ (e) [ˈpotʔokɛm] pod okem “under the eye” × [ˈpotokɛm] “through a rivulet” → /#poT#okem#/ × /#potokem#/ (f) [ˈpɛtrˈʔapoʃtol̩ ] Petr apoštol “Peter apostle” × [ˈpɛtraˈpoʃtvalɪ] Petra poštvali “they set Peter at” → /#petr#apoŠtol#/ × /#petra#poŠtvali#/ (g) [ˈnɛjʔobliːbɛːɲɛjʃiː] nejoblíbenější “most likeable” → /#nej#oblībeňejšī#/ (h) [ˈbɛsʔotkladɲɛ] bezodkladně “without delay” → /#beS#oTkladňe#/ (i) [ˈpoʔuʒɪl] použil “he used” → /#po#užil#/ (j) [ˈnaʔopak] naopak “the other way round” → /#na#opaK#/ (k) [ˈnaʔuʧɪt] naučit “to learn someone” → /#na#uTšiT#/ (l) [ˈnɛʔuːstupɲiː] neústupní “insistent (pl.)” → /#ne#ūStupňī#/

These and similar examples show that the employment of the notions accent and diaereme is not only justified but necessary for the phonological analysis of Czech. Both have a phonological function and phonetic correlates. In short, accent and diaereme are types of contrastive para-phonotactic features whose function is to group phonotactic entities into more complex entities which assume identity of their own on the para- phonotactic level. They may also group para-phonotactic entities into more complex para-phonotactic entities. The point is that para-phonotactic entities of whatever complexity are ultimately analyzed into phonotactic bases and para-phonotactic features such as accent or diaereme. The phonotactic bases may be either single phonotagms like /soK/ sok or groups of two more phonotagms /suxem/ suchem. The later are then subject to further analysis of “syllabification” (see Section 3.5). Whatever the case, the topic of our work is the analysis of phonotactic bases of para-phonotactic entities corresponding to phonological forms of signs. Before leaving the topic of accent and diaereme, it is necessary to deal with one particular issue. Let us refer back the examples (2a–d). All involve one of the so-called non-syllabic propositions, namely s “with”, z “from”, k “to” and v “in”. At the phonetic

39 level the propositions are adjoined to the following syllable without forming a syllable on their own. Thus, [ˈsʔok] s ok is one phonic syllable, though consisting of two words grammatically. The only difference between [ˈsʔok] s ok and [ˈsok] sok is the presence of a glottal stop. While in other situations the glottal stop (or its equivalents) is optional, it is obligatory in prepositional syntagms with the non-syllabic prepositions s, z, k and v before a word beginning with a vowel (VSČ: 39, Zeman 2008: 74, Krčmová 2008: 203). However, it is not a realization of any phoneme in Czech. It occurs before a vowel at a morphological boundary, and so it could be interpreted as a realization of diaereme. Consequently, /#S#/, /#K#/ and /#F#/ are separate diaereme groups; we cannot analyze them as belonging to the next phonotagm (syllable) because we would fail thus to account for the differences in (2a–d). Once this analysis is entertained, two obvious questions arise: what is the base of these diaereme groups, and what is the status of these bases? The bases are the archiphonemes /S/ for both s and z, /K/ for k and /F/ for v. That these are archiphonemes is proven by the fact that the voicing of the prepositions is always predictable from the context. Actually, they behave like any other word-final voicing archiphonemes. An in- stant proof comes from the confrontation of s “with” with z “from”: while these are two distinct prepositions, they are realized alike as [s] before a vocoid or a voiceless obstru- 9 ent or as [z] before a voiced obstruent, i.e. both have the phonological form /S/. The answer to the second question is more complicated. The bases of the diaereme groups /#S#/, /#K#/ and /#F#/ are, respectively, /S/, /K/ and /F/, but these cannot be phonotactic bases because they are not phonotactic entities. By definition, a phonotactic entity is a syntagmatic entity which means it is capable of functional ordering. How- ever, the /S/, /K/ and /F/ occurring in the bases of /#S#/, /#K#/ and /#F#/ are not capable of standing in ordering relations with any entity in phonology. Following Gardner – Hervey (1983), we propose calling such bases a-phonotactic bases10. Although the matter deserves more attention, the point of our discussion is that the phonological forms of

9 They are even sometimes confused and interchanged; cf. VSČ: 53–4, Palková (1997: 331). 10 In Bičan (2008b) we viewed them as phonotagms with empty nuclei. However, as Barry He- selwood (personal communication) pointed out, the nucleus cannot be by definition empty (see also El-Shakfeh 1987 who agrees with this). We correct our view here.

40 the prepositions s, z, k and v are hereby excluded from the domain of phonotactics, and will not be considered in any further in this work.

3.3 Signs and their types

A full phonological description should contain a phonematic, a phonotactic as a well as a para-phonotactic part. In addition, such as an analysis should be complemented with a description of allophony containing statements of realizations of phonological entities by means of their allophones. Properly speaking, allophony does not belong to phonology because the latter, as a theory of phonological systems, accounts for the systematic character of semiotic entities which have only form (i.e. of figurae) by providing a deductive classification in terms of their decomposability or combinability. On the other hand, allophony puts figurae in relation with their physical substance, and as such it is part of so-called signum theory (Mulder – Hervey 1972, Mulder 1989: 151ff.) where the ontological nature of semiotic entities is set (Hervey 1982). Phonological entities, i.e. distinctive features, phonemes, phonotagms or para-phonotactic features, are either forms of signa or constituents of these forms. To be precise, they are forms (or their constituents) of allomorphs of signa because signa are classes of allomorphs, each representing a “variant” of a signum. The signum theory is sketched in Figure I together with the correspondences between signum theory and systemology. The latter is a cover term for phonology and grammar because both center on a deductive classification of semiotic entities, phonology on classification of figurae and grammar of signa. A signum is a semiotic entity with a form and an information-value. There are two types of signa: signs and symbols, the difference between them lying in the convention- ality of their information-value. Whereas the information-value of signs is wholly fixed and conventional, this does not hold for symbols because their information-value may only be partially fixed (in the case of proper names like John, Theresa, London) or not fixed at all being purely occasional (in the case of nonce words slithy or chortle used in Lewis Carroll’s poem Jabberwocky from Through the Looking-Glass). The point is that in the case of symbols “[their] information-value is not wholly determined by fixed con-

41 ventions, but at least partly by separate definitions for each separate operation” (Mulder – Hervey 1980: 183), which means that someone or something has to decide that slithy has the meaning it has, and that John is a name of a male person or London a name of a certain city. The difference between signs and symbols is very useful (at least in Czech) because phonological forms of the latter are in several ways different to those of the former, particularly as regards phoneme combinations they contain (e.g. /Kf/ in the place name Kfely or /rSK/ in the place name Kursk). Throughout this work we will specify whether a certain combination is attested only in phonological forms of symbols. How- ever, the basis of our analysis will still be phonological forms of signs.

SYSTEMOLOGY

Grammar Phonology

para-phonotactic enti- sentences, syntagms, ties, phonotagms, pho- pleremes (words), nemes, distinctive fea- monemes tures ↕ ↕

correspondence correspondence signum (sign)

= class of allomorphs allomorph has phonological form = class of allophones allophone has phonetic form = class of images image ≈ speech-sound

Level of signum Level of figura Phonetics

SIGNUM THEORY Figure I: Signum theory and its relation to systemology

A sign is defined as a conjunction of an expression and a content. These are two aspects of sign: An expression is the converse of a content and a content the converse of an expression. They imply one another, and they were implied by the sign just as the latter implies them. These mutual implications allow us to speak about signs as if they were just their expression because their contents are automatically implied. From the perspective of its expression, a sign is defined as a self-contained class of allomorphs

42 with the same distinctive function. An allomorph of a sign is a phonological form (accounting for its formal aspect) which has a distinctive function in grammar (setting its identity). A phonological form is defined as a self-contained class of allophones where each of these allophones has the same distinctive function in phonology. The difference between distinctive function in phonology and distinctive function in grammar lies in the fact that the former is given by the set of oppositions between semiotic entities having only the formal nature whereas the latter is given by the set of oppositions between semiotic entities with both form and information-value. An allophone is a phonetic form which has a distinctive function in phonology. A phonetic form as a notion belongs to phonetics rather than to signum-theory proper. It covers simple as well as complex entities such as sounds and their combinations or their components (“sound gestures”). A phonetic form represents a generalization of some phonetic facts: it is a class of impressionistically similar images. An image is finally a model for a single speech event occurring at the given time and place. The signum theory deals with signs without specifying their complexity or analyz- ability. However, from the perspective of grammar, it is useful to distinguish between different types of signs, of various complexities. We can have minimal and maximal signs and those in-between (Mulder – Hervey 2009). The minimal ones are monemes and they are roughly the same as what is usually called morphemes. Unordered complexes of monemes are pleremes; they are at the same time minimum syntactic entities. Ordered complexes of pleremes are then syntagms. They are the maximum entities of syntax. If pleremes or syntagms are accompanied by sentence-forming features, they produce sentences. They are the maximal possible signs. We agree with Mulder and Hervey who do not regard sentences as syntactic entities of higher complexity, but as syntactic entities that are accompanied by some additional sentence-forming features. These features are most commonly intonations, but they also may be so-called sentential markers such as ka marking questions in Japanese or ma with the same function in Chi- nese (Mulder 1989: 368ff.). Accordingly, the necessary condition for sentences to be well-formed is not that they contain a predicate (“verb phrase”) and a subject (“noun phrase”), but that they contain sentence-forming features which make them sentences. In effect, sentences are the most complex signs and they are in addition self-contained

43 vehicles for conveying messages. They are linguistic models that directly apply to entire speech events, i.e. “[e]very spatio-emporally concrete speech-event is, in its entirety, a realization of at least one ‘sentence’” (Hervey 1978: 38). For instance, an event transcribed as [strom] is a realization of a sentence if it is accompanied by appropriate intonation, and if it conveys a certain message (like “to je strom [it is a tree]” as opposed to “je to strom? [is it a tree]”). In this case the base of the sentence would correspond to a single plereme with a phonological form /Strom/.11 Let us return to pleremes, the minimum syntactic entities. They are not necessarily the same as what others call words. Pleremes (being signs) are classes of allomorphs, but when speaking about words, we generally mean one specific allomorph. So, for example, we would label Czech jsem “I am”, jsi “you are”, je “he is” as words, even though they are only allomorphs of the plereme BÝT “to be”. For the sake of convenience, we will follow this practicem but without incorporating the notion word into our analysis. In fact, it is irrelevant for phonological analysis whether jsem, jsi and je are allomorphs of one plereme or separate pleremes as long as their phonological forms are different. Instead, we will employ the term phonological form for a phonological unit showing “a maximum agreement with the phonological form of certain types of word- expression” (Mulder 1968: 55). For us, a phonological word will be a continuous phonological form of an allomorph of a plereme, and we will declare continuous only those phonological forms which correspond to simple para-phonotactic entities. This is because, as explained in the previous section, the base of a simple para-phonotactic entity is either a single phonotagm such as /noS/ or a conglomeration of two or more phonotagms such as /oko/ or /koleno/ (cf. nos, oko, koleno in Czech), whereas the base of a complex para-phonotactic entity is a conglomeration of two or more other, less complex para-phonotactic entities. In Czech in particular, phonological words coincide with diaereme groups, i.e. with such phonological units which are circumscribed by diaeremes (usually realized as pauses, real or potential, or with glottal stops), and which do not contain any smaller diaereme groups. Thus, the phonological form /#beS#oTkladňe#/ of

11 This work is not going to discuss or argue with the theory of sentence. An interested reader can consult Gardner – Hervey (1983), Gardner (1985) or Hervey (1990) with references.

44 bezodkladně is not a phonological word because it is not continuous; instead, it contains two phonological words /#beS#/ and /#oTkladňe#/. Being a phonological form, a phonological word is of course a class of allophones. Each of these allophones has a phonetic form. It is another way of saying that a phonological word can be realized in one or several different ways. As we analyze only standard Czech, not its substandard varieties, the whole matter can be simplified. A characteristic feature of this variety is that its pronunciation is codified by the othoepic norm which as a rule allows only one way of how allomorphs of signs can be pronounced. Accordingly, phonological words will usually have only one phonetic form which corresponds to the orthoepically correct pronunciation. For instance, les has a phonological form /#leS#/ (= phonological word) which has only one orthoepically correct phonetic form [lɛs]. Yet some words may have several possible pronunciations, all of them orthoepically correct. An example is tramvaj with the phonological form /tramvaj/ which has two phonetic forms [tramvaj] and [traɱvaj]. As such situations are not numerous, we can assume that any Czech word has one phonological form, and that this phonological form has one phonetic form (realization).

3.4 Accidental gaps vs. structural restrictions

Although it has been necessary to define the sign and related notions, we will deal mostly with their phonological forms. Following Hervey (1978), we can distinguish several types of them before getting to the main topic of this section. First, there are directly attested phonological forms. These are phonological forms of allomorphs of signs in any given language, and are consequently directly attested in the data, so to speak. In principle, we can list all directly attested phonological forms and make a phonological description of them. These forms represent the basis for our analysis and should be considered in its total. As, in fact, the only directly attested forms can be those of allomorphs of sentences (see the previous section), all other phonological forms are indirectly attested. However, for the sake of simplicity, we will also regard phonological forms of allomorphs of pleremes directly attested because they can

45 function as bases of sentences (Gardner – Hervey 1983). Thus, phonological forms of words listed in dictionaries of standard Czech are directed attested phonological forms. Second, there are indirectly attested phonological forms. These are functional components of directly attested phonological forms. The forms of allomorphs are seldom simple—they are generally complexes of several phonological entities. For example, the phonological form /#leS#/ of Czech les is built of the phonotagm /leS/ and the respective para-phonotactic features. The phonotagm /leS/ is built of the components (phonemes) /l/, /e/ and /S/; these phonemes are ultimately built of bundles of distinctive features. However, all of these are merely analytical components of directly attested phonological forms, and are consequently only indirectly attested by their courtesy. Finally, we recognize potential phonological forms. These forms are neither directly attested as phonological forms of allomorphs nor indirectly attested as their analytical components. Nevertheless, they are still regarded as phonologically well-formed in a particular language. They are established by implication, by a method of extrapolation, which is “calculation from known terms of a series of other terms which lie outside the range of known terms” (Hervey 1978: 41, fn. 4). It is calculation of potential phonological forms from the set of directly and indirectly attested phonological forms. We will not go here into details because the rationale is well explained in Hervey’s paper. To put it otherwise, potential phonological forms are those which are well-formed, and which, should there ever be such a need, could function as forms of signs or of their analytical components. Let us compare two unattested phoneme combinations in Czech: /luS/ and /lluS/. The former is regarded as a well-formed phonotagm because it does not violate any structural rule; the fact it is not attested as a form of any sign is merely accidental (cf. /kluS/ attested in klus). The same cannot be claimed about the latter because it has a structure not allowed in this language. In particular, it violates a structural rule otherwise obeyed in all attested phonological forms in Czech: that no two or more identical phonemes can stand in close proximity within a single phonotagm. A question may now be rightly asked: Given a certain non-attested phoneme combination, is there a criterion of by which we can decide whether it is well-formed but accidentally missing or whether it is not well-formed and its non-occurrence is systematic? This problem has been mentioned by several linguists, though few of them have

46 tried to provide an answer (cf. the discussions in Fischer-Jørgensen 1952, Saporta – Ol- son 1952, Spang-Hanssen 1958, 1959, Vestergaard 1967, Mulder 1968 or Hervey 1978).12 Does non-attestedness of a combination imply its impossibility? No reasonable linguist would probably hold such a strong view. At the same time, however, no linguist would hold the opposite strong view, i.e. that every non-attested combination is a possible in a given language. If we reject both extremes, it must mean that there are some combinations which are missing accidentally (i.e. which are well-formed) and some which are missing systematically (i.e. which are ill-formed). The question now is: can we set a borderline between accidental non-occurrence and systematic non-occurrence of something? Fischer-Jørgensen (1952: 33) in agreement with Spang-Hanssen (1958 and 1959, Chapter 1) thinks that “it is theoretically impossible to fix a non-arbitrary borderline between law and accident. Laws may be stated as deviations from accidental distributions; and there are many degrees of deviation”. While there is some reason in their arguments, we think it is not only possible but above all desirable to have criteria by which some non-occurrences could be declared well-formed or ill-formed. We propose the following criterion which is based on the principle that a hypothesis is corroborated as long as it is not refuted:

Criterion of well-formedness A certain combination of phonemes is well-formed (i.e. possible) as long as it does not violate any structural rule established in the description.

As it is, the criterion does not, however, solve our problem. It just says that accidental is what cannot be considered systematic.13 The problem we now face is the delimita- tion of what a structural rule is. We proposed the following definition based on Vogt (1954: 31)14.

12 See also Bybee (2001: 89–93) on various experiments testing acceptability of non-occurrent phoneme combinations in English. 13 It is actually the opposite of Spang-Hanssen’s view because for him systematic is “what cannot reasonably be considered accidental” (1959: 26–5). 14 Vogt talked about distribution of phonemes, but we make the definition more general, i.e. we talk about entities and features. The latter may include positions of the distributional unit.

47 Definition of structural rule A structural rule is a statement concerning the distribution of entities in which at least one of the terms brought into relation by the statement is a class of entities or features. This class must be clearly and unambiguously defined.

The motivation behind this definition is obvious: as Mulder (1968: 198) noted “[n]o theory should include statements which are not simpler or more general than the corresponding statements of fact”. Any statement, if it is to be a structural rule, must generalize. Consequently, a statement like “/f/ does not occur before /ř/ in Czech” cannot be a structural rule because it does not generalize anything. On the other hand, a statement like “the semiconsonants /r/ and /l/ do not occur before /ř/ in Czech” can be a structural rule because it is a generalization—it concerns a well-defined class of phonemes. One may object that if the statement “/f/ does not occur before /ř/ in Czech” is not a structural rule because it does not concern a class of phonemes, then the structural rule must be a statement like “/f/ and /g/ do not occur before /ř/ in Czech” because here a class of phonemes is involved. However, this class is not unambiguously defined. There is no property the two phonemes have in common, something that would differentiate them from the other phonemes.15 On the contrary, the class of /r/ and /l/ is unambiguously defined because these are all members of the class of semiconsonants, and the rule applies to all of them. With the definition of a structural rule we now have a criterion for telling out whether something is systematically or accidentally missing. If no structural rule can be established other than an ad hoc statement about the non-occurrence in question, it is accidental (cf. Mulder 1968: 233). It is possible we have missed some important generalization, but until it is formulated, the non-occurrence is regarded accidental. Conse- quently, the missing combinations like /fř/ and /gř/ are regarded well-formed in Czech. The point is to demonstrate they are not.16

15 That these phonemes are sometimes regarded as marginal phonemes or phonemes of foreign origin is a pseudo-property. There is no way of proving it synchronically. 16 Finally, it should be remarked that for us something is a structural rule if it has no exceptions. For instance, Fudge (1969) mentions several restrictions limiting the combinability of phonemes

48 3.5 “Syllabication”

An analysis of the phonotactic structure must account for the constituency of all phonotagms. It is obvious that in order to achieve this, we need the phonotagms to be analyzed. We start with directly attested phonological forms corresponding to simple para- phonotactic entities, i.e. with what we have called phonological words. Their bases may correspond either to single phonotagms or to conglomerations of two or more phonotagms. Ignoring para-phonotactic features, we can say in a simplified manner that a phonological word is either mono-phonotagmic if it consists of a single phonotagm, or poly-phonotagmic if it consists of two or more phonotagms (cf. El-Shakfeh 1987: 95). There are two questions to consider now. First, on what ground do we decide whether a phonological word is mono-phonotagmic or poly-phonotagmic? Second, if it is poly- phonotagmic, how is it parsed to the individual phonotagms? The answer to the first question may sound banal at first: a phonological word consists of one phonotagm if it does not consist of two or more phonotagms. Since a phonotagm is a self-contained phonotactic entity, it follows that a phonological word ‘α’ can be divided to two phonotagms ‘βγ’ if and only if both ‘β’ and ‘γ’ are self-contained phonotactic entities. Now, something is a self-contained phonotactic entity if it is either directly attested as a phonotactic base of a word or if it does not violate structural rules otherwise valid for directly attested phonotactic entities. The first case can be illustrated on /jelen/ which is a phonological form of jelen. It can be divided to two phonotagms /je/ and /len/ (or /jel/ and /len/, see below) both of which are directly attested—cf. je and len (or jel and len). This proves /jelen/ is pluri-phonotagmic. The second case can be demonstrated on /klavīr/, the phonological form of klavír. This form can be divided to /kla/ and /vīr/17 where /kla/ is not directly attested, but /vīr/ is—cf. vír. However, despite its non-attestednesss, the latter form may be regarded to be a well-formed and self- contained phonotagm because it does not violate any structural rule. In other words, /kla/ is a potential phonological form in the sense it could be a well-formed phonologi- in English, but he immediately lists exceptions for some of them. Though we do not deny validity of such statements, they are not structural rules in our sense. 17 Or to /klaF/ and /vīr/ with functional amalgamation (to be explained below), but we exclude this possibility for the sake of argument.

49 cal word in Czech (i.e. the language could have a plereme with such a form). On the other hand, a form form like /jel/ cannot be further divided to two or more phonotagms, even though /je/ is directly attested (cf. je “he is”), because the remainder /l/ is not a well-formed phonotagm in Czech. Once we have determined whether a phonological word contains two or more phonotagms, it remains to answer the second question asked above: if a phonological word is poly-phonotagmic, how is it parsed to individual phonotagms? Had we used the term syllable instead of phonotagm, the question could be rephrased as follows: how do we syllabify phonological words? However, we do not use syllable in any phonological sense in this work (cf. Section 4.6), hence syllabification has been put into quotation marks in the title of this section. The process of syllabification, as it is usually understood, need not produce the same results as the process of division of poly-phonotagmic phonological words into phonotagms because a phonotagm need not be necessarily coextensive with a syllable. Certain ideas are nevertheless common to both procedures. A lot of literature has been devoted to the problem of syllabification, but we do not want deal with or solve all problems connected with it18. For our purposes, it will suffice to say that the main goal of syllabification can be summed up as follows (cf. Kahn 1976: 35–9): Given a phonological word consisting of a certain string of phonemes, syllabification is to divide the string into several well-formed sub-strings (i.e. syllables) so that every phoneme of the string belongs to some sub-string, and no two phonemes belong to two or more sub-strings. Though it seems clear and obvious, achieving this has not always proven so easy.19 If a phonological word is to be parsed, there must be criteria according which we decide where we set boundaries between the sub-strings. Vari- ous strategies have been offered, but it is generally agreed that the output of syllabifica-

18 See e.g. Pulgram (1970), Clements (1990), Blevins (1995), Goldsmith (2009). 19 There are moreover theoretical disagreements about the output of syllabification. First, it has been claimed that when phonological forms are parsed, not each and every segment must necessarily belong to some syllable. Such segments have been called extrasyllabic (cf. e.g. Clements – Keyser 1983; for arguments against extrasyllabicity see Hall 2002). We reject the idea of extrasyllabicity because extrasyllabic phonemes are not by assumption self-contained phonotactic entities. Second, it has been proposed that some segments may belong to both of the neighboring syllables in which case they are ambisyllabic. We will return to this notion in a moment.

50 tion should only be well-formed sub-strings. In particular, a condition sometimes called the Kuryłowicz condition is, explicitly or implicitly, invoked (see Bell 1976: 255):20

The Kuryłowicz condition If an intersyllabic sequence of consonants is analyzable into permissible word- initial and word-final clusters, then the syllable boundary does not fall between non-permissible clusters.

If translated into our terminology, this condition says that we should prefer a division which produces combinations of phonemes attested at the beginning and at the end of a word. First of all, we can safely assume that word boundaries are coextensive with phonotagm boundaries, that is to say, the beginning of a phonological word is also the beginning of a phonotagm, and the end of a phonological word is also the end of a phonotagm (cf. Pulgram 1970). Secondly, due to their direct attestedness, the combinations of phonemes found in these contexts are necessarily well-formed, i.e. permissible. Hence, if we parse a phonological word like /riPka/ rybka, the only thinkable division is the one into /riP/ and /ka/. If it were /riPk/ + /a/ or /ri/ + /Pka/, we would get non-permissible combinations of consonants in both cases—final or initial /Pk/, neither attested nor possible in Czech. It should be mentioned that if a combination is directly attested, it is of course well-formed, but a combination is also well-formed if it does not violate any structural rule. Thus, when parsing /naTxnöT/ nadchnout, the only possible division is the one into /naT/ and /xnöT/, even though /xn/ is not attested at the beginning of any Czech word. However, no rule can be stated to prevent the existence of such a combination (cf. attested /hn/, /xm/, /xň/ in hnout se, chmury, chňapnout). In some situations, however, the Kuryłowicz condition falls short. One21 can be encountered when the phonological word /jelen/ jelen is to be parsed because there are two divisions possible. The first is to /jel/ and /en/, and the second to /je/ and /len/, both being in accord with the Kuryłowicz condition. We do not of course have any consonant clusters here, but the point is whether either /jel/ or /je/ is a permissible word-final pho-

20 After Kuryłowicz (1948) who first exploited it systematically, though, as he notes, the idea has been known for a long time. 21 Another situation will be discussed in Chapter 6.

51 notagm, and whether either /len/ or /en/ is a permissible word-initial phonotagm. All of these are permissible by being directly attested (cf. je, jel, len and en (a name of a letter n), respectively). The medial /l/ could thus belong to both phonotagms, and if it is to be assigned to one or the other, the choice between two mentioned divisions, if a choice is to be made, must be conditioned by criteria other than the Kuryłowicz condition. Linguists have approached this problem variously (see the discussion in Bell 1976). Many have proposed that the division /je–len/ should be preferred because it yields an open syllable /je/ instead of a closed one /jel/ in the case of /jel–en/. This preference for open syllables has been supported by a claim that open syllables are more common than the closed ones in the world’s languages (Pulgram 1970: 66ff.).22 In a more general vein it has been proposed that syllabification should conform to the principle known as Maximal Syllable Onset Principle (Selkirk 1982) stating that phonemes or phoneme combinations between two syllables should be assigned to the onset of the second syllable rather than to the coda of the first syllable provided that the assignment to the onset is allowed by the phonotactics of a particular language. However, others have not gone so far to assume the universal validity of open syllables or the superiorness of syllable onsets, and as an alternative they have proposed that the division that to be preferred is the one whose outputs are syllables which are statistically more common in a given language (e.g. O’Connor – Trim 1953). Since this method was applied to Czech by Kučera (1961: 81ff.), we can say that the phonological word /jelen/ is to be divided to /je/ and /len/ because syllable types ‘CV’ and ‘CVC’ are claimed to be more common in this language than types ‘CVC’ and ‘VC’. Be it as it may, what decides where the border falls are criteria other than distributional. Yet it is obvious that the /l/ in /jelen/ can belong to both phonotagms as it can occur at the end of the first as well as at the beginning of the second (cf. jel and len). This brings us to another solution—the one which declares that the medial /l/ belongs to both phonotagms. From the perspective of distribution, it is in fact immaterial whether /jelen/ is to be divided to /je–len/ or /jel–en/, as either is possible. Hence, just as a wall between two rooms belongs in fact to both rooms, so the /l/ belongs at the same time to

22 Some (e.g. Jakobson – Halle 1956, Malmberg 1967, Clements – Keyser 1983) even claim that open syllables exist in all languages, but this is a pure speculation; it all depends on how the syllable is defined. Such statements may in fact be unfalsifiable (cf. Hyman 2011).

52 both phonotagms in this form. Such a situation has been called ambisyllabicity (Hockett 1955, Kahn 1976). Following Mulder (1968: 178ff.), we prefer calling it functional amalgamation. Of course, the problem is a little more complex than we present it here, there being no agreement whether ambisyllabicity is to be allowed into a phonological analysis (see e.g. Jensen 2000), but from a purely distributional perspective, the concept is justified in a phonotactic analysis such as ours. It is because our foremost goal here is to describe the structure of phonotagms in Czech. As there is no a priori reason why we should prefer the division /je–len/ over /jel–en/, it is completely irrelevant whether the medial /l/ belongs to the first phonotagm or to the second one. Therefore, we prefer dividing /jelen/ to /jel/ and /len/ and saying that /l/ belongs to both phonotagms, i.e. it is an instance of functional amalgamation. Whether on some other point in the analysis we select a criterion (e.g. the Maximal Syllable Onset Principle) which would prefer the division /jel–en/ over /je–len/ or vice versa is another matter. In general, every single inter-nuclear peripheral phoneme is functionally amalgamated in Czech because every peripheral phoneme can occur both at the beginning and at the end of a phonotagm, hence the phonological words /lano/ lano, /jaro/ jaro, /řījen/ říjen are divided to /lan/ and /no/, to /jar/ and /ro/, and to /řīj/ and /jen/, respectively. In the case of inter-nuclear occlusives and fricatives, there is functional amalgamation between an archiphoneme and the respective occlusive or fricative. Thus, /kňiha/ is parsed to /kňiX/ and /ha/ with functional amalgamation between /X/ and /h/23. It should be remembered that archiphonemes are defined as phonemes in a sub-system representing two or more phonemes of the overall system. In this situation the archiphoneme /X/ represents, in the first phonotagm, the phoneme /h/.24 Other examples are /laso/ laso divided to /laS/ and /so/, /koza/ koza to /koS/ and /za/ or /ďeťi/ děti to /ďeŤ/ and /ťi/. Now, if all single inter-nuclear phonemes are functionally amalgamated in Czech, inter- nuclear combinations may or may not be amalgamated. It depends on whether they are allowed both at the beginning and at the end of a phonotagm. For instance, in the already mentioned form /riPka/ there is not functional amalgamation because /Pk/ cannot occur at the beginning or at the end of a phonotagm. On the other hand, /Sk/ in /miSka/

23 On this cf. also Hervey (1978: 51ff.), El-Shakfeh (1987: 97ff.). 24 Cf. also the process /kňiha/ kniha “book (nom. sg.)” → /kňiX/ knih “book (gen. pl.)”.

53 miska is functionally amalgamated because the combination /SK/ can occur at the end of a phonotagm (cf. /voSK/ vosk) as well as at its beginning (cf. /Skāla/ skála). To sum it up: a phonological word is poly-phonotagmic if it can be divided to two or more well-formed phonotagms. A phonotagm is well-formed if it does not violate any structural rule. The well-formedness is judged against attested phonological forms of words, hence the division of phonological words into individual phonotagms is governed by the condition that only those phonotagms possible at the beginning and at the end of phonological words (where the boundaries are certain) are allowed. In case a phoneme or a phoneme combination can be assigned to both of two neighboring phonotagms, they are instances of functional amalgamation, i.e. they belong to both phonotagms. We have not, as is commonly done in works on syllabification, sought for criteria for the decision whether an inter-nuclear phoneme should be unequivocally assigned to the first or to the second phonotagm because this is not relevant for our analysis. What is important is whether it can belong to either or both.

54 Chapter Four DISTRIBUTIONAL UNIT: THEORETICAL BACKGROUND

4.1 Distributional unit as a domain of distribution

If a linguist wants to describe the distribution of some linguistic items, he has to do it in reference to a certain distributional domain upon which this could be related (already pointed out by Trubetzkoy 1939: 224). Without wishing to provide a detailed overview here, we content ourselves with saying that, in order to account for the distribution of phonemes, linguists have employed several such domains like a morpheme (e.g. Trubetzkoy 1939, Uhlenbeck 1950), a word (e.g. Mathesius 1929, Vachek 1940, Marti- net 1956), or a contour (Garvin 1948)1 and perhaps even others. Still most linguists seem to agree that the most economical phonological distributional domain is the syllable (e.g. Pike 1947, Fischer-Jørgensen 1952, O’Connor – Trim 1953, Haugen 1956b, Selkirk 1982, Zec 2007 and others). However, this idea has occasionally been challenged by linguists who have pointed to inadequacies of the syllable-based phonotactics (Bell 1976, Stediade 1999 and in particular Blevins 2003).2 The major argument is that in some languages phonological sequencing generalizations cannot be adequately accounted for upon the frame of a single syllable. Blevins (op. cit.) discusses a number of such languages. An ample example may be provided by Australian languages. Accord- ing to Dixon (1980: 159), words are typically at least disyllabic here, and have either the structure ‘C1VC2C3V(C4)’ or ‘C1VC5V(C4)’. He further specifies (ibid.):

1 Cf. also Sigurd (1965: 14–6; 1968). Garvin’s contour is a unit containing one stress; it may consist of one, two or even more words in Kutenai, an isolated language spoken in Canada (see also Haugen 1956a who uses the syllable for the description of this language). 2 Also e.g. Fudge (1969: 266): “the syllable is a NECESSARY unit, not an ALL-SUFFICIENT one” (it is an allusion to Kohler 1966 who denied any usefulness of the syllable in phonology).

55 “It is not thus possible, for an Australian language, to give a structure C1V(C2) for syllables, and then to describe a word as a sequence of these syllables. The possi-

bilities at C3 may be similar to those at C1, but they never coincide; similar re-

marks apply to C2 and C4. And the occasional occurrence at C5 of phonemes that are found in no other position further discredits a monosyllabic model.”

Thus, the domain of phoneme distribution must be more complex than one syllable here. Other languages such as Japanese, Italian or even English (though for other reasons) may require other distributional domain than the syllable. Another language where the syllable fails is Gokana of a Niger-Congo branch (Hyman 2011).3 This necessary preamble has brought us to one of the reasons why our description of the phonotactics of Czech is based not on the syllable, but on the concept known as distributional unit originally introduced by Mulder (1968) who was one of the few to realize the limits of the syllable-based phonotactics. Its definition unit runs as follows (adapted from Mulder – Hervey 2009):

Definition of distributional unit Distributional unit is a self-contained bundle of positions in phonotactics, where positions are divisions within a phonotactic construction, such that in every such division an entity, as an immediate constituent of that construction, can stand and alternate (i.e. commute) with other entities, or with ∅ (i.e. be left out).

The distributional unit is meant to be a model of the phonotactic aspect of languages. In many it may be coextensive with a syllable as traditionally understood, but others require distributional units of larger complexity because, as we have just mentioned, the distribution and combinations of phonemes cannot be exhaustively accounted for on the domain of a single syllable there. This is the case of Czech, even though in the majority

3 Accordingly to Hyman (op. cit.), the syllable is hardly used in Gokana in any way. To use his terminology, it is not a domain of distributional constraints on phonemes, of phonological, morphological or allomorphic rules, of prosodies or of prosodic grouping. Though Gokana has certainly phonic syllables, phonologically there is very little need to operate with such a unit. It is, however, possible to apply the distributional unit to this language.

56 of cases the distributional unit is coextensive with the phonic syllable here. Yet we are convinced of the advantages it has in comparison with other models as we hope to show in our work. The distributional unit is set up in such as way so as to ensure that phonotactic descriptions are relatively simple, adequate to the data and maximally exhaustive. The positions, of which it is a self-contained bundle, their functions and quantity are postulated for each language individually to reflect the fact that phonemes in every language display different distributional and combinational potentials.

4.2 Positions

The notion position is crucial for understanding what a distributional unit is. It may be viewed as a slot that can be filled with or occupied by a phoneme, one at a time, or it can be empty, depending on a situation. No single instance of a phoneme can occur in more than one position at a time, even though several instances of one phoneme can occur in more positions like in Czech /jej/ jej where /j/ occurs in a pre-nuclear and post- nuclear position. However, these are two instances of the same phoneme in different positions, not one phoneme in two positions. Actually, an alternative definition of the phoneme is “a simultaneous bundle of distinctive features in phonology which does not extend over more than one position” (Mulder 1968: 26). It is like a person who can only sit on one chair, not on two, even though he may sit on several chairs at different times. We have said that positions may also be empty; in this work we will mark this situation by the symbol ‘∅’. However, the zero is here not an entity of any kind, but it is what Haas (1957)4 has called operation of omission: a position is empty if a phoneme occurring there in a certain phonotactic construction can be omitted, so that the result is still a self-contained construction. For instance, the position in which the final /T/ occurs in /jeT/, a phonological form of Czech jet, can be empty because its omission produces /je/ which is well-formed (cf. the word je). But the same cannot, however, be said about /e/ because its omission produces /jT/ which not well-formed in Czech. Returning

4 Incidentally, Haas may not agree with our use of empty positions, explicitly rejecting them (op. cit.: 48), but then, his understanding of positions may differ to ours.

57 to our chair simile, we could say that even though a person may sit upon a chair, it happens that the chair is sometimes empty, but it is a chair nonetheless. In phonology it is convenient to regard the omission of a phoneme from a position as it were occupied by ∅ (cf. Mulder 1968: 126–7). A position is occupied by it (i.e. is empty) if it is not occupied by a phoneme, but it holds that every position which happens to be empty in a certain situation must be occupied by at least one phoneme under different circumstances. In other words, there must exist a choice between the presence of a phoneme and its absence. Positions which are always empty are prohibited because they would hardly have any purpose. The idea of empty positions has one important consequence: once the distributional unit is set up in terms of the positions it consists of, the number of positions remains constant. What changes is the way positions are occupied by phonemes or left empty. It is this particular aspect of Mulder’s conception of position, and ultimately his idea of the distributional unit, that makes it different to the way other linguists use this notion.5 The term position has been widely used in phonology (see e.g. Trubetzkoy 1939, Fischer-Jørgensen 1952, Haugen 1956b or Kučera 1961 analyzing Czech), though seldom specifically defined. In general, it stands for a certain relative placement of a phoneme, mostly with regard to the beginning of a syllable or a word or to the peak of a syllable. Hence e.g. Czech /v/ in /vaS/ vaz and /k/ in /kaS/ kaz are said to occur in the first position of a syllable. This is not true under our approach. Although it is not denied that /v/ and /k/ are the very first phonemes in the forms given, the subsequent analysis will have shown that they do not occur in one and the same position. Due to the existence of forms like /kvaS/, they must occur in different positions because it is not possible that they occur in the same position in /kvaS/. The forms /kaS/ and /vaS/ differ in the way particular positions of the distributional unit are filled with a phoneme or left empty. Strictly speaking, the phonotactic representation of the words kaz and vaz is /∅∅k∅∅a∅S∅/ and /∅∅∅v∅a∅S∅/. The zeros are for practical reasons usually not written, but they are indicators that the given positions can be occupied by phonemes

5 Among the few linguists who also operate with empty positions is Fudge (1969): “[English] sixths [is represented] as /s.i.kSθS/, where the symbol . indicates the selection of zero” (p. 269).

58 under other circumstances; cf. /FSk∅∅a∅S∅/ vzkaz, for example, where the first two zeros of /∅∅k∅∅a∅S∅/ kaz are replaced by /F/ and /S/. The purpose of the distributional unit is to account for the structure of phonotactic entities. As explained in Section 3.1, there are two types of them: either single phonemes like /a/ (cf. the word a in Czech) or phonotactic constructions like /kuŠ/ (cf. the word kuš) consisting of three phonemes. The occurrence of every phoneme is describable in two ways: first, in terms of its co-occurrence with other phonemes in a phonotactic construction (or in terms of their absence as in /a/), and second, in terms of its occurrence in a position of the distributional unit where it commutes with other phonemes. So, the phoneme /k/ does not only co-occur with /u/ and /Š/ in /kuŠ/, but also stands in a certain position of the distributional unit just as /u/ and /Š/ stands in certain, but other positions. In this position /k/ is commutable with other phonemes, e.g. with /t/ as in /tuŠ/ tuš, or it can be completely left out as in /uŠ/ už. It is assumed that every phoneme is assignable to at least one position of the distributional unit, and that the number of the positions is finite. The distributional unit is then an underlying model behind all phonotactic constructions in a given language, and its positions are expressions of distributional and combinational properties of particular phonemes. However, it is not a mere taxonomic tool without any wider application, as it may seem at the first sight. It is a common and wrong belief that structuralist models are not capable of explaining anything, merely describing it. The distributional unit formalizes phonotactic descriptions, presents them in a clear and systematic way, and besides that, it functions as a formal device for predicting (“generating” or still better, for calcu- lating because we do not guess) well-formed but unattested phonotactic constructions. To put it otherwise, the model also accounts for potential phonotactic constructions which are not actually attested, but which are nevertheless well-formed in the sense they could function as full-fledged phonological forms (or their parts) of words in a language if ever there is such a need. Linguists are widely agreed that a phonological theory should consider the question why some combinations of phonemes are permitted and others not (see e.g. Fischer-Jørgensen 1952, Vogt 1954, Hooper 1972, Selkirk 1982, Goldsmith 2009), but, as mentioned in Section 3.4, only a handful of them have provided a detailed procedure for answering this. Under the present approach the distribu-

59 tional unit offers a key to such an answer: since it captures the phonotactic structure of all attested phonotactic constructions of a given language, it follows that those unattested phonotactic constructions which do not conform to it cannot be considered well- formed. An example can be a confrontation of two constructions /tvje/ and /pvje/, both unattested in Czech. The first of them is well-formed, but the second is not, even though both are of similar constituency. As becomes clear in the next chapter, /p/ and /v/ are mutually exclusive, and cannot combine with one another. The structure of the distributional unit is such that its output can be /tvje/, but never /pvje/, which makes the latter ill-formed. It thus offers a principled basis for distinguishing accidental gaps from non- accidental gaps (cf. Chomsky 1964: 31). Having said this, it should be obvious that functionalist models cannot be accused of lacking predictive power, and are thus on a par with the other, mainstream linguistic approaches. In order for it to be applicable at all, the distributional unit must comprise at least two positions, but there is no restriction on their maximal number as this depends on particular languages. A distributional unit with one position is useless because all items would then occur in that position, and no distributional properties could be derived from it, so at least two positions are necessary. However, languages where a distributional unit comprises only two positions appear to be very rare. As there is usually a certain degree of isomorphism between the distributional unit and syllables, it is reasonable to suppose that a language where syllables have always the structure ‘(C)V’ or ‘V(C)’, the distributional unit most likely consists of only two positions there, one for the phoneme realized as a vocoid, the other for the phoneme realized as a contoid. While ‘VC’-only languages appear to be very rare (Arrernte, an Aboriginal language of Australia, is claimed to be one; Breen – Pensalfini 1999, Tabain – Breen – Butcher 2004),6 several languages have been cited to be the ‘(C)V’-only ones: Hua, a Papuan language, and Cayuvava, a Boliv- ian language (see Blevins 1995). In the latter the only attested syllable types are said to

6 This is a phonological interpretation. Phonetically (“on the surface”), Arrernte does not only have only ‘VC’ syllables because it has words beginning with a contoid and ending in a vocoid, though there are good reasons to interpret the final vocoids as non-phonological, and to say that all contoid-beginning words actually begin with the vowel /e/.

60 be ‘V’ and ‘CV’.7 Another example cited is a Senari dialect of Senufo, a branch of Ni- ger-Congo languages (cited e.g. by Clements – Keyser 1983: 29 without a source). Ac- cording to the analysis by Mills (1984), the language contains only the syllables ‘V’, ‘C’ (i.e. a syllabic contoid), ‘CV’ and ‘CRV’ (i.e. containing a contoid with a secondary release). Although additional research is still needed, the data suggests that these languages could have distributional units with merely two positions, one nuclear and the other peripheral. More common are probably languages with three positions. Rotokas, a Papuan language often cited as possessing the smallest phoneme inventory, may have the distributional unit of no more than three positions because the only attested syllables are of the structure ‘V’, ‘VV’, ‘CV’ and ‘CVV’ (Firchow – Firchow 1969). The ‘VV’ and ‘CVV’ syllables contain long vowels and diphthongs; more research is needed for deciding whether they correspond to combinations of two phonemes or to single phonemes. The latter being the case, the Rotokas distributional unit would probably have only two positions. Another example may be Hawaiian with the syllable structure similar to Rotokas and also having diphthongs; in this language the distributional unit appears to have three positions (a conclusion based on Schütz 1981). In general, more than three positions are required for describing the phonotactic structure of the world’s languages, simply because many languages allow consonantal combinations before or after a vowel. Pekingese Chinese, Thai and San Martín Quechua have four positions (according to Mulder 1968, 1987, and Howkins 1972, respectively). Sudanese Arabic has five positions (according to Dickins 2007), and so does Yulu (according to Gabjanda 1976). The distributional unit in Russian has eight and in French nine (according to Rastall 1993). As we will show, Czech can be analyzed with nine positions. The main distributional unit in English has ten positions (according to El- Shakfeh 1987). More complex distributional units may still occur, e.g. in Georgian which is known to have a very rich and complex structure of pre-nuclear and post- nuclear combinations (Vogt 1958, Butskhrikidze 2002).

7 The source from which Blevins drew her information about Hua is Haiman (1980). However, our inspection of that work has shown the language has also other syllable types (of the same opinion is also Zec 2007: 169, 192). The source for Cayuvava is Key (1961) which confirms her assertion, though Key mentions that the ‘CCV’ syllables also occur in the language.

61 Phonemes within phonotactic constructions are, during the analysis, are assigned to particular positions according to phonotactic properties they possess; a position is an expression of these properties. However, some phonemes may display unique phonotactic properties, and it will not possible to assign them to any of the already established positions without violating the consistency of the description. In that case it will usually be necessary to introduce a so-called archi-position (Mulder 1968). It accounts for the suspension of a syntagmatic difference between two or more adjacent positions, ac- commodating phonemes with unusual distributions into the distributional unit. As positions in phonology are sequenced in a linear order, an archi-position can only be a product of the suspension of the difference between two or more immediately adjacent positions. Formally, it is the intersection of these positions possessing only those phonotactic properties common to them. We will return to archi-positions in the next chapter.

4.3 Functional dependency: nuclear and peripheral entities

It has been said that phonemes constitute phonotactic constructions. Yet phonotactic constructions are not just complexes of phonemes; they are complexes of phonemes between which certain relations hold. Such constructional relations are called tactic relations (because phonemes are phonotactic entities). Several types are recognized, but the most important one for phonology is relation of subordination. It is an asymmetrical relation between two entities such that one entity is functionally dependent on another, which means that it is dependent on the other for the role it fulfills in a phonotactic construction. The dependent entity is sub-ordinate to the governing one which is, consequently, is declared to be the nucleus (or nuclear entity) of the construction while the dependent one is given the status of a peripheral entity (Mulder – Hervey 2009). Since all phonemes are assumed to occupy some position, nuclei stand in nuclear position, and peripheral entities in peripheral positions. There is generally only one nuclear position within the distributional unit. It is a position upon which the other positions are dependent, and as there can be only one king in a kingdom, so there can be only one governing position. Yet, for some languages, it is desirable to introduce a so-

62 called semi-distributional unit with subordinate nuclear position just as the office of viceroy is useful in large empires. We will return to this problem in Chapter 6. Peripheral positions are gathered around the nucleus as planets gather around the sun (or like peels of an onion, Mulder 1989: 293); the distributional unit is a centripetal model. However, as speech occurs in space and time, positions are arranged in a certain sequence. This logical necessity should not be misunderstood as implying that the linear sequencing is necessarily functional; it is only a fact we cannot ignor. Accordingly, there may be two kinds of peripheral positions: those occurring before the nucleus (i.e. pre-nuclear) and those after it (i.e. post-nuclear).8 However, a language need not have both types. Though not necessarily identical, there are usually strong correlations between the distributional unit and syllables (see Section 4.6), and it is likely that the absence of either type of peripheral positions would be encountered in languages with no syllable onset or those with no syllable coda. Since in the already mentioned Cayuvava there are only syllables with the structure ‘V’ and ‘CV’, it would be a language without any post-nuclear position. Coda-less languages are in fact commonly cited (cf. Hyman 2008: 103, note 13), but not all of them must be devoid of post-nuclear positions. Many (e.g. Rotokas or Hawaiian, see above) contain long vocoids or diphthongs, and these may correspond to single phonemes or combinations of two; if the latter, there probably is a post-nuclear position. But Cayuvava has only open syllables and no complex syllable cores. As regards onset-less languages, they are extremely rare, if existing at all. Ar- rernte mentioned above is claimed to be one. It all depends on the phonological interpretation of the data. Languages need not have both types of peripheral positions, but it appears that pre-nuclear positions are more probable than post-nuclear ones. With the difference between nuclear and peripheral entities, we can now distinguish between phonemes which always function as nuclei of phonotactic constructions, and phonemes which always function as peripheral entities. The former will be called vowels, whereas the latter consonants. These are then two basic classes of phonemes. Even though the traditional terms are used here, it should be obvious that vowels and consonants are in this conception purely distributional and phonological notions, which fact

8 They have also been called explosive and implosive (by Mulder 1968; after de Saussure (1972: 79ff.). In Section 5.2 we will assign a slightly specialized meaning to the term explosive.

63 does not, of course, deny they have phonetic correlates. However, there need not be one-to-one correspondence between phonological vowels and phonological consonants and phonetic vowels (i.e. vocoids) and phonetic consonants (i.e. contoids), respectively. Reasons are various. First of all, there may exist a third class of phonemes which is not only is logically possible, but also widely found in the world’s languages: a class of phonemes capable of occurring in nuclear as well as non-nuclear positions (cf. Bell 1978). Such phonemes can be called semivowels or semiconsonants (Mulder 1989: 226ff.). Czech is a language where this category has its justification because it possesses so-called “syllabic” /r/ and /l/ (see Sections 2.5 and 7.2). And so do other languages: English, for example, has a class of semivowels (El-Shakfeh 1987). Another reason is the fact that even though there is no known language which would not contain contoids and vocoids, there are languages where either the distributional category of consonants or that of vowels is missing. For instance, in Tashlhiyt Berber (a language spoken in Morocco; Dell – Elmedlaoui 2002), every contoid can be a syllable core as well as a syllable margin.9 Provided that this corresponds to our dichotomy nuclear vs. peripheral, the language would then have only semiconsonants and vowels, but no phonological consonants, i.e. phonemes only functioning as peripheral entities. On the other hand, Sudanese Arabic has only consonants and semiconsonants because vocoids and their corresponding glides can be united into single phonemes (Dickins 2007). In fact, there is some evidence that a similar union can be thought of even for Tashlhiyt Berber in which case that language would, distributionally, have only semiconsonants!

4.4 Occurrence dependency

There is one important circumstance concerning peripheral entities: they are occurrence- dependent on the nuclear entities, which means that the latter condition the occurrence of the former. To put it simply, peripheral entities cannot occur without the nuclear, governing entities. The presence of the governing entity is compulsory, it being the

9 Bella Coola, a Salishan language spoken in Canada, is also often cited as a language where every contoid can be syllabic, though alternative analyses have been proposed (Bagemihl 1991).

64 identity element of a phonotactic construction. The occurrence dependency is a necessary criterion for determining whether an entity is peripheral or not. It would be logically absurd to maintain the contrary, i.e. that there are entities which are not dependent on the occurrence of other entities, but which are still dependent on them for their function in the phonotactic construction. However, even though the occurrence dependency is necessary criterion, it is not a sufficient one. Although peripheral entities are always dependent on the occurrence of nuclear ones, it does not imply the latter are necessarily occurrence independent (Mulder 1989: 288–93). Despite being functionally superior, so to speak, a nuclear entity may still require the presence of another entity to assert its function. Accordingly, there are two types of sub-ordination: The first one can be called sub-ordination with unilateral occurrence dependency. It refers to a situation when a peripheral, subordinate entity a is dependent on the occurrence of a nuclear, governing entity b, but b is not at the same time dependent on the occurrence of a (Mulder – Hervey 2009). So whereas a requires the presence of b, b does not require the same of a, simply because b can occur without a. In a word, the a can omitted in a phonotactic construction. This type of sub-ordination holds between vowels and consonants in Czech. Consonants must always be accompanied by vowels, but vowels need not be accompanied by consonants. The latter are omittable with the resulting phonotactic construction still being well-formed (cf. /ta/, a form of ta, as opposed to /a/, a form of a). However, the opposite does not hold: if a vowel is omitted, the result is not a well-formed construction (cf. /ta/ ta as opposed to /t/ which is not well-formed, i.e. it is not a phonological form of any Czech word). The second type of sub-ordination can be called sub-ordination with bilateral occurrence dependency. It refers to a situation when a peripheral, subordinate entity a is dependent on the occurrence of a nuclear, governing entity b, and b is at the same time dependent on the occurrence of a (Mulder – Hervey op. cit.). In effect, neither of them can be left out, though they can be replaced by other phonemes. This is a case of nuclear /r/ and /l/ in Czech (see Section 7.2). Although they can function as nuclei, they require the presence of at least one consonant, so that a given phonotactic construction is well- formed. It is because /r/ and /l/, unlike vowels (cf. /a/ a above), cannot form a phonological word by themselves, and must be preceded by at least one consonant.

65 4.5 Semi-distributional unit

In Section 4.3 we hinted that it is possible that a distributional unit had more than one nuclear position. This situation is, however, only thinkable under some special circumstances because if there is a nuclear position, it follows from its very nature that there can be only one. To repeat, a nuclear position is a position where governing entities of phonotactic constructions occur, the other entities being dependent of the nuclear entities for their phonotactic function. However, it is possible—and sometimes even desirable—to operate with several degrees of nuclearity, with one major, governing nucleus and several minor nuclei subordinated to it. In fact, several degrees of nuclearity have already been recognized for syntax where the notion position (in a grammatical sense) is also utilized (Mulder 1989: 278). On the other hand, for phonotactics, the possibility of having several degrees of nuclearity has hardly explicitly recognized, although Mulder’s analysis of Japanese poly-syllabic words suggests it is thinkable (1987: 41ff. and 1989: 268–71). In this connection he mentions the term semi-distributional unit as a model accounting for dependent syllables in poly-syllabic words such as nippon10. It was probably El-Shakfeh (1987) who first explored this idea in full. In his analysis of the phonotactics of English he introduced a distinction between major-type and minor-type phonotagms. The former are basic, independent phonotagms corresponding to autonomous stressed syllables with unreduced vocalic elements, whereas the latter account for unstressed syllables prone to reduction of their vocalic elements to schwas. An example may be /nrreitr/, a phonological form of narrator, where the underlined parts are minor- type phonotagms corresponding to unstressed syllables (/reit/ being a major-type phonotagm). The peculiar behavior of unstressed syllables in English is a well-known phenomenon. For one thing, they cannot occur without accompaniment of the stressed ones whereas the latter can. This occurrence dependency means that the unstressed syllables cannot correspond to separate phonotagms because phonotagms are by definition self- contained (i.e. autonomous) entities.

10 The second syllable in this word is dependent on the first because of the gemination: the first phoneme of the second syllable is the same at the last phoneme of the first syllable. In essence, the situation is similar to the one cited for the Australian languages in Section 4.1 above. – On the term semi-distributional unit, see also Mulder (1968: 58).

66 However, for the convenience and simplicity of the description, the unstressed syllables in English can be viewed as if they were indeed independent and self-contained, i.e. as if they corresponded to genuine phonotagms. To distinguish them from the really independent and self-contained phonotagms, El-Shakfeh called such units minor-type phonotagms. Consequently, the structure of minor-type phonotagms can be described in a similar vein as that of the genuinely independent phonotagms realized as stressed syllables, which are called major-type phonotagms. Now, even though El-Shakfeh does not use the term, the underlying bundle of positions behind all minor-type phonotagms can be called a semi-distributional unit. The nucleus of the semi-distributional unit is dependent on the nucleus of the main distributional unit, the latter functioning as the ultimate nucleus of the overall distributional unit. We will return to this problem in Chapter 6 where we construct a semi-distributional unit as a means for dealing with certain phonotagm-like combinations of phonemes in Czech.

4.6 Phonotagms and syllables

The distributional unit is an abstract network of positions mutually connected via phonotactic relations between their members. A position is a slot that can be occupied either by a phoneme from a certain paradigmatic class or left empty. A particular instance of a distributional unit where all positions are either occupied or empty is a phonotagm. Phonotagms attested in a language differ from each other in the way positions of the underlying distributional unit are filled; see Figure I above. The distributional unit may then be compared to syllable templates with which some linguists operate (e.g. Selkirk 1982), and phonotagms to concrete phoneme sequences. Although our approach is different, the basic idea is the same: the structure of phoneme sequences (i.e. phonotagms), whose number is in principle infinite, can be described upon one underlying syllable template (i.e. the distributional unit). Despite there being obvious similarities, we will not use the term syllable in any phonological sense in this work. The reason is simple: over the time it has become rather overused in linguistic literature (cf. Bell – Hooper 1978): it stands for a great

67 many things and the dissimilar uses of it get confused with each other (see Hála 1956, 1961, Awedyk 1975, van der Hulst – Ritter 1999, Meynadier 2001, Duamnu 2008, Goldsmith 2009, Cairns – Raimy 2011 for overviews). It is used by phoneticians and phonologists alike, and it may denote everything from a certain “real” portion of utterance to a very abstract linguistic model (cf. e.g. Lowenstamm 1996). In order to avoid terminological confusion we prefer using the term phonotagm as a name of a unit of phoneme distribution and the term distributional unit as its underlying model (“template”). We will use syllable only as a name of a certain unit of physiological organization of speech.11 This notion is, however, far from being unproblematic; it is well- known that phoneticians have long struggled with how the (phonic) syllable is to be properly defined, if it can be defined at all (see Krakow 1999). It is not in our strength to resolve it here, but we will, for the purposes of this work, assume that something like a phonic syllable does exist (see also Cholin 2011 for psycholinguistic evidence of the syllable). Once the syllable is recognized as a phonetic entity and the phonotagm as a phonological entity, it is legitimate to ask what the relationship between them is. At the level of segments it is reasonable to consider relationships between speech sounds and phonemes because the latter serve as descriptive models of the former. Similarly, at a higher level phonotagms may be viewed as being in essence descriptive models of syllables. But just as there is no one-to-one correspondence between speech sounds and phonemes, neither should we expect it to be between syllables and phonotagms. This is to say, although it is very common for one phoneme to be realized by one speech sound, it is not a rule.12 Likewise, although phonotagms appear to be most commonly realized by singular syllables—at least in languages we know of—, this should not be understood as a universal rule (cf. also Pike 1947: 65, Fudge 1969: 254). Accordingly, we may think of three possible relationships between phonotagms and syllables:

Relationhips between phonotagms and syllables (α) One phonotagm is realized by exactly one syllable.

11 We could have employed the term phonological syllable, thus distinguishing it from a phonic syllable. This strategy has been chosen by several linguists, e.g. Pike (1947), Rosetti (1963). 12 As our analysis of the Czech affricates has proven (see Section 2.3).

68 (β) One phonotagm is realized by a unit of greater complexity than one syllable. (γ) One phonotagm is realized by a unit of lesser complexity than one syllable.

Out of these possibilities, the first one (α) appears as an ideal; it is probably the most common situation, though particular languages may exploit other options. Rotokas with its simple phonotactic structure seems to be one where it always holds. The same is probably true for Cayuvava (see the discussion and references in Section 4.2). And it is almost exclusively the relationship which holds in Czech, with the exception of phonological words like /Stārl/ stárl which is distributionally a single phonotagm corresponding in realization to two syllables (a point discussed in detail in Section 7.2). Aside from that, phonological forms of Czech words consisting of just one phonotagm are realized by one syllable.13 Moreover, in words built of two or more phonotagms the number of phonotagms is the same as that of syllables by which the forms are realized, for example, in náhrdelník whose four-phonotagm phonological from /nāhrdelňīK/ is realized by four syllables [naː.hr.dɛl.ɲiːk̩ ]. On the other hand, there are languages where not all vocalic phonic segments are realizations of nuclear phonemes. It is the case of British English for which the second relationships (β) holds, though the first one is encountered, too (e.g. [naɪt] night corresponding to one phonotagm /nait/). In this language a phonotagm may be realized by a unit of greater complexity than one syllable because some schwas occurring in the unstressed syllables are merely epenthetic elements fully predictable from the context; see the examples under (1) (taken from El-Shakfeh 1987: 378; cf. also Hervey 1978: 50–2, Fudge 1969: 269 and Heselwood 2007). Consequently, such syllables cannot correspond to separate phonotagms because we would ascribe distinctive function to something not capable of distinguishing anything.

(1) [riːdʒəntəl] (three syllables) → /rIidŠntl/ (one phonotagm) (cf. regental) [fɔːlʧənz] (two syllables) → /forltŠnz/ (one phonotagm) (cf. falchions) [sɪmpəltənz] (three syllables) → /siNplTnz/ (one phonotagm) (cf. simpletons) [bʌmblədəmz] (three syllables) → /brNblTmz/ (one phonotagm) (cf. bumbledoms)

13 Unless words like msta, rty or lhát are disyllabic; see Section 8.7.

69 Another interesting case is found in Piro, an Arawakan language spoken in Peru, which, according to Matteson (1965), has only syllables of the structure ‘CV’, ‘CCV’ and ‘CCCV’. However, what Matteson calls syllables rather corresponds to our phonotagms because in this language every consonant which does not stand right before a vowel is realized either as syllabic or with a short epenthetic vowel (op. cit.: 129–34, Lin 1997: 405–6), e.g. /šwamkalo/ [šəwam̩ kalo] “spider”. As this is completely predictable, the consonants cannot correspond to separate phonotagms, i.e. /šwamkalo/ presumably consists of three phonotagms but of five phonic syllables. Finally, as regards the last relationship (γ), i.e. a phonotagm being realized by a unit of lesser complexity than one syllable, its possibility is above all dependent on how the phonic syllable is defined. If it is declared, as is sometimes done, to be the smallest pro- nounceable unit of speech, it is obvious that a singular phonotagm could never be realized by something less than a syllable. It may even be argued that anything uttered by a human is a syllable, no matter what phonetic segments such an utterance contains, even interjections such as pst.14 If so, the commonly held opinion that a syllable should contain a vocalic element is not valid. However, this is not something specific to interjections only. Linguists have long been aware of languages with words completely realized without a single vocoid or even without a single voiced element. Two such languages are usually mentioned: the already referred-to Bella Coola and Tashlhiyt Berber. The latter has words like those given in (2) where voiceless obstruents function as syllable cores (examples from Ridouane 2008, syllable cores underlined)15.

(2) [fk] “give” (one syllable) [tf.tk.tstt] “you strained it” (three syllables) [ts.sk.ʃf.tstt] “you dried it” (four syllables)

14 A thing rejected e.g. by Hála (1956: 54, 1961: 114–5). A detailed overview of opinions on this problem is found in Awedyk (op. cit.). The whole point is that whether something is a syllable or not depends on the theory we choose, as recently aptly demonstrated by Hyman (2011). 15 The opinions on the exact nature of words like these are not unitary. Coleman (2001) suggested they may in fact contain epenthetic vocalic segments. However, Ridouane (op. cit.) presented strong evidence that they are completely devoid of vocoids or of any voiced segment.

70 A priory, there is no reason to assume the notions distributional unit and phonotagm are not applicable to languages like Bella Coola or Tashlhiyt Berber, too. A rather pre- liminary analysis by the present writer indicates that the distributional unit in Tashlhiyt Berber consists of three positions, one pre-nuclear, one nuclear and one post-nuclear, though further research is still needed. The point is that phonotagms are in this language realized by utterances which lack any vocalic element, but which are still regarded as genuine syllables by its users. This suggests that, there being no limits on what sounds a syllable must contain, a single phonotagm is always realized by a syllable. A different situation may arise when phonotagms do not occur in isolation but as parts of phonological forms, for it is common that words are built of more than one phonotagm. It is then possible that these forms are realized by a sequence of syllables whose number is lesser than the number of the phonotagms. They may, so to speak, coalesce into one syllable. As of now we can think this possible only in the case of underarticulation, i.e. when articulation of one syllable is neglected in rapid and careless speech as in Czech vole “dude, man (voc. sg.)”, phonologically two phonotagms /vole/, which is often un- derarticulated, via [voɛ], to sound like monosyllabic [wɛ], instead of orthopic [volɛ]. To sum it up, a phonotagm is always realized by at least one syllable, but it may be realized by more than one syllable because all non-functional, predictable features (e.g. certain schwas in unstressed syllables in English) are left at the level of phonetic realization without being accommodated in any way to phonological representation of words.16

4.7 Constructing a distributional unit

The purpose of the distributional unit is to be a model upon which occurrences and combinations of phonemes can be described in a non-contradictory, non-redundant and non-deficient manner. It must contain no fewer positions than is necessary for account for all occurrences of phonemes, and, likewise, it cannot contain extra positions to avoid redundancy. Mulder (1993, 1996) has developed a methodology allowing analysts to

16 It does not mean such features are ignored altogether; they are only dealt with, in a systematic manner, elsewhere, in description of the allophony, which is, however, not at issue in this work.

71 consider and operate with only those elements that are essential for describing certain phenomena. In short, in a description one puts forth hypotheses about speech phenomena. These hypotheses must be in principle refutable, i.e. they must be testable against available data. The goal is to refute the hypotheses. But should they remain unrefuted, they are corroborated and become part of the description. If refuted, they are replaced by new hypotheses which are further tested. Refuted hypotheses are actually very important for a description because they provide evidence of what is not possible. On the other hand, corroborated hypotheses leave us with a doubt they could be refuted one day by relevant data. A linguistic analysis is not a straightforward procedure, but rather a series of conjectures and refutations, to borrow Popper’s phrase (Popper 2002). Before showing in detail on Czech how a distributional unit is arrived at, we would like to give a general outline of the way the analysis may be conducted. With every language being unique, the actual procedure will of course differ from a language to a language. Two points should be stressed. First, the heuristic methods presented here are just a convenient way of working in practice and not any sort of discovery procedures. We do not discover the distributional unit in the data; we apply this theoretical model on the data with an aim to get a better understanding of them. Second, although we act here as if the phonotactic analysis of a language were independent of the phonematic one (as presented in Chapter 2) and the other way round, it is certainly not true. The paradigmatic identity of phonemes is determined hand in hand with their syntagmatic identity. It is only for presentational reasons that the two are separated. When describing the phonotactics of a certain language, we may proceed in the following way: As there must be, for logical reasons, at least two positions within a distributional unit, our initial hypothesis is such that it contains two positions, and that no more are necessary for an exhaustive description of all phoneme occurrences within phonotactic constructions in a given language. This hypothesis will be held until refuted. It will usually be very easily refuted, as it is rare that only two positions would do the job17. Once it is refuted, we launch a new hypothesis: that no more than three positions are necessary. If even this one is refuted, we replace it with a new hypothesis and try refuting it again. We proceed this way until arriving at a hypothesis about a certain

17 Rotokas or Cayuvava could be such languages; see Section 4.2.

72 number of positions we are not able to refute by relevant examples. This will usually result in our knowing how many positions are necessary, though we should still and always consider the adequacy of our findings. In practice, however, it is not necessary to go this long way; as Mulder (1987: 36) has noted, there is a way to cut it short, but this shortcut has rational background behind it. To begin with, we should consider whether there is some ground for postulating the nuclear position which is a position where nuclear entities occur. Although it is theoretically possible to have a distributional unit without the nuclear position, this situation appears to be quite rare. One must remember that nuclear entities are those upon which the others are functionally dependent and those which cannot be left out. So we should first try to find out whether a language under investigation has words whose phonological forms are built of a single phoneme because these will usually18 be the nuclear entities. Czech or French has such words. Even if the language has no such words, the diag- nostic test for nuclearity will still be the same: the presence of a nuclear entity is compulsory for a phonotactic construction to be well-formed. Let us take English as an example. Unless long vowels like the one in awe or diphthongs like the one in eye are regarded as singular phonemes (cf. Malone 1936), vowels must always be accompanied by a consonant in this language. Yet it holds that vowels can never be omitted from a phonotactic construction.19 Naturally, the whole thing may sometimes be quite complicated, and we have to consider carefully the choice of nuclear entities and of the nuclear position. Some linguists have tried to define vowels and consonants on distributional criteria but without any recourse to nuclearity. O’Connor – Trim (1953) suggested one method, and applied it successfully to English. But although the same method was applied to French by Arnold (1955–6), the same linguist eventually showed it fails on languages such as Greek and Polish (Arnold 1964). Greenberg (1962) proposed another and a more sophisticated method, but even this one fails in languages like Hottentot or

18 We say “usually” because in Czech the prepositions s, z, k, v, though built of a single phoneme, do not contain any nuclear entity; this is a very special case, see Section 3.2. 19 We simplify here a little bit because the classes of vowels and consonants are established only after the nuclear position has been set up, never before! Our point is that although there may be no English word with the form /a/, the phoneme /a/ is a nuclear entity because it and its com- munants cannot be left out (cf. /pat/ pat against /at/ at against ill-formed /pt/).

73 Guaraní (Bell 1976). The bottom line is that it may be difficult for some languages, if possible at all, to define vowels and consonants as separate distributional classes of phonemes. However, the distribution of phonemes in such languages should be in principle describable even without these notions. The point is to set up a grid of positions functioning as slots where phonemes occur and commute with each other. Should the nuclear position be established or not, the next step in analysis is determination of the set of phonemes capable of occurring in a certain position. The commutation test is the best procedure at this point. Once we have determined that /a/ in a nuclear entity in English /pat/ (see footnote 19), we can list its commutants; they are /e/, /i/, /o/, /u/ and /r/ (cf. pet, pit, pot, put and putt, respectively). The occurrence of any of them is compulsory, but the same could not be said about /p/ and its commutants. Pho- nemes occurring in the nuclear position in these examples will be, for the time being, declared vowels; this will be later corrected because some of are in fact semivowels (namely /r/) capable of occurring in non-nuclear positions, too. As there must be at least two positions within a distributional unit, it is obvious that if one is declared nuclear, the other(s) must be peripheral unless we introduce a semi- distributional unit, but even in that case there will be only one nucleus nucleorum. Due to the linearity of speech, there may be two kinds of peripheral positions, pre-nuclear and post-nuclear, so the next step is evaluation of the possibility of setting up pre- nuclear and/or post-nuclear positions, and determination of their number. This can be easily found out—and this is the shortcut alluded to above—by considering the maximum number of peripheral phonemes capable of occurring in between the beginning of a phonological word and the first nuclear entity, and the maximum number of peripheral phonemes capable of occurring in between the last nuclear entity and the end of a phonological word. Since the assumption is that only one phoneme occurs in a position at a time, their number corresponds to the number of pre-nuclear and post-nuclear positions. By now we have gained a rough idea of the form of the distributional unit. It may still be altered by further considerations and circumstances peculiar to a given language such as by the introduction of archi-positions and/or a semi-distributional unit. With the positions of the distributional unit set up, our task is to map onto them attested phonotactic constructions. When doing this, we can adopt these two principles:

74 Principle of Exclusive Membership A phoneme assigned to a certain position does not occupy any other position unless proven otherwise.

Principle of Mutual Commutability Phonemes mutually commutable in equivalent contexts do not occupy different positions unless proven otherwise.

The first principle says the following: Having assigned a phoneme /α/ to a position ‘β’ in a construction A, we assume it occurs in the very same position in any other construction B it belongs to unless this is in contradiction with other facts. It derives from the general principle of simplicity of descriptions: we should not introduce new elements (new statements, features, entities, concepts etc.) into a description if it is not necessary. It is simpler to assume that a phoneme always occupies the same position than that it occupies several ones. However, hypotheses of this kind are often refuted. For instance, we cannot hold the hypothesis about the English phoneme /t/ due to the existence of the two different constructions /tip/ and /pit/ (cf. tip and pit). The phoneme must occur in at least two positions, one pre-nuclear and one post-nuclear. But we can still assume that /t/ in /tip/, /t/ in /trip/ and /t/ in /Strip/ (cf. trip and strip) occur in the same pre-nuclear position in which case this hypothesis will not be refuted. The second principle makes recourse to the mutual commutability of phonemes in functionally equivalent contexts. We will assume that if two or more phonemes are mutually commutable, they occupy the same position. It follows from the fact a position is a slot where a phoneme occurs and commutes with other phonemes. In a phonotactic construction a phoneme not only occurs in a certain position but also in the presence of other phonemes. A certain other context is functionally equivalent to it provided that it is identical to it or that it represents the same (Mulder 1968: 119). For example, in a construction /na/ the phoneme /n/ occurs in a pre-nuclear position and in the presence of /a/. In constructions /ma/ and /ta/ the phonemes /m/ and /t/ occur in functionally equivalent contexts because they also occur in a pre-nuclear position and in the presence of /a/. The hypothesis will thus be that the mutually commutable phonemes /n/, /m/ and /t/ oc-

75 cur in the same position. However, we would at once like to make clear that by intro- ducing this principle we do not want to suggest that all phonemes which are mutually commutable in equivalent contexts must always occupy the same position. Although our initial hypothesis may be that /n/, /m/ and /t/ occur in the same pre-nuclear position in Czech due to their mutual commutability, it will later prove that such a hypothesis cannot be held. The existence of the form /tma/ (cf. tma) indicates that at least /t/ and /m/ occur in two different positions. On the other hand, we will not be able to refute the hypothesis that /m/ and /n/ do not occur in the same position. In this case the Principle of Mutual Commutability is the way we can assign both to the same position. We have briefly outlined how a distributional unit can be constructed, and how phonemes can be assigned to particular positions. Every description of a certain language has of course its own intricacies, and it is not possible to review them all here. In the next chapter we will demonstrate how a distributional unit is constructed in practice.

76 Chapter Five DISTRIBUTIONAL UNIT: APPLICATION TO CZECH

5.1 Nuclear position

In the previous chapter we outlined the theoretical and methodological background of the construction of the distributional unit, and in this chapter we construct one for Modern Standard Czech. The analysis is presented here in several steps, though its actual course was of course not so straightforward; we offer here only the results. Following the principles introduced at the end of the previous chapter, we will first determine whether there are any grounds for establishing the nuclear position where nuclear entities would occur (see Sections 4.3 and 4.4 on the rationale behind this). The examples (1) prove there are. As these phonological words are built of one phoneme only, it is evident that /a/, /o/, /i/ and /u/ are not dependent on any other item, either for their phonotactic function or for their occurrence. Consequently, they are nuclear entities, and we assume they occur in the nuclear position.

1 (1) /a/ a, realized [ʔa] /o/ o, realized [ʔo] /i/ i, realized [ʔɪ] /u/ u, realized [ʔu]

Having assigned /a/, /o/, /i/, /u/ to the nuclear position, we will assume, by the Principle of Exclusive Membership (see Section 4.7), that they do not occupy any other position as long as this hypothesis withstands refutation. It has withstood, and we adopt

1 The glottal stop is only a concomitant feature and a realization of diaereme (see Section 3.2).

77 it as a descriptive statement. Hence, in constructions like /na/ na, /on/ on, /sin/ syn, /kruX/ kruh and others, the phonemes /a/, /o/, /i/, /u/ are assumed to occupy the very same nuclear position. We will now determine which phonemes are commutable with them in functionally equivalent contexts (cf. the Principle of Mutual Commutability we discussed in Section 4.7). Compare the examples in (2).

(2) /paST/ past ~ /poST/ post ~ /pāST/ past ~ /prST/ prst ~ /plST/ plst ~ /pīST/ píst ~ /pūST/ půst /paS/ pas ~ /pāS/ pás ~ /peS/ pes ~ /puS/ pus ~ /pōS/ póz ~ /prS/ prs /vaS/ vaz ~ /vāS/ vás ~ /veS/ ves ~ /viS/ viz ~ /vīS/ víz ~ /voS/ vos ~ /vūS/ vůz ~ /vöS/ vous /ar/ ar ~ /ēr/ ér ~ /ir/ Ir ~ /är/ aur ~ /ër/ eur

The phonemes commutable with /a/, /o/, /i/, /u/ in equivalent contexts are /e/, /ā/, /ē/, /ō/, /ī/, /ū/, /ä/, /ë/, /ö/ and /r/, /l/. The latter two have been deliberately separated from the others because the subsequent analysis will reveal that /r/, /l/ are different to them. Initially, we assumed for all of these phonemes that they always occupied the same position like /a/, /o/, /i/, /u/, but this hypothesis is refuted for /r/ and /l/ once we considered examples like the following ones:

(3) /praK/ prak ~ /parK/ park /kloT/ klot ~ /kolT/ kolt

Since only one phoneme may occupy a position at a time, the sequences /ra/, /ar/, /lo/ and /ol/ prove that either /a/ and /o/, or /r/ and /l/ occur in the nuclear position in these forms, which refutes the hypothesis “/r/, /l/, /a/ and /o/ always occur in the nuclear position”. Thus, either all or some of them occur in at least one other position. By further considerations we have concluded that it is much simpler to assume that it is /a/ and /o/ that always occur in the nuclear position, while /r/ and /l/ are phonemes that can occur in the nuclear position as well as in non-nuclear positions. This has proven to be the most adequate and simple solution. Consequently, by commutations in (2) we

78 establish the class of vowels which includes /a/, /ā/, /e/, /ē/, /o/, /ō/, /ī/, /ū/, /ä/, /ë/, and the class of semiconsonants which includes /r/, /l/. Vowel are nuclear entities occurring only in the nuclear position; they are either directly occurrence independent, i.e. /a/, /o/, /i/, /u/, or commutable in functionally equivalent contexts with /a/, /o/, /i/, /u/. Semiconsonants are also nuclear entities by virtue of being commutable in functionally equivalent contexts with /a/, /o/, /i/, /u/, but because they must occur in at least one non- nuclear position, they also have the capacity of functioning as peripheral entities. By further commutations we have determined that the remaining phonemes are consonants, namely /p/, /b/, /f/, /v/, /m/, /t/, /d/, /s/, /z/, /n/, /ť/, /ď/, /š/, /ž/, /ň/, /k/, /g/, /x/, /h/, /ř/, /j/ (including their archiphonemes). These phonemes always and only function as peripheral entities and require the presence of a nuclear entity. Now, since the semiconsonants can function both as vowels and consonants, we will prefer in the rest of this work to talk about nuclear phonemes when dealing with vowels and semiconsonants in the nuclear function, and about peripheral phonemes when dealing with consonants and semiconsonants in the peripheral function.

5.2 Peripheral positions

Initially, we assume there is only one nuclear position within the distributional unit. This assumption will be corrected in the next chapter where we introduce the semi- distributional unit with its own nuclear position, but even that one will be dependent on the nuclear position of the main distributional unit, which is the ultimate nucleus. In Czech peripheral phonemes can occur before and after a nuclear phoneme (cf. /lem/ lem × /mel/ mel (imper. sg. of mlít)). By physical necessity, phonemes (and hence the positions) are arranged in a linear sequence. Therefore, there are two domains of the distribution of peripheral entities in this language: the pre-nuclear context and the post-nuclear context.2 The former includes all positions and archi-positions we

2 As we do not operate with the syllable here, its traditional division to the onset and rhyme, and the division of the rhyme to the nucleus and coda are irrelevant here. In fact, in Czech there do not appear to be any exceptional relation between the nucleus and the post-nuclear context to necessitate introduction of the notion rhyme. See Chapter 13.

79 postulate for phonemes capable of occurring before the nuclear entity; the latter all positions and the one archi-position postulated for phonemes capable of occurring after it. We can call the domain of the distribution of nuclear phonemes the nuclear context; it will include the nuclear position and the nuclear archi-position (see below). For the time being, we will assume mutual independency of all of these contexts, but this assumption will be challenged in Chapter 13. By examining combinations at the beginning of phonological words before a nuclear phoneme and at their end after a nuclear phoneme, we can determine the maximum number of phonemes preceding and following it. See the examples (4).

(4) /FSkvjeT/ vzkvět, realized [fskvjɛt] /vojSK/ vojsk, realized [vojsk]

As every phoneme must occur in just one position, we now have a rough idea about the constituency of the distributional unit: it contains one nuclear position, and it must contain at least five pre-nuclear positions and at least three post-nuclear positions. This is still a hypothesis to be further tested, but so far it has withstood refutations in the sense that more positions than these ten are not necessary, and fewer are insufficient for the exhaustive description the distributional properties of the Czech phonemes. Some modifications will be necessary later, though. Our initial hypothesis about the structure of the distributional unit is schemed in Figure I alongside with the way the initial and final parts of /FSkvjeT/ and /vojSK/ are assigned to the particular positions. There are two rows in the scheme of the distributional unit and two types of labels. In both ‘n’ stands for the nuclear position; the other markings are for peripheral positions. The more distant from the nucleus, the more peripheral a position is. In the first row the positive and negative numbers indicate the degree of peripherality of non-nuclear positions. Pre-nuclear positions ‘–1’, ‘–2’ and ‘–3’ (along with their archi-positions) will be called explosive, and post-nuclear positions ‘+1’, ‘+2’ and ‘+3’ (along with their archi-position) will be called implosive. Thus, in the second type of labeling the minus sign has been replaced by ‘e’ for positions ‘–1’, ‘–2’ and ‘–3’, and the plus sign by ‘i’. Now, in order to explain why pre-nuclear positions ‘–4’ and ‘–5’ have not been

80 relabeled as ‘e4’ and ‘e5’ but as ‘pre1’ and ‘pre2’, we must go ahead in our exposition. The mark ‘pre’ stands here for pre-explosive, and in our conception a pre-explosive position will be such a pre-nuclear position which can only be occupied by a phoneme provided that at least one of the explosive positions ‘e1’, ‘e2’, ‘e3’ or their archi- positions is occupied by a phoneme. Otherwise, a pre-explosive position must be empty. Pre-explosive positions ‘pre1’ and ‘pre2’ function as extensions to explosive positions, as it were. We will return to them in Chapter 8.

Pre-nuclear context /F/ /S/ /k/ /v/ /j/ /e/ ↓ ↓ ↓ ↓ ↓ ↓ –5 –4 –3 –2 –1 +1 +2 +3 ‘n’ ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘i1’ ‘i2’ ‘i3’ ↑ ↑ ↑ ↑ /o/ /j/ /S/ /K/ Post-nuclear context Figure I: Assignment of phonemes to positions of the pre-nuclear and post-nuclear context.

Figure I visualizes the default variant of the main distributional unit in Czech. It is a model upon which the distribution and combinations of the phonemes in Czech can be exhaustively and completely described for all possible phonotagms3. All phonemes in any phonotactic construction can be assigned to at least one position or archi-position and, once assigned there, their occurrence cannot be affected by the occurrence of phonemes outside the given construction. In this respect the model has so far proven its mettle, and has not been invalidated by relevant evidence that more positions would have to be introduced or that some positions would in fact be redundant. We say default because there are some reservations as to under what conditions all the positions are valid in particular situations. They will be discussed in the due course. The question to be resolved now is how to assign phonemes to the individual positions. This is done through a series of hypotheses, their corroborations or refutations, and postulations of new hypotheses. We already assigned some phonemes to the nuclear position. As to the non-nuclear context, it is reasonable to base our further analysis on the forms where all positions are filled. For the pre-nuclear context, the best

3 That is, all possible major-type phonotagms. We will explain this in the next chapter.

81 starting point is /FSkvjeT/. The same procedure will be applied on the post-nuclear context where we start with /vojSK/ or any other phonological word ending in three peripheral phonemes. Let us note that although there are many examples of phonotagms where all three post-nuclear positions are filled with a phoneme, we are only aware of two of them where the same is true for the pre-nuclear context. Besides /FSkvjeT/, there is also /STkvjel/ stkvěl, an archaic variant of /Skvjel/ skvěl.

‘–5’ ‘–4’ ‘–3’ ‘–2’ ‘–1’ ‘n’ Phonological form ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ /F/ /S/ /k/ /v/ /j/ /e/ /FSkvjeT/ vzkvět ∅ ∅ /k/ /v/ /j/ /e/ /kvjeT/ květ ∅ ∅ ∅ /v/ /j/ /e/ /vjeT/ vjet ∅ ∅ ∅ ∅ /j/ /e/ /jeT/ jet ∅ ∅ /k/ /v/ ∅ /e/ /kveT/ from kvetoucí ∅ ∅ ∅ /v/ ∅ /e/ /veT/ from vetchý

∅ /S/ /k/ ∅ ∅ /e/ /SkeT/ sket (gen. pl. of sketa) ∅ ∅ /x/ /v/ /j/ /e/ /xvjeT/ chvět ∅ ∅ /h/ /v/ /j/ /e/ /hvjeST/ hvězd ∅ ∅ /d/ /v/ /j/ /e/ /dvje/ dvě ∅ ∅ /s/ /v/ /j/ /e/ /svjeT/ svět ∅ ∅ /z/ /v/ /j/ /e/ /zvjeT/ zvěd ∅ ∅ /s/ /v/ /r/ /a/ /svraP/ svrab ∅ ∅ /z/ /v/ /l/ /ā/ /zvlāŠŤ/ zvlášť ∅ ∅ /s/ /v/ ∅ /ē/ /svēST/ svést ∅ ∅ /z/ /v/ ∅ /ā/ /zvāT/ zvát ∅ ∅ /d/ /v/ ∅ /a/ /dva/ dva /F/ /S/ /d/ ∅ ∅ /ā/ /FSdāT/ vzdát ∅ ∅ /t/ /v/ ∅ /ā/ /tvā/ tvá /F/ /S/ /t/ ∅ ∅ /ā/ /FStāT/ vstát ∅ /S/ /t/ ∅ /r/ /a/ /StraX/ strach

Figure II: Application of the Principles of Exclusive Membership and Mutual Commutability.

The five pre-nuclear phonemes of /FSkvjeT/ have been assigned to five pre-nuclear positions, and we assume they always and only occur there as long as this is not proven untenable. The phoneme /j/ has been assigned to position ‘e1’, /v/ to position ‘e2’, /k/ to ‘e3’, /S/ to ‘pre1’, and /F/ to ‘pre2’. By assumption, they occupy the very same positions also in phonological words like /kvjeT/, /vjeT/, /jeT/, /kveT/, /veT/, and /SkeT/ whose phonotactic structure can be plotted onto the distributional unit in the manner of Figure II. Even though not containing the same number of phonemes, the phonotagms

82 are still mapped onto the same grid of positions. It is because the remaining positions are actually empty. However, it should be remembered that a position may only be empty if and only if it is occupied by a phoneme under other circumstances, which Figure II clearly proves. Now, having assigned /F/, /S/, /k/, /v/, /j/ to their positions, we find out which phonemes are commutable with them because, following the Principle of Exclusive Membership, we can assume the commutants occupy the same positions. For example, /h/, /x/, /d/, /s/ and /z/ can be assigned to the same position like /k/ due to their commutability with it in phonological words like /xvjeT/, /hvjeST/, /dvje/, /svjeT/ and /zvjeT/. Taking the last two, we can find out which phonemes are commutable in functionally equivalent contexts with /j/ (/r/, /l/ in /svraP/ and /zvlāŠŤ/), and so on. The way these forms are mapped onto the distributional unit is also illustrated in Figure II. We keep assigning phonemes to positions in a similar manner. However, even if it looks straightforward, there may be—and usually are—certain situations requiring special treatment. One of them concerns /r/ and /l/. Due to the mutual commutability with /j/, they have been assigned to position ‘e1’ (see Figure II). Our initial hypothesis is that they always occur in this pre-nuclear position4. However, the assumption will have to be corrected as soon as we come across phonological forms like those given in (5a). If /r/ and /l/ occupy ‘e1’ here, where does /z/ occur? Previously, it has been assigned to ‘e3’ which, however, precedes ‘e1’, not follows it. The position following ‘e1’ is the nuclear position occupied by /i/ in /rzi/ and by /e/ in /lze/. Consequently, if /r/, /l/ occupy ‘e1’ in these forms, /z/ would have to remain unassigned, which is not possible. This proves that /r/, /l/ cannot occur in /rzi/ and /lze/ in position ‘e1’.

(5) (a) /rzi/ rzi (gen. sg. of rez) (b) /zrīT/ zrýt /lze/ lze /zle/ zle (c) /rvjeT/ from rvěte (se) (d) /StraX/ strach lStnī/ lstný /zvlāŠŤ/ zvlášť

There are two possibilities to interpret this: One is that our initial hypothesis about the occurrence of /r/, /l/ in pre-nuclear position ‘e1’ was wrong, and these phonemes

4 As shown earlier, they also occur in the nuclear position, but this is irrelevant now.

83 occupy another position. The second is that they do indeed occupy this position, but are in addition capable of occurring in yet another. Further analysis and evidence of the forms given in (5b), and also of those in (5c) and (5d) will decide that the second alternative is to be preferred. Hence, /r/ and /l/ occupy two pre-nuclear positions, namely in ‘e1’ and ‘pre2’. The final analysis is given in Figure III. It reflects several facts summarized as follows: In the pre-nuclear context the phonemes /r/ and /l/ can be both preceded and followed by a peripheral phoneme or a group of peripheral phonemes. If they are preceded, the phoneme which in turn follows them is always a nuclear phoneme (cf. /StraX/). On the other hand, if they are followed by a peripheral phoneme in the pre-nuclear context, they are always the first phonemes of the phonotagm, i.e. they cannot be at the same time preceded and followed by a peripheral phoneme. In fact, they are the very first phonemes of a phonological word. In this case the resulting phonotagms are realized by so-called side-syllables (for more see Section 8.7).

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ Phonological form /r/ ∅ /z/ ∅ ∅ /i/ /rzi/ rzi ∅ ∅ /z/ ∅ /r/ /ī/ /zrīT/ zrýt /l/ ∅ /z/ ∅ ∅ /e/ /lze/ lze ∅ ∅ /z/ ∅ /l/ /e/ /zle/ zle

Figure III: Analysis of the difference between /rzi/ and /zrīT/ and between /lze/ and /zle/.

‘–1’ ‘–2’ ‘–3’ ‘n’ Phonological form ‘i1’ ‘i2’ ‘i3’ /u/ /m/ /Š/ /T/ /kumŠT/ kumšt /a/ /r/ /K/ /T/ /farKT/ from infarkt /e/ /K/ /S/ /T/ /teKST/ text /o/ /j/ /S/ /K/ /vojSK/ vojsk /o/ /m/ /S/ /T/ /pomST/ pomst /ā/ /P/ /S/ /T/ /zāPST/ zábst /ē/ /T/ /S/ /T/ /pēTST/ péct /i/ /n/ /T/ /S/ /prinTS/ princ /u/ /l/ /T/ /S/ /sulTS/ sulc /ā/ /T/ /S/ /P/ /zāTSP/ zácp

Figure IV: Phonotactic analysis of three-phoneme post-nuclear combinations.

We have sketched the basics of how phonemes are assigned to particular pre-nuclear positions. There are still other problems to deal with, and we will do so in the next

84 section. We will now briefly move to the post-nuclear context for which we have postulated three positions. The assignment of phonemes into them is simpler because there are at least 19 post-nuclear combinations of three phonemes, and their mapping is very straightforward. Figure IV gives some examples.

5.3 Archi-positions

The differences /zrīT/ × /rzi/ and /zle/ × /lze/ has illustrated one of the cases where the data necessitated corrections of our initial hypotheses. Another instance is when we encounter phonemes with special distribution whose mapping onto the so-far established grid of positions will not prove possible, and we will have to introduce for them an archi-position. We have already touched upon it briefly in Section 4.2. In a way, archi-positions are syntagmatic parallels of archiphonemes. Like archiphonemes are products of the suspension of a paradigmatic difference between two or more phonemes, so are archi-positions products of the suspension of a syntagmatic difference between two or more positions. The usefulness of this notion is best shown on an example of the phonemes /ť/ and /ď/ in the pre-nuclear context. Their special distributional and combinational properties are summarized as follows:

Phonotactic properties of /ť/, /ď/ in the pre-nuclear context (α1) They can be preceded by one or two peripheral phonemes, but no more. (α2) They are always followed by a vowel, never by another consonant or a semiconsonant (i.e. never by a peripheral phoneme).

To begin with, let us assume that /ť/ and /ď/ have been assigned to position ‘e3’ because of their commutability with the phonemes already assigned there, namely with /t/ and /d/, cf. /TŠte/ čte ~ /TŠťi/ čti, /FSdāT/ vztát (se) ~ /FSďel/ from vzdělaný.5 This hypothesis will soon have to be abandoned. The process of assignment of phonemes to particular positions is not a game whose goal is merely to fill up all slots without any

5 The front vowels /i/, /e/ do not condition the choice of the preceding phoneme, see Chapter 13.

85 consequences. On the contrary: by assigning a phoneme to a position we imply it possesses the phonotactic properties ascribed to the given position; they define the position in question (see below). The following are two fundamental phonotactic properties of position ‘e3’; they are shared by all phonemes belonging to ‘e3’:

Phonotactic properties of position ‘e3’ (β1) A phoneme belonging to ‘e3’ can be preceded by up to two phonemes belonging to positions ‘pre1’ and ‘pre2’, respectively. (β2) A phoneme belonging to ‘e3’ can be followed by up two phonemes belonging to positions ‘e2’ and ‘e1’, respectively.6

The properties can now be compared with two facts about /ť/ and /ď/ given above. Whereas (α1) is compatible with (β1), there is a discrepancy between (α2) and (β2): /ť/ and /ď/ can never be followed by any consonant in the pre-nuclear context—on the contrary, they are always followed by a vowel. Failing thus to conform to the second distributional property, they cannot be without contradiction assigned to position ‘e3’. The initial hypothesis is refuted. We must launch a new hypothesis about the occurrence of /ť/ and /ď/. They cannot of course occur in ‘pre1’ and ‘pre2’, as this would violate (α1), so we are left with ‘e2’ and ‘e1’, but once again, the assignment of the phonemes to either of them carries with it a condition they have the properties of these positions. However, before long we find out it cannot be ‘e2’. Phonemes belonging there have the capacity of being followed by one phoneme from position ‘e1’, which is again not the case for /ť/ and /ď/ due to (α2). The only option left is to assign them to ‘e1’ which is the position immediately next to the nucleus. However, although this is in accord with (α2), it violates (α1) as /ť/, /ď/ can only be preceded by up to two peripheral phonemes, whereas phonemes occurring in ‘e1’ can be preceded by to five peripheral phonemes (cf. /j/ in /FSkvjeT/). Moreover, there are two other significant factors. First, neutralization of voicing takes place before

6 In general, a phoneme /α/ occurring in position β can be preceded by and combined with one or more phonemes belonging to the position(s) preceding β (if they are any such positions). Similarly, a phoneme /α/ occurring in position β can be followed by and combined with one or more phonemes belonging to the position(s) following β (if there are any such positions)

86 /ť/ and /ď/, which never happens before any phoneme belonging to ‘e1’ (cf. /Sťīn/ stín × /sňīT/ snít). Second, all phonemes or combinations preceding /ť/ and /ď/ can be easily mapped onto positions ‘pre1’ and ‘pre2’, and we have no reasons not to assign them there in accord with the Principle of Exclusive Membership. The neutralization and the ability to be preceded by just two phonemes from ‘pre1’ and ‘pre2’ is a thing /ť/ and /ď/ share with phonemes belonging to position ‘e3’. But the ability to be always followed by a vowel is a thing share with phonemes belonging to ‘e1’. They have the properties of ‘e3’ and ‘e1’, yet they cannot belong to either (and not even to ‘e2’). At the same time, we cannot leave them unassigned to any pre-nuclear position. In order to find a way out, it is now the point where the concept archi-position comes in. An archi-position, formally defined as the intersection of two or more positions, may be viewed as a position in a sub-system which represents two or more positions in the overall system. By a sub-system is here meant a particular constellation of entities applicable to a certain situation, here to the occurrence of /ť/ and /ď/ for which the difference between positions ‘e3’, ‘e2’ and ‘e1’ (which were postulated to capture certain phonotactic properties) has been made redundant—it does not have any functional validity for them. The purpose of archi-positions is to account for the distribution of phonemes in certain contexts with the distribution equivalent to two or more phonemes in other contexts. For /ť/ and /ď/, we posit an archi-position which we call ‘E1’ and which is the intersection of positions ‘e3’, ‘e2’ and ‘e1’ (formally: ‘e3∩e2∩e1’). It is equivalent to these three positions representing them in a given sub- system. Let us note that functional equivalency does not imply functional identity, and ‘E3’ is not identical to any of positions ‘e3’, ‘e2’ or ‘e1’; it is only their representative. The way /ť/ and /ď/ are plotted onto the distributional unit is schematized in Figure V. Another pre-nuclear archi-position we find necessary to introduce is ‘E2’ accounting for the distribution of /p/, /b/, /f/. It is the product of the suspension of the difference between positions ‘e2’ and ‘e3’ (formally: ‘e2∩e3’). These phonemes can be preceded by up to two phonemes only—the phonemes that can be mapped onto positions ‘pre1’ and ‘pre2’. Second, they can only be followed by one phoneme which can be mapped onto position ‘e1’; see again Figure V. From this follows that /p/, /b/, /f/ cannot be without contradiction assigned to positions ‘e3’ or ‘e2’, and the difference between

87 them is suspended. Note that /f/ is only attested to be preceded by one phoneme, either from ‘pre1’ or from ‘pre2’ (only in the place name Kfely).

The overall distributional unit, pre-nuclear context ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ A contextual distributional unit, pre-nuclear context ‘pre2’ ‘pre1’ ‘e3∩e2∩e1’ / ‘E3’ ‘n’ Phonological form /T/ /Š/ /ť/ /i/ /TŠťi/ čti /F/ /S/ /ď/ /e/ /FSďel/ from vzdělaný

‘pre2’ ‘pre1’ ‘e3∩e2’ / ‘E2’ ‘e1’ ‘n’ Phonological form ∅ ∅ /p/ /r/ /a/ /praX/ prach /T/ /Š/ /p/ /j/ /e/ /TŠpjeT/ čpět ∅ ∅ /b/ /ř/ /e/ /břeX/ břech /F/ /S/ /b/ ∅ /u/ /FSbuŤ/ vzbuď ∅ ∅ /f/ /ň/ /u/ /fňuK/ from fňukat ∅ /S/ /f/ ∅ /ö/ /SföK/ from sfouknout /K/ ∅ /f/ ∅ /e/ /Kfeli/ Kfely

Figure V: Phonotactic analysis of /ť/, /ď/, and of /p/, /b/, /f/ in the pre-nuclear context.

We will now move to the post-nuclear context for which we have postulated three positions on the basis of forms like /vojSK/, /pomST/ and others. Having analyzed further the distribution of phonemes in this context, we have come to a conclusion that one archi-position should be introduced here: archi-position ‘I’. It is necessary for dealing with the distribution of /m/, /n/, /ň/, /r/ and /l/ in forms like /jilm/ jilm, /hejn/ hejn, /Tšerň/ čerň, /xejr/ chejr (SSJČ) and /Ktejl/ from koktejl, respectively. Namely, we have to deal with the post-nuclear combinations /lm/, /jn/, /rň/, /jr/ and /jl/. The first phonemes of these combinations are /l/, /j/ and /r/; they can be assigned to position ‘i1’7. On the other hand, the second phonemes, /m/, /n/, /ň/, /r/ and /l/ cannot be without contradiction assigned to any of post-nuclear positions ‘i2’, and ‘i3’. The reason is simple: in these particular combinations they cannot be further followed by another phoneme or even preceded a phoneme. Our conclusion is that the difference between positions ‘i2’ and ‘i3’ is suspended resulting in archi-position ‘I’ (formally: ‘i2∩i3’). The way the phonemes are mapped into this archi-position is given in Figure VI. One point should be mentioned: archi-position ‘I’ is filled with a phoneme if and only if

7 Where we have already assigned them in the previous analysis, see Figure IV above.

88 position ‘i1’ is not empty. This is because the phonemes /m/, /n/, /ň/, /r/ and /l/ normally occur in position ‘i1’ where they can be followed by one or two peripheral phonemes, but never preceded by any. On the other hand, when occurring in ‘I’, they are preceded by exactly one peripheral phoneme, but never followed by any.

The overall distributional unit, post-nuclear context ‘n’ ‘i1’ ‘i2’ ‘i3’ A contextual distributional unit, post-nuclear context ‘n’ ‘i1’ ‘i2∩i3’ / ‘I’ Phonological form /i/ /l/ /m/ /jilm/ jilm /e/ /j/ /n/ /hejn/ hejn /e/ /r/ /ň/ /Tšerň/ čert /e/ /j/ /l/ /tejl/ from koktejl /e/ /j/ /r/ /xejr/ chejr

Figure VI: Phonotactic analysis of /m/, /n/, /ň/, /l/ and /r/ in the post-nuclear context.

The very last archi-position we postulate is ‘N’. It results from the suspension of the difference between pre-nuclear position ‘e1’ and nuclear position ‘n’ (formally: ‘e1∩n’). It is a slot where /r/, /l/ occur if they function as nuclear entities of phonotagms. Since archi-position ‘N’ is a product of the suspension of the difference between pre-nuclear position ‘e1’ and nuclear position ‘n’, it shares in fact the properties of both, and so nuclear /r/ and /l/ occurring here belong at the same time to the pre-nuclear and to the nuclear context. The reason why archi-position ‘N’ has been introduced derives from the following properties of nuclear /r/, /l/. Consequently, nuclear /r/, /l/ are mapped onto the distributional unit in the way given in Figure VII.

Phonotactic properties of nuclear /r/, /l/ (α3) They cannot be preceded by any of the phonemes belonging to position ‘e1’, i.e. by /m/, /n/, /ň/, /r/, /l/, /j/ or /ř/. (β3) They can be preceded by one, two, three or four phonemes all of which can be mapped onto positions ‘e2’, ‘e3’, ‘pre1’ and ‘pre2’. (γ3) They can be followed by one, two or three phonemes belonging to positions ‘i1’, ‘i2’ or ‘i3’.

89 The overall distributional unit ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘i1’ ‘i2’ ‘i3’ A contextual distributional unit ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1∩n’ / ‘N’ ‘i1’ ‘i2’ ‘i3’ Phonol. form ∅ ∅ /s/ /M/ /r/ /T/ ∅ ∅ /sMrT/ smrt /T/ /Š/ /t/ /v/ /r/ /T/ ∅ ∅ /TŠtvrT/ čtvrt /F/ /S/ /t/ ∅ /r/ /T/ /Š/ ∅ /FStrTŠ/ vstrč ∅ ∅ /p/ /r/ ∅ /S/ /T/ /prST/ prst ∅ ∅ /š/ /M/ /r/ /n/ /T/ /S/ /šMrnTS/ šmrnc

Figure VII: Phonotactic analysis of nuclear /r/ and /l/.

In conclusion, we would like to mention one point. Although an archi-position is an effective concept by which we can capture special distribution of some phonemes, it is not always possible or desirable to employ it without making arbitrary decisions. Unmotivated decisions should be avoided in an analysis, for otherwise we would loose control over it. For instance, /g/ (and it holds for /f/, too) has, due to historical reasons, rather limited distribution. It can be followed by one phoneme belonging either to ‘e2’ or to ‘e1’, and it can also be preceded by one phoneme belonging either to ‘pre1’ or to ‘pre2’. If we were to introduce archi-positions to capture these facts, our choices would be arbitrary. As /g/ displays similar distributional properties like the other velars occurring in position ‘e3’, we have decided to assign /g/ there as well.

5.4 Phonotactic properties and collocational restrictions

We continue, in the manner outlined above, analyzing self-contained combinations of phonemes in an attempt to determine which positions or archi-positions these phonemes occupy. We stick to the Principle of Exclusive Membership and the Principle of Mutual Commutability. We assume that, firstly, a phoneme, once determined to occur in a certain position, always occurs there. Secondly, we assume that phonemes with which it commutes also occur in that position. Although such assumptions will often have to be modified, they are good starting hypotheses. A phoneme initially assumed to occur in a certain position may later prove to occur in another position or in an archi-position. Still, it is quite unlikely that each and every one phoneme would occur in a different

90 position. To say it otherwise, we will sooner or later find out that there is a certain limited set of phonemes ascribable to one position; this set will be called position class (Mulder 1968). It will be abbreviated as pos, so e.g. pos ‘e2’ stands for the position class for pre-nuclear position ‘e2’. Nuclear position classes are dealt with in Chapter 7; pre-nuclear ones in Chapter 8; and post-nuclear ones in Chapter 9. A position class may include ∅ which means that it can also be empty; conditions under which a position is empty are discussed in the next chapter. For logical reasons, a position class must have at least two members, either two phonemes or a phoneme and ∅ (see Section 5.2). For the ease of reference we refer to Figure VIII summarizing, in advance, position classes of what we will call main distributional unit in the next chapter.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘i1’ ‘i2’ ‘i3’ k g x h r l P T K P T K v P T K t d s z m n ň S Š F S Š X T M j r l S Š š ž j ř F X M j r l S Š ∅ i e a o u P T K Ť v t h ř ∅ r l m n ň ř ī ē ā ō ū ∅ (b) ∅ p b f ∅ ë ä ö ∅ ∅ ∅ ť ď m n ň r l

Figure VIII: Main distributional unit and its position classes. Phonemes stretching over several positions belong to an archi-position. ∅ means a position can be empty.

If a phoneme belongs to a certain position, it possesses, in principle, by this virtue all phonotactic properties of that position. By a phonotactic property we mean distributional and combinational capacities of a phoneme, that is, its occurrences in various positions of the distributional unit, and its ability to combine with other phonemes from other positions. Position classes are sets of phonemes which share certain phonotactic properties. A position is then an expression of these properties or the intension of a certain class of phonemes with similar distribution. Every position is defined by a unique set of phonotactic properties. For a phoneme to belong to a certain position, it is necessary that it display phonotactic properties unambiguously determining its membership in the position in question. Thus, even though both /o/ and nuclear /r/ can be preceded by up to four peripheral phonemes (cf. /PŠtroS/ pštros and /TŠtvrT/ čtvrt), and followed by up to three peripheral phonemes (cf. /vojSK/ vojsk and /šMrnTS/ šmrnc), they belong to different positions because /r/ cannot be preceded by

91 phonemes from pos ‘e1’ whereas /o/ can (cf. e.g. /noS/ nos), i.e. there is a phonotactic property which /o/ possesses, but /r/ does not. Some phonemes can belong to more positions provided that they display phonotactic properties of more positions. The sum of their phonotactic properties is then equal to the sum of the properties of the individual positions. Actually, all phonemes of Czech except for vowels and /Ť/ belong to more than one position classes. For example, /j/ is a member of three position classes, pos ‘e1’, pos ‘pre1’ and pos ‘i1’ because it displays phonotactic properties of all the three positions. It can stand right before a nuclear phoneme, and be preceded by other peripheral phonemes (cf. /svjeT/ svět) or it can stand before a peripheral phoneme and be the first phoneme in the phonotagm (cf. /jSte/ jste) or it can stand right after a nuclear phoneme and be followed by other peripheral phonemes (cf. /vojSK/ vojsk). For each of these possibilities, there are other phonemes possessing the same properties (cf. /r/ in /svraP/ svrab, /rti/ rty and /verST/ verst). Above we deliberately wrote that a phoneme can occur in a certain position, it possesses in principle all phonotactic properties of the position in question because each and every one phoneme need not be actually attested to display each and every one phonotactic property. The absence of some may be accidental if there is not enough evidence to declare it regular; we must find out whether it concerns some well-defined class of phonemes (see Section 3.4). For instance, all phonemes of pos ‘e1’ (i.e. /m/, /n/,

/ň/, /j/, /ř/, /r/, /l/) have the ability to be preceded by two peripheral phonemes ‘C1C2’ where ‘C1’ is from pos ‘pre2’ (i.e. /P/, /F/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/, /v/, /t/,

/h/) and ‘C2’ from pos ‘e2’ (i.e. /v/, /M/). However, this particular property is only attested for /ň/, /j/ and /l/, cf. /vMň/ in vmněstnat, /rvj/ in rvěte and /vMl/ in vmlátit. Yet, there is no reason to think the other phonemes do not have this property, too, and that no Czech word could begin with the combinations like /vMn/, /vMř/ or /vMr/. They are missing by accidence because it is not possible to discern some regularity behind their absence. On the contrary, the absence of the combination /vMm/ is not accidental because it is predictable from another fact. If all phonemes from pos ‘e1’ have the capacity to be preceded by all phonemes from pos ‘e2’ (i.e. /v/, /M/), it is only the combination /Mm/ is not found. The nasal /m/ is a labial, and if we examine combinability of labials in the pre-nuclear context, we find out that no labial is attested

92 to be preceded by /M/. We can regularize this fact by proposing that no labial can be preceded by a nasal in the pre-nuclear context, and by failing to refute this hypothesis, we adopt it as a descriptive statement about the distribution of labials. Statements like these, that is, unrefuted hypotheses about co-occurrences of certain classes of phonemes, will be called collocational restrictions (a term adopted from Fudge 1969; analyzing Czech, Kučera 1961 speaks about sequential constraints). They will be introduced in the course of the analysis, and eventually, for the sake of reference and clarity, numbered and summarized in Appendix A. However, we have not attempted to provide any explanations why these restrictions hold for Czech, though many of them can no doubt be explained externally, either by general phonetic restrictions or by the historical development of Czech. The mentioned impossibility of /Mm/ (i.e. of [mm]) can be explained by the fact that no Czech word can begin with two identical contoids. Similarly, the impossibility of diphthongal vowels /ë/, /ä/, /ö/, which are realized as [ɛu̯], [au],̯ [ou̯], to be followed by /j/ can be explained by saying that two glides (i.e. [u̯]/[w] and [j]) cannot stand next to each other in Czech. But other restrictions cannot perhaps be readily explained. It is hard to say why the nasals /m/, /n/, /ň/ cannot be preceded by /b/ if they can by /p/ (cf. /pnöT/ pnout and /pňi/ pni), but maybe an explanation can be found in the history of Czech. There are two restrictions we would like to introduce just at this point; they concern the co-occurrence of identical phonemes. No two identical phonemes (“geminates”) can stand in close proximity in Czech within a single phonotagm; it includes even archiphonemes (= restriction RG1a in Appendix A). Geminates are possible across phonotagm boundaries as in /dvojjaziTšnī/ dvojjazyčný. This restriction must be further specified due to neutralization of voicing: Within a single phonotagm a voiceless and voiced occlusive, or a voiceless and voiced fricative cannot stand in close proximity with a voicing archiphoneme of the same place and manner of articulation (= RG1b). It means that combinations like /Tt/, /Fv/ are also not allowed within a single phonotagm, though they are again possible across phonotagm boundaries, cf. /TštvrTtōn/ čtvrttón.

93 5.5 Postscript

Originally,8 we operated with four post-nuclear positions on the basis of the form /borŠTŠ/ borsč “borsch”. Eventually, this analysis has been turned down, and the form /borŠTŠ/ excluded from out analysis. The decisive reason for this was the fact that it is the only example of a nuclear phoneme followed by four post-nuclear phonemes. On the other hand, there are at least 19 three-phoneme post-nuclear combinations. With the four post-nuclear positions, the phonotactic analysis of these combinations proved complicated and quite redundant because we had to either postulate several archi- positions or declare many positions empty. That is, there being four post-nuclear positions, either one of the phonemes had to occur in an archi-position or one position had to be empty in a form like /vojSK/. The establishment of archi-positions complicated the analysis, and empty positions probably introduced a great degree of inconsistency. Once the form /borŠTŠ/ was removed from consideration, the whole analysis became neater. Other reasons, though not decisive, support the exclusion. The phonotagm /borŠTŠ/ is a phonological form of boršč introduced to Czech from Russian, and thus it may resemble the structure of Russian rather than that of Czech. It also suggested by the fact that the combinations /rŠT/ and /ŠTŠ/, included in /rŠTŠ/, are not otherwise attested in Czech. The latter point is more important than the former because the origin of words is actually irrelevant for a synchronic analysis. The least decisive, though still worth consideration, is that fact that boršč ends in four peripheral phonemes only because the affricate [ʧ] has been analyzed as a two-phoneme combination. Even though it is well justified, such an analysis may be viewed with suspicion by others. The exclusion of boršč does not of course remove this suspicion, but it makes, hopefully, our phonotactic analysis more agreeable. The inclusion of boršč necessitates a statement like this: “phonotagms in Czech can end in up to four peripheral phonemes”, whereas its exclusion allows for a much more intuitively satisfactory statement “phonotagms in Czech can end in up to three peripheral phonemes”.

8 See Bičan (2011b), originally presented at the conference Formal Description of Slavic Languages 8 in Potsdam, Germany, 2–5 December 2009.

94 Chapter Six PHONOTAGMS

6.1 Major-type phonotagms

The purpose of this chapter is to work out a difference between a class of major-type phonotagms and a class of minor-type phonotagms. The former are genuine phonotagms that can be exhaustively described on the distributional unit we have posited in the preceding chapter. To repeat, this unit contains nine positions: five pre- nuclear, one nuclear and three post-nuclear ones. This is its default variant. The syntagmatic difference between some of these positions becomes redundant in some contents resulting in an archi-position. We have recognized four such archi-positions. A distributional unit containing at least one is a contextual variant of the distributional unit. As there are four archi-positions, their combinations produce 9 contextual variants; see Figure I. The type marked with ‘0’ is the default variant. Those marked with ‘1’ are contextual variants with at least one archi-position. Number ‘2’ designates variants with two archi-positions. Finally, number ‘3’ stands for the variant with three archi-positions. All attested major-type phonotagms in Czech can be mapped on the default variant of the distributional unit or on any of its contextual variants. To distinguish it from the semi-distributional unit we introduce below, we will call this type the main distributional unit. Although all major-type phonotagms are instances of either the default variant of the main distributional unit or of one of the contextual variants, it does not mean that all positions have to be always occupied by a phoneme. In the default variant it is only the nuclear position that is compulsorily occupied; the others may be empty. In fact, the nuclear position must not be empty in any variant of the distributional unit. In the subtypes involving one or more archi-positions the archi-positions must be occupied, for otherwise there would be no reason to operate with them. In subtypes ‘1d’ and ‘2d’

95 at least one of positions ‘e3’ or ‘e2’ must be occupied because nuclear /r/, /l/ are always preceded by at least one consonant (see Section 7.2). Archi-position ‘I’ is only feasible when ‘i1’ is not empty; that is why there are not more variants with this archi-position.

0 ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘i1’ ‘i2’ ‘i3’

1a ‘pre2’ ‘pre1’ ‘E1’ ‘n’ ‘i1’ ‘i2’ ‘i3’ 1b ‘pre2’ ‘pre1’ ‘E2’ ‘e1’ ‘n’ ‘i1’ ‘i2’ ‘i3’ 1c ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘i1’ ‘I’ 1d ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘N’ ‘i1’ ‘i2’ ‘i3’

2a ‘pre2’ ‘pre1’ ‘E1’ ‘n’ ‘i1’ ‘I’ 2b ‘pre2’ ‘pre1’ ‘E2’ ‘e1’ ‘n’ ‘i1’ ‘I’ 2c ‘pre2’ ‘pre1’ ‘E2’ ‘N’ ‘i1’ ‘i2’ ‘i3’ 2d ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘N’ ‘i1’ ‘I’

3 ‘pre2’ ‘pre1’ ‘E2’ ‘N’ ‘i1’ ‘I’

Figure I: Main distributional unit and its contextual variants.

Let us now give examples of major-type phonotagms where each position is either filled with a phoneme or empty. Ideally, we should provide examples for every of its variants, but this would not be much illustrative because there are many ways the particular positions can empty and/or filled with a phoneme. For example, the phonotagms /a/, /ta/, /sen/ and /FSkvjeT/ are all examples of the default variant of the main distributional unit, but it is apparent they are quite different. Therefore, we have decided for another approach. We will list phonotagms according to the number of positions filled with a phoneme, not according to which positions are filled. Figure II summarizes all thinkable types of phonotagms. ‘F’ stands for “followed by”, ‘P’ for “preceded by”, and the number designates by how many peripheral phonemes a nuclear phoneme is followed or preceded. Note that ‘V’ stands for any nuclear phoneme, i.e. for vowels as well as nuclear /r/, /l/. The shading indicates that no example has been found for the respective type. The examples are given in Figure III. The absence of any example for the type ‘CCCCCV’ must be accidental because there is no reason a nuclear phoneme could not be preceded by five peripheral phonemes while not followed by any such phoneme. On the other hand, the absence of the types ‘CCCCCVCCC’, ‘CCCCCVCC’ and ‘CCCCVCCC’ can be regularized in one

96 important generalization: there is no phonotagm in which all nine positions of the distributional unit are filled. The maximum number of the filled positions is seven, cf. /SxlamST/ from schlamstnout and /FStřīTS/ vstříc. Another way of putting it is to introduce a collocational restriction according which no phonotagm in Czech can contain more than six peripheral phonemes (= RG2a)1; it must of course contain one nuclear phoneme. The rule has also one important supplement: as we see immediately, if the nuclear phoneme is either /r/ or /l/, there is no phonotagm of the types ‘CCCCVCC’ and ‘CCCVCCC’. Consequently, we propose that no phonotagm in Czech can contain more than five peripheral phonemes if /r/ or /l/ is its nucleus (= RG2b).

F0 F1 F2 F3 P0 ‘V’ ‘VC’ ‘VCC’ ‘VCCC’ P1 ‘CV’ ‘CVC’ ‘CVCC’ ‘CVCCC’ P2 ‘CCV’ ‘CCVC’ ‘CCVCC’ ‘CCVCCC’ P3 ‘CCCV’ ‘CCCVC’ ‘CCCVCC’ ‘CCCVCCC’ P4 ‘CCCCV’ ‘CCCCVC’ ‘CCCCVCC’ ‘CCCCVCCC’ P5 ‘CCCCCV’ ‘CCCCCVC’ ‘CCCCCVCC’ ‘CCCCCVCCC’

Figure II: Logically possible phonotagms types. Examples for the shaded types have not been found. ‘F’ means “followed by” and ‘P’ “preceded by”.

Figure III gives examples of the phonotagm types listed in Figure II; more examples can be found in Appendix D. We have preferred phonotagms attested in mono- phonotagmic words. In case no such word has been found, examples of phonotagms from poly-phonotagmic words are provided. The italicized types are attested both for vowels and nuclear /r/, /l/; the others have been found only for vowels. As we just mentioned, the types ‘CCCCCV’, ‘CCCCCVCC’, ‘CCCCVCCC’ and ‘CCCCCVCCC’ are not attested at all, and the types ‘CCCCVC’ and ‘CCCVCCC’ are not possible for nuclear /r/, /l/. The types ‘V’, ‘VC’, ‘VCC’ and ‘VCCC’ are also not possible for nuclear /r/, /l/ because these phonemes must be preceded by at least one consonant if they are to be nuclear entities. Moreover, they cannot be preceded by five peripheral phonemes; this fact is already encoded in the distributional unit, see Figure I above for the variants involving archi-position ‘N’. The types ‘CCCV’ and ‘CCCCV’ are also not

1 All restrictions are numbered and summarized in Appendix A.

97 possible for the nuclear /r/, /l/, that is, if nuclear /r/, /l/ occur at the end of a phonological word, they cannot be preceded by three or more peripheral phonemes (see Section 7.3).

Phonotagm type Example ‘V’ /a/ a, /i/ i, /o/ o ‘CV’ /ta/ ta, /se/ se, /hr/ hr (cf. být hr) ‘CCV’ /Sto/ sto, /dvö/ dvou, /Str/ from mistr, /Sxl/ from uschl ‘CCCV’ /MSda/ mzda, /Tkvī/ tkví, /Sklo/, /TStl/ from octl ‘CCCCV’ /STkvī/ stkví, /lStnī/ lstný ‘CCCCCV’ not found but possible ‘VC’ /on/ on, /aŠ/ až, /ēr/ ér ‘CVC’ /ten/ ten, /noS/ nos, /köT/ kout, /vrT/ vrt, /hlT/ hlt ‘CCVC’ /sňīT/ snít, /SdāT/ zdát, /SprX/ sprch, /Splň/ splň ‘CCCVC’ /StraX/ strach, /Xřtān/ chřtán, /Skvrn/ skvrn ‘CCCCVC’ /FSplāT/ vzplát, /PStruX/ pstruh, /TŠtvrT/ čtvrt ‘CCCCCVC’ /FSkvjeT/ vzkvět, /STkvjel/ stkvěl ‘VCC’ /erP/ erb, /aKT/ akt, /uST/ uzd ‘CVCC’ /pūST/ půst, /tanK/ tank, /prST/ prst, /plST/ plst ‘CCVCC’ /zmāST/ zmást, /SporT/ sport, /sMrŠŤ/ smršť ‘CCCVCC’ /FznēST/ vznést, /SkvoST/ skvost, /FStrTŠ/ vstrč ‘CCCCVCC’ /FSkvēST/ vzkvést, /TStnoST/ ctnost, /FStřīTS/ vstříc ‘CCCCCVCC’ not found and not possible ‘VCCC’ /ūTST/ úct ‘CVCCC’ /vojSK/ vojsk, /pomST/ pomst, /verST/ verst ‘CCVCCC’ /StēTST/ stéct, /zřīTST/ zříct, /šMrnTS/ šmrnc ‘CCCVCCC’ /SxlamST/ from schlamstnout, /StlöTST/ stlouct ‘CCCCVCCC’ not found and not possible ‘CCCCCVCCC’ not found and not possible

Figure III: Examples for phonotagm types from Figure II. Italicized types are attested both for vowels and nuclear /r/, /l/; non-italicized ones only for vowels.

To conclude this section, we should note that we have not examined the frequency of the occurrence of the phonotagm types or of individual phonotagms (or combinations within them) because our primary and foremost interest is the description of the structure of phonotagms. That is, our focus is only to determine whether a certain type is attested and/or possible, and some other type is not attested and/or possible, not whether some types are more often occurring than others, which is beyond the scope of this work. Yet, the research of this kind has already been conducted for Czech by others, particularly by Ludvíková (1972a, 1972b, 1976, 1978; cf. also Těšitelová et al.

98 1985: 21–8).2 However, we can hardly use results of this research due to different theoretical and methodological foundations. The most notable difference is the way poly-phonotagmic words are parsed to individual phonotagms (i.e. how words are syllabified). Ludvíková uses morphological criteria with which we cannot agree. She would apparently parse the word prostát “to spend standing” to pro and stát because the former is a prefix, but the word prostat “prostate (gen. pl.)” to pros and tat because there is not morphological boundary (Ludvíková 1972b, Těšitelová et al. 1985: 22–3). But apart from the length of /a/, these words are phonologically and phonetically identical, and should be treated alike. In our approach both contain the same types of phonotagms because a phonotagm is for us a unit of phonological division of phonological forms, not a unit of morphological division of signs. A phonological form and a sign are ontologically different entities, the former being just one aspect of the latter (see Section 3.3). Consequently, it should be obvious that our analysis of the frequency of phonotagm types would be different to that by Ludvíková.

6.2 Minor-type phonotagms

In Section 3.6 we dealt with the way phonological words are divided into individual phonotagms. We adopted the so-called Kuryłowicz condition as a helpful guideline for such an analysis. We also mentioned there are situations where the condition falls short. One example was the form /jelen/ jelen which could be, according to this condition, divided to /je–len/ as well as to /jel–en/. In this case the problem can be resolved by assuming functional amalgamation (i.e. distributional “ambisyllabicity”) of the medial /l/, i.e. by saying that the phoneme belongs to both phonotagms. Yet in other situations even the functional amalgamation is of no avail. Certain phonological words cannot be

2 And recently also by Volín – Churaňová (2010) who have examined whether occurrences of peripheral combinations in phonological forms of words, stress groups (presumably corresponding to our accent groups) and prosodic phrases (presumably corresponding to phonological forms of sentences) are random in Czech texts. Their conclusion is that they “are not entirely random, although they do not depart substantially from hypothetical unconstrained distributions” (p. 61).

99 parsed to phonotagms permitted at the beginning and at the end of phonological words, even though they are intuitively divisible. This problem is known from many languages (see Bell 1976 for a discussion), and we encounter it in Czech, too. What we have in mind is the division of the words such as obvaz, odjet or škrábnout. From all we know so far, the phonological words /obvaS/, /odjeT/ and /ŠkrābnöT/ appear to contain two phonotagms, as there are two vowels in each of them; the point is only how to set boundaries between them. Figure IV summarizes three possible divisions. None of them is in accord with the Kuryłowicz condition, though.

/obvaS/ /odjeT/ /ŠkrābnöT/ S1 /ob–vaS/ /od–jeT/ /Škrāb–nöT/ S2 /o–bvaS/ /o–djeT/ /Škrā–bnöT/ S3 /obv–aS/ /odj–eT/ /Škrābn–öT/

Figure IV: Three possible divisions (“syllabifications”) of obvas, odjet and škrábnout.

We will start with the S3 divisions given in Figure V. They produce phonotagm-final combinations /bv/, /dj/ and /bn/, but none of them is attested. Moreover, no combination of a similar structure is found, so we may safely conclude they are not possible (they are actually disallowed by the distributional unit). The S3 divisions thus produce ill-formed phonotagms. The output of the S2 divisions is also unattested combinations—this time phonotagm-initial ones—, but in this case they have at least some parallels in attested combinations like /tv/, /zj/ or /pn/ (cf. tvůj, zjev, pnout). Notwithstanding this fact, the combinations /bv/, /dj/ and /bn/ are still problematic because they violate the following collocational restrictions otherwise valid for all attested pre-nuclear combinations:

(α) In the pre-nuclear context a labial fricative cannot be preceded by another labial (= restriction RE1 in Appendix A). (β) If /j/ is preceded by a fricative, it can be no other than a labial or an alveolar one (cf. RG12a in Appendix A). (γ) A nasal cannot be preceded by /b/ (= RG8 in Appendix A).

100 These rules have been put forth as hypotheses, and so far have not been refuted unless we apt for the S2 divisions. The inevitable conclusion is such that either we remove the just given rules from our description, or we reject the S2 divisions. Before deciding which solution is the most appropriate, let us consider the S1 divisions. The S1 divisions are probably the ones a native speaker would regard appropriate, and those that other analysts of Czech would prefer. What is more, they would not quite probably be viewed as problematic at all.3 However, there is a problem with them— they violate the Kuryłowicz condition. To be more specific, the S1 divisions produce the forms /ob/, /od/ and /Škrāb/, but these are not permissible word-final phonotagms because the occlusives and fricatives are subject to neutralization of voicing at the end of words. Even if we did not operate with neutralization, the problem would persist. It is not neutralization which makes the mentioned phonotagms impermissible, but the fact they end in a voiced occlusive because there is no attested phonotagm which would end in one. It might be objected that in e.g. /hodňī/ hodní the first phonotagm does end in a voiced occlusive, but the truth is that we do not know that—we can only decide that! We are aware of no phonetic or phonological signal indicating the form should be divided to /hod/ and /ňī/ instead of /ho/ and /dňī/4; the choice of the division is thus a matter of our analysis. The only place we are sure there is an end of a phonotagm is the end of a phonological word, but no Czech word ends with a phonologically voiced occlusive or fricative. At most, the voicing can be predicted from the context, which is not the case of /b/ and /d/ in /obvaS/, /odjeT/ and /ŠkrābnöT/, cf. /kapverTSkī/ kapverdský, /matjes/ matjes (SSJČ) and /napnöT/ napnout. In short, the alleged phonotagms /ob/, /od/ and /Škrāb/ violate the following rule:

3 This is not quite true at least about Kučera (1961) who must have been aware of the problem. When listing phonemes capable of occurring at the end of the syllable, he includes both voiceless and voiceless obstruents, even though he notes that the difference between them is suspended in this context. It is because “this will later simplify the procedure for dividing interludes into codas and onsets” (p. 80). This does not solve the problem, though. 4 The morphological boundary, often invoked as a criterion, is irrelevant. In všední (historically vše+dní) it is elsewhere, and so it is in povodní (it is povodeň+í with the e being lost).

101 Rule of phonotagm-final neutralization of voicing There is no functional difference between voiceless and voiced occlusives, and between voiceless and voiced fricatives at the end of a phonotagm. (In other words: this opposition is neutralized giving rise to the voicing archiphonemes.)

As we see, none of the divisions given in Figure V yield permissible phonotagms. Therefore, the forms /obvaS/, /odjeT/ and /ŠkrābnöT/ are either single phonotagms, as we are not able to divide them to smaller phonotagms, or the Kuryłowicz condition should be abandoned. However, neither option appears satisfactory. The condition is a very reasonable principle. In fact, it says that our analysis should only generate well- formed phonotagms, and that the well-formedness can only be judged against the attested data. If the condition is retained, we must conclude that the discussed phonological words are mono-phonotagmic, not poly-phonotagmic. This is, however, problematic: were the forms /obvaS/, /odjeT/ and /ŠkrābnöT/ single phonotagms, the analysis we have so far carried out would have to be rejected because the distributional unit with five pre-nuclear, one nuclear and three post-nuclear positions cannot account for them. In /ŠkrābnöT/ it is most apparent: if /ö/ is the nucleus, the phonemes /Š/, /k/, /r/, /ā/, /b/, /n/ would have to occur in six pre-nuclear positions. Though not denying that this kind of analysis could not be carried out, we think it would be quite complicated. Therefore, we suggest instead making advantage of the concepts semi-distributional unit and minor-type phonotagm briefly introduced in Section 4.5. To start with, let us return to the divisions in Figure IV. Though all are problematic, we regard the S1 divisions as the least controversial ones. They are not only intuitively acceptable (which is a virtue but not a decisive point) but, and this is decisive, they do not produce any impermissible phoneme combinations. The only problem with them is that the alleged phonotagms /ob/, /od/ and /Škrāb/ end in a voiced occlusive, thus violating the above rule of phonotagm-final neutralization of voicing. Were they /oP/, /oT/ and /ŠkrāP/, they would be normal phonotagms containing one obligatory nuclear phoneme, and one or more optional peripheral phonemes. Except for the immunity to the neutralization of voicing, the combinations /ob/, /od/ and /Škrāb/ resemble genuine phonotagms. To account for combinations of this kind, we propose calling them minor-

102 type phonotagms thereby distinguishing them from the genuine, major-type phonotagms. To put it formally, a major-type phonotagm will be defined as a self- contained phonotactic entity capable of occurring at any place in a poly-phonotagmic phonological word, that is, as the very first phonotagm, as a medial phonotagm, or as the very last phonotagm. On the other hand, a minor-type phonotagm will be defined as the entirety of the phenomena which do not fall within the range of the main distributional unit (El-Shakfeh 1987: 513). In our particular case a minor-type phonotagm will be a phonotagm-like combination of phonemes occurring only before a major-type phonotagm and not capable of occurring on its own. Minor-type phonotagms are thus occurrence-dependent on the major-type phonotagms. If the structure of major-type phonotagms is describable on the main distributional unit we postulated in the preceding chapter, the structure of minor-type phonotagms will be accounted for by the so-called semi-distributional unit as a special attachment to the main distributional unit. We can now reformulate the neutralization rule given above:

Rule of phonotagm-final neutralization of voicing (revised) There is no functional difference between voiceless and voiced occlusives, and between voiceless and voiced fricatives at the end of a major-type phonotagm. (In other words: this opposition is neutralized giving rise to the voicing archiphonemes.) In minor-type phonotagms the difference is retained.

In the rest of this section we construct the semi-distributional unit. Before doing that, we must first know how minor-type phonotagms are identified. It derives from what has been said so far: A phonological word which contains two nuclear phonemes, but which cannot be divided to two major-type phonotagms is necessarily built of a minor-type phonotagm and a major-type phonotagm. It cannot be divided because the form contains an inter-nuclear phoneme combination not attested at the beginning or at the end of any directly attested phonotagm. To put it schematically, the phonological word must have the following structure:

(1) ‘(Cn)V(Cn)C1C2(Cn)V(Cn)’

103 In the formula ‘Cn’ stands for any number of peripheral phonemes with parentheses indicating they are optional. The symbols ‘C1C2’ stand for a combination of peripheral phonemes not permissible either at the beginning or at the end of major-type phonotagm. The necessary condition is that the phoneme ‘C1’ is either a voiceless or voiced occlusive or fricative, from which follows that ‘C2’ is a peripheral phoneme not triggering neutralization of voicing. Consequently, ‘C2(Cn)V(Cn)’ is a major-type phonotagm and ‘(Cn)V(Cn)C1’ is a minor-type one. Having analyzed phonological words in Czech, we conclude that the problematic ‘C1C2’ combinations are those listed in Figure V; none of them can occur at the beginning or at the end of a phonotagm. The bold face indicates the extension of minor-type phonotagms.

‘C1C2’ Occurring in /pv/ /kapveTSkī/ kapverdský /pm/ /SklepmiStr/ sklepmistr /bm/ /obmikaT/ obmykat, /obmeziT/ obmezit /fm/ /hofmiStr/ hofmistr, /šēfmexaniK/ šéfmechanik /ťm/ /objeťmi/ oběťmi, /huťmiStr/ huťmistr /ďm/ /loďmi/ loďmi /bn/ /ŠkrābnöuT/ škrábnout, /drobnī/ drobný /bň/ /Škrābňi/ škrábni, /drobňī/ drobní, /neblbňi/ neblbni /ťn/ /rtuťnatī/ rtuťnatý /ďn/ /Mňeďnatī/ měďnatý /bM/ /obMňen/ obměn, /obMňeKTšiT/ obměkčit /kM/ /šikMňe/ šikmě, /takMňeř/ takměř /gM/ /enigMňe/ enigmě, /zëgMňe/ zeugmě /xM/ /draxMňe/ drachmě /bv/ /obvaS/ obvas, /obvoT/ obvod /fv/ /šēfvijednavaTŠ/ šéfvyjednavač /tj/ /matjeS/ matjes /dj/ /odjeT/ odjet, /podjatī/ podjatý

Figure V: Peripheral combinations across minor-type and major-type phonotagms.

Although we have said that minor-type phonotagms have to be of the general structure ‘(Cn)V(Cn)C1’, Czech apparently avoids minor-type phonotagms ending in more than one peripheral phoneme, hence their structure is rather ‘(Cn)VC1’. This becomes clear after we examine the following words:

104 (2) pouštní, an adjective derived from poušť [pou̯ʃc] výpovědní, an adjective derived from výpověď [viːpovjɛc] závěťmi or závětmi or závětěmi, instr. pl. of závěť [zaːvjɛc] výpůstmi [viːpustmi] or výpustěmi [viːpuscɛmi], instr. pl. of výpusť [viːpusc] záštěmi [zaːʃcɛmɪ], instr. pl. of zášť [zaːʃc] žerděmi [ʒɛrɟɛmi], instr. pl. of žerď [ʒɛrc]

They demonstrate a few interesting things. First, if an adjective is derived from a word ending in a palatal occlusive, the palatals ť and ď are in spelling changed to the alveolars t and d as in poušť → pouštní and vypověď → výpovědní. Since the Czech orthography tends to reflect the pronunciation, it means that the words could be realized as [pou̯ʃtɲiː] and [viːpovjɛdɲiː], respectively, that is, with alveolar stops [t] and [d] before -ní. However, despite the spelling, they could also be realized as [pou̯ʃcɲiː] and [viːpovjɛɟɲiː], i.e. with palatal stops [c] and [ɟ] before -ní. There is free variation between alveolars and palatals, a phenomenon already mentioned in Chapter 2 in connection with words like matně, hodně and kontě (VSČ: 61–2, Zeman 2008: 109, Krčmová 2008: 205).5 And just as we interpret these words as /matňe/, /hodňe/ and /konťe/, we must interpret pouštní, výpovědní and similar words as /pöŠtňī/ and /vīpovjedňī/. They are subject to the following allophonic rule:

Realizations of /t/, /d/, /n/ before /ť/, /ď/, /ň/ If, within a phonological word the alveolar occlusives /t/, /d/ or the alveolar nasal /n/ stand before the palatal occlusives /ť/, /ď/ or the palatal nasal /ň/, the former may be realized either as alveolar stops [t], [d], [n], respectively, or as palatal stops [c], [ɟ], [ɲ], respectively.

A corollary of this rule is the following important point: While the inter-nuclear combinations /ťm/, /ďm/, /ťn/, /ďn/ (cf. Figure V) are attested and possible, the inter- nuclear combinations /ťň/, /ďň/ are not. Likewise, while the pre-nuclear combinations

5 These pronunciation guides do not mention the words pouštní and výpovědní, but they mention hutník derived from huť for which both [ɦutɲiːk] and [ɦucɲiːk] are acceptable.

105 /tm/, /dm/, /tň/, /dň/, /tn/, /dn/ are attested and possible, the pre-nuclear combinations /ťm/, /ďm/, /ťň/, /ťň/, /ťn/ and /ďň/ are not. The other interesting thing about the examples (2) is what happens when instrumental plurals are derived from nouns ending in a palatal occlusive. The suffix used is either -mi or -emi. The latter seems to be preferred, and is only used for words like zášť or žerď. However, if -mi is used, two things happen. First, if a word ends only in a palatal, the palatal may be retained as in loďmi or changed to the alveolar as in závětmi6. Second, if it ends in two consonants the second of which is a palatal, this palatal appears to be always changed to the alveolar as in výpustmi7. In words ending in two consonants, however, it is more common that the suffix -emi is used instead; cf. zášť or žerď under (2). The spelling of výpustmi suggests it is pronounced as [viːpustmɪ] instead of [viːpuscmɪ], though the latter pronunciation probably occurs in substandard Czech.8 Consequently, we claim that the phonological representation of this word is /vīpuStmi/ instead of /vīpuSťmi/. If it were the latter, the form would have contained a minor-type phonotagm /puSť/ because the combination /ťm/ is not a possible phonotagm-initial or phonotagm-final combination. But since the form is /vīpuStmi/, it poses no problem as both /Stm/ and /tm/ are attested phonotagm-initial combinations. The purpose of this digression was to show that there are no minor-type phonotagms ending in two or more peripheral phonemes. It means that while major-type phonotagms may end in none or in one, two or three peripheral phonemes (cf. /je/ je, /leF/ lev, /leSK/ lesk, /vojSK/ vojsk), minor-type phonotagms must end in one and only one peripheral phoneme which, in addition, is either a voiceless or voiced occlusive or fricative. This very fact is what distinguishes major-type phonotagms from the minor-type ones, but their constituency does not differ in any other way apart from that. Both contain an obligatory nuclear phoneme which may be preceded by none or by several peripheral phonemes (cf. /ob-/ from obmykat and /Škrāb-/ from škrábni). Admittedly, no minor- type phonotagm beginning with a combination of four or five peripheral phonemes has been found, but we may still, for the sake of simplicity of the overall analysis, assume

6 závěťmi is also possible but not *lodmi. See PMČ: 260–1, MSČ: 180–1 on this issue. 7 Or, instead, the instr. pl. suffix -emi is used. This option is the only possible one for zášť, žerď or poušť. 8 To the best of our knowledge, no study deals with this phenomenon.

106 that the pre-nuclear phonemes in minor-type phonotagms are distributed over the same grid of the five pre-nuclear positions postulated for the major-type phonotagms. The phonotactic analysis of minor-type phonotagms will then look like the one in Figure VI where the top row is the default variant of the semi-distributional unit. The difference between certain pre-nuclear positions becomes suspended in some situations giving rise to archi-positions as in the main distributional unit.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘im’ Example ∅ ∅ ∅ ∅ ∅ /o/ /b/ /ob-/9 ∅ /Š/ /k/ ∅ /r/ /ā/ /b/ /Škrāb-/ /r/ ∅ /t/ ∅ ∅ /u/ /ť/ /rtuť-/ ∅ ∅ /d/ ∅ /r/ /o/ /b/ /drob-/ ∅ ∅ ∅ /M/ /ň/ /e/ /ď/ /Mňeď-/ ∅ ∅ /b/ /l/ /b/ /blb-/

Figure VI: Phonotactic analysis of minor-type phonotagms

In Figure VI the post-nuclear position is marked as ‘im’, the subscript indicating it is the post-nuclear position in minor-type phonotagms. The presence of this position in contrast with positions ‘i1’, ‘i2’ and ‘i3’ is the only difference between the semi- distributional unit and the main distributional unit. Considering the combinations in

Figure VI, we conclude ‘im’ may be occupied by one of the following phonemes:

(3) /p/, /b/, /f/, /ť/, /ď/, /t/, /d/, /k/, /g/, /x/

The phonemes /f/ and /x/ are the only fricatives to occur in ‘im’; their voiced counterparts /v/ and /h/ do not occur here. The voiced labial fricative /v/ cannot do so because there is no inter-nuclear combination whose division would assign it to this position. For instance, unlike /pm/, /bm/ and /fm/, the inter-nuclear combination /vm/ as in /revma/ revma can be parsed to /reF/ and /vma/, since /vm/ is attested at the beginning of phonotagms, cf. /vmīxaT/ vmíchat. As to /h/, we can speculate that it might be occur in this position because we see no reason a combination /hMň/ could not

9 The dash will indicate it is a minor-type phonotagm.

107 be possible if /xMň/ is. But until further evidence is available, /h/ will not be considered as a member of the position class ‘im’. The default variant of the semi-distributional unit and its contextual variants are given in Figure VII. The occurrence of phonemes in the particular positions is subject to the same conditions like the main distributional unit. Concluding our discussion, we would like to remark that minor-type phonotagms do not only occur at the beginning of a phonological form like in /obvaS/, but also between two phonotagms as in /objeťmi/. Here, the minor-type phonotagm /bjeť-/ is in between two major-type phonotagms /oP/10 and /mi/, the latter being the one upon which it is dependent. Other examples are /neblbňi/ or /enigMňe/. In general, however, the occurrence of minor-type phonotagms is limited in phonological forms of Czech words, constituting indeed only a minor subset of phonotactic phenomena.

0 ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘im’

1a ‘pre2’ ‘pre1’ ‘E1’ ‘n’ ‘im’

1b ‘pre2’ ‘pre1’ ‘E2’ ‘e1’ ‘n’ ‘im’

1c ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘N’ ‘im’

2 ‘pre2’ ‘pre1’ ‘E2’ ‘N’ ‘im’

Figure VII: Semi-distributional unit and its contextual variants.

10 With functional amalgamation between /P/ and /b/ (see Section 3.5).

108 Chapter Seven DISTRIBUTION IN THE NUCLEAR CONTEXT

7.1 Position class ‘n’

Phonemes occurring in the nuclear context are those capable of functioning as nuclear entities, which in Czech are vowels and semiconsonants. As the nuclear entity serves as the identity element of a phonotagm, it can never be omitted. Position ‘n’ is the nuclear position sui generis; it is occupied only by vowels which always function as nuclear entities; its position class is given under (1). On the other hand, nuclear archi-position ‘N’, dealt with in the next section, is occupied by the semiconsonants, which can function both as a nuclear and peripheral entities.

(1) pos ‘n’ ∈ {/i/, /e/, /a/, /o/, /u/, /ī/, /ē/, /ā/, /ō/, /ū/, /ë/, /ä/, /ö/}

All of the phonemes under (1) possess in principle all phonotactic properties of position ‘n’, though as we see below, some additional collocational restrictions play their role. Phonotactic properties of the position are schematized in Figure I. Examples are provided in Appendix D. Compare it with Figure II in the previous chapter. In dealing with phonotactic properties of individual positions in this and the following chapters we provide for every position class a list of phonotactic properties pertinent to it. In a phonotagm a vowel may be preceded (= ‘P’) by no other phoneme (= ‘P0’), in which case it is the first phoneme of the phonotagm, or it may be preceded by one (= ‘P1’), two (= ‘P2’), three (= ‘P3’), four (= ‘P4’), or five (= ‘P5’) peripheral phonemes. If a vowel is preceded by one or more peripheral phonemes, it is possible to specify to which positions these belong, and this can be done with subscripts. Code ‘P1e2’ means that a vowel can be preceded by one phoneme from pos ‘e2’. Code ‘P2pre1e3’ means it

109 can be preceded by two phonemes, one of them being from pos ‘pre1’ and the other from pos ‘e3’ (e.g. /a/ in /Stan/ stran). Furthermore, a vowel can be followed (= ‘F’) by no peripheral phoneme (= ‘F0’), or it may be followed by one (= ‘F1’), two (= ‘F2’), or three (= ‘F3’) peripheral phonemes. It is again possible to specify by subscripts to which positions these belong, so e.g. ‘F2i1i2’ means a vowel is followed by two phonemes belonging, respectively, to ‘i1’ and ‘i2’ (e.g. /a/ in /tanK/ tank).

F0 F1 F2 F3 /e/ /i/ /e/ /a/ /o/ /u/ P0 /ī, /ē/ /ā/ /ō/ /ū/ /ī/ /ī/ /ē/ /ā/ /ō/ /ī/ /ē/ /ā/ /ō/ /ë/ /ä/ /ö/ /ö/ /ë/ /ä/ /ö/ /ë/ /ä/ /ö/ (/i/) P1 /ū/ /ō/ /ō/ /ū/ /ë/ /ë/ /ä/ /ö/ /i/ /e/ /a/ /o/ /u/ P2 /ō/ /ā/ /ō/ /ū/ /ë/ /ë/ /ö/ /ë/ /ä/ /ë/ /ä/ /i/ /e/ /o/ /u/ P3 /ō/ /ō/ /ō/ /ī/ /ā/ /ō/ /ū/ /ë/ /ä/ /ë/ /ä/ /ë/ /ä/ /ö/ /ë/ /ä/ /i/ /a/ /o/ /u/ /i/ /e/ /a/ /u/ P4 /ō/ /ū/ /ō/ /ū/ /ā/ /ō/ /ū/ – /ë/ /ä/ /ë/ /ä/ (/ö/) /ë/ /ä/ /ö/ /i/ /a/ /o/ /u/ P5 – /ī/ /ē/ /ā/ /ō/ /ū/ – – /ë/ /ä/ /ö/

Figure I: Phonotactic properties of pos ‘n’; phonemes listed are not attested to possess a given property. Phonemes in brackets have the respective property only in marginal words. Empty cells mean the property is attested for all members of pos ‘n’, ‘–’ that it is for none.

The specification to which the preceding and following phonemes belong is very useful when we examine phonotactic properties of peripheral phonemes in Chapters 8 and 9 because we can limit the precedence and antecedence to the particular context we deal with. That is to say, when dealing with phonotactic properties of a pre-nuclear position, we can examine by which positions it can be preceded or followed in the pre- nuclear context only, not in the context of the whole phonotagm. Similarly, we can examine precedence and antecedence of post-nuclear positions in the post-nuclear context only. Consequently, /t/ in /StraX/ stran will be said to possess properties ‘P1pre1’ and

‘F1e1’, i.e. it can be preceded by a phoneme from pos ‘pre1’, and followed by a pho-

110 neme from pos ‘e1’. On the contrary, doing the same for nuclear phonemes has proven rather bothersome and lengthy because they display vast combinational possibilities. Vowels may be preceded by no phoneme, or by one phoneme, or by two, three, four, or five phonemes. These preceding phonemes can be distributed over several pre-nuclear positions, and if we also count the situation when a vowel is preceded by no phoneme, there are 40 possibilities in total.1 At the same time, vowels can be followed by no phoneme, or by one phoneme, or by two or three phonemes; these following phonemes can be distributed over several post-nuclear positions, and there are in total 10 possibilities (including also when not followed any phoneme). In total, there are 400 ways how vowels can be preceded and followed by peripheral phonemes, and it is for this reason we have chosen not to account for all of these possibilities, but only to examine the capacity of nuclear phonemes to be preceded and followed by a given number of peripheral phonemes without specifying to which positions these phonemes belong. Let us return to Figure I. Each cell of the table corresponds to one phonotactic property. In principle, any one phoneme belonging to a position class should automatically possess every single property, but this situation is seldom encountered.2 The figure lists phonemes for which a given phonotactic property is not attested. If a cell contains a dash ‘–’, the property is not attested for any member of the position class. On the contrary, if it is empty, the property is attested for all members of the class. The phonemes listed as not possessing a certain property may lack this property by accidence, or they may be regularly incapable of possessing it. The difference between accidental gaps and structural restrictions rests upon our ability to show a certain absence is pertinent to a well-defined set of entities in which case we can introduce a structural law accounting for this absence; failing to do so, we must conclude the absence is merely accidental. The distribution of vowels is in many ways limited, but the limitations are mostly accidental. For instance, it must be a mere accident that there is no phonological word where /i/ is preceded by four peripheral phonemes because its long counterpart /ī/ is attested in such a situation (cf. /lStnī/ lstný). Aside from the regular restrictions to be discussed presently, it is obvious that Czech tends to avoid, first, phonotagms beginning

1 For example, a vowel being preceded by three phonemes, these can belong to ‘e3’, ‘e2’, ‘e1’ or to ‘pre1’, ‘e2’, ‘e1’ or to ‘pre2’, ‘e2’, ‘e1’ and so on. 2 This ideal is only found for archi-positions ‘E3’ and ‘I’; see Chapters 8 and 9.

111 with a vowel, second, phonotagms ending in three peripheral phonemes, and third, phonotagms beginning with four or five peripheral phonemes. The most defective distribution has diphthongal vowels (especially /ä/ and /ë/) and /ō/. Finally, it is apparent that vowels prefer phonotagms which begin with one or two peripheral phonemes, and end in none or one or two peripheral phonemes.3 The absence of property ‘P5’/‘F0’ corresponding to phonotagm type ‘CCCCCV’ was already mentioned in the previous chapter. We have concluded it is accidental. If it were not, we would assume that if a vowel is preceded by five peripheral phonemes, it must be obligatorily followed by at least one peripheral phoneme as in /FSkvjeT/ vzkvět, or which equals to the same, that a form like /FSkvje/ is not possible in Czech. We do not find such an assumption appropriate. But on the contrary, as there is no evidence a long or diphthongal vowel could be preceded by five peripheral phonemes, we do find appropriate to regard this fact as structurally conditioned, and accordingly we introduce this collocational restriction: A long vowel cannot be preceded by more than four peripheral phonemes (= restriction RN2)4. What is more, examining Figure I, we can say that diphthongal vowels are even more restricted: they cannot be preceded by more than three peripheral phonemes (= RN5). Figure I also shows that no vowel is attested to possess properties ‘P4’/‘F3’, ‘P5’/‘F2’ and ‘P5’/‘F3’. This fact was already regularized in the previous chapter by a rule stating that no phonotagm can contain more than six peripheral phonemes (= RG2a).

7.2 Nuclearity of /r/, /l/

Vowels are not the only nuclear entities in Czech. The phonemes /r/, /l/ we called semiconsonants also assume this function, but unlike the vowels they also function as peripheral phonemes. If /r/, /l/ have two phonotactic functions, it is natural to ask under what circumstances they fulfill these roles. Although the answer was already suggested

3 Let us note that /ä/, /ö/ could possess the property ‘P0’/‘F0’ in the forms of the onomatopoeia au, ou, and /e/, /ī/, /ē/, /ā/, /ō/, /ū/ could be included as forms of the names of the letters e, í, é, á, ó, ú. The property ‘P1’/‘F0’ could be attested for /ū/ in /bū/, a form of the onomatopoeic bú. 4 All collocational restrictions are numbered and summarized in Appendix A.

112 in preceding chapters, it has not been fully considered. As the available phonetic research has shown there is no relevant phonetic difference between allophones of the nuclear /r/, /l/, and those of the non-nuclear /r/, /l/ (see the references in Section 2.5), we can hardly claim that /r/, /l/ are nuclear when realized as syllabic [r̩], [l̩], and non- nuclear when realized as non-syllabic [r], [l]. The nuclearity of /r/, /l/ does not appear to have any phonetic cues in syllabicity; it is a purely phonotactic property. Phonemes are nuclear if and because they comply with the definition of a nuclear entity: when there is an asymmetrical relation between two entities such that one is functionally dependent on the other, the latter is a nuclear entity while the former is a peripheral entity. A particular characteristic of a peripheral entity is that the presence of a nuclear entity is a necessary condition for the occurrence of a peripheral entity, whereas the presence of a nuclear entity need not be dependent on any other entity (see Section 4.4). We underline need not because the occurrence of nuclear entities may sometimes be dependent on the occurrence of peripheral entities, but the occurrence of peripheral entities is always dependent on the occurrence of nuclear entities. In Czech vowels are nuclear entities because they are not occurrence-dependent on any other phonemes—they can occur on their own (see the previous section). On the other hand, the semiconsonants /r/, /l/ are always dependent on the presence of another phoneme because we do not find a word whose phonological form would contain only them. Yet they function as nuclear entities. A phoneme whose nuclearity is not immediately obvious may be regarded nuclear if it is commutable, in functionally equivalent contexts, with phonemes which are undoubtedly nuclear (see Sections 4.7 and 5.1). This assumption allows us to determine and test not only the nuclearity of vowels, but also that of nuclear /r/, /l/; see the commutations such as these:

(2) (a) /paST/ past ~ /pūST/ půst ~ /prST/ prst ~ /plST/ plst (b) /vaK/ vak ~ /vīK/ vík ~ /vlK/ vlk (c) /ten/ ten ~ /tun/ tun ~ /trn/ trn

In the phonotagms /paST/, /vaK/ and /ten/ the vowels can be replaced only by other vowels or by /r/ or /l/; if they are replaced by any other phoneme, the result will not be a

113 well-formed phonotagm (e.g. /pmST/, /vnK/, /třn/). It is this very fact that makes /r/, /l/ nuclear in the phonotagms under (2). Thus, the semiconsonants are nuclear when occurring between two peripheral phonemes. But this is not the only context where they have this function. As traditionally asserted (Frinta 1909: 107, 115, Kučera 1961: 72), they are also nuclear when occurring after a consonant at the end of a phonological word; see (3) for examples. Here, they are commonly regarded syllabic, thus potentially corresponding to a separate phonotagm.

(3) (a) /vītr/ vítr, /obr/ obr, /hadr/ hadr, /miStr/ mistr (b) /hnětl/ hnětl, /mandl/ mandl, /māvl/ mávl, /roStl/ rostl

However, the nuclearity of /r/, /l/ is not so straightforward in this case. The commutation test is of no help here because, for example, /r/ in /vītr/ can be replaced both by a vowel (cf. /vīte/ víte) and a peripheral phoneme (cf. /vīTS/ víc, with contextual neutralization of voicing). It may be argued that there is a difference in these two commutations: /vīte/ contains two phonotagms whereas /vīTS/ only one. This is true, but it does not prove whether /vītr/ is built of two phonotagms or of one. For something to be a phonotagm it must be a self-contained phonotactic entity, which means it must be not dependent on anything else. If /vītr/ is to be built of two phonotagms one of which contains a nuclear /r/, it must be proven that the form can indeed be parsed to two phonotagms. The possible divisions are /vī–tr/, /vīT–tr/ (with functional amalgamation of /T/ and /t/) and /vīt–r/. The last is at once rejected because /vīt/ is not a well-formed phonotagm—occlusives and fricatives undergo neutralization of voicing at the end of major- type phonotagms, and there is a priori no reason to regard /vīt/ as a minor-type phonotagm. On the other hand, in the first two divisions the phonotagms /vī/ and /vīT/ are well-formed (cf. the words ví and výt); both assume /tr/ as the second and potential phonotagm. The onus is now to determine whether /tr/ is really a well-formed phonotactic construction, i.e. a phonotagm. As argued in Section 3.4, well-formedness is instantly proven by direct attestation. Such a proof is not available here because there is no word whose phonological form would be just /tr/. However, this is not an uncommon situation. The form /kolo/ (cf.

114 kolo) is built of two phonotagms, /kol/ and /lo/, even though the latter is not attested as a phonological word in Czech; yet it is a well-formed phonotagm because it does not violate any phonological rule we can think of. Is there, in the case of /tr/, a phonological rule it violates? The only rule we think, judging from the available evidence, it may violate is this: If the semiconsonants /r/, /l/ are nuclear, they must be both preceded and followed by at least one peripheral phoneme. The rule is satisfied in the forms like /prST/, /plST/, /trn/, /vlK/; see (2) above. If it is accepted, the nuclearity of /r/ in /tr/ must be rejected, and /vītr/ is a single phonotagm simply because it cannot be parsed to two well-formed phonotagms. The only way to reject such a rule is either to dismiss it as inappropriate or to find counter-evidence. The appropriateness is a matter of judgment and of the degree to which we want to make our analysis intuitive with the fact that vítr is after all regarded as disyllabic (in poetry, for example). But since the phonotagm is not necessarily co-extensive with the syllable (see Section 4.6), we should rather look for the counter-evidence, and that could perhaps be found in the word hr5. Admittedly, it occurs only as a part of the idiom být hr “to be rash, hasty”, and is in origin an onomatopoeic expression. Its acceptance may thus be questionable because it creates precedence for admission of other onomatopoeic expressions, which might not be quite desirable due to their phonotactic structure being quite different to the rest of the Czech vocabulary (Fidler 2010). Yet if we rejected /hr/ hr, there would be no evidence for the well-formedness of /tr/, and /vītr/ would then have to be interpreted as a mono-phonotagmic phonological word. However, if /vītr/ is a single phonotagm with /ī/ as the only nuclear phoneme, our whole analysis of the phonotactics of Czech would have to be re-worked. The combinations /tr/ in /vītr/ or /Str/ in /miStr/ (cf. mistr) would have to be plotted onto the post-nuclear positions of the distributional unit. In /miStr/ the phoneme /S/ would have to occur in ‘i1’, /t/ in ‘i2’ and /r/ in ‘i3’, but the previous analysis has not shown they occur here. Furthermore, our conclusion about neutralization of voicing in the post-nuclear context would also have to reformulated because there is not such neutralization before /r/ or /l/ (cf. /vītr/ ~ /hadr/). All in all, the hypothesis of /vītr/ being a single phonotagm brings about more complica-

5 We could, at least hypothetically, also have schl as a variant of schnul “he was getting dry”. However, such a form is apparently avoided and not used, even though its prefixed derivate uschl “he got dry” is possible. There is also dialectal /hl/ hl, a variant of hnul “he moved”.

115 tions than simplicity. Consequently, we dismiss it as inappropriate, and assume it is a bi-phonotagmic phonological word, which is confirmed by the form /hr/. The same holds for the other forms under (3). There are still other problems with the nuclearity of /r/, /l/. Let us now consider the following examples:

(4) (a) /Stārl/ stárl “he grew old” (b) /tirl/ Tyrl, /karl/ Karl (both proper names) (c) /umrlTse/ umrlce, /umrlTšī/ umrlčí, /Štamprlka/ štamprlka, /povrlŠťī/ povr- lští6

(4a) and (4b) should be qualitatively identical, all ending in /rl/; yet they are reportedly evaluated differently by some native speakers: for some, (4a) is disyllabic whereas (4b) are monosyllabic. There must be some confusion here—in stárl the final l is a suffix,7 which might create an impression of a structure different to Tyrl and Karl where it is not a suffix; however, the difference must be morphological, not phonological.8 At any rate, the real point is this: are the phonological words /Stārl/, /tirl/ and /karl/ mono- phonotagmic or pluri-phonotagmic? If the latter is the case, they should presumably contain the phonotagm /rl/, but is this really a well-formed phonotagm? We do not find any word with this form or even any suggestion similar to hr that it could be possible at all. Having accepted /tr/ as a potential phonotagm, we might accept /rl/, too, but there is a difference between them: while /t/ is a consonant, i.e. a phoneme occurring solely in the non-nuclear context, /r/ is a semiconsonant capable of occurring in the nuclear as well as in the non-nuclear context. They are therefore phonologically different. Not hav-

6 The word umrlce is gen. sg. of umrlec, and umlčí is an adjective derived from it; both are listed in Czech dictionaries. Štamprlka, a diminutive of štamprle, is not mentioned in any dictionary, but can be found in the Czech National Corpus and through Google.com (as of 13/04/2011). Similarly, povlrští can be found there, too; it is an adjective derived from the place name Povrly. I would like to thank to Pavel Šmerk for drawing my attention to words like these. 7 stárl is probably the only word of its kind if we do not count its derivate zestárl. The words under (4b) are proper names borrowed from other languages. 8 If it were phonological, Czech would then have two separate phonemes: nuclear/syllabic /r/, and non-nuclear/non-syllabic /r/. See also Section 8.8.

116 ing any evidence /rl/ is a well-formed phonotagm, we conclude that /Stārl/, /tirl/, /karl/ cannot be divided to two phonotagms, and they are thus mono-phonotagmic. In this case, however, there is no obstacle for such an evaluation because /rl/ is a post-nuclear combination comparable to /lm/, /jn/, /řň/, /jr/, /jl/ (cf. jilm, hejn, čerň, chejr, koktejl). First of all, /rl/ is not expandable9 because no peripheral phoneme can come before /r/ or after /l/, and this is true for /lm/, /jn/, /rň/, /jr/, /jl/ as well. Second, in /Stārl/, /karl/ and /tirl/, the /r/ can be mapped onto position ‘i1’ where it does indeed occur, and the /l/ onto archi-position ‘I’ where it occurs, too (see Chapter 9). This was not possible with /vītr/ or /miStr/. In short, /Stārl/, /tirl/ and /karl/ are single phonotagms, though disyllabic in realization. Now, to reflect the conclusions of the preceding paragraphs, we propose the following collocational restriction (= restriction RN7a in Appendix A) which is not violated by /tr/ or by /hr/, but is violated by /rl/ because /r/, /l/ are semiconsonants, not consonants.

Restriction on the occurrence of nuclear semiconsonants A nuclear semiconsonant must be preceded by at least one consonant.

Turning now our attention to (4c), we can say these may be most problematic of all. We may query how many syllables the mentioned words contain. If we rely on the traditional assertion that /r/ and /l/ are syllabic between two consonants, and if we realize that they are usually placed among consonants (i.e. the class of semiconsonants is not generally recognized), it is obvious that both /r/ and /l/ should be syllabic here, as both are between two consonants. This is, however, in contradiction with how the words umrlce and umrlčí are treated in verses such as the one under (5) where the first and the third lines agree in the number of syllables (nine) and where the second and the four lines also agree provided that umrlčí is trisyllabic.10

9 See Section 10.6 on expandability. Note that /tr/ in /vītr/ is expandable, cf. /Str/ in /miStr/. 10 Taken from the collection Dokument by Vladimír Holan, Prague, 1949, p. 15. Other examples of the use of these words in verse can be gained by looking in the database of PSJČ available online.

117 (5) Tak také trest, trest z nekonečna, do chvíle zbalen, sevře pěst, a rána střelná, bodná, sečná umrlčí rasu umí smést…

But even if we accept that syllabic is only one of the phonemes /r/ and /l/ in these words, we should still wonder which one it is. Whether this can be resolved on phonetic or other grounds is far from certain. However, we do not operate with syllabicity but with nuclearity, and so we should look at those words from that perspective. If we accept that a nuclear semiconsonant must be preceded by at least one consonant (see above), we can conclude that the forms /umrlTse/, /umrlTšī/, /Štamprlka/ and /povrlŠťī/ are built of three phonotagms, and that nuclear is the /r/ in them. If the /l/ were nuclear instead, it would violate restriction RN7a, as it is not preceded by a consonant here— remember that /r/ is a semiconsonant for us! Once the restriction is accepted, the forms of these words are parsed to /um–Mrl–lTse/, /um–Mrl–lTšī/, /ŠtamP–prl–lka/ and /poF– vrl–lŠťī/.11 Although this produces the phonotagms /Mrl/, /prl/ and /vrl/ which have no attested parallel, we see no reason to declare them ill-formed. The alternative is to regard them as parts of the preceding phonotagms, i.e. to say parse the forms to /umrl– lTse/, /umrl–lTšī/, /Štamprl–lka/, but this would complicate considerably the whole analysis: first, restriction RN7a would have to be abandoned, and second, the distributional unit would have to contain four post-nuclear positions to accommodate /mprl/ in /Štamprl/. On the other hand, /Mrl/, /prl/ and /vrl/ do not violate any obvious restriction and conform to the structure of the distributional unit. To conclude this section, we can say that the semiconsonants /r/, /l/ function as nuclear entities only if they are preceded by a consonant, and only if they are not at the same time followed by a vowel, i.e. if they are followed by a peripheral phoneme, or if they occur at the end of a phonological word.

11 With functional amalgamation in all cases. Note that /umrlTse/ is parsed to /um–Mrl–lTse/, not to /um–mrl–Tse/ because there is neutralization of place of articulation of the nasals in the pre-nuclear context, see Section 2.7. Hence, there is functional amalgamation between /m/ and its archiphoneme /M/. The same is true for the division of words like hamr, humr, žánr, všiml, i.e. they are parsed to /ham–Mr/, /hum–Mr/, /žān–Mr/ and /Fšim–Ml/, respectively.

118 7.3 Position class ‘N’

If /r/, /l/, which constitute a separate position class given under (6), function as nuclear entities, they do not occupy position ‘n’ like vowels, but archi-position ‘N’; the reasons were explained in Section 5.3. However, as archi-position ‘N’ is the product of cancellation of the difference between pre-nuclear position ‘e1’ and nuclear position ‘n’, /r/ and /l/ belong both to the pre-nuclear and to the nuclear context. This agrees with their nature: like vowels, they have the capacity to be the identity elements of phonotagms, but they have at the same time strong affinities to consonants as regards their combinability and occurrence in peripheral contexts.

(6) pos ‘N’ ∈ {/r/, /l/}

The phonotactic properties of pos ‘N’ are examined in a similar way as those of vowels above; they are listed in Figure II. As nuclear /r/, /l/ must be obligatorily preceded by at least one consonant (see the previous section), property ‘P0’ is redundant for them. Similarly, they can be preceded by no more than four peripheral phonemes; hence ‘P5’ is also redundant. Examples are given in Appendix D.

F0 F1 F2 F3 P1 – P2 /l/ P3 – /l/ /l/ – P4 – /l/ – –

Figure II: Phonotactic properties of pos ‘N’; phonemes listed are not attested to possess a given property. Empty cells mean the property is attested for all members of pos ‘N’, ‘–’ that it is for none.

The distribution of the semiconsonants is not perfect, but there are more regularities in comparison to the vowels. The distribution of /l/ is markedly limited in comparison with /r/. Phonotactic properties ‘P3’/‘F3’, ‘P4’/‘F2’ and ‘P4’/‘F3’ are impossible for nuclear /r/, /l/, a point already discussed in Section 6.1 (no phonotagm can contain more than five peripheral phonemes if /r/ or /l/ is its nucleus). The non-attestation of proper-

119 ties ‘P3’/‘F0’ and ‘P4’/‘F3’ can be regularized by the following restriction: If nuclear /r/, /l/ occur at the end of a phonological word, they cannot be preceded by three or more peripheral phonemes (= restriction RN7b in Appendix A)12. On the other hand, the non-attestation of property ‘P1’/‘F3’ is accidental. Although this may be viewed as an arbitrary decision, it is supported by the fact that property ‘P2’/‘F3’ is attested for /r/ (cf. /šMrnTS/ šmrnc), and we see no reason why nuclear /r/, /l/, when followed by three peripheral phonemes, could not be preceded just one peripheral phoneme if they can be preceded by two of them.

12 The forms like /šamStr/ šamstr cannot be parsed to /šamS–MStr/ because the combination /MSt/ or a similar combination of the same type is not attested before nuclear /r/.

120 Chapter Eight DISTRIBUTION IN THE PRE-NUCLEAR CONTEXT

8.1 Position class ‘e1’

The distribution in the pre-nuclear context is divided among five positions ‘e1’, ‘e2’, ‘e3’, ‘pre1’ and ‘pre2’, and among two archi-positions ‘E2’ and ‘E3’. By definition, only consonants and semiconsonants occur here. Position ‘e1’ is the first before the nucleus, and so if a phoneme occurs here, it always stands before a vowel; its position class is given in (1). As neither phoneme belonging to pos ‘e1’ is phonologically voiceless or voiced, neutralization of voicing does not take place before any of them1. From the phonematic perspective, the class is constituted by the nasals and by the sonants; it can also be empty.

(1) pos ‘e1’ ∈ {/m/, /n/, /ň/, /j/, /ř/, /r/, /l/, ∅}

Except for the nasals, all phonemes of pos ‘e1’ occur in yet another pre-nuclear position. The phonemes /j/, /r/ and /l/ also occur in ‘pre2’, and /ř/ also occurs in ‘pre1’. We will return to /ř/ when discussing pos ‘pre1’ below in Section 8.6. In the pre-nuclear context the nasals /m/, /n/, /ň/ occur only in ‘e1’, which means they always stand before a vowel. In other situations they are represented in the pre-nuclear context by their archiphoneme /M/ which occurs in ‘e2’ and ‘pre2’. Figure I illustrates the occurrence of the nasals and of /j/, /r/, /l/ in ‘e1’ as compared to ‘pre2’. The conditions of the occurrence of /M/, /j/, /r/, /l/ in ‘pre2’ will be discussed in Section 8.7.

1 It may take place before /ř/, but not if it occurs in ‘e1’; see Sections 2.6 and 8.6.

121 Occurrence in ‘e1’ Occurrence in ‘pre1’ α /moTS/ moc β /smīX/ smích /m/ γ /StmīF/ from stmívat se δ /FSdmōT/ vzdmout ε – α /noS/ nos β /vnuK/ vnuk /n/ γ /SxnöT/ schnout ζ /Mše/ mše δ /lStnī/ lstný η /Mdlī/ mdlý ε – θ /Mstnī/ mstný α /ňiTS/ nic β /Mňel/ měl /ň/ γ /ShňīT/ shnít δ /hřMňeT/ hřmět ε – α /jen/ jen ζ /jmēn/ jmén β /pjeT/ pět η /jsme/ jsme /j/ γ /kvjeT/ květ θ – δ /FSpjer/ vzpěr ε /FSkvjeT/ vzkvět α /rāj/ ráj ζ /rtuŤ/ rtuť β /druX/ druh η /rvjeT/ from rvěte /r/ γ /sMraT/ smrad θ – δ /PŠtroS/ pštros ε – α /lom/ lom ζ /lvi/ lvi β /xlaP/ chlap η /lSťiF/ from lstivý /l/ γ /FplöT/ vplout θ /lStnī/ lstný δ /FStlaK/ vztlak ε –

Figure I: Differences in the occurrence of /m/, /n/, /ň/, /j/, /r/, /l/ in ‘e1’ and ‘pre1’. A dash means that a respective example has not been found.

Figure I demonstrates that /m/, /n/, /ň/, /j/, /r/ and /l/ can appear right before a vowel, and be the first phonemes in the phonotagms (cf. the α examples). Moreover, they can occur before a vowel, and be at the same time preceded by one (= the β examples), two (= the γ examples) or three peripheral phonemes (= the δ examples). The approximant /j/ can in addition be preceded by four peripheral phonemes (cf. the ε example, the only one); no other peripheral phoneme is preceded by four peripheral phonemes. In all of the mentioned situations these phonemes occur in ‘e1’. To put it simply: when standing

122 before a vowel, /m/, /n/, /ň/, /j/, /r/ and /l/ occur in the position ‘e1’. On the other hand, when /j/, /r/, /l/ stand before a peripheral phoneme in the pre-nuclear context, they occur in the position ‘pre2’ in which case they are the first phonemes in the phonotagm. The nasals do not have this capacity because the opposition between them is neutralized before a peripheral phoneme in the pre-nuclear context. The neutralization results in the archiphoneme /M/ (Section 2.7). If this archiphoneme occurs before a peripheral phoneme, and is at the same time the first phoneme of the phonotagm, it occurs in ‘pre2’ as well (see the examples ζ–θ for /M/). If /M/, /j/, /r/, /l/ occur in ‘pre2’, they can precede one peripheral phoneme (= the ζ examples)2 or two peripheral phonemes (= η); /M/ and /l/ stand in addition before three peripheral phonemes (= θ). We can view position ‘e1’ as a sum of phonotactic properties common to the members of its position class. It can be expressed by the ability of phonemes to be preceded (coded ‘P’) or followed (coded ‘F’) by a certain number of phonemes (e.g. ‘P2’ means “preceded by two phonemes”). Furthermore, it is possible to specify to which positions these preceding and following phonemes belong, so code ‘P3pre1e3e2’ means that a phoneme can be preceded by three peripheral phonemes which belong, respectively, to ‘pre1’, ‘e3’ and ‘e2’ (e.g. /j/ in /Skvjel/ skvěl). Although for reasons explained in Chap- ter 7 we did not provide this specification for nuclear phonemes, it is useful to do so for pre-nuclear phonemes if we limit the precedence and antecedence just to the pre-nuclear context. Hence, we will be interested in whether a phoneme occurring in the pre-nuclear context can be preceded and/or followed by other pre-nuclear phonemes, by how many phonemes, and to which positions these phonemes belong. In addition, however, it may be useful to specify whether pre-nuclear phonemes can be followed by vowels or by nu- 3 clear semiconsonants; for example, property ‘F1e1n’ means that a phoneme is followed by a phoneme from ‘e1’ which is itself followed by a vowel from ‘n’ (e.g. /t/ in /tvūj/ tvůj); ‘F1e2N’ means a phoneme is followed by a phoneme from ‘e2’ which is itself followed by a nuclear semiconsonant from ‘N’ (e.g. /s/ in /sMrK/ smrk). Phonotactic properties of the phonemes belonging to pos ‘e1’ are given in Figure II. Some examples of their distribution were already given above in Figure I; others can be

2 /M/ can stand before a vowel, but only if it is preceded by /ř/ in which case it occurs in ‘e2’. 3 Note that number 1 here means “followed by one peripheral phoneme”.

123 found by examining distributional types DE1a, DE2a, DE2b, DE2c, DE2d, DE2e, DE3a, DE3b, DE3c, DE3d, DE3e, DE3f, DE3g, DE4a, DE4a, DE4b, DE4c, DE4d, DE4e and DE5 discussed in Chapter 11. The phonemes listed in Figure II are again those not attested to have a certain property; the same is true for all other figures in this chapter reproducing phonotactic properties. This negative way will help us to introduce collocational restrictions limiting the distribution of sets of phonemes.

F0n P0

P1e2

P1e3

P1E2 /m/

P1pre1 –

P1pre2 (/j/) /ř/ /r/ /l/

P2e3e2 /m/

P2pre1e2 /m/ /n/ /ň/ /ř/ /r/ /l/

P2pre2e2 /m/ /n/ /ř/ /r/

P2pre1e3 /j/

P2pre2e3 /j/

P2pre1E2 /m/ /n/ /ň/

P2pre2E2 /m/ /n/ /ň/

P2pre2pre1 –

P3pre1e3e2 /m/ /n/ /ř/ /r/ /l/

P3pre2e3e2 /m/ /n/ /ř/ /r/ /l/

P3pre2pre1e2 /m/ /n/ /j/ /ř/ /r/ /l/

P3pre2pre1e3 /j/

P3pre2pre1E2 /m/

P4pre2pre1e3e2 /m/ /n/ /ň/ /ř/ /r/ /l/

Figure II: Phonotactic properties of pos ‘e1’; phonemes listed are not attested to have a given property. A dash means a property is not attested for any member the position class; an empty cell means a property is attested for all members of the class.

Properties ‘P1pre1’ and ‘P2pre2pre1’ are not attested for any phoneme of pos ‘e1’. Both refer to a situation when such a phoneme would be preceded by a phoneme from pos ‘pre1’, that is, by /T/, /S/, /Š/ or /ř/ (see Section 8.6). Out of these, they could only be preceded by /ř/, but no such combination is attested. A nasal can be preceded by /ř/ as in /hřMī/ hřmí, but the /M/ is the nasal archiphoneme resulting from neutralization of the opposition /m/ ~ /n/ ~ /ň/ (i.e. the place of articulation is predicable for nasals when

124 standing after /ř/); this archiphoneme belong pos ‘e2’, not to pos ‘e1’ (see the next section). This is accord with a generalization that the phonemes from pos ‘e2’ can be preceded by /ř/, whereas the phonemes from pos ‘e1’ cannot. Accordingly, we propose the following collocational restriction: A sonant cannot stand in close proximity with /ř/ in the pre-nuclear context (= restriction RE5 in Appendix A). The situation is different in the post-nuclear context when /ř/ can be preceded a sonant, namely by /j/ as in /kejř/, the forms of the surname Kejř. Restriction RE5 partly overlaps with another, a more general one: /ř/ cannot occur in the same section of the distributional unit with a semiconsonant (= RG13a), be they in close proximity or separated by other phonemes. The same section refers here either to the pre-nuclear or post-nuclear context. Hence, pre-nuclear combinations like /řvj/ or /hřMň/ are possible, but those like /řl/, /řvr/ or /hřMl/ are not. What is more, as explained in Section 7.3, by being a product of cancellation of the difference between positions ‘e1’ and ‘n’, archi-position ‘N’ belongs both to the pre-nuclear context and to the nuclear context. Consequently, the mentioned restriction holds true even if the semiconsonant is nuclear, that is, a nuclear semiconsonant cannot be preceded by a combination containing /ř/. Figure II allows us to introduce other restrictions on the combinability of phonemes from pos ‘e1’. As property ‘P4pre2pre1e3e2’ is attested for none of them, we conclude that a non-nuclear semiconsonant and /ř/ cannot be preceded by more than three peripheral phonemes (= RE7a).4 It does not include the nasals because there is some hint they can after all be preceded by three peripheral phonemes. If the verb vzdmouti (PSJČ, SSJČ), a variant of vzedmouti, is to have a 2nd person plural imperative, it would most likely be vzdměte, i.e. /FSdMňete/. To the best of our knowledge, however, the word is not attested, but it is at least an indication of the nasals’ ability to be preceded by four peripheral phonemes. No such evidence exists for non-nuclear /r/, /l/ and for /ř/, though. As is obvious from Figure II, /ř/, /r/, /l/ are not attested to be preceded by a phoneme from pos ‘pre2’, i.e. by /P/, /F/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/, and /b/, /v/, /t/, /h/ (see Section 8.7). We will ignore the last four because they have a very special distribution. Now, as already mentioned, sonants cannot be preceded by voicing archipho-

4 A nuclear semiconsonant has this capacity, cf. /TŠtvrT/ čtvrt.

125 nemes, so the only phonemes by which /ř/, /r/, /l/ could be preceded are /M/, /j/, /r/ or /l/. They are preceded by /M/ as in /Mřenka/ mřenka, /Mrak/ mrak or /MloK/ mlok, but the /M/ occupies here position ‘e2’ because it can be further preceded by another peripheral phoneme, e.g. by /s/ from pos ‘e3’ as in /sMraT/ smrad. They cannot be preceded by /j/. On the other hand, /r/ and /l/ can be followed by /j/, though only in /ljuba/, a phonological form of the proper name Ljuba, and in /rjazaň/, a form of the (foreign) city Rjazaň. It is noteworthy that in the post-nuclear context /j/ can, on the contrary, be only followed by /r/ and /l/, cf. /chejr/ chejr and /koKtejl/ koktejl. We can capture this fact by the following restriction: If a semiconsonant occurs in the same section of the distributional unit with /j/, the semiconsonant occurs in a more peripheral position (= RG16). The more peripheral position is here the one more distant from the nucleus. Let us return to the impossibility of /ř/, /r/, /l/ to be preceded by the phonemes from pos ‘pre2’. The inability of /ř/ to be preceded by /r/ or /l/ derives from restriction RG13a introduced above, and the inability of /r/, /l/ to be preceded by themselves can be cap- tured by the following general restriction: The pre-nuclear section of the distributional unit cannot contain two instances of a semiconsonant (= RG15). This is to say that if a semiconsonant occurs either in a pre-nuclear position or in archi-position ‘N’, which belongs to the pre-nuclear section as well (see above), it cannot be preceded by another semiconsonant. It is important to specify that the restriction applies only to the pre- nuclear section because the semiconsonants can combine in the post-nuclear one, and moreover, a nuclear semiconsonant can be followed by another semiconsonant (but not the same one); it is a consequence of our analysis of /Stārl/ stárl and /ūMrl/ from úmrlce as single phonotagms (see Section 7.2). Finally, we introduce this restriction partly overlapping with the previous one: the same section of the distributional unit cannot contain two identical sonants (= RG14), i.e. if it contains /r/, it cannot contain another /r/; if it contains /j/, it cannot contain another /j/, and so on, even if they would be separated by another phoneme. This time the restriction holds for the pre-nuclear and post- nuclear context.

Properties ‘P3pre1e3e2’, ‘P3pre2e3e2’ and ‘P3pre2pre1e2’ are also not attested for /r/, /l/ and for /ř/, which suggests another restriction: a non-nuclear semiconsonant and /ř/ cannot be preceded by a combination of the type ‘CCC1’ where ‘C1’ is from pos ‘e2’, i.e. /v/ or

126 /M/ (= RE7b) as opposed to the nasal /ň/ in /hřMňeT/ hřmět.5 In other words, if they are preceded by three peripheral phonemes ‘CCC1’, ‘C1’ is always either from pos ‘E2’ or from ‘e3’, i.e. it is an occlusive other than palatal or a fricative other than /v/. Unlike /ř/, /r/, /l/, the approximant /j/ can be preceded by a phoneme from pos ‘pre2’, but only by /r/ or /l/ in the above-mentioned /ljuba/ and /rjazaň/. It can only be preceded by /M/ as in /MjaXkī/, the form of the surname Mjachký, but we assume that /M/ occurs here in ‘e2’, not in ‘pre2’ because the combination /Mj/ parallels the structure of similar combinations of the ‘e1’ phonemes with the ‘e2’ phonemes (see Section 11.2 under DE2a for the combinations). Apart from that, the distribution of /j/ is rather limited. It can only be preceded by a phoneme from pos ‘e2’ or pos ‘E2’ (with one exception), which in effect means it can only be preceded by the labial occlusives /p/, /b/, the labial fricatives /f/, /v/, and the nasal /M/. If these phonemes precede /j/, they can be further preceded other phonemes as in /FSpj/ from vzpěrač. The alluded-to exception is when /j/ is preceded by the alveolar fricatives /s/, /z/ (cf. /sj/, /zj/ in sjet and zjev) which belong to pos ‘e3’; however, in that case there one important circumstance: the /s/ and /z/ are not found to be preceded by another peripheral phoneme when preceding /j/. Conse- quently, we propose the following restrictions: /j/ cannot be preceded by a phoneme from pos ‘e3’ unless it is an alveolar fricative (= RG12a); if /j/ is preceded by an alveolar fricative, the latter is never preceded by any other peripheral phoneme (= RG12b). Finally, /m/ cannot be preceded by a phoneme from pos ‘E2’, i.e. by /p/, b/, /f/, irrespective of whether the latter is preceded by any other peripheral phoneme (= RE15). On the other hand, it can be preceded by the labial fricative /v/ as in /vmīsiT/ vmísit.

8.2 Position class ‘e2’

Position ‘e2’ is the second position before a vowel or the first position before a semiconsonant occurring in ‘N’. Its position class is given under (2); it can also be empty.

(2) pos ‘e2’ ∈ {/v/, /M/, ∅}

5 A nuclear semiconsonant has again this capacity, cf. /Skvrn/ skvrn.

127 /v/ is a voiced labial fricative; it has different phonotactic properties than its voiceless counterpart /f/; the latter belongs to pos ‘E2’. /M/ is the nasal archiphoneme resulting from neutralization of the opposition between /m/, /n/ and /ň/. It is important to note that neutralization of voicing does not take place before /M/ or /v/, i.e. both voiceless and voiced consonants can stand before them (see Section 2.6). They also occur in ‘pre2’. The difference between these occurrences is demonstrated in Figure III. However, the occurrence of /v/ in ‘pre2’ is rather limited; we will return to it in Section 8.7.

Occurrence in ‘e2’ Occurrence in ‘pre2’ α – ι /Mše/ mše α’ /MlŠ/ mlž κ /MSta/ msta β – λ /MStnī/ mstný β’ /sMrT/ smrt γ /hřMī/ hřmí γ’ – /M/ δ – δ’ – ε /MraK/ mrak ζ /sMňer/ směr η /hřMňeT/ hřmět θ – α /vīŠ/ výš ι – α’ /vrX/ vrch κ /vMňeST/ from vměstnat β /tvūj/ tvůj λ – β’ /tvrS/ tvrz γ /SkvoST/ skvost γ’ /Skvrn/ skvrn /v/ δ /FSkvēST/ vzkvést δ’ /TŠtvrT/ čtvrt ε /vraT/ vrat ζ /svjeT/ svět η /Skvjel/ skvěl θ /FSkvjeT/ vzkvět

Figure III: Differences in the occurrence of /M/, /v/ in ‘e2’ and ‘pre2’. A dash means that a respective example has not been found.

/v/ occurs in ‘e2’ when standing right before a vowel or before a nuclear semiconsonant. In both cases it can be the first phoneme in the phonotagm (cf. the α and α’ examples in Figure III) or it can be preceded by one (= the β and β’ examples), two (= γ and

128 γ’), or three peripheral phonemes (= δ and δ’). The archiphoneme /M/ is in this respect limited because it can occur only before a nuclear semiconsonant in which case it can be the first phoneme in the phonotagm (cf. the α’ example for /M/), or it might preceded by one peripheral phoneme (cf. the β’ example). The only situation when /M/ occurs before a vowel is when preceded by /ř/, as the difference between the places of articulation is neutralized here for nasals (Section 2.7). /v/ and /M/ also occur in ‘e2’ when standing before a peripheral phoneme from ‘e1’, i.e. when preceding /m/, /ň/, /n/, /j/, /r/ and /l/. In that case they can be the first phonemes in the phonotagm (cf. the ε examples) or they can be preceded by one (= ζ), or two (= ζ), or three peripheral phonemes (= η). The phoneme /v/ can in addition be preceded by four phonemes (cf. the θ example, the only one). However, when preceding a phoneme which does not belong to pos ‘e1’, they occur in ‘pre2’, and are the very first phonemes of the phonotagms. The archiphoneme /M/ can precede one, two, or three peripheral phonemes (= ι, κ and λ), whereas /v/ can only precede two phonemes (= κ). As we will see in Section 8.7, this occurrence of /v/ is quite exceptional.

F0n F0N F1e1n P0 /M/

P1e3 /M/

P1pre1 /M/ – –

P1pre2 /M/ –

P2pre1e3 /M/ /M/

P2pre2e3 /M/ – 6 P2pre2pre1 /v/ – /v/

P3pre2pre1e3 /M/ /M/ /M/

Figure IV: Phonotactic properties of pos ‘e2’; phonemes listed are not attested to have a given property. See Figure II for explanations.

Phonotactic properties of this position are given in Figure IV. Once again, the phonemes listed are not attested to have the property in question. Let us note that /M/ could have property ‘P3pre2pre1e3’/‘F1e1n’ if the form /FSdMňete/ vzdměte is accepted (see Sec- tion 8.1). Examples of the distribution of /v/ and /M/ have already been given in Figure III, and more can be found in Chapter 11 under distributional types DE1b, DE2a, DE2f,

6 This property is attested for /v/ in /břve/, the form of the place name Břve.

129 DE2g, DE2h, DE3a, DE3a, DE3b, DE3c, DE3h, DE3i, DE3j, DE4a, DE4b, DE4c and DE5. A nuclear semiconsonant is not found to be preceded by a two peripheral phonemes ‘C2C1’ where ‘C2’ is from a pre-explosive position (i.e. either from ‘pre1’ or from ‘pre2’) and ‘C1’ /v/ or /M/. As neutralization of voicing does not take place before these phonemes, ‘C2’ can only be a sonant /j/, /ř/, /r/, /l/ or the nasal /M/. The impossibility of the nuclear semiconsonants to be preceded by such combinations derives from restrictions RG13a, RG14, RG15 and RG16 (introduced in the previous section), and from restriction RN9a to be introduced in Section 8.7 according which a nuclear semiconsonant cannot be preceded by a peripheral combination of the type ‘NCn’ where ‘Cn’ is any number of peripheral phonemes.

8.3 Position class ‘e3’

Position ‘e3’ is the third pre-nuclear position. Its position class is given in (3). From the phonematic perspective, it contains all voiceless and voiced occlusives except for /p/, /b/, /f/, /v/, /ť/, /ď/, which belong to other positions. The position can also be empty.

(3) pos ‘e3’ ∈ {/t/, /d/, /s/, /z/, /š/, /ž/, /k/, /g/, /x/, /h/, ∅}

The phonemes from pos ‘e3’ do not occupy any other pre-nuclear position, though their archiphonemes, which represent them, do. The archiphonemes /T/, /S/, /Š/ occupy ‘pre1’ and ‘pre2’, and /K/, /X/ occupy ‘pre2’. Phonotactic properties of this position are given in Figure V. The pre-nuclear occurrences of /t/, /d/, /š/, /ž/, /s/, /z/, /k/, /g/, /x/, /h/ can be easily deduced from it. They occur in ‘e3’ when standing before a vowel, before nuclear /r/ or /l/, before a phoneme from pos ‘e1’, or before a phoneme from pos ‘e2’; at the same time they can be preceded by one phoneme from pos ‘pre1’ or pos ‘pre2’, or by two phonemes which occur, respectively, in ‘pre1’ and ‘pre2’. We do not give here special examples; they can found under distributional types DE1c, DE2b, DE2f, DE2i, DE2j, DE3a, DE3d, DE3e, DE3h, DE3i, DE3k, DE4a, DE4b, DE4d, DE4f and DE5 discussed in Chapter 11.

130 Property ‘P2pre2pre1’/‘F2e2e1n’ is attested only for /k/ (cf. /FSkvjeT/ vzkvět); it is not attested for alveolar occlusives unless we accept the form /FSdMňete/ vzdměte (see Sec- tion 8.1). No such evidence is available for the alveolar fricatives /s/, /z/, and they are, moreover, not attested to have property ‘P1pre1’/‘F2e2e1n’. This suggests that they cannot be followed by a combination of the type ‘C2C1’ where ‘C2’ is from pos ‘e2’ and ‘C1’ from pos ‘e1’ and be at the same time preceded by a phoneme from pos ‘pre1’ irrespective of whether the latter is preceded by another peripheral phoneme or not. Now, if ‘e3’ is occupied by /s/ or /z/, position ‘pre1’ can only be occupied by /T/ because /S/, /Š/, /ř/ cannot preceded them.7 In effect, it means that the combinations /Ts/, /Tz/, /Tš/, /Tž/ cannot be followed by two peripheral phonemes; we regularize this into restriction RE8.

F0n F0N F1e2n F1e2N F1e1n F2e2e1n /d/ 8 P0 (/š/) (/ž/) /ž/ /š/ /ž/ /k/ /g/ /x/ /h/ /g/ /z/ /z/ /d/ /z/ /z/ /d/ /s/ /z/ P1pre1 /ž/ /ž/ /š/ /ž/ /ž/ /š/ /ž/ /g/ /g/ /h/ /g/ /x/ /h/ /g/ /d/ /z/ /d/ /s/ /t/ /d/ P1pre2 /š/ /ž/ /š/ /ž/ – /š/ /ž/ /k/ /g/ /g/ /x/ /h/ /g/ /k/ /g/ /x/ /h/ /z/ /d/ /s/ /z/ /d/ /z/ /d/ /z/ /s/ /z/ /t/ /d/ /s/ /z/ P2pre2pre1 /ž/ /š/ /ž/ /š/ /ž/ /ž/ /š/ /ž/ /g/ /k/ /g/ /x/ /h/ /g/ /h/ /k/ /g/ /x/ /h/ /g/ /g/ /x/ /h/

Figure V: Phonotactic properties of pos ‘e3’; phonemes listed are not attested to have a given property. See Figure II for explanations.

When standing before a nuclear semiconsonant, /s/ and /z/ are also not found to be preceded by two peripheral phonemes. The same is true for all velars as well. We regularize this into the following: a nuclear semiconsonant cannot be preceded by a combination of type ‘CCAF’ or of type ‘CCK’ where ‘AF’ is an alveolar fricative and ‘K’ a velar (= RN9g and RN9h). On the contrary, labials, alveolar occlusives and palatals can be, cf. /v/ in /StvrŤ/ stvrď, /t/ in /FStrTŠ/ vstrč and /š/ in /FTšrtaťi/ včrtati (PSJČ).

7 It is due to restrictions RG1b, RG3 and RE10b; see Appendix A. 8 The reason /š/ and /ž/ are parenthesized is explained in Section 13.1.

131 The palatal fricatives /š/, /ž/ are not attested to be followed by two phonemes from ‘e2’ and ‘e1’ irrespective of whether they are preceded by other peripheral phonemes or not. We propose that a palatal fricative cannot be followed by two peripheral phonemes

‘C2C1’ where ‘C2’ is from pos ‘e2’, and ‘C1’ is from pos ‘e1’ (= RE9); compare it against /s/ in /svjeT/ svět. Thus, /š/, /ž/ can either be followed by a phoneme from pos ‘e2’ (cf. /švagr/ švagr) or a phoneme from pos ‘e1’ (cf. /žlutī/ žlutý).

8.4 Position class ‘E2’

Archi-position ‘E2’ is an outcome of suspension of the difference between positions ‘e3’ and ‘e2’. Its position class is given under (4). It means that the syntagmatic difference between these positions, which is otherwise valid for other phonemes, is not valid for /p/, /b/ or /f/. These phonemes, except for /b/, do not occur in any other pre-nuclear position, though their archiphonemes do: both /P/ and /F/ occupy position ‘pre2’. The phoneme /b/ can marginally occur in ‘pre2’, too, in /břve/, the form of the place name Břve. From the phonematic perspective, /p/ and /b/ are labial occlusives, and /f/ is a voiceless labial fricative. As we saw Section 8.2, its voiced counterpart /v/ belongs to a different position class, and is thus phonotactically different.

(4) pos ‘E2’ ∈ {/p/, /b/, /f/}

We have already mentioned that no archi-position in Czech can be empty, as there would be no reason for operating with it. Since the emptiness of a position refers to the fact that a phoneme occurring there can be omitted without producing an ill-formed phonotagm (see Section 4.2), this claim must be substantiated when we are confronted with the omission of /p/ in /praK/ (cf. prak) resulting in /raK/ (cf. rak). However, in /raK/ it is not archi-position ‘E2’ that is empty, but positions ‘e2’ and ‘e3’ because both or one of them can be filled with a phoneme, which is proven by /zvraT/ zvrat, /vraK/ vrak and /zraK/ zrak. Thus, when a phoneme occurring in an archi-position can be omitted, it means the positions which the archi-position represents are empty.

132 Phonotactic properties of archi-position ‘E2’ are given in Figure VI. We have said that an archi-position is represented the intersection of two or more adjacent positions. This can now be expressed in a different way: if every position is characterized by a unique set of phonotactic properties, an archi-position is characterized by the intersection of phonotactic properties of two or more adjacent positions. A phoneme occurring in an archi-position has only those phonotactic properties common to two or more adjacent positions. Therefore, phonotactic properties of the phonemes belonging to ‘E2’ (i.e. to ‘e3∩e2’) are those and only those that the phonemes from pos ‘e3’ and pos ‘e2’ have in common; this can be easily verified if one makes the intersection of Figures IV and V. To given an example: /t/ belongs to ‘e3’, /v/ to ‘e2’ and /p/ to ‘E2’. Now, the combinational abilities of /p/ are equal to those which /t/ and /v/ have in common. /t/ and /v/ can be followed a phoneme from ‘e1’ (cf. /trām/ trám and /vraX/ vrah), and so can /p/ (cf. /prām/ pram). They can be preceded by a phoneme from ‘pre1’ (cf. /Stāt/ and /řvāT/ řvát) or from ‘pre2’ (cf. /rti/ rty and /lvi/ lvi), and so can /p/ (cf. /SpāT/ spát and /lpī/ lpí). They can also be preceded by two phonemes from ‘pre1’ and ‘pre2’ (cf. /FStāT/ vstát and /břve/ Břve (a place name), and so can /p/ (cf. /FSpažiT/ vzpažit). The phonemes /t/ and /v/ have other properties, but there is no other they have in common, and none that /p/ would also have. We do not provide here any examples of the distribution of /p/, /b/, /f/ because they can be gained from distributional types DE1e, DE2c, DE2k, DE2l, DE3f, DE3g, DE3m and DE4e in Chapter 11. There are no obvious collocational restrictions for this position class. The lacunae for /f/ are accidental from the synchronic perspective.

F0n F0N F1e1n P0

P1pre1 /f/ 9 P1pre2 (/f/) /f/ /f/

P2pre2pre1 /f/ /b/ /f/ /f/

Figure VI: Phonotactic properties of pos ‘E2’; phonemes listed are not attested to have a given property. See Figure II for explanations.

9 /f/ has this property in /Kfeli/, the form of the place name Kfely.

133 8.5 Position class ‘E3’

Archi-position ‘E3’ results from suspension of the difference between positions ‘e3’, ‘e2’ and ‘e1’. It can be occupied by the phonemes in (5). It holds once again that the syntagmatic difference between the mentioned positions is not relevant for these phonemes. /ť/ and /ď/ are palatal occlusives; they do not belong to any other pre-nuclear or even to any other post-nuclear position. Their archiphoneme /Ť/ only occurs in post- nuclear position ‘i3’. Thus, if a palatal occlusive appears in a phonological word, either it stands before a vowel (in which case the difference between /ť/ and /ď/ is valid), or it is the last phoneme in the phonotagm (in which case it is the archiphoneme /Ť/).

(5) pos ‘E2’ ∈ {/ť/, /ď/}

Phonotactic properties of pos ‘E3’ are given in Figure VII. They represent the intersection between the properties of pos ‘e3’, pos ‘e2’ and pos ‘e1’. Both phonemes /ť/ and /ď/ have all properties; therefore, there are no collocational restrictions. The range of their distributional and combinational possibilities can be deduced from distributional types DE1e, DE2m, DE2n and DE3l given and discussed in Chapter 11.

F0n F0N P0

P1pre1

P1pre2

P2pre2pre1

Figure VII: Phonotactic properties of pos ‘E3’. An empty cell mean a property is attested for all member of the position class.

8.6 Position class ‘pre1’

The last two pre-nuclear positions ‘pre1’ and ‘pre2’ will be called pre-explosive positions whereas the previously discussed ones are explosive positions. The latter term also includes the archi-positions. We can now propose the following condition:

134 Condition for the pre-explosive positions to be occupied by a phoneme The pre-explosive positions (‘pre1’, ‘pre2’) can be occupied by a phoneme if and only if at least one of the explosive positions (‘e1’, ‘e2’, ‘e3’ or ‘E2’ or ‘E3’) is occupied by a phoneme.

In other words, if all explosive positions are empty, so are ‘pre1’ and ‘pre2’. It is because the latter are occupied by two types of phonemes. First, voicing archiphonemes occur here, and they can, by definition, only occur before occlusives or fricatives. All explosive positions being empty, there is logically no occlusive or fricative before which they could occur. Second, the pre-explosive positions are occupied by phonemes whose primary place of occurrence is an explosive position (for example, ‘pre2’ can be occupied by /r/ which occurs primarily in ‘e1’). Position ‘pre1’ is the first pre-explosive position or the fourth position from nuclear position ‘n’ or the third position from archi-position ‘N’. The phonemes belonging there are listed under (6); the position can also be empty.

(6) pos ‘pre1’ ∈ {/T/, /S/, /Š/, /ř/, ∅}

/T/, /S/, /Š/ are products of neutralization of /t/ ~ /d/, /s/ ~ /z/, and /š/ ~ /ž/. They occur in ‘pre1’ as well as ‘pre2’. However, their occurrence in the latter is conditioned, and this condition can be stated as follows:

Condition for the occurrence of /T/, /S/, /Š/ in ‘pre2’ /T/, /S/, /Š/ occur in ‘pre2’ if position ‘pre1’ is already occupied by a phoneme.

To put it otherwise, when /T/, /S/, /Š/ stand before a phoneme from pos ‘pre1’, i.e. before /T/, /S/, /Š/ or /ř/, they occur in ‘pre2’. In all other cases they occur in ‘pre1’. The α, β and γ examples in Figure VIII illustrate the occurrence of /T/, /S/, /Š/ in ‘pre1’. The examples with apostrophes illustrate situations when they are preceded by a phoneme from ‘pre2’. Finally, the δ and ε examples illustrate their occurrence in ‘pre2’.

135 /ř/ also belongs to ‘pre1’, though it has already been assigned to ‘e1’. Figure IX compares these occurrences. It belongs to ‘e1’ when standing right before a vowel. It can then be preceded by no other phoneme (= the α example), or it can be preceded by one (= β), two (= γ), or three (= δ). It can also stand before a consonant in which case it belongs to ‘pre1’. If so, it can be the very first phoneme in the phonotagm (= ε), or it can be preceded by one consonant (= ζ). When occurring before two consonants, it can again be the first phoneme in the phonotagm (= η), or it can be preceded one consonant (= θ). It is worth noting that /ř/ never occurs before more than two consonants.

Occurrence in ‘pre1’ Occurrence in ‘pre2’ α /Txoř/ tchoř δ /TřMen/ třmen α’ /FTkāT/ vtkát ε /TŠpjeT/ čpět β /TknöT/ tknout /T/ β’ /FTšleň/ from včlenit γ /Tkvjel/ tkvěl γ’ /STkvjel/ stkvěl α /SpāT/ spát δ /STkāT/ stkát α’ /FStaň/ vstaň ε /STkliF/ from ztklivět β /SplaF/ splav /S/ β’ /PStruX/ pstruh γ /Skvjel/ from skvělý γ’ /FSkvjeT/ vzkvět α /ŠťīT/ štít δ /ŠTkāT/ štkát α’ /FŠdi/ vždy ε /ŠTknöT/ štknout β /ŠtvāT/ štvát /Š/ β’ /PŠtroS/ pštros γ – γ’ –

Figure VIII: Differences in the occurrence of /T/, /S/, /Š/ in ‘pre1’ and ‘pre2’. A dash means that a respective example has not been found.

Occurrence in ‘e1’ Occurrence in ‘pre1’ α /řāT/ řád ε /řvāT/ řvát β /tři/ tři ζ /Křťin/ from křtiny /ř/ γ /StřeT/ střed η /řvjeT/ from řvěte δ /PStřeň/ pstřeň (PSJČ) θ /hřMňeT/ hřmět

Figure IX: Differences in the occurrence of /ř/ in ‘e1’ and ‘pre1’.

136 There is one peculiar thing about /ř/ occurring in ‘pre1’; we already touched upon it in Section 2.6. It concerns the nature of the consonant which appears before it in ‘pre2’. First of all, even though ‘pre2’ can be occupied by /M/, /j/, /r/, /l/ (see the next section), these phonemes cannot stand before /ř/ provided that it occupies ‘pre1’. The phonemes /j/, /r/, /l/ cannot combine with /ř/ in any way, and /M/ can only stand before /ř/ only if /ř/ occurs in ‘e1’ (cf. /Mřenka/ mřenka). In short, if /ř/ occurs in ‘pre1’, it may only be preceded by occlusives or by fricatives. Now, the latter are subject to neutralization of voicing if this /ř/ precedes either a voiceless or voiced consonant (except for /v/), cf. /Křťin/ křtin and /XřbeT/ hřbet. But if /ř/ is followed by a phoneme for which the voicing opposition is irrelevant, there is no neutralization for the occlusives and fricatives (cf. /hřMī/ hřmí). The point is that all phonemes which precede /ř/ in these situations belong to ‘pre2’ irrespective of whether the voicing opposition is relevant for them or not. Thus, both /X/ from /XřbeT/ and /h/ from /hřMī/ occur in ‘pre2’. Whether neutralization of voicing takes place in ‘pre2’ is conditioned by the neighboring phonemes. On the other hand, it always takes place in position ‘pre1’.

F1e3n F1e3N F1e2n F1e2N F1e1n F1E2n F1E2N F1E3n /T/ /S/ /T/ /T/ P0 – – /ř/ /Š/ /ř/ /ř/ /T/ /S/ /T/ /T/ /T/ P1 – – pre2 /ř/ /Š/ /Š/ /ř/

F2e3e2n F2e3e2N F2e3e1n F2e2e1n F2E2e1nF3e3e2e1n /T/ /S/ /T/ P0 /ř/ /ř/ /ř/ /Š/ /ř/ /Š/ /ř/ /S/ /T/ /S/ /T/ P1 pre2 /ř/ /ř/ /ř/ /Š/ /ř/ /Š/ /ř/

Figure X: Phonotactic properties of pos ‘pre1’; phonemes listed are not attested to have a given property. See Figure II for explanations.

Phonotactic properties of ‘pre1’ are given in Figure X. Examples demonstrating the pre-nuclear occurrence of the phonemes from pos ‘pre1’ can be found under distributional types DE2d, DE2g, DE2i, DE2k, DE2m, DE3b, DE3d, DE3f, DE3h, DE3j, DE3k, DE3l, DE3m, DE4a, DE4b, DE4c, DE4d, DE4e, DE4f and DE5 discussed in Chapter 11. Again, the phonemes listed are those not attested to have a given phonotactic prop-

137 erty. Some properties are not attested for any phoneme. It should be noted that properties ‘F0n’ and ‘F0N’ are not included because they are by definition impossible: the phonemes from pos ‘pre1’ cannot stand before a vowel or a nuclear semiconsonant. The impossibility of properties ‘F1e2N’ and ‘F1e1n’ for all members of this position class follows from restrictions RG13a and RE5 introduced in Section 8.1. On the basis of Figure X we can propose yet other collocational restrictions: /ř/ cannot be followed by more than two peripheral phonemes (= RG11) as opposed to /S/ in /FSkvjeT/ vzkvět or /T/ in /STkvjel/ stkvěl. But if it is followed by a combination of type

‘C1C’, then ‘C1’ can be no other phoneme than one from pos ‘e2’ (= RE10a), cf. /hřMňeT/ hřmět. If the nucleus of the phonotagm is /r/ or /l/, position ‘pre1’ cannot be occupied by /ř/; it follows from restriction RG13a introduced in Section 8.1. The alveolar occlusive /T/ is not found to be followed by a combination of type

‘C1C’ provided that ‘C1’ is a phoneme from pos ‘E2’, but it is found to be followed by a combination of type C1C’ provided that ‘C1’ is from ‘e3’, cf. /Tklivī/ tklivý. Also, /T/ cannot be followed by a phoneme from pos ‘E2’ if the latter stands right before a nuclear semiconsonant. These limitations are, however, more general and hold for all occlusives. We propose the restriction that an occlusive (occurring either in ‘pre1’ or

‘pre2’) cannot be followed by a combination of the type ‘C2C1’ where ‘C2’ is from pos

‘E2’ or from pos ‘e2’, and ‘C1’ is a phoneme from pos ‘e1’ (= RE12). Another restriction affecting occlusives is this: a nuclear semiconsonant cannot be preceded by a combination of or including the type ‘OC1’ where ‘O’ is an occlusive and ‘C1’ is from pos ‘E2’ (= RN9i); vowels have this capacity, cf. /TbāT/ dbát.

8.7 Position class ‘pre2’

Pre-explosive position ‘pre2’ is the very last pre-nuclear position. When a phoneme occurs here, it is the first one in the phonotagm, as no other phoneme can come before it except for the very rare accidental appendices (see below). The members of pos ‘pre2’ are given under (7); the position can also be empty. We will return to /v/ and /b/ below. The phonemes /t/ and /h/ occur here only if they precede /ř/ as in /třMen/ třmen and

138 /hřMī/ hřmí, which we discussed in the previous section. The position class includes all voicing archiphonemes except for /Ť/ which does not occur in the pre-nuclear context.

(7) pos ‘pre2’ ∈ {/P/, /F/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/, ∅} + {/v/, /t/, /h/, (/b/)}

From the perspective of their occurrence, the phonemes belonging to pos ‘pre2’ can be divided into two groups: The first are those not occurring in any other pre-nuclear position; namely /P/, /K/, /F/, /X/. They can never be preceded by any other phoneme within a single phonotagm and are always its first phonemes. The second, larger group is constituted by phonemes which can occur in another pre-nuclear position. These are /T/, /S/, /Š/, /ř/ we discussed in the previous section, /M/, which also occurs in ‘e2’, and /j/, /r/, /l/, which also occur in ‘e1’. However, /M/, /j/, /r/, /l/ have a special status in ‘pre2’; they occur here only if all of the following conditions hold. Note that condition (γ) says that the occurrence of /r/, /l/ in nuclear archi-position ‘N’ precludes the occurrence of /M/, /j/, /r/, /l/ in ‘pre2’.

Condition of the occurrence /M/, /j/, /r/, /l/ in ‘pre2’ (α) They precede a consonant; in the case of /M/ it must not be a nasal or /j/ or /ř/.10 (β) They appear at the beginning of a phonological word. (γ) The nucleus of the phonotagm they belong to is a vowel.

It follows from the Condition for pre-explosive positions to be occupied by a phoneme given at the beginning of Section 8.6 that if /M/, /j/, /r/ or /l/ occurs in ‘pre2’, at least one explosive position is also filled with a phoneme. These phonemes are part of several unique word-initial combinations summarized in Figure XI. They have stirred much discussion among linguists. Their peculiarity derives from two related issues. First, on the phonetic level there are disputes whether the initial [m], [j], [r] and [l] constitute separate syllable cores here, i.e. whether they form a separate syllable or not,

10 If /M/ occurs before these phonemes as in /MňeSto/ město, /MjaXkī/ Mjachký (surname) or /Mřenka/ mřenka, it belongs to ‘e2’. The same is true when /M/ occurs before /r/ or /l/ as in /MraK/ mrak or /MlīT/ mlít, but these are semiconsonants, not consonants.

139 though the general consensus appears to be that they do not (Hála 1956: 62, 1961: 123; Palková 1997: 154; Duběda 2005: 132; Krčmová 2008: 177). The problem springs from disagreements about the precise definition of the core of the phonic syllable. Generally, it is assumed that it is the segment associated with the peak of sonority (however the 11 sonority is defined) . Taking [msta] as an example, we can say that such a peak is represented here by the vocoid [a]. However, the initial [m] corresponds to another peak of sonority, a secondary one, because the immediately neighboring sounds [s] and [t] are allegedly less sonorous. The word msta must then be disyllabic under the sonority theory. Mutatis mutandis, the same holds for all words in Figure XI. This is, however, in contradiction with how they are treated in Czech (e.g. in verse)—they are not taken to contain any extra syllables. To reflect (but not to solve) this controversy, the segments [m], [j], [r], [l] have been called examples of pobočná slabika (discussed in detail in Skaličková 1954, 1958, and Hála 1956: 59–64, 1961: 120–4). The English term might be side syllable (literary translation, after Kučera 1961) or secondary syllable (free translation, after Krámský 1976: 57–70).

With /M/ With /j/ With /r/ With /l/ /Mz/ mze (SSJČ) /jm/ jméno /rv/ rvát se /lp/ lpí /Mš/ mše /jd/ jdu /rm/ rmoutit /lb/ lbový (SSJČ) /Mž/ mžitky /jď/ jdi /rt/ rty /lv/ lvi /Mh/ mhouřit /js/ jsi /rd/ rdousit /ln/ lnout /Mdl/ mdlý /jh/ jho /rť/ rtěnka /lň/ lněný /MSť/ mstít se /jMň/ jmění /rď/ rdít se /lz/ lze /MSt/ msta /jsm/ jsme /rz/ rzi /lž/ lži /MSď/ mzdě /jSt/ jste /rž/ ržát /lk/ lkát /Mzd/ mzda /rvj/ rvěte se /lh/ lhát /MStň/ mstně /rTs/ rci (archaic) /lpj/ lpět /MStn/ mstný /rTš/ rčení /lSť/ lstivý (/MTs/ Mcely) (/rt/ Rtepy) /lStm/ lstmi /lStň/ lstně (SSJČ) /lStn/ lstný (SSJČ) (/lTš/ Lčovice) (/lŠt/ Lštěň)

Figure XI: Known “side-syllabic” combinations. Some are found in place names only.

11 As repeatedly pointed out (e.g. by Clements 1990, Butt 1992, Harris 2006), it is hard (if possible at all) to define sonority on phonetic grounds. Cf. also Lebrun (1966).

140 The second issue with the combinations in Figure XI is related to the first. Even if the segments [m], [j], [r], [l] are viewed as not corresponding to separate syllable cores, the combinations are still problematic because they violate the so-called Sonority Se- quencing Principle (= SSP) according which sonority increases in a syllable toward the core, and decreases toward the margins (see Clements 1990 and references therein). Thus, we get to the same problem through a different path. Now, SSP is sometimes claimed to be a universally valid principle of syllable organization, even though it is well-known that there are many languages which violate it, and Czech is one of them.12 Various strategies have been offered to redeem the principle (Blevins 2006: 334); the idea of side syllables is one of them. One other strategy is to treat the offending phonemes as extrasyllabic, i.e. as not belonging to any syllable. Once they are removed from the domain of the syllable, it does indeed follow SSP. However, in our view the only purpose of such strategies is to evade one obvious fact: languages do not necessarily follow SSP. What is more, the idea of sonority is by itself rather dubious and controversial, and we do not operate with SPP for these and other reasons. The “side-syllabic” /M/, /j/, /r/, /l/ can be explained differently. To begin with, we reject the idea that /M/, /j/, /r/, /l/ would form a separate syllable (i.e. phonotagm) in the combinations in Figure IX. If it were so, they would have to be nuclear entities, but this is not the case. Let us illustrate it on /rti/ rty. Had it consisted of two phonotagms, it should be possible to parse it to them. There would be only two possible divisions given under (8). However, no matter how /rti/ is parsed, /r/ in (8a) and /rT/ in (8b) will never be well-formed phonotagms (although /ti/ is, cf. ty). They violate the distributional rule introduced in Section 7.2: a semiconsonant is the nucleus of a phonotagm if and only if preceded by at least one consonant; this rule is otherwise valid for all phonological words in Czech. Consequently, the division (8a) and (8b) are in conflict with the Kuryłowicz condition (see Section 3.5). The rule could of course be reformulated, but in order to keep our description as simple as and as consistent as possible, we will not introduce new elements into it unless necessary. Failing to prove /rti/ contains two phonotagms, we conclude it is a single phonotagm.

12 For a long list of other languages violating SSP, see Clements (op. cit.) and Blevins (2006).

141 (8) (a) /r/ + /ti/ (b) /rT/ + /ti/ (with functional amalgamation of /T/ and /t/)

The whole problem can also be approached from a different perspective: /rti/ is a single phonotagm because it holds that if one of the phonemes /M/, /j/, /r/ or /l/ appears at the beginning of a phonological word before a consonant, which in the case of /M/ is not a nasal, /j/ or /ř/, the phonological word contains at least one vowel. This is in fact a restatement of the Conditions of the occurrence of /M/, /j/, /r/, /l/ in ‘pre2’ given above. Simply said, the occurrence of word-initial /M/, /j/, /r/ or /l/ presupposes the occurrence of at least one vowel. On the contrary, the occurrence of a vowel does not presuppose the occurrence of word-initial /M/, /j/, /r/ or /l/, and so the latter cannot be nuclear phonemes of a separate phonotagm (see Section 4.3). We likewise reject the idea that /M/, /j/, /r/, /l/ are extrasyllabic. We do not see a reason why they should not belong to any phonotagm just because they violate SSP. The point is to determine whether they can be mapped onto the so-far established network of positions. Figure XII demonstrates that it is perfectly possible to assign them to position ‘pre2’ because they parallel the distribution of other phonemes belonging there.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ Phonological word /l/ /S/ /t/ ∅ /n/ /ī/ /lStnī/ lstný /T/ /S/ /t/ ∅ /m/ /i/ /TStmi/ ctmi (inst. pl. of čest) ∅ /S/ /t/ ∅ /m/ /ī/ /StmīF/ from stmívat /F/ /S/ /t/ ∅ ∅ /ā/ /FStāT/ vstát /M/ /S/ /t/ ∅ ∅ /a/ /MSta/ msta /j/ /S/ /t/ ∅ ∅ /e/ /jSte/ jste /M/ /S/ /ť/ /i/ /MSťi/ msti (se) /l/ /S/ /ť/ /i/ /lSťi/ lsti /r/ /T/ /š/ ∅ ∅ /e/ /rTšeň/ from rčení /F/ /T/ /š/ ∅ ∅ /e/ /FTšer/ from včera

Figure XII: Assignment of “side-syllabic” /M/, /j/, /r/, /l/ to ‘pre2’.

Let us now turn the attention to /t/, /h/, /v/ which also occur in ‘pre2’. The occurrence of /t/ and /h/ is in fact restricted to one particular case: when /ř/ occurs in ‘pre1’ as in the forms /třMen/ or /hřMī/ dealt with in the previous section. We could add one more phoneme here: /b/ from /břve/, a phonological form of the place name Břve, which attests

142 the occurrence of /b/ in ‘pre2’. As regards the occurrence of /v/ in the same position, let us first examine the examples in (9).

(9) (a) /vMlāťiT/ vmlátit (b) /vMňeStnaT/ vmněstnat (c) /vMrŠťiT/ vmrštit

When interpreting the pre-nuclear combinations /vMl/ and /vMň/, it is reasonable to assign /M/ to ‘e2’, and /l/ and /ň/ to ‘e1’ since these are the positions where they otherwise occur. We are left with /v/ which normally occurs in ‘e2’, but this position is already taken up by /M/ here. We must assign /v/ to another position which may be either ‘e3’, ‘pre1’, ‘pre2’ or their potential archi-position ‘pre2∩pre1∩e3’. However, before choosing one of the solutions, we must consider several facts. First of all, no other phoneme can precede /v/ in the pre-nuclear combinations /vMl/ and /vMl/ (no such combinations are attested). Similarly, no phoneme can come between /v/ and /M/. These facts suggest that an archi-position should be introduced here as the phonotactic difference between positions ‘pre2’, ‘pre1’ and ‘e3’ is apparently suspended for /v/ in the forms (9). This should have been a preferred solution, if it were not for one important fact: /v/ would be the only phoneme to occur in such an archi-position. Phonemes with which /v/ commutes in the context before /Ml/ or /Mň/ belong to other positions: cf. /s/ in /sMň/ from směna or /t/ in /Mň/ from tmě, both occurring in ‘e3’. However, it is not possible that an archi-position is introduced for only one entity. We may have a position which is occupied by one phoneme provided that it can also be empty,13 but we cannot have an archi-position occupied just by one phoneme, as archi-positions can never be empty. Such an archi-position would be not functional because it would not offer a possibility of an alternative choice. Where no choice can be made, there is no function (Mulder 1989: 83ff.). Therefore, we must assign /v/ either to ‘e3’, ‘pre1’ or ‘pre2’. Out of these choices, we regard ‘pre2’ as the optimal one. The phoneme /v/ does not undergo neutralization of voicing, and so we can reject ‘pre1’ because the latter is a position where

13 /s/ occurring in English /smail/ smile is the only phoneme occupying the given position, but the position can also be empty, as /mail/ mile proves.

143 the voicing opposition is always suspended; as we said in the previous section, in ‘pre2’ it depends on the neighboring consonants.14 Moreover, a labial, namely the labial fricative archiphoneme /F/, has already been shown to belong to ‘pre2’ whereas no labial consonant occurs in ‘pre1’ or ‘e3’. Therefore, it makes more sense to assign /v/ to ‘pre2’. Like in the case of the alveolar occlusive in /třMen/ třmen vs. /Třťin/ třtin, and of the velar fricative in /hřMī/ hřmí vs. /Xřtān/ chřtán, the voicing of the labial fricative will be dependent on the neighboring phonemes. The phonotactic analysis of these forms is given in Figure XIII. Let us note that when neutralization of voicing is not operative in ‘pre2’, position ‘e3’ is always empty.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ Phonological word /t/ /ř/ ∅ /M/ ∅ /e/ /třMen/ třmen /T/ /ř/ /ť/ /i/ /Třťin/ třtin /h/ /ř/ ∅ /M/ ∅ /ī/ /hřMī/ hřmí /X/ /ř/ /t/ ∅ ∅ /ā/ /Xřtān/ chřtán /b/ /ř/ ∅ /v/ ∅ /e/ /břve/ Břve /P/ /S/ /t/ ∅ /r/ /u/ /PStruX/ pstruh /v/ ∅ ∅ /M/ /l/ /a/ /vMlaŤ/ vmlať /v/ ∅ ∅ /M/ /ň/ /e/ /vMňeST/ from vměstnat /v/ ∅ ∅ /M/ /r/ /vMrŠŤ/ vmršť /F/ /S/ /t/ ∅ ∅ /ā/ /FStāT/ vstát

Figure XIII: Phonotactic analysis of pre-nuclear combinations when the opposition between voiceless and voiced consonants in valid in ‘pre2’.

Having discussed conditions of the occurrence of phonemes in ‘pre2’, we can now look at Figure XIV which lists phonotactic properties for this position. Once again, we do not provide here any examples because these can be gained from distributional types DE2e, DE2h, DE2j, DE2l, DE2n, DE3c, DE3e, DE3g, DE3i, DE3j, DE3k, DE3l, DE3m, DE4b, DE4c, DE4d, DE4e, DE4f and DE5 discussed in Chapter 11. When compiling the tables in Figure X, we have not taken into account the phonemes /t/, /h/, /v/

14 Although there are no combinations like /fMň/ or /fMl/ proving that the opposition /f/ ~ /v/ is indeed valid before /M/ here, this fact should not be taken as evidence for the neutralization. Had we interpreted it so, we would have lost the very cause of this neutralization: in Czech the opposition between voiceless and voiced consonants is not valid before a voiceless or voiced consonant (except for /v/). The archiphoneme /M/ is neither voiceless nor voiced. Hence, there cannot be neutralization of voicing.

144 and /b/ because their occurrence is very limited. /t/ and /b/ has only property ‘F2pre1e2’

(cf. /třMen/ and /břve/), /h/ only properties ‘F2pre1e2’ and ‘F3pre1e2e1’ (cf. /hřMī/,

/hřMňeT/), and /v/ only properties ‘F1e2N’ and ‘F2e2e1’ (/vMrŠťiT/, /vMlāťiT/). Note also that properties ‘F0n’, ‘F0N’, ‘F1pre1n’ and ‘F1pre1N’ are not included because they are by definition impossible: the phonemes from pos ‘pre1’ and ‘pre2’ cannot stand before a vowel or a nuclear semiconsonant.

F1e3 F1e3N F1e2 F1e2N F1e1 F1E2 F1E2N F1E3 /P/ /P/ /F/ /P/ /F/ /P/ /P/ /T/ /S/ /T/ /S/ /T/ /S/ /T/ /S/ /T/ /S/ /T/ /S/ /T/ /S/ /Š/ /Š/ /Š/ (see /Š/ /Š/ /Š/ /Š/ P0 # /X/ /K/ /X/ /K/ /X/ text) /K/ /X/ /K/ /X/ /K/15 /M/ /j/ /M/ /j/ /M/ /M/ /j/ /M/ /j/ /M/ /r/ /l/ /r/ /r/ /l/ /l/

F2pre1e3 F2pre1e3N F2pre1e2 F2pre1e2N F2pre1e1 F2pre1E2 F2pre1E2N F2pre1E3 /P/ /P/ /P/ /P/ /S/ /S/ /T/ /S/ /S/ /Š/ (see /Š/ /Š/ /Š/ P0 – – # /K/ /X/ text) /K/ /K/ /X/ /X/ /M/ /j/ /M/ /j/ /M/ /j/ /j/ 16 /l/ /r/ /l/ /r/ /l/ /r/ /l/ /r/

F2e3e2 F2e3e2N F2e3e1 F2e2e1 F2E2e1 F3pre1e3e2 F3pre1e3e2N /P/ /P/ /P/ /F/ /P/ /P/ /P/ /F/ /T/ /S/ /T/ /S/ /T/ /S/ /T/ /S/ /Š/ /Š/ /Š/ /Š/ /Š/ /Š/ P0 – # /K/ /X/ /K/ /X/ /K/ /X/ /K/ /X/ /K/ /X/ /K/ /X/ /M/ /j/ /M/ /M/ /j/ /M/ /j/ /M/ /j/ /r/ /l/ /r/ /l/ /l/ /r/ /r/ /l/ /r/ /l/

F3pre1e3e1 F3pre1e2e1 F3pre1E2e1 F3e3e2e1 F4pre1e3e2e1 /P/ /P/ /P/ /S/ /T/ /S/ /T/ /Š/ /Š/ /Š/ P0 (see text) # /K/ /X/ /K/ /X/ /K/ /X/ /K/ /X/ /j/ /M/ /j/ /M/ /j/ /M/ /j/ /r/ /r/ /l/ /r/ /l/ /r/ /l/

Figure XIV: Phonotactic properties of pos ‘pre2’; phonemes listed are not attested to have a given property. See Figure II for explanations.

15 /K/ has this property in /KťiŠ/, the form of the place name Ktiš. 16 /l/ has this property in the place names Lčovice and Lštěň.

145 On the basis of Figure XIV we propose the following collocational restrictions limiting the distribution of phonemes from pos ‘pre2’: An occlusive cannot be followed by more than three peripheral phonemes as opposed to the fricative /S/ in /STkvjel/ stkvěl; the same applies to the nasals and the semiconsonants (= RG10a). To put it otherwise, only fricatives can be followed by four peripheral phonemes because /j/ and /ř/ cannot be followed by more than two peripheral phonemes (= RG11; see also the previous section). The phonemes /M/, /j/, /r/, /l/ are not attested to be followed by two types of combination; first, by a combination of the type ‘CC1(C)’ where ‘C1’ is from pos ‘E2’ or from pos ‘e2’; second, by a combination of the type ‘CCC1(C)’ where ‘C1’ is from pos ‘e2’; in both cases ‘(C)’ is an optional peripheral phoneme. Compare it with /F/ in /FSpřīm/ from vzpřímený or with /S/ in /STkvī/ stkví. Therefore, we propose that a nasal and a sonant17 cannot be followed by such combinations (= RE11a and RE11b). The approximant /j/ cannot also be followed by a phoneme from pos ‘E2’ irrespective of whether the latter is further followed by a phoneme from pos ‘e1’ or not (= RE16). A velar cannot be followed by more than two peripheral phonemes (= RG9), and a palatal cannot be followed by more than three (= RG10b) as opposed to the labial /F/ in /FSkvjeT/ vzkvět and the alveolar /T/ in /TŠtvero/ čtvero. A velar cannot also be followed by a combination of the type ‘C1C’ where ‘C1’ is from pos ‘e3’ or from pos ‘e2’ or from pos ‘E2’ (= RE13a). The same restriction holds for /P/ (= RE13b). The occlusives /P/, /K/ are further subject to restriction RE12 introduced in the previous section. Finally, a nuclear semiconsonant cannot be preceded by a combination of the type

‘NCn’ where ‘N’ is a nasal and ‘Cn’ any number of peripheral phonemes (= RN9a).

8.8 Accidental appendices

Position ‘pre2’ is the last we have found necessary to introduce for describing the distribution of peripheral phonemes in the pre-nuclear context. Pre-nuclear combinations found in phonological words in Czech can be adequately described with these five pre- nuclear positions and/or with their archi-positions. They are discussed in Chapter 11.

17 The restriction applies even to /ř/, which is also a sonant, see Figure X.

146 There are, however, a handful of combinations that cannot be analyzed with the pre- nuclear position so far introduced. Although their occurrence is very limited, they are still existent. See the examples in (10); there may be others.

(10) /StřMen/ střmen /zMdlelī/ zmdlelý /STŠtvernāsobiT/ zčtvernásobit

The pre-nuclear combinations attested in these forms pose a problem for our analysis. As explained in the previous two sections, /t/ in /StřM/, /M/ in /zMdl/ and /T/ in /STŠtv/ must occur in ‘pre2’ which is the last pre-nuclear position or, which equals to the same, the very first position of the distributional unit. This tentative analysis is given in Figure XV. It is apparent that all of these phonemes are preceded by another one, namely by /S/ or /z/ which remain unassigned to any position. Out hypothesis about there being only five pre-nuclear positions seems to find a refutation. However, let us first look more closely at the words in question.

? ‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ Example /S/ /t/ /ř/ ∅ /M/ ∅ /e/ střmen /z/ /M/ ∅ /d/ ∅ /l/ /e/ zmdlelý /S/ /T/ /Š/ /t/ /v/ ∅ /e/ zčtvernásobit

Figure XV: Tentative analysis of /StřM/, /zMdl/, /STŠtv/, /zlh/ and /zrd/.

According to SSJČ and PSJČ, střmen is an archaic variant of třmen “stirrup”; it is also a place name. Another variant is střemen. Similarly, SSJČ and PSJČ claim that zmdlelý is an archaic variant of zemdlelý “quailed”.18 The word zčtvernásobit has also a variant with an inserted e: zečtvernásobit “to quadruple”, but whereas PSJČ asserts that this latter variant is more common, SSJČ claims the opposite: zčtvernásobit is more common. It is hard to say which of them is indeed more common because both are quite rare as one can confirm by searching them in the Czech National Corpus or through Google.com19. At any rate, it is obvious that the forms under (10) are somewhat acci-

18 Cf. also the verse (14) below. 19 The situation as of 20 June 2011.

147 dental, so to speak, and apparently even hard to pronounce as their variants with an inserted e suggest. We do not think necessary to introduce a new pre-nuclear position for /S/ and /z/ in these words. If it were introduced, it would be empty in all Czech words except for these deviant ones, and when filled by a phoneme, it would only be /z/ or its archiphoneme /S/.20 Yet we feel it is necessary to account for the occurrence of these phonemes in some way, and accordingly we propose to introduce the term accidental appendix (Mulder 1994: 51).21 It will refer to the accidental occurrence of phonemes (strictly speaking, only of /z/ and its archiphoneme /S/) at the beginning of phonotagms when position ‘pre2’ is already occupied by a phoneme. Thus, the question mark in Figure XV does not correspond to a separate position but to these appendices. Now, in the rest of this section we will deal with a couple of other problematic phonological forms reproduced as follows:

(11) /zlhoStejňeT/ zlhostejnět /zrdösiT/ zrdousit

The mentioned words do not appear to have any variant, but they are somewhat different to the previous ones. The word zlhostejnět is a perfective of the imperfective lhostejnět “to become indifferent”. The word zrdousit is derived from rdousit “to choke, to strangle (imperfective)”. Its meaning is “to put an end to something (especially emo- tional attitude) by force” (PSJČ) or “to choke, to strangle (perfective)” (SSJČ). The problem is this: both lhostejnět and rdousit are examples of side-syllabic /l/ and /r/, that is, of words where these phonemes are not nuclei of separate phonotagms, even though their realizations [l] and [r] correspond to peaks of sonority (see Section 8.7). The question that must now come to one’s mind is: are /l/ and /r/ in zlhostějnět and zrdousit nuclei of separate phonotagms? If they are, the forms under (11) pose no problem because they do not contain any deviant phoneme combination. However, if they are not nuclei, then the combinations /zlh/ and /zrd/ are unique in the pre-nuclear context.

20 Unless we accept the word vmžíti, an apparent hapax legomenon, mentioned in the database of PSJČ. In this word the new position would be occupied by /v/ because /M/ would have to belong to ‘pre2’. 21 Mulder (ibid., 1989: 228ff.) has noted similar accidental appendices in English.

148 To begin with, we should note that this problem cannot, as of now, be solved on the phonetic grounds because, to the best of our knowledge, no phonetic research has been dedicated to it.22 What is more, it is not certain whether there is even any relevant phonetic difference between syllabic and non-syllabic [r], [l] in Czech (cf. Section 2.5). Yet one cue might be the behavior of these words in verse, as suggested by Ziková (2008: 143, note 88)23 who mentions the following examples (syllable peaks underlined):

(12) Vlk si zuby brousil, že by ji hned zrdousil.

The argument is that the line Vlk si zuby brousil has the same number of syllables as the line že by ji hned zrdousil because the latter is “součástí říkanky s pravidelným šes- tislabičným rytmem [part of a rhyme with a regular six-syllable rhythm]” (ibid.). How- ever, provided that it is taken from the poem Vlk vlku vlkem by valesan24, this is not true, as becomes clear when we look at the whole verse:

(13) Na rozcestí slípka bílá (8 syllables) bezstarostně lenošila. (8 syllables) Vlk si zuby brousil, (6 syllables) že by ji hned zrdousil. (? 6 syllables) Slípka se však nezalekla (8 syllables) a pěkně mu to od plic řekla. (9 syllables) Že jí [sic! ji] vlk nemá žrát, (6 syllables) jinak bude litovat. (7 syllables) Vlk se lekl, až mu změkl (8 syllables) a raději utekl. (7 syllables) Tak chudák vlčici vzal (7 syllables) a táhli spolu dál. (6 syllables)

It is apparent that the lines hardly agree in the number of syllables, and accordingly there is no a priori reason the line že by ji hned zrdousil should contain six syllables, i.e. that zrd is not a separate syllable. However, while this invalidates Ziková’s example, it

22 Confirmed by Pavel Machač, Institute of Phonetics, Charles University (personal communication). 23 Ziková is one of the few to even mention this problem. Another reference can be found in Renský (1960: 93) and Short (1985: 40) for both of whom the l in zlhostejnět is not syllabic. 24 Available online (as of 20 June 2011): .

149 does not necessarily invalidate her argument.25 There is another example of the use of zrdousit in verse, from the poem Váhy by Bohumil Adámek26:

(14) Různý kroj se v zástupech těch kasá, mihá; (12 syllables) a kde nezrdousila písně klopot tíha, (? 12 syllables) chvílemi již v řeči cizí (8 syllables) domorodý zvuk se ztápí, mizí. (10 syllables)

The first two lines agree in the number of syllables provided that zrd in nezrdousila is not syllabic. Although the next two lines do not agree in the number of syllables, the whole structure of the verse is in accord with remaining verses of the poem (i.e. the first lines contain 12 syllables, the second lines 12 syllables, the third lines 8 syllables, and the fourth lines 10 syllables). Therefore, the author apparently meant zrd not to be syllabic. Another verse, this time containing zlhostejnělo, a form of zlhostejnět, may be cited.27 If the zlh is not a separate syllable here, the third line agrees with the second one in the number of syllables (the first one agrees and rhymes with the fourth one).

(15) Srdce lidu raněné, (7 syllables) k odvetě byť na čas zmdlelo, (8 syllables) ochladlo a zlhostejnělo, (? 8 syllables) nikdy nezapomene. (7 syllables)

Considering these examples, we conclude that zrdousit and zlhostejnět are indeed treated as not containing a syllabic r and l. If we approach it from a different angle, it means there is a difference between the words under (16a), which contain allegedly

25 Another argument is that r is zrdousit and l in zlhostejnět are not stressed, the stress being on the adjacent vowel (Ziková ibid., Short ibid.), i.e. that the vowels have some phonetic features making them different to the r or l. However, this is dubious because no stable phonetic correlates of the Czech stress has been found, the stress being rather potential or derived from the overall sound properties of the whole foot (Palková 1997: 277ff., Duběda 2002, Palková – Volín 2003, Palková 2004, Duběda – Votrubec 2005). 26 From the collection Horské ovzduší, Prague, 1902, p. 69. 27 From the book Král Ječmínek by František S. Procházka, Prague, 1906, p. 106.

150 non-syllabic r and l, and those under (16b), which contain allegedly syllabic r and l. As we have said, it is not at all certain whether there is a phonetic difference between these words (and/or whether the difference is quantitatively the same as the one between (16a’) and (16b’)).

(16) Allegedly non-syllabic Allegedly syllabic (a) zrdousit “to choke” (b) zrcadlit “to mirror” zlhostejnět “to become indifferent” slza “tear” (a’) rty “lips” (b’) krtek “mole” lsti “trick” plsti “felts”

We regard the phonetic proof superior because the treatment in verse may be influenced by the fact that we know the words zrdousit and zlhostejnět are derived from rdousit and lhostejnět where the r and l are allegedly non-syllabic. However, would we treat them alike if we did not have this information? We can perform a thought experi- ment: Suppose there is a Czech verb rcadlit (there is no reason that such a word could not exist); the initial r would be allegedly non-syllabic. It should be possible to coin a derivate from it by the perfective prefix z-, and the result will be zrcadlit, presumably trisyllabic. Now, this word (let us mark it zrcadlit2) should, with all that we know, be treated differently in verse than zrcadlit1 “to mirror” (quadrisyllabic) because it, by assumption, contains a different number of syllables. And would, in addition, these words be pronounced differently, zrcadlit1 as [zrʦadlɪt̩ ], i.e. with syllabic [r̩], and zrcadlit2 as [zrʦadlɪt], i.e. with non-syllabic [r]? Put this way, it may not seem so plausible now because it would mean there is an opposition, i.e. a functional difference, between syllabic [r̩] and non-syllabic [r], that is, there are two /r/ phonemes in Czech. If so, the whole phonematic system should be reconsidered. Until further evidence is available, we reject such an analysis. For us, therefore, zrcadlit2 would have the same phonotactic structure like zrcadlit1, i.e. /zrTsadliT/ with nuclear /r/ because it stands between two consonants which the context where semiconsonants are nuclear in Czech. This holds for the /l/ in /zlhoStejňeT/, too. Whether these /r/ and /l/ are syllabic or not remains to be seen. As again shown, syllabicity does not necessarily coincide with nuclearity.

151 Chapter Nine DISTRIBUTION IN THE POST-NUCLEAR CONTEXT

9.1 Position class ‘i1’

The distribution in the post-nuclear context is divided among three positions ‘i1’, ‘i2’, ‘i3’, and one archi-position ‘I’; by definition, only consonants and semiconsonants occur here. Position ‘i1’ is the first post-nuclear position; if a phoneme occurs here, it stands right after a vowel or after a semiconsonant. It can also be empty. The membership of pos ‘i1’ is given under (1). All of these phonemes also occur in another position with an exception of /j/ which occurs only in ‘i1’. /P/, /T/, /K/ also occurs in ‘i2’ and ‘i3’; we will return to this in Section 9.2. /m/, /n/, /ň/, /r/, /l/ also belong to pos ‘I’, but ‘i1’ is the primary place of their occurrence in the post-nuclear context; see Section 9.4. It is worth noting that all phonemes occurring in pre-nuclear position ‘e1’ occur in post- nuclear position ‘i1’. The only exception is /ř/ which belongs to ‘i2’ in the post-nuclear context. Phonotactic properties of pos ‘i1’ are given in Figure I. Examples of the occurrence of /P/, /T/, /K/ in ‘i1’ can be found in Figure II, and those for the occurrence of /m/, /n/, /ň/, /r/, /l/ in Figure VII below. For additional examples see distributional types DI1a, DI2a, DI2b, DI2c and DI3 given and discussed in Chapter 12.

(1) pos ‘i1’ ∈ {/P/, /T/, /K/, /m/, /n/, /ň/, /j/, /r/, /l/, ∅}

/P/, /T/, /K/ and /m/, /n/, /ň/ are not attested to be followed by a phoneme from pos ‘I’, i.e. by /m/, /n/, /ň/, /r/, /l/. This is a result of a more general limitation which can be accounted for by this collocational restriction: An occlusive, a fricative and a nasal cannot be followed by a semiconsonant or by a nasal in the post-nuclear context (= restriction RI1 in Appendix A). In other words, only a sonant can be followed a semiconso-

152 nant or by a nasal, cf. /j/ in /xejr/ chejr and /r/ in /Tšerň/ čerň. As seen, if a semiconsonant is the nucleus of a phonotagm, it is never followed by /j/; accordingly, we adopt this as a collocational restriction, namely RN10a (see also the next section on /ř/).

F0 F1i2 F1i3 F1I F2i2i3 /P/ /T/ /K/ /P/ /T/ /K/ P0n /m/ /n/ /ň/ /m/ /n/ /ň/ /ň/ 1 /l/ /P/ /K/ /P/ /T/ /K/ P0N /m/ /ň/ – – /m/ /ň/ /j/ /j/ /r/ /l/ /j/ /r/ /l/

Figure I: Phonotactic properties of pos ‘i1’; phonemes listed are not attested to have a given property. A dash means a property is not attested for any member the position class; an empty cell means a property is attested for all members of the class.

Occurrence in ‘i1’ Occurrence in ‘i2’ Occurrence in ‘i3’ α /xlaP/ chlap δ /alP/ alb θ (/vīSP/ výsp) β /pePř/ pepř ε /TserPT/ fr. excerpt ι /zāTSP/ zácp /P/ γ /zāPST/ zábst ζ /SkriPT/ skript η (/hiPŠ/ Hybš) α /jeT/ jet δ /nārT/ nárt θ /buřT/ buřt β /moTř/ modř ε /pūjTŠ/ půjč ι /kumŠT/ kumšt /T/ γ /pēTST/ péct ζ – η /puTŠ/ puč α /bīK/ býk δ /lajK/ laik θ /haFK/ fr. tomahavk β /moKř/ mokř ε /junKT/ fr. adjunkt ι /vojSK/ vojsk /K/ γ /teKST/ text ζ /aKT/ akt η /ŠtaKŠ/ fr. jakžtakž

Figure II: Differences in the occurrence of /P/, /T/, /K/ in ‘i1’, ‘i2’ and ‘i3’. A dash means that a respective example has not been found.

9.2 Position class ‘i2’

Position ‘i2’ is the second post-nuclear position. Its position class is given in (2); it can also be empty. We have already mentioned that /P/, /T/, /K/ also belong to ‘i1’ and ‘i3’; Figure II illustrate differences in their distribution. /S/, /Š/ can also occur in ‘i3’, and this is exemplified in Figure III and discussed below.

1 The phonemes /m/, /n/, /l/ have this property in the archaic words jsemť, onť, bylť.

153 (2) pos ‘i2’ ∈ {/P/, /T/, /K/, /S/, /Š/, /F/, /X/, /ř/, ∅}

/P/, /T/, /K/ occur in ‘i1’ when standing right after a nuclear phoneme, and when being at the same time the last phonemes in the phonotagm (cf. the α examples in Figure II). They also occur in this position when they stand right after a nuclear phoneme, and when they are followed by one phoneme belonging to ‘i1’ (= β) or by two phonemes belonging, respectively, to ‘i1’ and ‘i2’ (= γ). They occur in ‘i2’ when position ‘i1’ is already occupied by a phoneme; in other words, they occur in ‘i2’ when they are preceded by a phoneme from ‘i1’ in which case they may be the last phonemes in a phonotagm (see the δ examples). They also occur in ‘i2’ when both preceded and followed by one peripheral phoneme (= ε). They are, furthermore, two special cases of their occurrence in ‘i2’. First, if followed by /T/, the phonemes /P/, /K/ occur in ‘i2’, not in ‘i1’ (= ζ) even if position ‘i1’ is empty2. Second, if followed by /Š/, the phonemes /P/, /T/, /K/ also occur in ‘i2’, not in ‘i1 (= η), again even if position ‘i1’ is empty. The reason is this: a phoneme belongs to a certain position only if it displays phonotactic properties of that position. It is a property of ‘i2’ that phonemes occurring there can be both preceded and followed by one phoneme. Now, the differences in (3) show that when /P/, /K/ are followed by /T/ or when /T/ is followed by /Š/, they may be preceded by another peripheral phoneme, namely by a phoneme from pos ‘i1’.

(3) (a) /SkriPT/ skript (a’) /TserPT/ from excerpt (b) /aKT/ akt (b’) /junKT/ from adjunkt (c) /puTŠ/ puč (c’) /pūjTŠ/ půjč

The same does not, however, apply when they are, for instance, followed by /ř/ because the combinations /Př/, /Tř/, /Kř/ cannot further expanded from the left (i.e. there are no combinations /CPř/, /CTř/, /CKř/ where ‘C’ is any peripheral phoneme). As regards the combinations /PŠ/ (attested only in the surname Hybš) and /KŠ/, it should be noted that they cannot also be expanded from the left (i.e. there are no combinations /CPŠ/ and /CKŠ/ where ‘C’ is any peripheral phoneme), but for the sake of the overall simplicity

2 This does not concern /T/ because Czech does not allow geminates.

154 we treat them in the same manner like /TŠ/ which can be expanded (cf. /pūjTŠ/). Con- sequently, we can sum it up in the following points:

Conditions for the occurrence of /P/, /T/, /K/ in ‘i2’ (α) /P/, /T/, /K/ occur in ‘i2’ if they are preceded by /m/, /n/, /ň/, /j/, /r/, /l/. (β) /P/, /K/ occur in ‘i2’ if they are followed by /T/. (γ) /P/, /T/, /K/ occur ‘i2’ if they are followed by /Š/.

Finally, /P/ /T/, /K/ occur in ‘i3’ if they are the last phonemes in the phonotagm and are at the same time preceded by one phoneme from ‘i2’ (= the θ examples in Figure II). They also belong to ‘i3’ if they are preceded by two phonemes from ‘i1’ and ‘i2’, respectively (= ι). We can summarize this by the following condition:

Condition for the occurrence of /P/, /T/, /K/ in ‘i3’ /P/, /T/, /K/ occur in ‘i3’ only if they follow a phoneme from pos ‘i2’ (i.e. /P/, /T/, /K/, /S/, /Š/, /F/, /X/, /ř/).3

A note should be given to /SP/ given in Figure II under (θ). This combination is attested only in výsp, gen. pl. of výspa. Although výsp can be found in the Czech National Corpus and through Google.com (a situation as of 20 June 2011), there are disagreements whether it is correct in Standard Czech. The most recent grammar (MSČ, 2010, p. 173) says the only correct genitive plural is výsep, and so does MČ2 (1986, p. 325)4. On the other hand, PMČ (1996, p. 257) gives only výsp. Phonologically, there is no reason to declare /SP/ invalid. First, it fits the pattern of /ST/, /SK/ and /SŤ/, i.e. it is of the type alveolar fricative + occlusive. If /SP/ were rejected, /P/ would be the only occlusive which could not be preceded by /S/ in the post-nuclear context.5 Second, /SP/ is in fact attested as a part of the post-nuclear combination /TSP/ (found in zácp, gen. pl. of

3 Actually, it is only /T/ which can precede /P/ and /T/; see the previous condition. 4 výsep is also mentioned as the only genitive plural of výspa in Internet Language Reference book; see (accessed 20 June 2011). 5 Labial occlusives can be preceded by /S/ in the pre-nuclear context, cf. /SpāT/ spát.

155 zácpa, also given in Figure II). Considering these facts, we regard the combination /SP/ as structurally possible, only perhaps accidentally missing. Let us now move to /S/ and /Š/. Their post-nuclear occurrences are exemplified in Figure III. If they stand right after a nuclear phoneme, and are at the same time the last phonemes in a phonotagm, they occur in ‘i2’ (= the α examples). They occur in this position even if they are not the last phonemes of a phonotagm, in which case they can be followed by just one phoneme from ‘i3’ (= β), that is, they are never followed by two peripheral phonemes in the post-nuclear context. They also occur in ‘i2’ if they are preceded by a phoneme from ‘i1’ (= γ) or if they are both preceded by one peripheral phoneme ‘i1’, and followed by one peripheral phoneme from ‘i3’ (= δ).

Occurrence in ‘i2’ Occurrence in ‘i3’ α /noS/ nos ε /prinTS/ princ β /moST/ most ζ – /S/ γ /kurS/ kurz η – δ /teKST/ text α /muŠ/ muž ε /pūjTŠ/ půjč β /poŠT/ pošt ζ /jiXŠ/ jichž /Š/ γ /jenŠ/ jenž η /puTŠ/ puč δ /kumŠT/ kumšt

Figure III: Differences in the occurrence of /S/, /Š/ in ‘i2’ and ‘i3’. A dash means that a respective example has not been found.

In all other situations /S/, /Š/ occur in ‘i3’. First, they occur there if they are preceded by two peripheral phonemes, one from ‘i1’ and the other from pos ‘i2’ (= ε). In addition, /Š/ occurs in ‘i3’ if it is preceded by one phoneme from pos ‘i2’. The alveolar fricative /S/ cannot be preceded by a phoneme from pos ‘i2’. Although it combines with /P/, /T/, /K/ in the examples like (4a), (4b) and (4c), we assign it to position ‘i2’ in these combinations because they be further followed by /T/ belonging to ‘i3’ as the examples (4a’), (4b’) and (4c’) demonstrate (the first /P/, /T/, /K/ occur in ‘i1’).

(4) (a) /ziPS/ zips (a’) /zāPST/ zábst (b) /moTS/ moc (b’) /moTST/ moct (b’’) /sulTS/ sulc (c) /koKS/ koks (c’) /teKST/ text (c’’) /SfinKS/ Sfinx

156 The whole reasoning is the same as that behind the analysis of /P/, /T/, /K/ in /PŠ/, /TŠ/, /KŠ/ discussed above. However, there is another problem, namely the forms (4b’’) and (4c’’). They show that /TS/ and /KS/ can also be preceded by another phoneme to produce /lTS/ and /nKS/6. However, for reasons of simplicity, it makes more sense to prefer the first interpretation, i.e. to assign /P/, /T/, /K/ in /PS/, /TS/, /KS/ to ‘i1’, and /S/ to ‘i2’ (where they primarily occur), and to say that the occurrence of /S/ in ‘i3’ is conditioned and limited to one particular situation summarizes as follows:

Conditions for the occurrence of /S/, /Š/ in ‘i3’ (α) /S/, /Š/ occur in ‘i3’ if they are preceded by two peripheral phonemes. (β) /S/, /Š/ occur in ‘i3’ if they are preceded by a phoneme belonging to pos ‘i2’.

The remaining three members of pos ‘i2’, /F/, /X/, /ř/, do not occur in any other post- nuclear position. They need not combine with any other peripheral phoneme, but if they do, they can be either preceded by a phoneme from pos ‘i1’, or followed by a phoneme from pos ‘i3’, but never at the same time! Figure IV demonstrates this behavior. In other words, none of /F/, /X/, /ř/ is attested in a three-phoneme post-nuclear combination.

Preceded by Followed by Standing alone a phoneme from ‘i1’ a phoneme from ‘i3’ /F/ /leF/ lev /salF/ salv /praFT/ pravd /X/ /vraX/ vrah /arX/ arch /neXŤ/ nechť /ř/ /tvāř/ tvář /pePř/ pepř /buřT/ buřt

Figure IV: The distribution of /F/, /X/, /ř/ in the post-nuclear context.

Phonotactic properties of pos ‘i2’ are given in Figure V. /P/, /T/, /K/ do not have properties ‘P0n’/‘F0’ and ‘P0N’/‘F0’ if they occur in ‘i2’, but if they occur in ‘i1’, they do have them (see Figure I above). The figure allows us to introduce several collocational restrictions limiting the distribution of the phonemes from pos ‘i2’. One restriction was already mentioned above: in the post-nuclear context a labial fricative, a velar

6 /nKS/ is attested only in /SfinKS/, a phonological form of the brand name Sfinx.

157 fricative7 and /ř/ cannot be preceded by a peripheral phoneme and at the same time followed by a peripheral phoneme (= RI6 in Appendix A). /F/, /X/ and /ř/ are also not found to be followed by a peripheral phoneme when standing right after a nuclear semiconsonant as opposed to /S/ in /prST/ prst. This follows from two restrictions. First, if a semiconsonant is the nucleus of a phonotagm, /ř/ cannot occur in ‘i2’ (= RN10a). This restriction rules situations when /ř/ would immediately follow a nuclear semiconsonant, and when /ř/ would be separated from a nuclear semiconsonant by another peripheral phoneme (i.e. combinations /Př/, /Tř/ and /Kř/ can occur only after a vowel). Second, there is a more general restriction which includes all labials and all velars: a nuclear semiconsonant cannot be followed by a combination of or including the types ‘LC’ or ‘KC’ (= RN10c), i.e. where a labial (= ‘L’) or a velar (= ‘K’) is followed by another peripheral phoneme as opposed to the alveolar /T/ in /StrTŠ/ strč and the palatal /Š/ in /sMrŠŤ/ smršť.

F0 F1i3 /P/ /T/ /K/ P0n

/P/ /T/ /K/ /P/ /K/ P0N /F/ /X/ /ř/ /ř/

P1i1n /F/ /X/ /ř/ /P/ /T/ /P/ /K/ P1i1N /Š/ /F/ /X/ /S/ /Š/ /F/ /X/ /ř/ /ř/

Figure V: Phonotactic properties of pos ‘i2’; phonemes listed are not attested to have a given property. An empty cell means a property is attested for all members of the position class.

7 These are the fricatives which do not occur in any other post-nuclear position, whereas the other two fricatives /S/ and /Š/ are the ones which occur in two post-nuclear positions. Thus, /F/ and /X/ constitute a well-defined class of phonemes.

158 9.3 Position class ‘i3’

Position ‘i3’ is the last post-nuclear position; if a phoneme occurs here, it is the last phoneme in the phonotagm. Its position class pos ‘i3’ is listed under (5); it can also be empty. Phonotactic properties of this position are given in Figure VI.

(5) pos ‘i3’ ∈ {/P/, /T/, /K/, /S/, /Š/, /Ť/, ∅}

The occlusives /P/, /T/, /K/, and the fricatives /S/, /Š/ cannot have properties ‘P0n’,

‘P0N’, ‘P1i1n’ and ‘P1i1N’ if they occur in ‘i3’, but only if they occur in ‘i2’; their occurrence in ‘i3’ is conditioned as explained above. On the other hand, the occlusive /Ť/ occurs only in ‘i3’, which means that it is always the last phoneme in the phonotagm. Its occurrence is, however, limited. It can occur on its own (cf. /xuŤ/ chuť), and it can be preceded by one phoneme either from pos ‘i1’ (cf. /pojŤ/ pojď) or from pos ‘i2’ (cf. /pöŠŤ/ poušť), but it is not found to be preceded by two peripheral phonemes, which is a thing it differs to the other occlusives.

The absence of property ‘P2i1i2N’ for all occlusives (i.e. /P/, /T/, /Ť/, /K/) can be accounted for by the following restriction: a nuclear semiconsonant cannot be followed by a combination of the type ‘CCO’ where ‘O’ is an occlusive (= RN10d) as opposed to e.g. /S/ in /šMrnTS/ šmrnc.

F0 P0n /P/ /T/ /K/ /S/ /Š/ P0N /P/ /T/ /K/ /S/ /Š/

P1i1n /P/ /T/ /K/ /S/ /Š/ P1i1N – 8 P1i2n /P/ /S/

P1i2N /P/ /T/ /K/ /S/ /Ť/

P2i1i2n /Ť/ P2i1i2N /P/ /T/ /K/ /Ť/ Figure VI: Phonotactic properties of pos ‘i3’; phonemes listed are not attested to have a given property. A dash means a property is not attested for any member the position class.

8 /P/ has this property in /hiPŠ/, the form of the surname Hybš.

159 9.4 Position class ‘I’

Position ‘I’ is the only archi-position in the post-nuclear context. The phonemes occurring there are given under (6); it can never be empty. The reason it has been introduced was explained in Section 5.3.

(6) pos ‘I’ ∈ {/m/, /n/, /ň/, /r/, /l/}

As see in Section 9.1, all of these phonemes already belong to ‘i1’; their occurrence in ‘I’ is conditioned. Figure VII illustrates when /m/, /n/, /ň/, /r/, /l/ occur in ‘i1’ and when in ‘I’. They occur in the latter only if they follow a phoneme from ‘i1’ (cf. the δ example). In other cases they occur in ‘i1’, i.e. when they occur on their own (= α), or when they are followed by one (= β), or by two other phonemes (= γ). Note that /ň/ is not found to be followed by two peripheral phonemes. The phonotactic property of position ‘I’, the only one, is given in Figure VIII. All members of pos ‘I’ have it,9 and there is no need for collocational restrictions.

Occurrence in ‘i1’ Occurrence in ‘I’ α /zem/ zem /m/ β /lamP/ lamp δ /jilm/ jilm γ /pomST/ pomst α /sīň/ síň /ň/ β /zuňK/ from šizuňk δ /Tšerň/ čerň γ – α /jen/ jen /n/ β /banK/ bank δ /hejn/ hejn γ /junKT/ from adjunk α /mor/ mor /r/ β /kurS/ kurz δ /xejr/ chejr γ /verST/ verst α /pil/ pil /l/ β /pulS/ pulz δ /Ktejl/ from koktejl γ /sulTS/ sulc Figure VII: Differences in the occurrence of /m/, /n/, /ň/, /r/, /l/ in ‘i1’ and ‘I’. A dash means that a respective example has not been found.

9 If they did not, they would not naturally belong to this position.

160 F0

P1i1 Figure VIII: Phonotactic property of pos ‘I’. It is attested for all member of the class.

9.5 Position class ‘im’

With pos ‘im’ we move to the domain of the semi-distributional unit introduced in Chap- ter 6. It differs from the basic distributional unit in having one and only one post- nuclear position; otherwise, there is no major difference between the units. The phonemes occurring in this position are given under (7). It can be occupied by all voiceless or voiced occlusives, though the voiceless alveolar occlusive /t/ occurs here only very scarcely (probably only in matjes). Moreover, it can be occupied by the voiceless labial fricative /f/, but not by its voiced counterpart /v/. We can see that the phonemes /p/, /b/ and /f/ constitute again a special distributional class (cf. pos ‘E2’ in Chapter 8). The occurrence of the other fricative /x/ is also rather limited (probably only in drachmě). The position can never by empty. If it were, there would be no reason for it.

(7) pos ‘im’ ∈ {/p/, /b/, /f/, /ť/, /ď/, /t/, /d/, /k/, /g/, /x/}

Since ‘im’ is the only post-nuclear position of the semi-distributional unit, it follows that the phonemes belonging there share only one phonotactic property, namely, they can only be preceded by a nuclear phoneme, and followed by no other peripheral phoneme. However, since minor-type phonotagms are by definition always appended to a major-type phonotagm, the phonemes belonging to pos ‘im’ are obligatorily followed by a peripheral phoneme belonging to the adjacent major-type phonotagm; this phoneme is /m/, /n/, /ň/, /v/ or /j/ or it is /M/ provided that the latter is followed by another peripheral phoneme (cf. /šikMňe/ šikmě); details are discussed in Chapter 6.

161 Chapter Ten PROPERTIES OF PERIPHERAL COMBINATIONS

10.1 Peripheral combinations and their length

This chapter examines properties of pre-nuclear and post-nuclear combinations (jointly called peripheral combinations) in Czech. An early attempt to describe word-initial combinations in Czech was made by Hattala (1870). A classification of peripheral combinations was done by Novotná (1972) for Czech or by Tolstaja (1968, 1974) and Sawicka (1974) for all Slavic languages. In addition, Czech peripheral combinations have been included in an analysis of such combinations in 104 languages by Greenberg (1978). Our aim is to supplement and expand these descriptions, but without trying to correct them, though corrected they certainly could be in some cases.1 The most obvious criterion for classification of combinations is one according to the number of phonemes in them; we will call this combination length (after Greenberg op. cit.). Czech has pre-nuclear combinations whose length ranges from 2 to 5 phonemes, and post-nuclear combinations of 2 and 3 phonemes. Figure I summarizes how many combinations of a given length have been included in our analysis. The pre- nuclear combinations exceed almost five times the post-nuclear ones. It is also notewor-

1 For instance, Greenberg’s (op. cit.: 259) interpretation of Czech /j/ as a voiced fricative is highly suspect, and was apparently invoked to save his generalization that “[v]oiced semivowels are not followed by obstruents in initial systems [i.e. in the pre-nuclear context] or preceded by obstruents in final systems [i.e. in the post-nuclear context]”. As Greenberg rightly notes (ibid.), Czech jde and jsou would violate this if /j/ is interpreted as a semivowel. In fact, it has also been disproven by Yateé Zapotec (Jaeger – Van Valin 1982; cf. also Bell – Saka 1983). In this respect it is also another of Greenberg’s generalizations that could be corrected, namely “[t]wo successive voiced sonants are always followed by a vowel in initial systems and preceded by a vowel in final systems” (ibid.). As /j/ should rather be interpreted as a sonant, the generalization is violated by Czech /jMň/ in jmění (nasals are sonants in Greenberg’s terminology).

162 thy that the number of three-phoneme pre-nuclear combinations is a bit higher than the number of pre-nuclear two-phoneme ones. The number of the combinations is rather maximal in the sense that our database includes also combinations attested in words probably no longer or at best seldom used in present-day Czech. Examples are /Txn/ in dchnouti (SSJČ), /Sdm/ in sdmýchnouti (PSJČ), /PStř/ in pstřeň (PSJČ) or /FShř/ in vzhřížiti (PSJČ). They have been included as examples of combinational capacities of Czech, which are vast. The numbers can even be increased by inclusion of combinations attested only in symbols2 and other marginal words like /jř/ in the surname Kejř, /lTŠ/ in the place name Telč or /Kf/ in the place name Kfely and others.3 We have not surveyed these words systematically, and so our database may contain only a portion of them.

Combination length Total Five Four Three Two Pre-nuclear 2 48 185 (192) 183 (189) 417 (431) Post-nuclear – – 19 (28) 65 (73) 84 (101)

Figure I: The number of peripheral combinations in our database sorted according to their length. Values in brackets include marginal combinations.

Another way of looking at phoneme combinations is through their phonematic constituency, as every phoneme belongs to a paradigmatic class which is defined by distinctive features its members share with each other. The phonemes /m/, /n/, /ň/ belong to one paradigmatic class because they all have the distinctive feature ‘nasal’. Likewise, /p/, /b/, /f/, /v/, /m/ belong to one paradigmatic class because they are all labials. In this chapter we make use of two types of classes, and consequently we subject peripheral combinations to two types of classification according to their phonematic constituency. First, they can be analyzed through manners of articulations of their constituents. We distinguish between occlusives (marked as ‘O’), fricatives (= ‘F’), nasals (= ‘N’) and sonants (= ‘R’), which includes /j/, /ř/ and non-nuclear /r/, /l/. Combination types sorted according to various manners of articulation will be called manner types. The second way of looking at the combinations is through the places of articulation of its constituents. We recognize a class of labials (= ‘L’), alveolars (= ‘A’), palatals (= ‘P’), velars (=

2 See Section 3.3 on the difference between signs and symbols. 3 These combinations can be, for the sake of convenience, called marginal combinations.

163 ‘K’)4 and isolated phonemes (= ‘I’). The last group includes sonants (they are isolated because they are phonologically neither labial nor palatal nor alveolar nor velar) and the archiphoneme /M/, which is also not specified as to the place of articulation as that has undergone neutralization. Combination types sorted according to various places of articulation will be called place types. In the following sections we examine phonematic constituency of all peripheral combinations included in our database. Before doing that, let us look at the combinability of each paradigmatic class.

Containing at least one ‘F’ ‘O’ ‘R’ ‘N’ Pre-nuclear 346 (354) 250 (260) 205 (211) 121 (123) Post-nuclear 49 (61) 56 (70) 46 (55) 23 (28) Total 395 (315) 306 (330) 251 (266) 144 (151)

Figure II: The number of peripheral combinations containing at least one occlusive (‘O’), one fricative (‘F’), one nasal (‘N’) or one sonant (‘R’). Values in brackets include marginal combinations. Arranged from the most common to the least common.

Figure II shows how many combinations contain at least one occlusive, one fricative, one nasal or one sonant. The most combinable are fricatives, although in the post- nuclear context there are a little more combinations with at least one occlusive than with one fricative. The least combinable are nasals; it is partly given by the fact Czech has only three of them, but this does not explain why there are over 100 more combinations with sonants, of which Czech has four. Out of the fricatives, the most combinable are alveolars, then labials, palatals, and the least combinable are velars. Out of the occlusives, the most combinable are alveolars, then velars, labials and the least combinable are palatals. Out of the sonants, the most combinable is /l/, then /r/, then /j/, and the least combinable is /ř/, though the difference between the last two is not very significant. Fi- nally, out of the nasals, the most combinable is the palatal nasal, then the nasal archiphoneme /M/, then the alveolar nasal, and the least combinable is the labial nasal, though the differences are not very significant (see also Figure IV).

4 Note that archiphonemes are put in between slashes, while letters standing for a certain class of phonemes are put into single quotation marks. Therefore, /P/ is the archiphoneme ‘labial occlusive’, whereas ‘P’ is a class of palatals.

164 It is worth noting that the pre-nuclear context allows combinations of two, three and four fricatives, though there is only one pre-nuclear combination of four fricatives, namely /FShv/ in vzchvívati (PSJČ). The post-nuclear context allows only combinations of two fricatives, but there is actually only one, /XŠ/ in jejichž. It should be said that pre-nuclear combinations of five fricatives, and post-nuclear combinations of three fricatives are already disallowed by the structure of the distributional unit. On the other hand, the distributional unit allows combinations of up to three occlusives,5 but it is only combinations of two that are attested (e.g. /Tk/ in tkát and /KT/ in akt). To capture this impossibility, we propose the following collocational restriction: three or more occlusives cannot stand in close proximity (= RG7 in Appendix A). Combinations of three nasals are not possible in any context, which derives from the distributional unit and from restriction RG1a, but combinations of two nasals occur in the pre-nuclear context (only two of them, /Mn/ in mnout and /Mň/ in město); the post-nuclear context disfavors even combinations of two nasals. The distributional unit also disallows combinations of three sonants, but those of two are possible, cf. /lj/ in the proper name Ljuba and /jr/ in chejr. Figure III shows how many combinations contain at least one labial, one alveolar, one palatal, one velar or one isolated phoneme. Out of them, the most combinable are alveolars. Then come isolated consonants, and after them are labials, then velars, and the least combinable are palatals. Out of the alveolars, the most combinable are fricatives, then occlusives, and the least combinable is /n/. Out of the labials, the most combinable are fricatives, then occlusives, and the least combinable is /m/. The high combinability of the labial fricatives derives mostly from the combinability of /v/ and /F/; /f/ is one of the least combinable peripheral phonemes in Czech (see Figure IV). Curiously enough, out of the velars, the most combinable are occlusives, and then fricatives (there is no velar nasal). Finally, out of the palatals, it is again fricatives which are most combinable, but then comes /ň/, and the least combinable are occlusives. Alveolars are the most combinable place of articulation. In the pre-nuclear context we find combinations of two, three and even four alveolars, although there is only one

5 For example, in the pre-nuclear context one occlusive can occur in ‘e3’, one in ‘pre1’ and one in ‘pre2’. No such combination is attested.

165 combination of four, namely /TStn/ in ctný. This exhausts possibilities of the distributional unit because combinations of five alveolars are precluded by its structure. In the post-nuclear context there are combinations of two and three alveolars, cf. /nT/ in bažant and /TST/ in péct. Combinations two labials and two palatals are possible in both contexts (cf. /vm/ in vmáčknout, /žň/ in žně, /mP/ in lamp, /ŠŤ/ in poušť), but those of three (or more) labials or palatals are not because the distributional unit does not allow such combinations. Noteworthy is the absence of any combinations of two or more velars in any context (see the next section).

Containing at least one ‘A’ ‘I’ ‘L’ ‘K’ ‘P’ Pre-nuclear 271 (275) 206 (212) 227 (231) 131 (134) 129 (135) Post-nuclear 48 (60) 46 (55) 25 (29) 23 (30) 20 (26) Total 319 (335) 252 (267) 252 (260) 154 (164) 149 (161)

Figure III: The number of peripheral combinations containing at least one labial (‘L’), one alveolar (‘A’), one palatal (‘P’), one velar (‘K’) or one isolated consonant (‘I’). Values in brackets include marginal combinations. Arranged from the most common to the least common.

In conclusion, let us look at Figure IV illustrating combinability of individual peripheral phonemes.6 The most combinable are /S/ and /F/, which could have been expected because they are often forms of the prefixes s-, z- and v-, either synchronically or in origin. On the other hand, the high combinability of /T/ is curious, that being no form of any grammatical element. Also interesting is the combinability of /v/; quantitatively, it is similar to that of sonants, and it is even more interesting because /v/ does not occur in the post-nuclear context. The least combinable is /Ť/; in general, palatal occlusives are the least combinable phoneme class in Czech. Then come /g/ and /f/ whose low combi-

6 The combinability of all Czech phonemes is also examined in Ludvíková – Kraus (1966) and Ludvíková (1968), but the authors have employed a different method and a different phonological theory (they do not operate with archiphonemes). Their distributional domain is a phonological form of a word within which they counted how many phonemes can appear before a given phoneme, and how many can appear after it. Their analysis has shown that /l/, /r/, /b/, /n/, /m/, /t/, /s/, /v/, /k/ are (in descending order) the most combinable peripheral phonemes. – It is also worth noting that according to the analysis by Volín – Churaňová (2010), who analyzed Czech spoken texts, around 55% of consonants (i.e. peripheral phonemes) participate in consonantal combinations, whereas the remaining 45% occurs as singletons.

166 nability could have been expected from historical circumstances—except for a handful of “native” words7, they have been entered into Czech through foreign borrowings.

/S/ /F/ /T/ /l/ /r/ /v/ /j/ /ř/ /Š/ /t/ /k/ Pre- 116 94 54 68 56 62 37 47 34 50 49 nuclear (117) (58) (71) (57) (63) (40) (48) (37) (50) Post- 23 7 33 10 18 16 5 13 – – – nuclear (32) (36) (16) (20) (17) (6) (16) 139 101 87 78 74 62 52 52 47 50 49 Total (149) (94) (87) (77) (63) (57) (54) (52) (50)

/ň/ /M/ /n/ /m/ /z/ /h/ /s/ /K/ /p/ /b/ /d/ Pre- 41 40 28 28 35 34 33 8 32 27 28 nuclear (42) (34) (10) (29) Post- 2 10 11 16 – – – – – – – nuclear (14) (12) (23) 43 40 38 39 35 34 33 24 32 27 28 Total (42) (42) (40) (34) (33) (29)

/x/ /š/ /P/ /ť/ /ž/ /X/ /ď/ /g/ /f/ /Ť/ Pre- 26 26 11 14 15 6 9 8 7 – nuclear (27) (13) (18) (8) Post- 10 7 5 – – – – – – – nuclear (13) (8) 26 26 21 14 15 13 9 8 7 5 Total (27) (26) (18) (8) (8)

Figure IV: The number of peripheral combinations containing at least one particular peripheral phoneme. Values in brackets include marginal combinations. Arranged from the most common to the least common.

10.2 Phonematic constituency of combinations of length 2

In this section we examine the phonematic constituency of combinations of length. Fig- ure V shows place and manner types of the pre-nuclear combinations, and Figure VI does the same for the post-nuclear ones. Shaded cells indicate that the respective combination is attested, and the blank ones that it is not (e.g. the shaded cell at the intersection of the ‘LL’ row and the ‘FN’ column means that there is an attested combination of

7 As argued in Section 13.1, the idea of “foreignness” and “nativeness” of words is highly dubious in a synchronic analysis.

167 two labials where the first is a fricative and the second a nasal, i.e. /vm/). The symbol ‘×’ indicates a combination is not possible in principle (e.g. ‘FR’ cannot be ‘LL’ because sonants are not specified as to the place of articulation). The same symbol is also used for combinations of two identical phonemes because no two identical phonemes can stand in close proximity within a phonotagm. If a shaded cell contains ‘VR’, a given combination occurs before or after nuclear /r/, /l/. If it contains ‘VR’ and is not shaded, the combination is attested only for nuclear /r/, /l/; there are not many such combinations, only pre-nuclear /sM/, /zM/, /šM/ and /vM/ (cf. smrt, zmrznout, šmrnc and vmrštit). Otherwise, the combination is attested only in the vicinity of vowels. The parentheses ‘()’ in a cell indicate that a certain type is attested only in marginal combinations; the respective cells are not shaded this time. The columns and row headed ‘No.’ indicate how many combinations of a given type are attested (parentheses include marginal combinations). Although Figures V and VI schematize which combination types are attested, much more interesting are those which are not. Some absences must be accidental, whereas others can be regularized. We will propose several collocational restrictions, and though postulated in connection with combinations of length 2, they are applicable even to those of length 3 or more, as becomes clear in the next section. Both pre-nuclear and post-nuclear combinations are limited by these restrictions: An alveolar fricative cannot be preceded by a palatal fricative (= RG3), although the reverse is possible in the pre-nuclear context, cf. /SženŠťilī/ zženštilý.8 A palatal occlusive cannot be preceded by an alveolar occlusive (= RG4) as opposed to the labial occlusive /P/ in /PďīT/ bdít. A palatal occlusive cannot also be followed by another peripheral phoneme, but this is already encoded in the distributional unit where the pre-nuclear /ť/, /ď/ stand right next to a vowel, and the post-nuclear /Ť/ occurs in the rightmost post- nuclear position. A velar occlusive cannot be preceded by a labial occlusive (= RG5) as opposed to the labial fricative at the end of /tomahaFK/ tomahavk. Finally, two or more velars cannot stand in close proximity (= RG6), whereas two labials, two alveolars or two palatals can, cf. /Fb/ in vbodnout, /St/ in stát and /Šť/ in štěstí.

8 Greenberg (op. cit.: 257) notes that “there is a strong tendency [in the 104 language he examined, Czech included] for different type sibilants, most commonly [s] and [š], not to combine”. See also Jaeger – Van Valin (1982: 135–6) on this.

168 ‘OO’ ‘OF’ ‘ON’ ‘OR’ ‘FO’ ‘FF’ ‘FN’ ‘FR’ 7 17 12 19 25 18 26 25 No. (9) (18)

‘LL’ 3 × VR × × ‘LA’ 9 × VR VR × ‘LP’ 10 (11) × ×

‘LK’ 5 × × VR × ×

‘LI’ 16 × × × × × VR

‘PL’ 6 × VR ×

‘PA’ 4 × VR × ‘PP’ 4 × × ×

‘PK’ 2 × VR × ×

‘PI’ 5 × × × × × × VR

‘AL’ 12 VR × VR VR ×

‘AA’ 8 × VR × VR × ×

‘AP’ 10 VR × ×

‘AK’ 6 (7) × × VR VR × ×

‘AI’ 16 × × × × × VR ‘KL’ 9 (10) × × ‘KA’ 7 × × ‘KP’ 5 (6) () × × ‘KI’ 11 × × × × ×

‘NF’ ‘NN’ ‘NR’ ‘RO’ ‘RF’ ‘RN’ ‘RR’ 4 2 3 11 10 4 No. (4) (2) ‘IL’ 8 × ‘IA’ 9 × × ‘IP’ 9 × × ‘IK’ 5 × × × × ‘II’ 3 (6) × × × × × ()

Figure V: Place and manner types of pre-nuclear combinations of length 2. See the text for explanations.

If the restrictions introduced so far held both the phonotagm as a whole, there are several others holding either for the pre-nuclear context, or for the post-nuclear one. Figure V shows that place type ‘IL’ is not attested for the pre-nuclear context provided that the ‘I’ is a nasal, i.e. /M/. This means that a labial cannot also be preceded by a nasal in the pre-nuclear context (= RE1) as opposed to alveolars, palatals or velars, cf. /Mze/ mze, /Mše/ mše and /MhöřiT/ mhouřit. There is yet another restriction on labials: In the pre-nuclear context a labial fricative cannot be preceded by another labial (= RE2). If combined with restriction RE15 introduced in Section 8.1 (i.e. /m/ cannot be

169 preceded by a phoneme from pos ‘E2’, i.e. by /p/, /b/, /f/), the consequence is that the only possible pre-nuclear combinations of two labials are /Fp/, /Fb/ and /vm/ (cf. vpadnout, vbodnout and vmísit) as well as combinations like /Fpr/ or /Fbj/ (cf. vpravit and vběhnout). Another restriction concerns fricatives other than labial: they cannot also be preceded by a velar fricative (= RE3) in the pre-nuclear context, i.e. only labial fricatives can be preceded by a velar fricative as in /hvjeSda/ hvězda; it is possible in the post-nuclear context, cf. /XŠ/ in /jejiXŠ/ jejichž. Finally, an alveolar occlusive cannot be preceded by a velar fricative (= RE4) in the pre-nuclear context as opposed to post- nuclear /XT/ in ksicht.

‘OO’ ‘OF’ ‘OR’ ‘FO’ ‘FF’ ‘NO’ ‘NF’ 2 5 3 9 1 5 7 No. (6) (10) (7) ‘LL’ 2 × × × ‘LA’ 5 × ‘LP’ 1 (3) () × () ‘LK’ 2 × ‘LI’ 1 × × × × × ×

‘PA’ 1 × VR

‘PP’ 1 × × VR × ‘PK’ 1 × ‘AL’ 1 (2) × ()

‘AA’ 4 × VR × VR ×

‘AP’ 3 (4) VR × VR ()

‘AK’ 2 × VR VR ‘AI’ 1 × × × × × × ‘KA’ 3 × × ‘KP’ 3 × ‘KI’ 1 × × × × × ×

‘RO’ ‘RF’ ‘RN’ ‘RR’ 13 11 6 3 No. (14) (12) (7) (4) ‘IL’ 9 × ‘IA’ 9 (10) × ‘IP’ 5 (7) × ‘IK’ 7 × × ‘II’ 3 (4) × × ×

Figure VI: Place and manner types of post-nuclear combinations of length 2. See the text for explanations.

170 The distribution in the post-nuclear context is more limited; we propose the following restrictions: An occlusive cannot be followed by an occlusive other than alveolar (= RI2a), that is, only the alveolar occlusive /T/ can be preceded by another occlusive as in /aKT/ akt or /SkriPT/ skript. Moreover, an occlusive cannot be followed by a labial fricative or by a velar fricative (= RI2b) as opposed to pre-nuclear /tv/ and /Tx/ in tvůj and tchoř. A fricative cannot be preceded by a labial fricative or by a palatal fricative (= RI4). An alveolar fricative cannot stand in close proximity with any fricative (= RI7) as opposed to pre-nuclear /Fs/ and /Sh/ in vsadit and shořet. In effect, the only attested post-nuclear combination of two fricatives is /XŠ/ in jejichž. A labial cannot be preceded by another labial unless it is a labial nasal (= RI5a), cf. /mP/ in /lamP/ lamp. Moreover, a labial cannot be preceded by a palatal or by a velar (= RI5b) as opposed to pre-nuclear /Šp/ in špatný and /Xb/ in hbitý. The phoneme /ř/ cannot stand in close proximity with a fricative or with a nasal (= RI8)9 as opposed to pre-nuclear /vř/ in /vřīT/ vřít. Finally, see Section 9.1 and restriction RI1 on the impossibility of occlusive, fricatives and nasals to be followed by a semiconsonant or a nasal.

10.3 Reducibility and resolvability

There is one interesting fact about pre-nuclear combinations of length 2: a nasal cannot be followed by an occlusive provided that the latter stands right before a nuclear phoneme. We conclude that a nuclear phoneme cannot be preceded by a peripheral combination of type ‘NO’ irrespective of whether these phonemes are further preceded by other peripheral phonemes (= restriction RN1). But type ‘NOR’ attested in /Mdl/ (cf. mdlý) is possible. Simply said, /Md/ and similar combinations do not occur. The same can be said about the combinations /PStr/, /PStř/, /PŠTř/: they are attested, but /PSt/ and /PŠt/ are not. This brings us to another classification of peripheral combinations, ac- 10 cording to their reducibility . Its definition is given below. ‘C1’ and ‘C2’ are certain

9 We have not found appropriate to assume neutralization of the manner of articulation here because this type of neutralization would not take place in any other context. 10 Though not using this term, Hattala (1870: 72) also distinguished Czech pre-nuclear combinations as to their reducibility.

171 specific peripheral phonemes, and ‘Cn’ stands for a sequence of two or more peripheral phonemes; it is to ensure the definition holds for combinations of lengths 3, 4 and 5, too. The point is whether the leftmost and rightmost phonemes can be dropped or not. A combination which is not reducible is irreducible. If it is both left-hand and right-hand reducible, it is completely reducible. Likewise, if it is both left-hand and right-hand irreducible, it is completely irreducible. Finally, the combinations ‘CnC2’ and ‘C1Cn’ can be called reductions of ‘C1CnC2’.

Reducibility of peripheral combinations

(α) ‘C1CnC2’ is reducible if it is left-hand reducible and/or right-hand reducible.

(β) ‘C1CnC2’ is left-hand reducible if ‘CnC2’ is attested.

(γ) ‘C1CnC2’ is right-hand reducible if ‘C1Cn’ is attested.

Pre-nuclear Only /třM/ /hřM/ /Křt/ /Xřt/ /Třp/ /Třť/ /Křť/ right-hand reducible /třMň/ /hřMň/ /MStn/ /MStň/ /FSpn/ /FSpň/ /TŠpj/ Only †/Mdl/ /Fřt/ /XTs/ /XTš/ /MSt/ /MSd/ †/lSť/ /MSť/ /MSď/ (†/lTš/ †/MTs/ †/lŠť/) left-hand reducible /PStr/ /PStř/ /PŠtr/ /TStm/ /TStn/ /TStň/ /lStm/ /Shvj/ Irreducible /lStn/ /lStň/

Figure VII: A list of irreducible or partly reducible combinations. A cross means a combination in systematically irreducible. Marginal combinations are parenthesized.

The vast majority of peripheral combinations in our database are reducible. Combi- nations of length 2 are not considered here because they cannot naturally be reduced to any other combination. Actually, there are only two completely irreducible combinations: /lStn/ in lstný and /lStň/ in lstně11. The remaining ones are reducible, but not all of them are both left-hand and right-hand reducible. Figure VII lists such combinations. In case a combination contains an archiphoneme like /M/ in /vMň/, we regard it reducible provided that there is attested at least one combination with a phoneme from which the archiphoneme arose; for /vMň/, it is /vm/ or /vn/ or /vň/. It is noteworthy that only pre- nuclear combinations are partly or wholly irreducible. All post-nuclear combinations are

11 These slightly archaic words are the only ones where the combinations are attested.

172 completely reducible although the left-hand reducibility of /TSP/ (in zácp, gen. pl. of zácpa) is thinkable only if /SP/ (in výsp, gen. pl. of výspa) is accepted (see Section 9.2). Closely related to the idea of reducibility is another concept: the one called resolvability; its definition is given below.12 A combination like /FStr/ is completely resolvable in Czech because /Fs/,13 /St/, /tr/, /FSt/, /Str/ are all attested, whereas /PStr/ is partly resolvable because /PSt/ is not attested.

Resolvability of peripheral combinations (α) A pre-nuclear or post-nuclear sequence of phonemes (i.e. a peripheral combination) of length m is resolvable if it contains at least one attested continuous subsequence of length m – 1.14 (β) It is partly resolvable if at least one but not all such continuous subsequences are attested. (γ) It is completely resolvable if all such continuous subsequences are attested.

The term resolvability was borrowed and the definition adopted from Greenberg (op. cit.: 250), although the original idea should be credited to Hjelmslev (1936: 53) who suggested that all peripheral combinations in all languages are completely resolvable. As we see from the Czech examples, and as was already pointed out by Fischer- Jørgensen (1952: 36) in connection with Russian and Kutenai, this is certainly not true. Yet, at least in Czech and in Greenberg’s databases all peripheral combinations are either partly or completely resolvable. To return to reducibility, it follows that if a combination is either left-hand or right-hand reducible, it is partly resolvable. Combinations of length 3 which are completely reducible are logically at the same time completely resolvable. On the other hand, the complete reducibility of combinations of length 4 or more does not imply their complete resolvability because a combination ‘C1C2C3C4’ is

12 Tolstaja (1974) mentions resolvability in connection with Czech, though without providing concrete examples. 13 We take /Fs/, not /FS/ or /Fz/, to be included in /FStr/ because the base of the combination is the voiceless /t/. Similarly, /Fz/ is included in /FShř/ because the base is here the voiced /h/. 14 This should of course be understood as that the combination is attested in the respective context, i.e. either in the pre-nuclear or post-nuclear context.

173 completely reducible if ‘C2C3C4’ and ‘C1C2C3’ are attested, but the latter combinations need not be themselves reducible to ‘C1C2’, ‘C2C3’ and ‘C3C4’. However, in our database all completely reducible combinations are also completely resolvable.15 Therefore, all combinations listed in Figure VII are those and only those that are partly resolvable, while the remaining one are completely resolvable.16 As we do not claim that our database contains all peripheral combinations of Czech, it may actually be the case that some combinations now declared partly reducible and/or resolvable will eventually prove to be completely reducible and/or resolvable. The onus is to decide whether there are certain structural restrictions which prevent combinations from being completely reducible and/or resolvable. We propose that a combination whose left-hand or right-hand reduction is not attested is at least in principle reducible if the reduction is not precluded by any structural rule. It allows us to distinguish between combinations like /Mdl/, and those like /Shvj/. Reductions of the first are not structurally possible and are thus regularly missing; in the case of /Mdl/, it is because Czech does not allow combinations of type ‘NO’, as mentioned at the beginning of this section. On the other hand, those of the second kind are structurally possible and are missing by accident; in the case of /Shvj/, it is because type ‘FFF’ is attested e.g. in /Sxv/, which is a reduction of /Sxvj/, a voiceless counterpart of /Shvj/. All combinations which are right-hand or left-hand reducible are accidentally irreducible provided that their respective reductions are structurally possible but missing by accidence. Combinations whose right-hand or left-hand reduction is not structurally possible by some regularity are systematically irreducible. After examining the combinations in Figure VII, we conclude that combinations which are only right-hand reducible are accidentally left-hand irreducible, there being no obvious restriction disallowing combinations like /řM/, /řt/ or /Stn/. Likewise, combinations which are only left-hand reducible are accidentally right-hand irreducible except for /Mdl/, /lSť/, and marginal /MTs/, /lŠť/ which we declare systematically right- hand irreducible. /Mdl/ and /MTs/ are so due to the impossibility of type ‘NO’ in the

15 The only combination for which it does not hold is post-nuclear /rŠTŠ/ in boršč. It was one of the reasons we excluded it from our database; see Section 5.5. 16 The number of completely resolvable combinations is much greater than that of partly resolvable ones. Greenberg (op. cit.) noted that this is true for all of the 104 languages he examined.

174 pre-nuclear context. The right-hand irreducibility of /lSť/ and /lŠť/ is also structurally conditioned because their reductions /ls/ and /lš/ (cf. footnote 13 above) violates the following restriction which is otherwise valid: A non-nuclear semiconsonant cannot be followed by a voiceless fricative in the pre-nuclear context (= restriction RE14). Finally, the two completely irreducible combinations /lStn/ and /lStň/ are irreducible by accidence because /lSt/, /Stn/ and /Stň/ do not violate any structural rule we can think of.

10.4 Phonematic constituency of combinations of length 3

The reason we discussed reducibility and resolvability before dealing with the phonematic constituency of combinations of length 3 or more is simple: If a combination

‘C1C2C3’ is completely reducible, the combinations ‘C1C2’ and ‘C2C3’ are attested, i.e. structurally possible. Thus, such a three-phoneme combination does not violate the collocational restrictions we postulated in connection with the two-phoneme combinations in Section 10.1. If a three-phoneme combination is partly reducible but only accidentally so, it does not violate them either. The only systematically irreducible, or to be more precise, the only systemically right-hand irreducible combinations are /Mdl/, /lSť/ and marginal /MTs/, /lŠť/. Consequently, their phonematic constituency is special and does not agree with the constituency of the two-phoneme combinations. If the phonematic constituency of two-phoneme combinations could be described quite simply by means of Figures V and VI, it is difficult to do the same for three- phoneme ones because they correspond to much more types. We have decided for a different approach. Figure VIII gives manner types of pre-nuclear combinations and Figure IX lists manner types of post-nuclear combinations of length 3. The types are sorted according to the number of combinations they underlie. Under every manner type there is a list of place types attested for the respective manner type. For example, manner type ‘OFF’ corresponds to two place types ‘AAL’ and ‘APL’ (namely to the combinations /Tsv/ and /Tšv/). If a type is attested to occur before nuclear /r/, /l/, it is in italics, which means that non-italicized types occur only before vowels.

175 ‘FOR’ ‘FFR’ ‘FFN’ ‘FOF’ ‘FFO’ ‘OFO’ ‘FON’ ‘FNR’ ‘OFN’ ‘FFF’ 42 29 17 14 10 8(11) 7 6 5 5 ‘LLI’ ‘LAI’ ‘LAL’ ‘LAL’ ‘LAL’ ‘LAP’ ‘LAL’ ‘LII’ ‘AAL’ ‘LAL’ ‘LAI’ ‘LPI’ ‘LAA’ ‘LAA’ ‘LAA’ (‘LPP’) ‘AAL’ ‘AII’ ‘APL’ ‘LAP’ ‘LKI’ ‘LKI’ ‘LAP’ ‘LAP’ ‘LAP’ ‘AAL’ ‘AKA’ ‘APP’ ‘LAK’ ‘ALI’ ‘ALI’ ‘LKL’ ‘LKL’ ‘LAK’ ‘AAP’ ‘AKL’ ‘AKA’ ‘AKL’ ‘AAI’ ‘API’ ‘LKP’ ‘AAL’ ‘LPA’ ‘APL’ ‘AKP’ ‘AKP’ ‘AKI’ ‘AKI’ ‘ALA’ ‘AAA’ ‘LPP’ ‘APA’ ‘PLI’ ‘KLI’ ‘ALP’ ‘AAP’ ‘LPK’ (‘APK’) ‘PAI’ ‘AKL’ ‘AKL’ ‘APP’ ‘PKI’ ‘AKA’ ‘PAL’ ‘KPP’ ‘AKP’ ‘PKL’ ‘PKA’ ‘KAA’ ‘PKP’ ‘KAP’

‘OFR’ ‘ORO’ ‘NFO’ ‘FNN’ ‘FOO’ ‘FRO’ ‘RFO’ ‘ROF’ ‘OON’ ‘OFF’ 4 4 4 4 3 3 2 (3) 2 (3) 2 2 ‘ALI’ ‘AIL’ ‘IAA’ ‘LIP’ ‘LAK’ ‘LIA’ ‘IAA’ ‘IAA’ ‘AKA’ ‘AAL’ ‘AAI’ ‘AIP’ ‘IAP’ ‘AIA’ ‘AAK’ ‘KIA’ ‘IAP’ ‘IAP’ ‘AKP’ ‘APL’ ‘API’ ‘KIA’ ‘AIP’ ‘PAK’ ‘KIL’ (‘IPP’) ‘KLI’ ‘KIP’

‘ONN’ ‘RFR’ ‘ORN’ ‘OOR’ ‘NOR’ ‘OOF’ ‘ROR’ ‘FRN’ ‘RFN’ ‘RNN’ 2 2 1 1 1 1 1 1 1 1 ‘AIP’ ‘ILI’ ‘AII’ ‘AKI’ ‘IAI’ ‘AKL’ ‘ILI’ ‘KII’ ‘IAL’ ‘IIP’

(‘ORF’) (‘NOF’) (1) (1) (‘LIL’) (‘IAA’)

Figure VIII: Manner types for pre-nuclear combinations of length 3.17

‘ROF’ ‘OFO’ ‘RFO’ ‘NFO’ ‘NOF’ ‘ROO’ ‘NOO’ 5 (7) 4 (6) 3 (6) 2 (3) 2 (3) 2 1 ‘IAA’ ‘LAA’ ‘IAA’ ‘LPA’ ‘AAP’ ‘ILA’ ‘AKA’ ‘IAP’ (‘LAK’) ‘IAK’ ‘LAA’ ‘AAA’ ‘IKA’ (‘IKA’) ‘AAA’ (‘AAK’) (‘AKA’) ‘AAP’ ‘AAK’ ‘KAA’

Figure IX: Manner types for post-nuclear combinations of length 3.

17 Manner types are given in the first rows. The numbers in the second rows indicate how many combinations are attested for a given type. Parentheses include marginal combinations. Under every manner type there is a list of place types attested for the respective type. Italicized types are attested also for nuclear /r/, /l/; the others are attested only for vowels. Similar notation is used in other figures in this chapter.

176 We can now draw several conclusions about the manner types. There are 32 such types attested for pre-nuclear combinations of length 3 out of which 2 are marginal combinations (i.e. ‘ORF’ and ‘NOF’, cf. /břv/ in Břve and /MTs/ in Mcely). However, these are not all logically possible combinations of the classes ‘O’, ‘F’, ‘N’, ‘R’—there are 64 logically possible manner types for three-phoneme combinations (i.e. 43 = 64), so attested is exactly one half for the pre-nuclear context. For the post-nuclear context, only 7 manner types are attested. The absence of some is already predicted by the structure of the distributional unit,18 and the absence of other types is taken care of by the collocational restrictions introduced in the preceding sections. Finally, the absence of still other combinations can be accounted for by additional restrictions. Again, though postulated in connection with combinations of length 3, they have general validity and hold for combinations of length 4 and 5, too. A nasal is not followed by a sonant and at the same time preceded by an occlusive as opposed to the fricative /s/ in /sMr/ (smrad), that is, a combination of or including type ‘ONR’ is impossible (= RG17a).19 A sonant is not preceded by a nasal and at the same time followed by a peripheral phoneme, so a peripheral combination of or including type ‘NRC’ where ‘C’ is any peripheral phoneme is impossible (= RG17b). If a sonant is preceded by a nasal, it is always followed by a vowel within a single phonotagm. Ac- tually, the only sonant which can occur between two peripheral phonemes is /ř/; it can be preceded by an occlusive and at the same time followed by it (cf. /Třť/ in třtina), or preceded by a occlusive and followed a fricative (only in /břv/ attested in the place name Břve), or be preceded by a fricative and followed by an occlusive (as in /Xřt/ in chřtán). However, /ř/ cannot be both preceded and followed by a fricative, which we can generalize into a restriction that a peripheral combination of or including type ‘FRF’ is impossible (= RG17c). Another impossible type is ‘NFR’ where a fricative is preceded by a nasal and at the same time followed by a sonant (= RG17d). Occlusives have this capacity (cf. /Mdl/ in mdlý), but are limited by another restriction as opposed to fricatives: a peripheral combination of or including type ‘RON’ is not possible

18 The distributional unit predicts that combinations of types like ‘ONO’, ‘FNF’, ‘NNN’, ‘ORR’, ‘FRR’, ‘NRR’, ‘RRO’, ‘RRF’, ‘RRN’ or ‘RRR’ are impossible. 19 This restriction in fact holds only for the pre-nuclear context because post-nuclear combinations of the type ‘ONR’ are precluded by the structure of the distributional unit.

177 (RG17e); compare it against ‘RFN’ attested in /jsm/ (jsme).20 Finally, a single peripheral phoneme cannot be preceded and at the same time followed by a nasal (= RG17f), which is particularly interesting if we consider that there exists type ‘NFON’ for combinations of length 4, though only attested in rather archaic words mstný and mstně.

‘ALI’ ‘AKI’ ‘LAI’ ‘LLI’ ‘LAL’ ‘AAL’ ‘AAI’ ‘LKI’ ‘LAP’ ‘AKL’ 18 12 9 8 8 7(8) 7 7 7 6 ‘OFR’ ‘OOR’ ‘FFR’ ‘FOR’ ‘FFO’ ‘OFO’ ‘OFR’ ‘FOR’ ‘OFO’ ‘OOF’ ‘FFR’ ‘FFR’ ‘FOR’ ‘FFF’ ‘OFF’ ‘FOR’ ‘FFR’ ‘FFO’ ‘FFF’ ‘FOR’ ‘FOR’ ‘FFN’ ‘OFN’ ‘FFF’ ‘FFN’ ‘FOF’ ‘FOF’ ‘FFN’ ‘FOF’ ‘FON’ ‘FON’ ‘FOF’ ‘FON’

‘AII’ ‘AIP’ ‘AKP’ ‘AKA’ ‘IAA’ ‘IAP’ ‘LAA’ ‘PLI’ ‘LAK’ ‘AAP’ 6 5 5 5 4(5) 4(5) 4 4 4 3 ‘ORN’ ‘ORO’ ‘OON’ ‘OON’ ‘ROF’ ‘ROF’ ‘FFN’ ‘FOR’ ‘FOO’ ‘OFO’ ‘FNR’ ‘ONN’ ‘OFN’ ‘OFN’ ‘RFO’ ‘RFO’ ‘FFO’ ‘FFF’ ‘FOF’ ‘FNN’ ‘FFN’ ‘FON’ ‘NFO’ ‘NFO’ ‘FOF’ ‘FFO’ ‘FON’ ‘FFN’ (‘NOF’)

‘APL’ ‘KLI’ ‘ILI’ ‘LPI’ ‘KIA’ ‘LKL’ ‘PKI’ ‘APP’ ‘LPP’ ‘API’ 3 3 3 3 2 2 2 2 1(2) 2 ‘OFF’ ‘OFR’ ‘ROR’ ‘FFR’ ‘ORO’ ‘FFN’ ‘FOR’ ‘OFO’ (‘OFO’) ‘OFR’ ‘OFN’ ‘FFR’ ‘RFR’ ‘FRO’ ‘FOF’ ‘OFN’ ‘FFO’ ‘FFR’ ‘OFO’

‘LPK’ ‘LKP’ ‘IAI’ ‘IAL’ ‘PAK’ ‘PKA’ ‘PKL’ ‘PKP’ ‘KAA’ ‘KAP’ 1 1 1 1 1 1 1 1 1 1 ‘FFO’ ‘FFN’ ‘NOR’ ‘RFN’ ‘FOO’ ‘FFN’ ‘FOF’ ‘FFN’ ‘FOF’ ‘FOF’

‘PAL’ ‘KPP’ ‘KIL’ ‘KII’ ‘KIP’ ‘AIL’ ‘AAA’ ‘AAK’ ‘ALA’ ‘IIP’ 1 1 1 1 1 1 1 1 1 1 ‘FOF’ ‘OFO’ ‘FRO’ ‘FRN ‘ORO’ ‘ORO’ ‘FOF’ ‘FOO’ ‘FFN’ ‘RNN’

‘ALP’ ‘AIA’ ‘APA’ ‘APK’ ‘LIA’ ‘LII’ ‘LIP’ ‘LPA’ ‘PAI’ (‘LIL’) 1 1 1 1 1 1 1 1 1 (1) ‘FFN’ ‘FNN’ ‘OFO’ ‘OFO’ ‘FRO’ ‘FNR’ ‘FNN’ ‘FFO’ ‘FOR’ (‘ORF’)

(‘IPP’) (1) (‘RFO’) Figure X: Place types for pre-nuclear combinations of length 3.

20 Type ‘RFN’ is further limited by restriction RG18i introduced in Section 11.4 sub DE3b.

178 In addition to these restrictions, there are a handful of others applicable either to the pre-nuclear context or to the post-nuclear context only: In the pre-nuclear context a sonant and a nasal cannot be followed by two occlusives or two fricatives (= RE6), but a sonant can be followed by two nasals, cf. /jMň/ in jmění and /hřMň/ in hřmět. In the post-nuclear context they can be followed by two occlusives, cf. /rKT/ in infarkt and /nKT/ in adjunkt. However, in the post-nuclear context two occlusives or two fricatives cannot be followed by any other peripheral phoneme (= RI3), whereas they can in the pre-nuclear context, cf. /Tkl/ in tklivý, /Tkn/ in tknout, /Fhl/ in vhled, /Sxň/ in shnít. Although Figures VIII and IX also list place types for combinations of length 3, it may be useful to look at them directly from the perspective of place types. Figures X and XI list, respectively, all attested place types for pre-nuclear and post-nuclear combinations. Under every place type there is a list of manner types it corresponds to. There are 125 (= 53) logically possible three-phoneme combinations of ‘L’, ‘P’, ‘A’, ‘K’, ‘I’, but only 61 of them are actually attested for pre-nuclear combinations and only 13 for post-nuclear combinations. As the combinability of phonemes according to their place type is quite diverse, we refrain from proposing any collocational restrictions pertain to place types. Note also that almost one half of the attested types are attested only once.

‘IAA’ ‘IAP’ ‘AAA’ ‘AAP’ ‘LAA’ ‘IAK’ ‘IKA’ ‘AKA’ ‘ILA’ ‘LPA’ 5 (6) 2 (3) 2 2 2 1 (3) 1 (2) 1 (2) 1 1 ‘RFO’ ‘ROF’ ‘OFO’ ‘OFO’ ‘OFO’ ‘RFO’ ‘ROO’ ‘NOO’ ‘ROO’ ‘NFO ‘ROF’ ‘NOF’ ‘NOF’ ‘NFO’ (‘ROF’) (‘NOF)

‘KAA’ (‘AAK’) (‘LAK’) 1 (2) (1) ‘OFO (‘OFO’) (‘OFO) (‘NFO’)

Figure XI: Place types for post-nuclear combinations of length 3.

10.5 Phonematic constituency of combinations of length 4 and 5

We now get to manner and place types of four-phoneme and five-phoneme pre-nuclear combinations; there are no combinations of these lengths in the post-nuclear. It follows

179 from the discussion in Section 10.3 that all of these combinations are at least partly resolvable, and all of them are at least partly reducible with exception of /lStn/, /lStň/. The irreducibility of combinations of length 4 has been recognized as accidental, and so the combinations of lengths 4 and 5 are subject to the same restrictions as those of length 3. Figure XII summarizes manner types for combinations of length 4. In total, there are 256 logically possible types (i.e. 44), but only 20 are actually attested. The absence of many is structurally conditioned which derives from two major facts. First, the distributional unit precludes some of them; this holds for types like ‘ONOF’, ‘FNOF’, ‘FOFO’ or ‘FFOO’. Second, others are not possible due to the restrictions we introduced in the previous sections in connection with combinations of length 2 and 3. And the absence of yet other types can be taken care of by the following additional restrictions. A peripheral combination of or including types ‘CFFN’ and ‘CFNR’ and where ‘C’ is any peripheral phoneme is not possible (= RG18a and RG18b). In other words, the attested combinations of types ‘FFN’ and ‘FNR’ (cf. /Sxn/ in schnout and /zMr/ in zmrazit) cannot be expanded from the left.21 Impossible are also peripheral combinations of or including type ‘OFFC’ (= RG18c), which in this case means that combinations of type ‘OFF’ (cf. /Tsv/ in cválat) cannot be expanded from the right. Further- more, a peripheral combination of or including types ‘COFN’22 or ‘OFNC’ are impossible (= RG18d); peripheral combinations of type ‘OFN’ (cf. /Tšm/ in čmárat) cannot then be expanded either from the right or from the left. A combination of or including type ‘C1FFF’ where ‘C1’ is an occlusive, a nasal or a sonant is impossible as well (= RG18e); thus, only ‘FFFF’ is possible (cf. /FSxv/ in vzchvívati se (PSJČ)). Other restrictions are these: Two occlusives cannot be followed by two peripheral phonemes of the same manner of articulation (= RG19) as compared to two fricatives which can, though apparently only in rare words, cf. the already mentioned /FSxv/ and /FsMň/ in vzměť (PSJČ). In effect, it means that a peripheral combination of or including types ‘OOFF’ and ‘OONN’ are impossible.23

21 See the next section on expandability. 22 But type ‘CFON’ is possible, cf. /TStnī/ ctný, /FSpnöT/ vzpnout or /MStnī/ mstný. 23 Types ‘OOOO’, ‘OORR’, and also ‘FFOO’, ‘FFRR’ are already disallowed by the distributional unit.

180 ‘FFOR’ ‘FFFR’ ‘OFOR’ ‘RFON’ ‘FFON’ ‘OFON’ ‘FFNN’ 11 6 4 3 3 3 2 ‘LALI’ ‘AKLI’ ‘APLI’ ‘IAAA’ ‘LAAL’ ‘AAAL’ ‘LAIP’ ‘LAAI’ ‘LALI’ ‘LAAI’ ‘IAAL’ ‘LALA’ ‘AAAA’ ‘LAKI’ ‘LAKI’ ‘LPAI’ ‘IAAP’ ‘LALP’ ‘AAAP’ ‘LPKI’

‘FOFR’ ‘FOON’ ‘NFON’ ‘OOFR’ ‘ORNN’ ‘OFOF’ ‘FOOR’ 2 2 2 1 1 1 1 ‘LAPI’ ‘PAKA’ ‘IAAA’ ‘AKLI’ ‘AIIP’ ‘APAL’ ‘AAKI’ ‘AKLI’ ‘PAKP’ ‘IAAP’

‘FRNN’ ‘FONN’ ‘FFOF’ ‘FOOF’ ‘FOFF’ ‘FFFF’ 1 1 1 1 1 1 ‘KIIP’ ‘AAIP’ ‘LAKL’ ‘AAKL’ ‘AAAL’ ‘LAKL’

Figure XII: Manner types for pre-nuclear combinations of length 4.

Let us briefly look at the place types attested for four-phoneme pre-nuclear combinations. Figure XIII lists those which are attested. Out of 625 logically thinkable combinations (i.e. 54), only 27 are attested.

‘LALI’ ‘LAKI’ ‘AKLI’ ‘LAAI’ ‘AAAL’ ‘LAKL’ ‘LAIP’ ‘IAAA’ 7 5 4 4 2 2 2 2 ‘FFFR’ ‘FFOR’ ‘OOFR’ ‘OFOR’ ‘FOFF’ ‘FFFF’ ‘FFNN’ ‘RFON’ ‘FFOR’ ‘FFFR’ ‘FOFR’ ‘FFOR’ ‘OFON’ ‘FFOF’ ‘NFON’ ‘FFFR’

‘IAAP’ ‘APAL’ ‘APLI’ ‘IAAL’ ‘AAIP’ ‘PAKA’ ‘LAAL’ ‘LALA’ 2 1 1 1 1 1 1 1 ‘RFON’ ‘OFOF’ ‘OFOR’ ‘RFON’ ‘FONN’ ‘FOON’ ‘FFON’ ‘FFON’ ‘NFON’

‘LALP’ ‘AAAA’ ‘AAAP’ ‘AAKI’ ‘AAKL’ ‘AIIP’ ‘LAPI’ ‘LPAI’ 1 1 1 1 1 1 1 1 ‘FFON’ ‘OFON’ ‘OFON’ ‘FOOR’ ‘FOOF’ ‘ORNN’ ‘FOFR’ ‘OFOR’

‘LPKI’ ‘PAKP’ ‘KIIP’ 1 1 1 ‘FFOR’ ‘FOON’ ‘FRNN’

Figure XIII: Place types for pre-nuclear combinations of length 4.

181 Finally, Figure XIV gives manner and also place types for the two five-phoneme pre- nuclear combinations. The phonematic constituency of these combinations is largely dependent on the structure of the distributional unit and on restrictions we have postulated in this chapter.

‘FFOFR’ ‘FOOFR’ 1 1 ‘LAKLI’ ‘AAKLI’

Figure XIV: Manner types for pre-nuclear combinations of length 5.

10.6 Further criteria: expandability, pairedness and mirror effect

In this section we mention three other possible classificatory criteria of peripheral combinations: expandability, pairedness and mirror effect. The evaluation of the combinations according to these properties can be found in Appendix B. Though we are not aware of any linguist sorting combinations according to it, expandability is in fact a logical reversal of reducibility; its definition runs as follows:

Expandability of peripheral combinations

(α) ‘Cn’ is expandable if it is left-hand expandable and/or right-hand expandable.

(β) ‘Cn’ is left-hand expandable if ‘C1Cn’ is attested. 24 (γ) ‘Cn’ is right-hand expandable if ‘CnC2’ is attested.

The classification of peripheral combinations is more intricate when it comes to their expandability than it was in the case of reducibility. Some are not expandable because their expandability is ruled out by the structure of the distributional. These ones can be called anti-expandable. First of all, combinations of length 5 are anti-expandable simply because no longer combinations are possible in Czech. Second, all pre-nuclear combinations ending in /r/, /l/, /j/, /ř/ are right-hand anti-expandable because these phonemes

24 ‘Cn’ is any peripheral combination. ‘C1Cn’ and ‘CnC2’ can then be called expansions of ‘Cn’.

182 occupy the rightmost pre-nuclear position, i.e. ‘e1’. This position can also be occupied by /m/, /n/, /ň/, but combinations ending in them can be expandable because the nasals can be represented by the archiphoneme /M/ in this case. So /tm/, /tň/, /tn/ are all right- hand expandable because /tMň/ is attested. On the other hand, combinations ending in /Mn/ or /Mň/ are right-hand anti-expandable because no other phoneme can be attached to them from the right. Similarly, all pre-nuclear combinations beginning with /P/, /K/, /F/, /X/ are left-hand anti-expandable because these phonemes occupy the leftmost pre- nuclear position, i.e. ‘pre2’.25 The same holds for /ť/, /ď/ which occupy archi-position ‘E3’. Similar reasoning can be applied to the post-nuclear context. Combinations whose expandability is not a priori structurally precluded, but whose expansions are not attested in our database can be called non-expandable, eight left- hand or right-hand or both. If a combination is non-expandable, it does not necessarily mean its potential expansion is not structurally possible; we should look out for further criteria to see whether the non-attestation is structurally determined. For instance, the non-expandability of post-nuclear /Př/, /Tř/, /Kř/ as well as that of /řT/, /řK/ is regular because /ř/ cannot be combined with more than one peripheral phoneme in the post- nuclear context (see Section 9.2). But the non-expandability of pre-nuclear /fj/ is merely accidental because other combinations of a labial with /j/ are expandable, that is to say, there is no reason a combination like /Sfj/ could not occur (cf. /Spj/, /Sbj/, /svj/, /zvj/).

Expandability Attested combinations Symbols Left-hand Right-hand Pre-nuclear Post-nuclear -- anti-exp. anti-exp. 85 (91) 30 (43) -X, -R, -R! anti-exp. 109 (115) 33 (35) X-, L-, L!- anti-exp. 126 (127) 15 (16) XX non-exp. non-exp. 34 1 (2) X-, XR non-exp. 63 (64) 12 -X, LX non-exp. 84 (90) 19 (23) LR exp. exp. 33 2 L-, LX, L!R, L!- exp. 93 6 (7) -R, XR, L!R, -R! exp. 56 15

Figure XV: Expandability of peripheral combinations. The symbols are explained in the text and in Appendix B. The numbers indicate how many combinations are attested; the values in parentheses include marginal combinations.

25 We ignore here what we called accidental appendices in Section 8.8.

183 Figure XV shows how many combinations are anti-expandable, non-expandable and expandable in our database. The numbers are quite diverse, but that there is a prevalence of anti-expandable combinations over non-expandable and over expandable combinations. Due this diversity we will not attempt to draw any conclusions about the expandability of peripheral combinations, though we admit it could be in principle possible. We content ourselves with speficifying whether a combination is expandable or not. This information is provided in Appendix B where the codes given in Figure XV are used; their usage is explained there as well. Code ‘-R!’ is used for pre-nuclear /Ps/ and /Pš/. They are right-hand expandable, but only if we add two peripheral phonemes to the right to get /PStr/ and /PŠtr/; combinations like /PŠt/ or /PŠt/ are not attested. Similarly, code ‘L!-’ is used for /pn/ and /pň/, and code ‘L!R’ for /tn/ and /tň/; none of them have an immediate left-hand expansion, but they can be expanded by addition of two peripheral phonemes to produce /FSpn/, /FSpň/, /TStn/ and /TStň/. The second criterion to be introduced in this section is pairedness; it concerns voiceless and voiced consonants occurring in a peripheral combination. There are four types of combinations according to the voicing of phonemes: voiceless combinations, voiced combinations, mixed combinations and indifferent combinations. We should remember that the voicing is here a phonological property, hence only occlusives and fricatives can be phonologically voiceless or voiced. The sonants /j/, /ř/, /r/, /l/, the nasals /m/, /n/, /ň/, /M/, and the voicing archiphonemes like /P/, /T/, /X/ are indifferent to this distinction, i.e. they are neither voiceless nor voiced. Voiceless combinations are those which contain one and only one voiceless consonant, e.g. /tr/, /St/, /PStr/. Voiced combinations are those which contain one and only one voiced consonant, e.g. /dr/, /FSb/, /rd/. Mixed combinations are those which contain both a voiceless and voiced consonant. The reason such combinations are occurring in Czech lies in the fact that there is no neutralization of voicing before /v/, hence /kv/, /gv/, /Stv/, /Sdv/ and others are all mixed combinations. Finally, indifferent combinations are those which contain neither a voiceless nor voiced consonant. Examples are /rm/, /lň/, /jMň/ and, due to neutralization of voicing, also all post-nuclear combinations. There are 208 (214 if marginal combinations are counted) voiceless combinations, 151 (154) voiced combinations, 49 (50) mixed combinations, and 10 (13) indifferent pre-

184 nuclear combinations. It is obvious there is a great prevalence of voiceless combinations, i.e. those containing one phonological voiceless consonant. Now, only voiceless or voiced combinations can be paired. Such a combination is paired if its corresponding voicing counterpart is also attested; /tr/ is paired because /dr/ is also attested. A combination is single if its corresponding voicing counterpart is not attested; /Ftř/ is single because /Fdř/ is not attested. The same is true for /Fhm/, /Mdl/ or /lbj/. An interesting question is whether combinations are which single are single because their corresponding voicing counterparts are not structurally possible or whether they are just accidentally missing. There are combinations of both kinds. For example, /pn/, /pň/ must be regularly single because their voicing counterparts /bn/, /bň/ can safely be regarded as structurally impossible (according to collocational restriction RG8). On the other hand, combinations like /Fhm/, /Ftř/ are single by accidence because there is no reason to think /Fxm/ and /Fdř/ cannot be valid combinations in Czech. Be it as it may, in Appendix B the paired combinations are marked with ‘P’ in the ‘P/S’ column, the single ones as ‘S’, and those which are neither are marked with ‘-’. There are moreover two combinations which are marginally paired because their counterparts are found only in marginal combinations. These are /Tsp/ (cf. /TSb/ in Dzbel) and /Pď/ (cf. /Pť/ in Ptice). Overall, there are 167 single pre-nuclear combinations (175 with marginal combinations), 239 (241) paired pre-nuclear combinations, and 10 (13) comhinations which are neither. Post-nuclear combinations are neither voiceless nor voiced. The very last criterion is the mirror effect; the idea was suggested by Sigurd (1965:

106). A pre-nuclear combination ‘C2C1’ has a mirror counterpart if a post-nuclear combination ‘C1C2’ is attested. Similarly, a post-nuclear combination ‘C1C2’ has a mirror counterpart if a pre-nuclear combination ‘C2C1’ is attested. Due to neutralization of voicing in the post-nuclear context, we ignore whether a combination is voiceless or voiced. Hence, post-nuclear /rT/ has a mirror counterpart because /dr/ is attested26. The idea holds also for combinations of length 3 or more, but pre-nuclear combinations of lengths 4 and 5 cannot have a mirror counterpart in Czech because the language allows only post-nuclear combinations of length 3. There are 95 (or 99 with marginal combina-

26 Attested is also /tr/, but the point is that at least one of such combinations is sufficient for de- claring that a post-nuclear combination has a mirror counterpart.

185 tions) pre-nuclear combinations and 57 (64) post-nuclear combinations which have a mirror counterpart.27 In addition, there are two pre-nuclear combinations whose mirror counterparts are marginal combinations, namely /Šp/ and /Šb/ (cf. /PŠ/ attested in the surname Hybš). Similarly, there are post-nuclear combinations whose mirror counterparts are marginal combinations, namely /jm/ (cf. /Mj/ in the surname Mjachký), /jl/ (cf. /lj/ in the proper names Ljuba) and /jr/ (cf. /rj/ in the place name Rjazaň). The remaining combinations do not have a mirror counterpart.

27 The numbers are not same because, as explained in the previous footnote, /tr/ and /dr/ are both mirror counterparts of /rT/.

186 Chapter Eleven PRE-NUCLEAR COMBINATIONS

11.1 Distributional types DE0 and DE1

To account of pre-nuclear combinations, we will distinguish between six major pre- nuclear distributional types: DE0, DE1, DE2, DE3, DE4 and DE5, with the number indicating how many pre-nuclear positions of the distributional unit are occupied by a phoneme. In addition, we will recognize for every major type, except for DE0 and DE5, several subtypes. Although types DE0 and DE1 do not underlie any combinations, they are included here for the completeness. Each and every attested pre-nuclear combination belongs to one of the remaining types or to their subtypes. If the occurrence of a certain combination is attested only in symbols1 and other marginal forms, it is placed in between round brackets in the tables to follow. If a combination is attested to occur also before nuclear /r/, /l/, it is italicized. If the possibility of a certain combination is precluded by a collocational restriction (all restrictions being listed in Appendix A), the respective table cells are marked by ‘–’. If a certain combination is missing by accidence, not by any rule, the cells are left empty. Under every distributional type we list phonemes which are not attested to occur in a given position in a particular distributional type (we use ‘~’). When a type is valid for the whole position class, ‘∀’ is used. Now, in distributional type DE0 none of the pre-nuclear positions is filled with a phoneme; all are empty. Phonotagms of this type are frequent in Czech (cf. /on/ on), which means that it allows phonotagms beginning with a vowel. In distributional type DE1 it is just one pre-nuclear position that is filled with a phoneme. Particular possibilities are given in Figure I. Note that positions ‘pre1’ and ‘pre2’ cannot be filled with a phoneme if the other pre-nuclear positions are all empty.

1 See Section 3.3 on the difference between signs and symbols.

187 Distributional type DE1 a ∅ ∅ ∅ ∅ e1 b ∅ ∅ ∅ e2 ∅ c ∅ ∅ e3 ∅ ∅ d ∅ ∅ E2 ∅ e ∅ ∅ E3

Figure I: Subtypes for distributional type DE1, i.e. when just one pre-nuclear position is occupied by a phoneme.

DE1a: ‘e1’ = ∀, i.e. {/m/, /n/, /ň/, /j/, /ř/, /r/, /l/}. DE1b: ‘e2’ = ∀, i.e. {/M/, /v/}. /M/ occurs only before a nuclear semiconsonant. DE1c: ‘e3’ = ∀, i.e. {/t/, /d/, /s/, /z/, /š/, /ž/, /k/, /g/, /x/, /h/}. DE1d: ‘E2’ = ∀, i.e. {/p/, /b/, /f/}. DE1e: ‘E3’ = ∀, i.e. {/ť/, /ď/}.

11.2 Distributional type DE2

This type accounts for situations when exactly two pre-nuclear positions are occupied by phonemes. Its subtypes are given in Figure II.

Distributional type DE2 a ∅ ∅ ∅ e2 e1 b ∅ ∅ e3 ∅ e1 c ∅ ∅ E2 e1 d ∅ pre1 ∅ ∅ e1 e pre2 ∅ ∅ ∅ e1 f ∅ ∅ e3 e2 ∅ g ∅ pre1 ∅ e2 ∅ h pre2 ∅ ∅ e2 ∅ i ∅ pre1 e3 ∅ ∅ j pre2 ∅ e3 ∅ ∅ k ∅ pre1 E2 ∅ l pre2 ∅ E2 ∅ m ∅ pre1 E3 n pre2 ∅ E3

Figure II: Subtypes for distributional type DE2, i.e. when just two pre-nuclear positions are occupied by phonemes.

188 DE2a: ‘e1’ = ∀, ‘e2’ = ∀; Figure III. All combinations are attested except for /Mm/ which is impossible due to restriction RE1 (see Appendix A). /Mj/ is attested only in the surname Mjachký.

/m/ /n/ /ň/ /j/ /ř/ /r/ /l/ /v/ /vm/ /vn/ /vň/ /vj/ /vř/ /vr/ /vl/ /M/ – /Mn/ /Mň/ (/Mj/) /Mř/ /Mr/ /Ml/

Figure III: Two-phoneme pre-nuclear combinations attested for type DE2a (see the beginning of this chapter for explanations; it holds for the similar tables in the chapter).

DE2b: ‘e1’ = ∀, ‘e3’ = ∀; Figure IV. Although both ‘e1’ and ‘e3’ may be filled with all phonemes belonging to those positions, not all conceivable combinations are attested and possible. The most limited are those with /j/, which derives from restrictions RG12a–b; simply said, in the pre-nuclear context /j/ can only be preceded by /p/, /b/, /f/, /v/, /s/, /z/. The combinations /šř/ and /žř/ are impossible to the following restriction (discussed in Bičan 2011c): /ř/ cannot occur in the same section2 of the distributional unit with a palatal fricative. This restriction rules out also combinations like /Štř/, /PŠtř/. On other hand, the absence of /gm/, /gň/, /gř/ is accidental from the synchronic perspective, though explainable diachronically. /xn/ is attested only in /naTxnöT/, a phonological form of nadchnout, which is parsed into /naT–xnöT/ (see Section 3.6 on “syllabification”).

/m/ /n/ /ň/ /j/ /ř/ /r/ /l/ /t/ /tm/ /tn/ /tň/ – /tř/ /tr/ /tl/ /d/ /dm/ /dn/ /dň/ – /dř/ /dr/ /dl/ /s/ /sm/ /sn/ /sň/ /sj/ /sř/ /sr/ /sl/ /z/ /zm/ /zn/ /zň/ /zj/ /zř/ /zr/ /zl/ /š/ /šm/ /šn/ /šň/ – – /šr/ /šl/ /ž/ /žm/ /žn/ /žň/ – – /žr/ /žl/ /k/ /km/ /kn/ /kň/ – /kř/ /kr/ /kl/ /g/ /gn/ – /gr/ /gl/ /x/ /xm/ /xn/ /xň/ – /xř/ /xr/ /xl/ /h/ /hm/ /hn/ /hň/ – /hř/ /hr/ /hl/

Figure IV: Two-phoneme pre-nuclear combinations attested for type DE2b.

2 The same section refers here either to the pre-nuclear or to the post-nuclear context; cf. also restrictions RG15, RG16 introduced in Section 8.1.

189 DE2c: ‘e1’ = ~{/m/}, ‘E2’ = ∀; Figure V. /m/ cannot occur in ‘e1’ due to restrictions RE1 and RE2. The absence of /bn/ and /bň/ can be regularized by a collocational restriction saying that a nasal cannot be preceded by /b/ (= RG8 in Appendix A).3 The absence of /fn/ and /fř/ is accidental from the synchronic perspective. Cf. the situation with /g/ above.

/n/ /ň/ /j/ /ř/ /r/ /l/ /p/ /pn/ /pň/ /pj/ /př/ /pr/ /pl/ /b/ – – /bj/ /bř/ /br/ /bl/ /f/ /fň/ /fj/ /fr/ /fl/

Figure V: Two-phoneme pre-nuclear combinations attested for type DE2c.

DE2d: an impossible type. This sub-type does not underlie any combination. It is due to the impossibility of a phoneme from ‘e1’ to be preceded by a phoneme from ‘pre1’; see Section 8.1.

DE2e: ‘e1’ = ~{(/j/), /ř/, /r/, /l/}, ‘pre2’ = ~{/P/, /F/, /T/, /S/, /Š/, /K/, /X/, /M/}; Figure VI. As voicing archiphonemes cannot precede phonemes from pos ‘e1’, the only phonemes attested for ‘pre2’ are /r/, /l/ and /j/; the phonotagms in which they occur are so- called side-syllables (see Section 8.7). /M/ cannot occur in ‘pre2’ provided that ‘e2’ is empty. Position ‘e1’ cannot be occupied by /ř/, /r/, /l/ due to restrictions RG13a, RG14 and RG15. The combinations /rj/ and /lj/ are attested only in the place name Rjazaň (Russian city) and the proper name Ljuba. The others are accidentally missing.

/m/ /n/ /ň/ /j/ /j/ /jm/ – /r/ /rm/ (/rj/) /l/ /ln/ /lň/ (/lj/)

Figure VI: Two-phoneme pre-nuclear combinations attested for type DE2e .

3 This is a specific of Modern Standard Czech because in Old Czech the combinations /bň/ and /bn/ were possible (Hattala 1870: 52).

190 DE2f: ‘e2’ = ∀, ‘e3’ = ∀; Figure VII. /M/ occurs here only before a nuclear semiconsonant (cf. /sMrT/ smrt), hence the combinations /sM/, /zM/, /šM/ are attested only before nuclear /r/ or /l/, although similar combinations exist for vowels, e.g. /sm/, /zn/, /šň/ (see Figure IV).

/v/ /M/ /t/ /tv/ /d/ /dv/ /s/ /sv/ /sM/ /z/ /zv/ /zM/ /š/ /šv/ /šM/ /ž/ /žv/ /k/ /kv/ /g/ /gv/ /x/ /xv/ /h/ /hv/

Figure VII: Two-phoneme pre-nuclear combinations attested for type DE2f.

DE2g: ‘e2’ = ~{/M/}, ‘pre1’ = ~{/T/, /S/, /Š/}; Figure VIII. As neutralization of voicing does not take place before /v/ or /M/, it is only /ř/ that can occupy ‘pre1’ here. /v/ is the only fricative before which /ř/ can occur, hence the following restriction holds: /ř/ cannot be followed by a fricative other than labial, that is, it can only be followed by a fricative occurring in ‘e2’ (= RE10b).4

/v/ /ř/ /řv/

Figure VIII: Two-phoneme pre-nuclear combinations attested for type DE2g.

DE2h: ‘e2’ = ∀, for ‘pre2’ see below; Figure IX. As explained above, ‘pre2’ cannot be occupied by a voicing archiphoneme. There- fore, only /M/, /j/, /r/, /l/ are expected to occupy ‘pre2’, though only /r/, /l/ are actually attested. The absence of /M/ derives from restrictions RE1 and RG1a. The combination

4 Admittedly, this fact could also be expressed by saying that the opposition between fricatives of various places of articulation is neutralized after /ř/ just as it is for nasals (cf. /hřMī/ hřmí). However, we prefer to say that /ř/ can be followed by a phoneme from pos ‘e2’ (i.e. /v/ and /M/) but never by a phoneme from pos ‘e1’ (i.e. /m/, /n/, /ň/, /j/, /ř/, /r/, /l/).

191 /vM/ is attested only in the word vmrštit, i.e. before a nuclear semiconsonant (for the occurrence of /v/ in ‘pre2’ see Section 8.7). The combinations /rM/ and /lM/, which could occur before a nuclear semiconsonant, are not possible due to restriction RG15.

/v/ /M/ /r/ /rv/ – /l/ /lv/ – /v/ – /vM/

Figure IX: Two-phoneme pre-nuclear combinations attested for type DE2h.

DE2i: ‘e3’ = ∀, ‘pre1’ = ∀; Figure X. Although the two positions may be filled with all phonemes belonging there, not all conceivable combinations are attested due to restrictions RG1b, RG3 and RE10b. Let us say at this occasion a few words about /T/. It occurs in ‘pre1’ only if ‘e3’ is occupied by alveolar or palatal fricatives or by the velars /k/ and /x/. It may also precede /b/, which occupies ‘E2’ (i.e. /Tb/ in dbát; it belongs under type DE2m). If /T/ is combined with the alveolar or palatal fricatives, we get, at the level of realization, the so- called affricates [ʦ], [ʧ], [ʣ] and [ʤ]. Phonological words with /Ts/ [ʦ] and /Tš/ [ʧ] are abundant; those with /Tž/ are also attested, though they are not numerous (e.g. /Tž/ in džus). The examples of /Tz/ are very rare; it is attested perhaps only in dzinkati (SSJČ), yet structurally possible. As we see below under subtype DE3j, /Ts/ and /Tš/, but not /Tz/ and /Tž/, can be preceded by /X/ or /S/. When /T/ precedes /k/, /x/, /b/, we get some unique combinations: /Tx/ occurs probably only in phonological forms of three Czech words: /Txoř/ tchoř “polecat”, /Txān/ tchán “father-in-law” and /Txīňe/ tchýně “mother- in-law” (the latter two are historically related). Neither of these combinations is expand- able5 by a phoneme from ‘pre2’ nor from ‘e2’ or ‘e1’. On the other hand, /Tk/ is expandable by a phoneme from ‘e2’ or ‘e1’ (cf. /Tkv/, /Tkvj/, /Tkn/, /Tkň/, /Tkl/) as well as by a phoneme from ‘pre2’ (namely /FTk/, /STk/, /ŠTk/, /ŠTkn/, /ŠTkň/, /STkv/, /STkvj/, /STkl/). The occurrence of these combinations is limited, though; they are attested probably only in phonological forms of several words and its derivatives (e.g.

5 See Section 10.6 on expandability.

192 those derived and related to tkát). The last combination /Tb/ probably occurs only in phonological form of one word only, in dbát (and its derivates, naturally).

/t/ /d/ /s/ /z/ /š/ /ž/ /k/ /g/ /x/ /h/ /T/ – – /Ts/ /Tz/ /Tš/ /Tž/ /Tk/ /Tx/ /Th/ /S/ /St/ /Sd/ – – /Sš/ /Sž/ /Sk/ /Sg/ /Sx/ /Sh/ /Š/ /Št/ /Šd/ – – – – /Šk/ /Šh/ /ř/ – – – – /řk/ – –

Figure X: Two-phoneme pre-nuclear combinations attested for type DE2i.

DE2j: ‘e3’ = ∀, ‘pre2’ = ~{/T/, /S/, /Š/, /X/};6 Figure XI. The distribution of /X/ in position ‘pre2’ is very limited in the subtype. In fact, it is attested only in six combinations: /Xb/, /Xť/, /XTs/, /XTš/, /Xřt/ and /Xřb/. In the pre- nuclear context velar fricatives cannot occur before any fricative other than /v/ due to restriction RE3. Two or more velars cannot stand in close proximity due to restriction RG6. Though not preceding fricatives other than /v/, a velar fricative can precede an occlusive, so the absence of /Xt/ and /Xd/ is accidental. The absence of combinations like /rs/, /rx/, /lš/ is regular due to restriction RE14 (a non-nuclear semiconsonant cannot be followed by a voiceless fricative).7 Regular is also the absence of /Mt/, /Md/, /Mk/ and /Mg/ due to restriction RN1.

/t/ /d/ /s/ /z/ /š/ /ž/ /k/ /g/ /x/ /h/ /P/ /Pt/ /Ps/ /Pz/ /Pš/ /Pž/ – – /Px/ /F/ /Ft/ /Fd/ /Fs/ /Fz/ /Fš/ /Fž/ /Fk/ /Fg/ /Fx/ /Fh/ /K/ /Kt/ /Kd/ /Ks/ /Kš/ – – – – /M/ – – /Mz/ /Mš/ /Mž/ – – /Mh/ /j/ /jd/ /js/ /jh/ /r/ /rt/ /rd/ – /rz/ – /rž/ – /l/ – /lz/ – /lž/ /lk/ – /lh/

Figure XI: Two-phoneme pre-nuclear combinations attested for type DE2j.

6 /T/, /S/, /Š/ cannot occur in ‘pre2’ provided that ‘pre1’ is empty; see Section 8.7. 7 This rule, at least in such a form, would not hold for descriptions not operating with archiphonemes because there is a pre-nuclear combination /lSť/ in lstivý realized as [lscɪviː]. For us, /S/ is not phonologically voiceless.

193 DE2k: ‘E2’ = ∀, ‘pre1’ = ∀; Figure XII /řb/ is attested only in řbuchan, a name of Cirsium oleraceum (from Old Czech třbu- chan; see Machek 1954: 256, Hattala 1870: 64).

/p/ /b/ /f/ /T/ /Tb/ /S/ /Sp/ /Sb/ /Sf/ /Š/ /Šp/ /Šb/ /ř/ /řb/

Figure XII: Two-phoneme pre-nuclear combinations attested for type DE2k.

DE2l: ‘E2’ = ∀, ‘pre2’ = ~{/P/, /T/, /S/, /Š/, /M/, /j/, /r/}; Figure XIII. /P/, /M/, /j/ cannot occur in ‘pre2’ due to restrictions RE1, RE2 and RE15. The absence of /r/ in ‘pre2’ is accidental in light of the existence of /lp/ and /lb/. The gaps for /Kp/ and /Xp/ are noteworthy because their voiced counterparts are attested. /Kf/ is attested only in a Czech place name Kfely.

/p/ /b/ /f/ /F/ /Fp/ /Fb/ – /K/ /Kb/ (/Kf/) /X/ /Xb/ – /l/ /lp/ /lb/

Figure XIII: Two-phoneme pre-nuclear combinations attested for type DE2l.

DE2m: ‘E3’ = ∀, ‘pre1’ = ~{/T/, /ř/}; Figure XIV. The absence of /ř/ in ‘pre1’ is accidental for /ř/ because this phoneme can occur before a palatal occlusive (cf. /Třť/ in třtina) and, when standing before an occlusive, it need not be preceded by a peripheral phoneme (cf. /řb/ in řbuchan, /řk/ in řka). Combin- ing these facts together, we conclude /řť/ and /řď/ are structurally possible. On the other hand, /T/ cannot precede a palatal occlusive due to restriction RG4.

/ť/ /ď/ /S/ /Sť/ /Sď/ /Š/ /Šť/ /Šď/

Figure IV: Two-phoneme pre-nuclear combinations attested for type DE2m.

194 DE2n: ‘E3’ = ∀, ‘pre2’ = ~{/T/, /S/, /Š/, (/K/), /M/, /l/}; Figure XV. The absence of the combinations with /l/ in ‘pre2’ is accidental in light of the existence of those with /r/, but those with /M/ are impossible due to restriction RN1. /Pť/ and /Kť/ are attested only in the place names Ptice and Ktiš, respectively.

/ť/ /ď/ /P/ (/Pť/) /Pď/ /F/ /Fť/ /Fď/ /K/ (/Kť/) /X/ /Xť/ /j/ /jď/ /r/ /rť/ /rď/

Figure XV: Two-phoneme pre-nuclear combinations attested for type DE2n.

11.3 Distributional type DE3

In this type exactly three pre-nuclear positions are occupied by phonemes. The subtypes are given in Figure XVI.

Distributional type DE3 a ∅ ∅ e3 e2 e1 b ∅ pre1 ∅ e2 e1 c pre2 ∅ ∅ e2 e1 d ∅ pre1 e3 ∅ e1 e pre2 ∅ e3 ∅ e1 f ∅ pre1 E2 e1 g pre2 ∅ E2 e1 h pre2 pre1 ∅ ∅ e1 h ∅ pre1 e3 e2 ∅ i pre2 ∅ e3 e2 ∅ j pre2 pre1 ∅ e2 ∅ k pre2 pre1 e3 ∅ ∅ l pre2 pre1 E2 ∅ m pre2 pre1 E3

Figure XVI: Subtypes for distributional type DE3, i.e. when just three pre-nuclear positions are occupied by phonemes.

195 DE3a: ‘e1’ = ~{/m/}, ‘e2’ = ∀, ‘e3’ = ~{/š/, /ž/, /g/}; Figure XVII. There are no combinations with /m/ in ‘e1’, but /svm/ and /zvm/ must be missing by accidence if they are compared against /zvn/ or /zvň/ (the first admittedly probably only in zvnaditi (PSJČ)). /š/, /ž/ cannot occur in ‘e3’ in this subtype due to restriction RE9, but the absence of /g/ is accidental from the synchronic perspective. Interpreting Figure XVII, we can propose some additional collocational restrictions. First, if ‘e1’ is occupied by a sonant, and ‘e2’ by /M/, then ‘e3’ cannot be occupied by an occlusive; this follows from restriction RG17a prohibiting combinations of type ‘ONR’. Second, if ‘e1’ is filled with /ř/ or with /r/, /l/, and ‘e2’ with /v/, then ‘e3’ cannot be filled with an alveolar occlusive or with any velar. This situation is more complex, but we introduce these restrictions to account for it: A peripheral combination of or including type ‘AOLFC1’ where ‘AO’ is an alveolar occlusive, ‘LF’ is a labial fricative, and 8 ‘C1’ is a nasal or /ř/ or a semiconsonant is impossible (= RG17g). And impossible is also a peripheral combination of or including type ‘KFC1’ where ‘K’ is a velar, ‘F’ is a fricative, and ‘C1’ is a nasal or /ř/ or a semiconsonant (= RG17h). The first of these restrictions is purposely less general because it was necessary to specify that the fricative is labial due to the existence of /Tšl/ in člověk and /Txn/ in dchnouti (SSJČ); on the other hand, no such specifications are necessary for the velars.

/n/ /ň/ /j/ /ř/ /r/ /l/ /tM/ /tMň/ – – – – /dM/ /dMň/ – – – – /dv/ – – /dvj/ – – – /sv/ /svj/ /svr/ /svl/ /zv/ /zvn/ /zvň/ /zvj/ /zvř/ /zvr/ /zvl/ /sM/ /sMň/ /sMr/ /sMl/ /zM/ /zMn/ /zMň/ /zMř/ /zMr/ /zMl/ /kv/ – – /kvj/ – – – /xv/ – – /xvj/ – – – /hv/ – – /hvj/ – – –

Figure XVII: Three-phoneme pre-nuclear combinations attested for type DE3a.

8 Since we speak about a peripheral combination, the semiconsonant is necessarily non-nuclear. Thus, this restriction is not violated by /tvrS/ tvrz where the /r/ is nuclear.

196 DE3b: ‘e1’ = ~{/m/, /n/, /ň/, /ř/, /r/, /l/}, ‘e2’ = ~{/M/}, ‘pre1’ = ~{/T/, /S/, /Š/}; Figure XVIII. With neutralization of voicing not taking place before /v/ and /M/, it is only /ř/ that can occupy ‘pre1’. Only one combination is attested for this type, though, but there is no reason a combination like /řMň/ could not be possible as well, cf. /hřMňet/ hřmět. On the other hand, the combinations like /řvr/ are impossible due to restriction RG13a, and combinations like /řvn/ are declared impossible due to a new collocational restriction saying that a combination of and including type ‘C1FN’ where ‘C1’ is a semiconsonant or /ř/, ‘F’ is a fricative and ‘N’ a nasal is impossible (= RG17i).

/j/ /řv/ /řvj/

Figure XVIII: A three-phoneme pre-nuclear combination attested for type DE3b.

DE3c: ‘e1’ = ~{/m/, /n/, /ř/, /r/}, ‘e2’ = ∀, for ‘pre2’ see below; Figure XIX. There are two “side-syllabic” combinations attested for this type; the absence of /l/ in ‘pre2’ is accidental, whereas that of /M/ is regular due to restrictions RG1a and RE1. The other combinations are impossible due to restrictions RG14, RG15 and RG16. Posi- tion ‘pre2’ cannot be occupied by a voicing archiphoneme. It is only occupied by /v/ on which see Section 8.7. The absence of /n/, /ř/ and /r/ in ‘e1’ is accidental because there is no reason to think combinations like /vMn/, /vMř/ and /vMr/ are impossible. On the contrary, /m/ cannot regularly occur in ‘e1’ due to restrictions RG1a and RG17i.

/ň/ /j/ /l/ /jM/ /jMň/ – – /rv/ – /rvj/ – /vM/ /vMň/ /vMl/

Figure XIX: Three-phoneme pre-nuclear combinations attested for type DE3c.

DE3d: ‘e1’ = ~{/j/}, ‘e3’ = ~{/z/, /ž/}, ‘pre1’ = ~{/ř/}; Figure XX. The absence of /j/ in ‘e1’ is regular due to restrictions RG12a–b. Similarly regular is the absence of /ř/ in ‘pre1’ due to restriction RE10a. But the absence of /z/, /ž/ in ‘e3’ is

197 accidental because there is no phonological property they would have in common so that they cannot constitute a well-formed class, and no rule can be introduced. A few more words should be said about phonemes from position ‘pre1’. The most combinable is /S/: it can precede an occlusive and a fricative (either from pos ‘e3’ or from pos ‘E2’) or a combination of an occlusive or a fricative with a phoneme from pos ‘e1’ or from pos ‘e2’ (cf. /Stm/, /Sdv/). However, not all combinations are possible: /S/ does not occur before /s/, /z/, and a few others are not attested: /Stn/, /Stň/, /Sdn/, /Sdň/. There does not seem to be any evidence these combinations could not be structurally possible. Most lacunae are found in combinations with /Š/. In general, the combinations with this archiphoneme are small in number, though it is mostly by accidence only (e.g. the absence of the combinations /Šg/ and /Šx/). Moreover, there are not many combinations where /Š/ is followed by two consonants. Besides /hn/ and /hň/, it can only be followed by /p/, /b/, /t/, /k/ which can in turn be followed only by /r/ or /l/. If /Š/ is followed by /t/, /k/, the latter can also be further followed by /v/ (cf. /Štv/, /Škv/), but these belong to subtype DE3i. Let us add that the combinations /Šhn/ and /Šhň/ are attested only in žhnout and žhni (imper. sg. of zhnout). It is possible to think of žhla (with /Šhl/) as the 3rd person singular past of this verb. However, even though žhla is found in the Czech National Corpus, its variant žhnul is much more common; it is also the one regarded as correct.

/m/ /n/ /ň/ /ř/ /r/ /l/ /Ts/ /Tsm/ /Tsl/ /Tš/ /Tšm/ /Tšň/ – /Tšl/ /Tk/ /Tkn/ /Tkň/ /Tkl/ /Tx/ /Txn/ /Tkn/ /St/ /Stm/ /Stř/ /Str/ /Stl/ /Sd/ /Sdm/ /Sdř/ /Sdr/ /Sdl/ /Sš/ /Sšl/ /Sk/ /Skm/ /Skn/ /Skň/ /Skř/ /Skr/ /Skl/ /Sg/ /Sgr/ /Sgl/ /Sx/ /Sxm/ /Sxn/ /Sxň/ /Sxř/ /Sxr/ /Sxl/ /Sh/ /Shm/ /Shn/ /Shň/ /Shř/ /Shr/ /Shl/ /Št/ – /Štr/ /Šk/ – /Škr/ /Škl/ /Šh/ /Šhn/ /Šhň/ –

Figure XX: Three-phoneme pre-nuclear combinations attested for type DE3d.

198 DE3e: ‘e1’ = ~{/j/}, ‘e3’ = ~{/g/}, ‘pre2’ = ~{/P/, /T/, /S/, /Š/, /K/, /X/, /r/, /l/}; Figure XXI. The non-occurrence of /j/ in ‘e1’ is regular due to restrictions RG12a–b, but that of /g/ in ‘e3’ is accidental. /P, /K/ and /X/ cannot occur in ‘pre2’ due to restrictions RE13a and RE13b. The non-occurrence of /r/, /l/ in ‘pre2’ is not straightforward. There are two combinations of the type semiconsonant – occlusive/fricative – sonant (/lpj/ and /rvj/; they belong to DE3c and DE3g, respectively), but the occlusive/fricative is here a labial and the sonant is /j/. Now, /j/ cannot be preceded by a consonant other than labial except for /s/ and /z/, and the labials belong to different position classes, i.e. to pos ‘E2’ and pos ‘e2’. Thus, lacking the evidence of the contrary, we regard the non-occurrence of /r/, /l/ in ‘pre2’ in this subtype as accidental.9 Position ‘pre2’ may also be occupied by /F/, though many combinations with it are missing. Notable is the non-occurrence of /Ftn/, /Ftň/, /Fdn/ and /Fdň/ which could be compared to the non-occurrence of the same combinations with /S/, i.e. of /Stn/, /Stň/, /Sdn/ and /Sdň/ on the one hand (see Figure XX), and /Spn/, /Spň/ on the other (see Figure XXII)10. Compare also /Skn/ and /Skň/ attested in sknotiti (SSJČ) and zknižnění (PSJČ), respectively (see Figure XX), against missing /Fkn/ and /Fkň/.

/m/ /n/ /ň/ /ř/ /r/ /l/ /Ft/ /Ftm/ /Ftř/ /Ftr/ /Ftl/ /Fd/ /Fdm/ /Fdr/ /Fdl/ /Fs/ /Fsm/ /Fsň/ /Fsl/ /Fz/ /Fzm/ /Fzn/ /Fzň/ /Fzř/ /Fzr/ /Fzl/ /Fš/ – /Fšr/ /Fšl/ /Fž/ – /Fžr/ /Fk/ /Fkř/ /Fkr/ /Fkl/ /Fx/ /Fxl/ /Fh/ /Fhm/ /Fhň/ /Fhř/ /Fhr/ /Fhl/ /Md/ /Mdl/ /js/ /jsm/

Figure XXI: Three-phoneme pre-nuclear combinations attested for type DE3e.

9 The database of PSJČ contains a word lsníti se found in a poem from 1873, but the word is not included into the printed version of the dictionary, nor into SSJČ. We have not considered it, eight. 10 But attested are /FSpn/ and /FSpň/!

199 DE3f: ‘e1’ = ~{/m/, /n/, /ň/}, ‘E2’ = ∀, ‘pre1’ = ~{/T/, /ř/}; Figure XXII. /m/ cannot occur in ‘e1’ due to restriction RE2, but the absence of /n/, /ň/ must be accidental because there are /pn/, /pň/ and /FSpn/, /FSpň/, i.e. there is no reason to think a combination like /Spn/ is impossible. The same reasoning applies to the next type. The absence of /ř/ and /T/ in ‘pre1’ is regular due to restrictions RE10a and RE12.

/j/ /ř/ /r/ /l/ /Sp/ /Spj/ /Spř/ /Spr/ /Spl/ /Sb/ /Sbj/ /Sbř/ /Sbr/ /Sbl/ /Sf/ /Sfr/ /Sfl/ /Šp/ /Špr/ /Špl/ /Šb/ /Šbr/ /Šbl/

Figure XXII: Three-phoneme pre-nuclear combinations attested for type DE3f.

DE3g: ‘e1’ = ~{/m/, /n/, /ň/}, ‘E2’ = ~{/f/}, ‘pre2’ = ~{/P/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/}; Figure XXIII. See the previous type on the nasals. The non-occurrence of /f/ in ‘E2’ is accidental from the synchronic perspective. /P/, /K/, /X/, /M/, /j/ cannot occur in ‘pre2’ due to restrictions RE13a, RE13b, RE1 and RE15. Finally, considering /lpj/, we regard the non- occurrence of /r/ in ‘pre2’ as accidental.

/j/ /ř/ /r/ /l/ /Fp/ /Fpj/ /Fpř/ /Fpr/ /Fpl/ /Fb/ /Fbj/ /Fbř/ /Fbr/ /Fbl/ /lp/ /lpj/

Figure XXIII: Three-phoneme pre-nuclear combinations attested for type DE3g.

DE3h: impossible type. See DE2d.

DE3i: ‘e2’ = ~{/M/}, ‘e3’ = ~{/z/, /ž/, /g/, /h/}, ‘pre1’ = ~{/ř/}; Figure XXIV. /M/ cannot occur in ‘e2’ due to restrictions RN9d and RN9e, and /ř/ cannot occur in ‘pre1’ due to restriction RG13a. The non-occurrence of /z/, /ž/, /g/, /h/ in ‘e3’ is accidental because these phonemes do not constitute a well-defined class for which a distributional rule could be introduced.

200 /v/ /Ts/ /Tsv/ /Tš/ /Tšv/ /Tk/ /Tkv/ /St/ /Stv/ /Sd/ /Sdv/ /Sk/ /Skv/ /Sx/ /Sxv/ /Št/ /Štv/ /Šk/ /Škv/

Figure XXIV: Three-phoneme pre-nuclear combinations attested for type DE3i.

DE3j: ‘e2’ = ~{/M/}, ‘e3’ = ~{/d/, /s/, /š/, /ž/, /g/, /x/, /h/}, ‘pre2’ = ~{/P/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/}; Figure XXV. There are only three combinations attested in all of which /F/ occurs in ‘pre1’, neither of them is very common—they are probably only found in vtvořiti (SSJČ), vzvolati (SSJČ), vkvapiti (PSJČ). In light of this the absence of /d/, /s/, /š/, /ž/, /g/, /x/ and /h/ in ‘e3’ is regarded as accidental. But the absence of /P/, /K/, /X/, /M/, /j/, /r/, /l/ in ‘pre2’ is regular due to restrictions RE13a, RE13b and RE11a.

/v/ /Ft/ /Ftv/ /Fz/ /Fzv/ /Fk/ /Fkv/

Figure XXV: Three-phoneme pre-nuclear combinations attested for type DE3j.

DE3k: ‘e2’ = ∀, ‘pre1’ = ~{/T/, /S/, /Š/}, for ‘pre2’ see below; Figure XXVI. The only phonemes attested to occur in ‘pre2’ are /b/, /t/ and /h/, but /b/ only in Czech place names Břve and Břvany. See Section 8.7.

/v/ /M/ /bř/ (/břv/) /tř/ /třM/ /hř/ /hřM/

Figure XXVI: Three-phoneme pre-nuclear combinations attested for type DE3k.

201 DE3l: ‘e3’ = ~{/z/, /g/}, ‘pre1’ = ∀, ‘pre2’ = ~{/P/, (/l/)}; Figure XXVII. The absence of /z/, /g/ in ‘e3’ and that of /P/ in ‘pre1’ is accidental (there is no reason to think that combinations like /STz/, /FSg/ or /Přt/ could not be structurally possible). The combination /Fřt/ is only attested in /oteFřte/, a phonological form of otevřte, which is parsed to /oT–teF–Fřte/ (see Section 3.6 on “syllabification”). /TŠk/ is only attested in the place name Čkyně11, /MTs/ only in the place name Mcely12 and /lTš/ only in Lčov- ice. It is also worth noting that no combination of /TS/ + a phoneme from pos ‘e3’ is attested in this type, even though /TStm/ exists (from ctmi). Combinations marked with a dash in Figure XXVII are impossible due to restrictions RG1b, RG3 and RE10b.

/t/ /d/ /s/ /š/ /ž/ /k/ /x/ /h/ /FT/ – – /FTs/ /FTš/ /FTk/ /FS/ /FSt/ /FSd/ – /FSš/ /FSk/ /FSx/ /FSh/ /FŠ/ /FŠd/ – – – /FŠk/ /Fř/ /Fřt/ – – – /TŠ/ /TŠt/ – – – (/TŠk/) /ST/ – – /STs/ /STš/ /STž/ /STk/ /ŠT/ – – /ŠTk/ /Kř/ /Křt/ – – – /XT/ – – /XTs/ /XTš/ /Xř/ /Xřt/ – – – /rT/ – – /rTs/ /rTš/ /lT/ – – (/lTš/) /MT/ – – (/MTs/) /MS/ /MSt/ /MSd/ – /jS/ /jSt/ –

Figure XXVII: Three-phoneme pre-nuclear combinations attested for type DE3l.

DE3m: ‘E2’ = ~{/f/}, ‘pre1’ = ~{/T/}, ‘pre2’ = {/P/, /S/, /Š/, /K/, /M/, /j/, /r/, /l/}; Figure XXVIII. From the synchronic perspective, the absence of /f/ in ‘E2’ is accidental, and so is the absence of /T/ in ‘pre1’ because there is no reason to think combinations like /FTb/ or /STb/ should be impossible. Similarly accidental is the absence of /P/, /K/, /S/, /Š/ in ‘pre2’, but the absence of /M/, /j/, /r/, /l/ is regular due to restrictions RE11a. The combination /TSb/ is attested only in the place name Dzbel (i.e. /TSbel/).

11 The database of PSJČ mentions also archaic čkáti “to wait”. It is obsolete now. 12 Its variant is Mčely which would give the combination /MTš/,

202 /p/ /b/ /FS/ /FSp/ /FSb/ /TS/ /TSp/ (/TSb/) /TŠ/ /TŠp/ /TŠb/ /Tř/ /Třp/ /Xř/ /Xřb/

Figure XXVIII: Three-phoneme pre-nuclear combinations attested for type DE3m.

DE3n: ‘E3’ = ∀, ‘pre1’ = ~{/T/}, ‘pre2’ = ~{/S/, /Š/, /X/, /j/, /r/}; Figure XXIX. The absence of /S/, /Š/, /X/, /j/ and /l/ in ‘pre2’ is accidental but that of /T/ in ‘pre1’ is regular due to restriction RG4. The combination /PŠť/ occurs only in Pština, which is a Czech name for Polish Pszczyna; /lŠť/ occurs only in the place name Lštěň.

/ť/ /ď/ /PS/ /PSď/ /PŠ/ (/PŠť/) /FS/ /FSť/ /FSď/ /FŠ/ /FŠť/ /TS/ /TSť/ – /TŠ/ /TŠť/ – /Tř/ /Třť/ /KŠ/ /KŠť/ /Kř/ /Křť/ /lS/ /lSť/ /lŠ/ (/lŠť/) /MS/ /MSť/ /MSď/

Figure XXIX: Three-phoneme pre-nuclear combinations attested for type DE3n.

11.4 Distributional type DE4

This pre-nuclear distributional type accounts for combinations in which as many as four positions that are occupied by phonemes. Its subtypes are given in Figure XXX.

203 Distributional type DE4 a ∅ pre1 e3 e2 e1 b pre2 ∅ e3 e2 e1 c pre2 pre1 ∅ e2 e1 d pre2 pre1 e3 ∅ e1 e pre2 pre1 E2 e1 f pre2 pre1 e3 e2 ∅

Figure XXX: Subtypes for distributional type DE4, i.e. when just four pre-nuclear positions are occupied by phonemes.

DE4a: ‘e1’ = ~{/m/, /n/, /ř/, /r/, /l/}, ‘e2’ = ∀, ‘e3’ = ~{/d/, /s/, /z/, /š/, /ž/, /g/}, ‘pre1’ = ~{/Š/, /ř/}; Figure XXXI. All combinations under this subtype involve /tMň/, /kvj/, /xvj/ or /hvj/ for which see DE3a; the same restrictions mentioned there apply here as well. The absence of /d/ from ‘e3’ is accidental (cf. /dMň/ in Figure XVII above), and so is that of /g/. The phonemes /s/, /z/, /š/, /ž/ cannot occur in ‘e3’ due to restrictions RE8 and RE9. Finally, the absence of /Š/ in ‘pre1’ is accidental13, but that of /ř/ is regular due to restriction RE10a.

/ň/ /j/ /Tkv/ – /Tkvj/ /StM/ /StMň/ – /Skv/ – /Skvj/ /Shv/ – /Shvj/ /Sxv/ – /Sxvj/

Figure XXXI: Four-phoneme pre-nuclear combinations attested for type DE4a.

DE4b: ‘e1’ = ~{/m/, /n/, /ř/, /r/, /l/}, ‘e2’ = ∀, ‘e3’ = ~{/t/, /d/, /š/, /ž/, /k/, /g/, /x/, /h/}, ‘pre2’ = ~{/P/, /T/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/}; Figure XXXII. The three combinations attested for this type are found in vsměsnati, vzměť and vzvě- děti, respectively. These words are recorded in PSJŠ but not in SSJŠ. We should perhaps conclude that there are no combinations falling under type DE4b in present-day Czech, yet it does not mean they are a priori impossible. Several more combinations such as /FsMl/, /Fsvl/, /FsMn/ were possible in older Czech (Hattala 1870: 84).

13 SSJČ and PSJČ list a dialectal word škvěčeti (i.e. /ŠkvjeTšeťi/) where /Š/ is followed by three peripheral phonemes.

204 /ň/ /j/ /FsM/ /FsMň/ – /Fzv/ /Fzvj/ /FzM/ /FzMň/

Figure XXXII: Four-phoneme pre-nuclear combinations attested for type DE4b.

DE4c: ‘e1’ = ~{/m/, /n/, /j/, /ř/, /r/, /l/}, ‘e2’ = ~{/v/}, ‘pre1’ = ~{/T/, /S/, /Š/}, for ‘pre2’ see below; Figure XXXIII. Neutralization of voicing does not obtain for ‘pre2’ here. Only two combinations have been found, though others are possible, e.g. /křMň/ or /xřMň/. The phonemes /ř/, /r/, /l/ cannot occur in ‘e1’ due to restrictions RG13a and RG14.

/ň/ /třM/ /třMň/ /hřM/ /hřMň/

Figure XXXIII: Four-phoneme pre-nuclear combinations attested for type DE4c.

DE4d: ‘e1’ = ~{/j/}, ‘e3’ = ~{/s/, /z/, /ž/, /g/}, ‘pre1’ = ~{/ř/}, ‘pre2’ = ~{/K/, /X/, /j/, /r/}; Figure XXXIV. The absence of /j/ in ‘e1’ is regular due to restrictions RG12a–b. The absence of /s/, /z/, /ž/, /g/ in ‘e3’ is viewed as accidental. We could have assumed that alveolar fricatives cannot be preceded by two peripheral phonemes and at the same time followed by a phoneme from pos ‘e1’, but we do not find such a restriction appropriate because of /STsv/ from scvaknout where /s/ is preceded by two phonemes and followed by a phoneme from pos ‘e2’; we see no reason it could not be followed by a phoneme from pos ‘e1’, too. Similarly accidental is the absence of /r/ in ‘pre2’ in light of /lStn/. But the absence of /K/, /X/ and /j/ in ‘pre2’ is regular due to restrictions RG9 and RG11. /TStm/ is attested only in ctmi, instr. pl. of čest “virtue”. /TStn/ is attested only in ctnost “virtuousness” and archaic ctný “virtuous” (SSJČ). Finally, /TSTň/ is attested only in ctní “virtuous (pl.)”. All these words are ultimately derivatives of the same root. 14 The Czech orthoepy allows two types of pronunciations for ctnost : [ʦtnost] and [ʦnost] (VSČ: 63, Zeman 2008: 123). The former is a full, unreduced pronunciation

14 Presumably also for ctný and ctní, but apparently not for ctmi.

205 recommended for recitation, and the latter is a simplified pronunciation supposedly preferred in casual speech. It may be questioned whether these combinations should be included in the inventory of Modern Standard Czech, but they certainly represent structural possibilities of the system. There are numerous accidental gaps in Figure XXXIV; many combinations are structurally possible, and several of them such as /FSkr/, /FStr/, /FSdr/, /FSdř/, /FShr/, /FShň/ or /FSxr/ were possible in older Czech (Hattala 1870: 83).

/m/ /n/ /ň/ /ř/ /r/ /l/ /PSt/ /PStř/ /PStr/ /PŠt/ – /PŠtr/ /FTš/ – /FTšl/ /FSt/ /FStř/ /FStl/ /FSd/ /FSdm/ /FSk/ /FSkř/ /FSkl/ /FSx/ /FSxl/ /FSh/ /FShř/ /FShl/ /FŠk/ – /FŠkr/ /ŠTk/ /ŠTkn/ /ŠTkň/ – /TSt/ /TStm/ /TStn/ /TStň/ /STk/ /STkl/ /lSt/ /lStm/ /lStn/ /lStň/ /MSt/ /MStn/ /MStň/

Figure XXXIV: Four-phoneme pre-nuclear combinations attested for type DE4d.

DE4e: ‘e1’ = ~{/m/}, ‘E2’ = ~{/f/}, ‘pre1’ = ~{/T/, /ř/}, ‘pre2’ = ~{/P/, /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/}; Figure XXXV. /m/ cannot occur in ‘e1’ due to restriction RE2; /T/, /ř/ in ‘pre1’ due to restrictions RE13a and RE10a; and /S/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/ in ‘pre2’ due to restrictions RG1b, RG3, RG9, RE11a and RE11b. The phoneme /f/ is accidentally missing in ‘E2’. Several other combinations occurred in older Czech, e.g. /FŠpl/ (Hattala op. cit.: 83–4).

/n/ /ň/ /j/ /ř/ /r/ /l/ /FSp/ /FSpn/ /FSpň/ /FSpj/ /FSpř/ /FSpr/ /FSpl/ /FSb/ – – /FSbj/ /FSbř/ /TŠp/ /TŠpj/

Figure XXXV: Four-phoneme pre-nuclear combinations attested for type DE4e.

206 DE4f: ‘e2’ = ~{/M/}, ‘e3’ = ~{/d/, /z/, /š/, /ž/, /g/, /h/}, ‘pre1’ = ~{/ř/}, ‘pre2’ = ~{/P/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/}; Figure XXXVI. /M/ cannot occur in ‘e2’ due to restrictions RN9d and RN9e; /š/, /ž/ in ‘e3’ due to restriction RE9; /ř/ in ‘pre1’ due to restriction RE10a; and /K/, /X/, /M/, /j/, /r/, /l/ in ‘pre2’ due to restrictions RG9 and RE11b. The absence of /d/, /z/, /g/, /h/ in ‘e3’ is accidental, and so is that of /P/ in ‘pre2’. Although there are only a few combinations attested, we assume more are structurally possible (e.g. /FStv/, /FSdv/ or /FShv/).

/v/ /FSk/ /FSkv/ /FSx/ /FSxv/ /TŠt/ /TŠtv/ /STs/ /STsv/ /STk/ /STkv/

Figure XXXVI: Four-phoneme pre-nuclear combinations attested for type DE4f.

11.5 Distributional type DE5

In this final distributional type all pre-nuclear positions are occupied by phonemes. The only possible configuration is given in Figure XXXVII.

Distributional type DE5 pre2 pre1 e3 e2 e1

Figure XXXVII: The only subtype for the distributional type DE5, i.e. when all pre-nuclear positions are occupied by phonemes.

DE5: ‘e1’ = ~{/m/, /n/, /ň/, /ř/, /r/, /l/}, ‘e2’ = ~{/M/}, ‘e3’ = ~{/t/, /d/, /s/, /z/, /š/, /ž/, /g/, /x/, /h/}, ‘pre1’ = ~{/Š/, /ř/}, ‘pre2’ = ~{/P/, /T/, /Š/, /K/, /X/, /M/, /j/, /r/, /l/}; Figure XXXVIII. /ř/, /r/, /l/ cannot occur in ‘e1’ due to restriction RE7a; /s/, /z/, /š/, /ž/ in ‘e3’ due to restrictions RE8 and RE9; /ř/ in ‘pre1’ due to restriction RG11; and /P/, /T/, /K/, /Š/, /X/, /M/, /j/, /r/, /l/ in ‘pre2’ due to restrictions RG10a, RG10b, RG9 and RG11. On the other hand, accidental is the absence of /m/, /n/, /ň/ in ‘e1’, of /M/ in ‘e2’, of /t/, /d/, /g/, /x/,

207 /h/ in ‘e3’, and of /Š/ in ‘pre1’. That is to say, there is no reason combinations like /FStvj/, /FShvj/, /FSxvj/ could not be possible. Moreover, a combination /FSdMň/ could be at least hypothetically possible, see Sections 8.1 and 8.3.

/j/ /FSkv/ /FSkvj/ /STkv/ /STkvj/

Figure XXXVIII: Four-phoneme pre-nuclear combinations attested for type DE5.

208 Chapter Twelve POST-NUCLEAR COMBINATIONS

12.1 Distributional types DI0 and DI1

Like in the case of the pre-nuclear combinations, we distinguish in our account of post- nuclear combinations between four major post-nuclear distributional types: DI0, DI1, DI2 and DI3, where the number indicates how many positions are filled with a phoneme. For all major types, except for DI0 and DI3, several subtypes are recognized. Every attested post-nuclear combination belongs to one and only one subtype. Types DI0 and DI1 are not of much importance for this chapter because neither produces a combination of phonemes, but they are discussed here for completeness. Under every type we list phonemes which are not attested to occur in a certain in a given subtypes. The notation used in this chapter is the same like the one in the previous chapter. In type DI0 none of the post-nuclear positions is occupied by a phoneme. Phono- tagms of this type are numerous in Czech (cf. /Sto/ sto). In type DI1 it is one and only one post-nuclear position that is filled with a phoneme. There are three possibilities how this can be achieved; see Figure I. The fourth type, DI1d, is actually impossible because ‘I’ can only be filled with a phoneme if ‘i1’ is filled with a phoneme (see Section 9.4).

Distributional type DI1 a i1 ∅ ∅ b ∅ i2 ∅ c ∅ ∅ i3 d ∅ I

Figure I: Subtypes for distributional type DI1, i.e. when just one post-nuclear position is occupied by a phoneme.

DI1a: ‘i1’ = ∀, i.e. {/P/, /T/, /K/, /m/, /n/, /ň/, /j/, /r/, /l/}.

209 DI1b: ‘i2’ = ~{/P/, /T/, /K/}. /P/, /T/, /K/ cannot occur in ‘i2’ if the other post-nuclear positions are empty (see Section 9.2). In this type the position can be filled with /F/, /X/, /ř/, /S/ or /Š/. DI1c: ‘i3’ = ~{/P/, /T/, /K/, /S/, /Š/}. This position can only be filled with /Ť/ because /P/, /T/, /K/, /S/, /Š/ cannot occur in ‘i3’ if the other post-nuclear positions are empty (see Section 9.4).

12.2 Distributional type DI2

This type accounts for the situation when exactly two post-nuclear positions are filled with a phoneme, the others being empty. It has four subtypes listed in Figure II.

Distributional type DI2 a i1 i2 ∅ b i1 ∅ i3 c i1 I d ∅ i2 i3

Figure II: Subtypes for distributional type DI2, i.e. when just two post-nuclear positions are occupied by phonemes.

DI2a: ‘i1’ = ∀, ‘i2’ = ∀; Figure III. /ř/ cannot combine with /r/, /l/, with nasals and with fricatives due to restrictions RG12a and RI8 (see Appendix A). In fact, in the post-nuclear context /ř/ can only combine with occlusives (see Figures III and VI) or with /j/, cf. /jř/, but the latter is attested only in the surname Kejř. As explained in Section 9.2, if the occlusives /P/, /T/, /K/ occur in ‘i1’, they cannot be followed by /P/, /T/, /K/ or /Š/. Although the combinations /PT/, /KT/, /PŠ/, /TŠ/ and /KŠ/ are attested, they belong under DI2d. The combinability of /P/, /T/, /K/ is further limited due to restriction RI1. The only accidental gaps in Figure III are those for combinations with nasals. From the realizational point of view, the following syllable-final sequences are attested: [mp], [ɱf], [ms], [mʃ], [mt], [mx], [nf], [nt], [ns], [nʃ], [ŋk], [ɲk]. Out of them, [mt], [nf] and [ɲk] deserve special attention. The first occurs probably only in [vɪkomt] vikomt “vis-

210 count” and in a couple of surnames like Klimt. Since it does not violate the constituency of the other combinations, /mT/ is a valid post-nuclear combination. The sequence [ɲk] occurs in more words: in kuňk (sound of frog) and žbluňk (sound of something falling into water), both being onomatopoeia, and in šizuňk and festuňk and similar words, all probably originally borrowed from German. The latter group has, according to SSJČ, also variants pronounced with [ŋk] and spelled šizunk and festunk. Still, /ňK/ can be viewed as a valid post-nuclear combination. The sequence [nf] is also problematic: it may be found in tonf, gen. pl. of tonfa “tonfa (side-handed police baton)”1. We say “may” because although the pronunciation [tonf] is probably the one preferred by the norm of pronunciation for Czech, it is [toɱf] we could actually be heard when the word is pronounced. Be it as it may, we assume that [nf] can occur, which gives us the ground for regarding /nF/ valid as well.

/P/ /T/ /K/ /m/ /n/ /ň/ /j/ /r/ /l/ /P/ – – – /mP/ /jP/ /rP/ /lP/ /T/ – – – /mT/ /nT/ /jT/ /rT/ /lT/ /K/ – – – /nK/ /ňK/ /jK/ /rK/ /lK/ /S/ /PS/ /TS/ /KS/ /mS/ /nS/ /jS/ /rS/ /lS/ /Š/ – – – /mŠ/ /nŠ/ /jŠ/ /rŠ/ /lŠ/ /ř/ /Př/ /Tř/ /Kř/ – – – (/jř/) – – /F/ – – – /mF/ /nF/ /jF/ /rF/ /lF/ /X/ – – – /mX/ /jX/ /rX/ /lX/

Figure III: Two-phoneme post-nuclear combinations attested for type DI2a.

/m/ /n/ /j/ /r/ /l/ /Ť/ (/mŤ/) (/nŤ/) /jŤ/ /rŤ/ (/lŤ/)

Figure IV: Two-phoneme post-nuclear combinations attested for type DI2b.

DI2b: ‘i1’ = ~{/P/, /T/, /K/, (/m/, /n/, /l/)}, ‘i3’ = ~{/P/, /T/, /K/, /S/, /Š/}; Figure IV. /P/, /T/, /K/, /S/, /Š/ cannot occur in ‘i3’ provided that ‘i2’ is empty. The combinations /mŤ/, /nŤ/, /lŤ/ are parenthesized because they occur in phonological forms of markedly archaic words jsemť, onť and bylť, these being jsem “I am”, on “he” and byl

1 Though listed in SN1, its inflection may be a matter of dispute. However, the form tonf occurs in the Czech National Corpus.

211 “he was” with the suffix -ť. They nevertheless show a combinatorial potential. The absence of the combinations with /P/, /T/, /K/ results from restriction RI10.

DI2c: ‘i1’ = ~{/P/, /T/, /K/, /m/, /n/, /ň/}, ‘I’ = ∀; Figure V. /P/, /T/, /K/, /m/, /n/, /ň/ cannot occur in ‘i1’ in this subtype due to restrictions RI1. /rr/, /ll/ are impossible due to restriction RG1a; /rl/ occurs in /Stārl/ stárl on which see Section 7.2. /ln/ is probably only attested in the foreign surname Lincoln.

/j/ /r/ /l/ /m/ /jm/ /rm/ /lm/ /n/ /jn/ /rn/ (/ln/) /ň/ /rň/ /r/ /jr/ – /rl/ /l/ /jl/ –

Figure V: Two-phoneme post-nuclear combinations attested for type DI2c.

DI2d: ‘i2’ = ∀, ‘i3’ = ~{(/P/), /S/}; Figure VI. The combinations marked with a dash in Figure VI are impossible due to restrictions RG1a, RI2, RI10, RI7 and RI8. The non-occurrence of /S/ in ‘i3’ is regular due to restrictions RI7 and RI8. Note that /PS/, /TS/ and /KS/ belong under DI2a as explained in Section 9.2. See there also on the combination /SP/ attested perhaps in výsp, gen. pl. of výspa. Noteworthy is the absence of /řP/ if we consider that /Př/ is attested as well as /Tř/ and /řT/, and /Kř/ and /řK/. /PŠ/ is only attested in the surname Hypš.

/P/ /T/ /K/ /S/ /Š/ /F/ /X/ /ř/ /P/ – – (/SP/) – – /T/ /PT/ – /KT/ /ST/ /ŠT/ /FT/ /XT/ /řT/ /K/ – – – /SK/ /FK/ – /řK/ /Š/ (/PŠ/) /TŠ/ /KŠ/ – – /XŠ/ – /Ť/ – – – /SŤ/ /ŠŤ/ /XŤ/

Figure VI: Two-phoneme post-nuclear combinations attested for type DI2d.

212 12.4 Distributional type DI3

In this last distributional type all three post-nuclear positions are occupied by a phoneme; the only possible configuration is given in Figure VII.

Distributional type DI3 i1 i2 i3

Figure VII: The only subtype for distributional type DI3, i.e. when all post-nuclear positions are occupied by phonemes.

DI3: ‘i1’ = ~{/ň/}, ‘i2’ = ~{/ř/, /F/, /X/}, ‘i3’ = ~{/Ť/}; Figure VIII. The absence of /ň/ in ‘i1’ is accidental, though it is curious that /ň/ as the only nasal not attested to be followed by two peripheral phonemes (cf. /mST/ in pomst and /nTS/ in princ). Also accidental and curious is the absence of /Ť/ in ‘i3’; it is not attested to be preceded by two peripheral phonemes. The non-occurrence of /ř/, /F/, /X/ in ‘i2’ is regular due to restriction RI6. Some combinations with /P/, /T/, /K/ in ‘i1’ are marked as impossible due to restriction RI3. It is worth noting there is no combination of type

‘OPFO’ where ‘O’ is an occlusive and ‘PF’ is a palatal fricative. So, while combinations like /TST/ or /KST/ are attested, combinations like /TŠT/ or /KŠT/ are not. To account for this regular absence, we introduce a new restriction RI9. Many combinations in Fig- ure VIII are attested only in proper names and other symbols; these are the surnames Marx, Holst (/rKS/, /lST/), the foreign place names, all of Russian origin. Magni- togorsk, Tobolsk, Smolensk, Vítebsk, Kuzněck (/rSK/, /lSK/, /nSK/, /PSK/, /TSK/2), the Czech place name Telč (/lTŠ/), and the brand name Sfinx (/nKS/). The Russian-origin words are especially doubtful and should not have probably been even included in our analysis, but they fit the pattern of the other combinations for this subtype, which must be a reason why Czechs do not have troubles pronouncing them.

2 /TSK/ is also attested in kuck which is, according to PSJČ, an onomatopoeic word for choking.

213 /P/ /T/ /K/ /m/ /n/ /j/ /r/ /l/ /PT/ – – – /rPT/ /TS/ – – – /nTS/ /jTS/ /rTS/ /lTS/ /TŠ/ – – – /nTŠ/ /jTŠ/ /rTŠ/ (/lTŠ/) /KT/ – – – /nKT/ /rKT/ /KS/ – – – (/nKS/) (/rKS/) /SP/ /TSP/ /ST/ /PST/ /TST/ /KST/ /mST/ /jST/ /rST/ (/lST/) /SK/ (/PSK/) (/TSK/) (/nSK/) /jSK/ (/rSK/) (/lSK/) /ŠT/ – – – /mŠT/

Figure VIII: Three-phoneme post-nuclear combinations attested for type DI3.

214 Chapter Thirteen COMBINATIONS OF NUCLEAR AND PERIPHERAL PHONEMES

13.1 ‘CV’ combinations

Nuclear phonemes do not combine with each other within a single major-type phonotagm because there can only be one such phoneme there. If a group of two nuclear phonemes occurs, they belong to separate phonotagms like those in /idea/ idea or in /abēovē/ abbéové (see Mathesius 1931b on such combinations). Within a single phonotagm they combine only with peripheral phonemes. In the chapter we examine this type of combinations. We start with combinations where the short vowels /i/, /e/, /a/, /o/, /u/ are preceded by a single peripheral phoneme. Having examined this, we conclude they can be preceded by all peripheral phonemes,1 as shown by the example in Figure I. This conclusion must, however, be confronted with other claims. There is a long tradition of phonological descriptions of Czech questioning the combinability of the back vowels /o/, /u/, /ö/ with the palatals /ť/, /ď/, /š/, /ž/, /ň/, with /j/, /ř/, and with /c/ and /č/, the latter two being affricates we have interpreted as /Ts/ and /Tš/, respectively. The idea goes back at least to Mathesius (1931a), and is repeated by other linguists (Ludvíková 1968: 61, Va- chek 1968: 89, Těšitelová et al. 1985: 18–9, Grygarová-Rechzieglová 1993: 265, Krčmová 2008: 111). Although Mathesius (ibid.) did not deny the ability of the back vowels to combine with the mentioned consonants, he argued that such combinations only occur in the following situations:2

1 Naturally, voicing archiphonemes cannot occur before nuclear phonemes, and /M/ can occur before vowels only if it is preceded by /ř/ as in /hřMī/ hřmí. 2 Let us add that Mathesius (ibid.) also questioned the ability of the velars /k/, /h/, /x/ to combine with the vowel /e/, but this claim can be at once dismissed because at least /k/ and /h/ do com-

215 (α) Across morphological boundaries (cf. rtuťový, sleďů, mužů, pňový, bojovat, křovina). (β) In words of foreign origin (cf. čokoláda, jód, žok). (γ) In “domestic” words of onomatopoeic origin or in emotionally colored words (cf. křupat, ťukat, fňukat, kňourat, šoupati, ďobati).

Although such observations have undoubtedly some importance, they are somewhat misguided and overestimated. To begin with, the presence of a morphological boundary is irrelevant for phonology unless it is phonetically signaled. Phonology accounts for the sound structure of a language, not for its morphological structure, and is independent of morphology (Hervey 1978). A statement that such combinations, e.g. /řo/ in křovina, may not belong to the same morpheme3 is something else than a statement that they belong to the same phonotagm or to two phonotagms, for phonotagms are entities of a different kind than morphemes. Secondly, how do we find out whether words are of foreign and/or onomatopoeic origin? Some linguists (Mathesius 1932: 231, Vachek 1968: 31, 89) are apparently of an opinion that some words are so phonologically marked that we can deduce they are synchronically foreign even without knowing their origin, but the idea is false. For example, once we know that the word jód was borrowed to Czech from a foreign language, we can state that it contains a combination /jō/. On the exami- nation of its distribution we find out that /jō/ occurs only in words about which we could say they were borrowed from foreign languages. Consequently, the conclusion is that /jō/ is a mark of foreignness. But could we arrive to this without a prior knowledge of the origin of the words it occurs in? We can at most list these words (and how can we be sure we have listed all of them?), but so can we list words containing e.g. /je/. One of them is jed; now, what is the difference between jed and jód? Unless we know their his-

bine with /e/ in “neutral” words such as ke, keř, kedluben, herec, hezký, hebký. It is only /xe/ which has limited combinability because it occurs mostly in words originally borrowed from foreign languages such as chemie (but there is also a genuine Czech place name Cheb, and also the word chechtat, although the latter, one might argue, is onomatopoeic in origin). 3 We deliberately say may not belong because the decision whether they do or do not is a matter of a morphological analysis which may differ from a linguist to a linguist. For example, black- bird consists of two morphemes in some analyses, while others treat it, more appropriately, as a single morpheme/moneme (Mulder – Hervey 1980: 122–44).

216 tory, there is none. The whole argument is furthermore dubious because the conclusion that /jō/ occurs only in words of foreign origin rests upon our ability to say they are indeed words of foreign origin. But what makes a certain word domestic in origin? Such a question cannot have a definitive answer, and the answer could not even be decisive because Modern Czech contains a number of words (such as kalhoty or košile) once borrowed from foreign languages, but since then fully “domesticated”.

/i/ /e/ /a/ /o/ /u/ /p/ pil pes pak pot puk /b/ byl bez baba bota bude /f/ film fena farma fonetika funkce /v/ vysoký ven vana voda vulkán /m/ myš med mast moč muset /t/ tyč teď tak to tu /d/ diktát den dar doma duch /s/ syn sen sad sok sud /z/ zima zem zase zorný zub /n/ nyní nes nad noc nutit /ť/ tis tělo ťal šťovík ťuhýk 4 /ď/ div děs ďas ďobat ) /š/ šik šel šaty šotek šum /ž/ živý žena žací žok žula /ň/ nic něco ňadra žňový šňupat /k/ kytka keř kamna koza kus /g/ gymnázium gel galantní golf guma /x/ chyba chemie chata chodba chuť /h/ hynout hezký had hod hudba /j/ jiný jen jak jogurt junák /ř/ řinout řez řada křoví křupat /r/ rytec rej rak roj ruka /l/ lys let lano loj luka

Figure I: Examples of words where ‘CV’ combinations with a short vowel are attested.

Thus, it is not relevant whether a certain combination is attested in a certain group of words only, but whether it is attested at all. To return to the combinability of short vowels with peripheral phonemes, we repeat that every short vowel can be preceded by every possible peripheral phoneme. Figure I gives examples of words in which short

4 Attested in shromažďuje; see the text.

217 vowels combine with the suspected palatals /ť/, /ď/, /š/, /ž/, /ň/ and with /ř/, /j/. Although some combinations are rare, they are still occurrent (like /ťa/ in ťal). Others are only attested in more complex combinations such as /ťo/ which has been found only as a part of /Šťo/ in šťovík. Similarly, /ňo/ is attested only as a part of /žňo/ in žňový, /ňu/ as a part of /Šňu/ in šňupat, /řu/ as a part of /křu/ in křupat, and /ďu/ found only in shro- mažďuje. It does not, however, mean that words beginning just with /ťo/, /ďu/, /ňo/, /ňu/, /řu/ are not possible in Czech. In fact, they (but except those beginning with /ňo/) exist, but are, some might argue, of onomatopoeic or other origin (cf. ťopkat, ďubka, ňuhňat, řupoň). If there is no restriction of the combinability of the short vowels with single pre- nuclear phonemes, the situation is more interesting in the case of the long vowels /ī/, /ē/, /ā/, /ō/, /ū/. Figure II gives examples of words where such combinations are attested. An empty cell means the respective combination has not been found. The dash indicates a given combination is not structurally possible; for these, we postulate several collocational restrictions below. As can be deduced from the figure, there is no problem with labials, alveolars, velars and with /r/, /l/: all combine with the long vowels. On the contrary, the combinability of the palatals /ť/, /ď/, /š/, /ž/, /ň/ and of /ř/, /j/ requires a com- mentary. First of all, /řē/ has been found only in the proper name Břéťa, which suggests it is possible at least marginally. Many other combinations are attested only word-internally, for example, /ťū/ in zeťů. If we assume that /zeťū/ is built of two phonotagms (since it contains two vowels), then it should be parsed (“syllabified”) either to /zeť–ū/ or to /zeŤ–ťū/, the latter with functional amalgamation. The first division is problematic for its assumption of /zeť/ which is not a well-formed major-type phonotagm in Czech—the difference between voiceless and voiced consonant is canceled at the end of such phonotagms. The second division is also problematic, this time for its assumption of /ťū/ whose well-formedness could be questioned, as it is otherwise unattested. If it is also declared ill-formed, our only option would be to conclude that /zeťū/ is not built of two phonotagms but of a single one because there would be no way of dividing to smaller phonotagms, and accordingly no way of proving it is two phonotagm. Such a solution is not quite desirable, though. It would mean that there are certain phonological words in

218 which vowels do not function as nuclear entities, that is, if /zeťū/ is a single phonotagm, /ū/ cannot be a nuclear phoneme here. Having considered all alternatives, we find most adequate to view /ťū/ as a well-formed phonotagm, i.e. to divide /zeťū/ to /zeŤ/ and /ťū/. The same reasoning could then be applied to /ďū/ in sleďů, /šū/ in hrabošů, /žē/ in pasažér, /žō/ in demižón and /žū/ in mužů. And also to /ňā/ in tučňák and /ňū/ in stupňů because /tuTšňāK/ and /Stupňū/ can only be parsed in such a way that /ňā/ and /ňū/ belong to one phonotagm.5 The same argument cannot, however, be used in the division of /frajēr/ frajér, /lēkařū/ lékařů and /pokojū/ pokojů which contain an inter-nuclear sonant. If /ť/ in /zeťū/ cannot stand the end of a phonotagm due to neutralization of voicing, then the sonants /j/ and /ř/ can do so (cf. /māj/ máj and /keř/ keř), and so it is possible to parse phonological forms of the given words to /fraj–ēr/, /lēK–kař–ū/ and /poK–koj–ū/. Hence, they cannot provide evidence that /jē/, /řū/ and /jū/ are possible combinations within one phonotagm. But there is equally no indication of the opposite, that is, we cannot postulate a collocational restriction which would declare /jē/, /řū/ and /jū/ impossible. Doing so requires that we can single out a unique property that either the phonemes /j/ and /ř/ or the phonemes /ē/ and /ū/ have in common. We cannot think of any. Furthermore, there is indirect evidence of /jēgroFki/ jégrovky (obsolete according to SSJČ) and /jūlinka/ Jůlinka (proper name) that at least /jē/ and /jū/ are allowed. Figure II also shows that there are certain missing combinations: /ťē/, /ťō/, /ďē/, /ďō/, /šō/6, /ňē/, /ňō/ and /řō/. The absence of /ťē/, /ťō/, /ďē/, /ďō/, /ňē/ and /ňō/ can be regularized by a simple collocational restriction: a long mid vowel (i.e. /ē/, /ō/) cannot be preceded by a palatal occlusive or the palatal nasal (i.e. by /ť/, /ď/, /ň/) (= restriction RN3 in Appendix A). This time we can single out unique properties these phonemes have in common. On the other hand, no such restriction applies to the missing /šō/ and /řō/ because /g/ and /ř/ do not constitute any well-defined class with /ť/, /ď/, /ň/ or with each other (i.e. they have phonologically nothing in common). Consequently, their absence is regarded as accidental.

5 The same procedure can be applied to the division of angína, kolegů given in Figure II. 6 Unless we accept kapišónek (SSJČ).

219 /ī/ /ē/ /ā/ /ō/ /ū/ /p/ pít péro pás pór původ /b/ být bérec bát se bójka bůh /f/ fík fén fáma fór fúze /v/ víla vést vánek kvóta vůle /m/ mýt mést mást móda může /t/ týl téct tát tón tůň /d/ dým déle dál dóm důl /s/ sít sect sám sója sůl /z/ zívat zévnatka zákon zóna zůstat /n/ nýbrž nést náš nóbl nůše /ť/ tít – šťáva – zeťů /ď/ dílo – ďábel – sleďů /š/ šít šéf šálit hrabošů /ž/ žít pasažér žárlit demižón můžů /ň/ níž – tučňák – stupňů /k/ kýchat kéž kámen kóta kůl /g/ angína génius gáže gól kolegů /x/ chýše schéma chátrat chór chůze /h/ hýbat se hélium házet cirhóza hůl /j/ jíst frajér jádro jód pokojů /ř/ říkat (Břéťa) řádek lékařů /r/ rýt réva rám rómština růže /l/ lýko léto láska lóže lůžko

Figure II: Examples of words where ‘CV’ combinations with a long vowel are attested. Paren- thesized examples are symbols.7 The dash indicates the combination is structurally impossible.

The last type of vowels is diphthongal vowels. Their ability to combine with pre- nuclear phonemes is exemplified by Figure III8. The combinability is quite limited, although these limitations concern mostly /ë/ and /ä/. Diachronically speaking, these two vowels have been introduced to Czech relatively recently through loan words, whereas /ö/ has arisen through regular development of Czech from Common Slavic. But facts like these should not have any bearings on a synchronic analysis. A more grievous problem is with /ë/, realized as [ɛu̯]. In the Czech orthography this sound is spelled eu. How-

7 See Section 3.3 on the difference between signs and symbols. 8 See above on žeťů for the argument behind “syllabification” of slaďoučký, kolegou, Poldauf and Dachau. The words Teuton, Poldauf, Šaur, Dachau and Jouza are proper names; their evidence is only indirect. The combination /ňä/ has only been found in the onomatopoeic expression mňau, and /jä/ only in onomatopoeic jau. Finally, it could be questioned whether ňouma belongs to Standard Czech; since it is commonly used, it has been included.

220 ever, not every eu sequence is pronounced as the [ɛu̯] diphthong. In many cases it is pronounced as two syllables (e.g. in muzeum, phonologically /muzeum/, not /muzëm/), and in some other cases there is apparently free variation between monosyllabic and disyllabic pronunciation (and hence between /ë/ and /eu/), though eu is probably always pronounced monosyllabically in some words (e.g. in euro). Unfortunately, the latter does not seem to be case of the words included in Figure III where all may be pronounced both with monosyllabic or disyllabic eu. Due to these problems, it is difficult to draw conclusions from Figure III because many combinations may be missing by accidence. Yet there is an apparent restriction on the combinability of /ë/ with palatals; no palatal has been found before this vowel, which may simply be a consequence of the fact that no words has been borrowed from a language where such a combination occurs. Since the other two diphthongal vowels can combine with palatals, we have decided not to introduce any collocational restriction here.

/ë/ /ä/ /ö/ /p/ terapeut pauza pouta /b/ bauxit bouda /f/ feudál fauna foukat /v/ vousy /m/ mauzoleum mouka /t/ (Teuton) tautologie touha /d/ deuterium (Poldauf) doufat /s/ sauna soud /z/ zeugma tezaurus zout se /n/ neuron nausea nouze /ť/ šťouchat /ď/ ďaur slaďoučký /š/ (Šaur) šourek /ž/ lžou /ň/ ňouma /k/ kauce koule /g/ gauner kolegou /x/ (Dachau) choutky /h/ heuristika hausbót houba /j/ (Jouza) /ř/ dřou /r/ raut roura /l/ leukémie laureát louka

Figure III: Examples of words where ‘CV’ combinations with a diphthongal vowel are attested.

221 /r/ /l/ /p/ prst plsť /b/ brk blb /f/ frnknout troufl /v/ vrch vlk /m/ – – /t/ trn pletl /d/ drvoštěp vedl /s/ srp slza /z/ zrno vezl /n/ – – /ť/ – – /ď/ – – /š/ (Kutlvašr) (Litomyšl) /ž/ (Pížl) /ň/ – – /k/ krk klk /g/ bagr triangl /x/ chrt uprchl /h/ hrb hlt /j/ – – /ř/ – – /r/ – – /l/ – /M/ mrkat mlž

Figure IV: Examples of words where ‘CV’ combinations with a semiconsonant are attested.

Finally, we will look at the combinability of the nuclear semiconsonants /r/ and /l/. Figure IV provides examples of words where they are preceded by a single pre-nuclear phoneme. The words troufl, pletl, vedl, vezl, Litomyšl, Pížl, bagr, triangl and uprchl can be “syllabified” in the same way like zeťů discussed above; see also Chapter 7 on the phonotagm-final nuclear /r/, /l/. The combinations /šl/, /šr/ and /žl/ have only been found in proper names such as Litomyšl, Kutlvašr and Pížl (there are a few others); /žr/ has not been found. So, nuclear /r/, /l/ can combine with the palatal fricatives /š/, /ž/, but apparently only in proper names. They cannot be preceded by /m/, /n/, /ň/ because of neutralization of place of articulation for the nasals (see Section 2.7), and they cannot also be preceded by /ť/, /ď/, which is already encoded in the distributional unit. Their inability to combine with /ř/ is due to restriction RG13a. Finally, /rr/ and /ll/ are impossible due to restriction RG1a. The dash is also used for combinations with /j/ because these are

222 impossible due to restriction RG17, i.e. if a semiconsonant occurs in the same section with /j/, the semiconsonant occurs in a more peripheral position.

13.2 ‘VC’ combinations

Let us now shift our attention to combinations of nuclear phonemes with single post- nuclear phonemes. Their combinability with the short vowels is quite straightforward because all combinations are attested; see Figure V. However, the combinability of the long vowels is not so neat, which one can see on Figure VI. There are several missing combinations, but failing to find any regularity in them, we conclude they are missing by accidence. Let us add that gróf, takéť, čikóš are archaic words according to SSJŠ. The phonological form /drūPki/ (i.e. drůbky) must be parsed to /drūP/ and /ki/ because /Pk/ is not a possible pre-nuclear combination. In comparison with the short and long ones, the diphthongal vowels are quite defective in their occurrence with post-nuclear phonemes; see Figure VII. The most defective is /ë/: we are aware of only two words where it is followed by a single peripheral phoneme, namely terapeut9 and eur (gen. pl. of euro). Since it is not possible to tell why only /T/ and /r/ could follow /ë/, the absence of the other ‘VC’ combinations must be regarded as accidental—with one important exception to be given presently. Similarly, the gaps in the combinations with /ä/ are accidental because no regularity can be dis- cerned. The words šturmhaub, lauf and šlauch are slightly archaic or colloquial, but they could perhaps be included (all are listed in SSJČ). The word Rosenbaum is a proper name with an indirect evidence value. Now, the exception we have alluded to is the absence of combinations with /j/ running through all diphthongal vowels. Accord- ingly, we propose this restriction: A diphthongal vowel cannot be followed by /j/ (= RN6) as opposed to the short and long vowels such as /a/ in /kraj/ kraj or /ā/ in /rāj/ ráj.

9 Provided that it is pronounced as [tɛrapɛu̯t], not as [tɛrapɛ.ut] (see Romportl et al. 1978).

223 /i/ /e/ /a/ /o/ /u/ /P/ hřib sklep chlap strop strup /F/ vliv lev splav slov smluv /m/ podzim lem klam strom rum /T/ byt let plat plot prut /S/ lis les klas kos kus /n/ syn ven stan slon kun /Ť/ sviť teď zaplať hoď chuť /Š/ myš lež až koš kuš /ň/ zhyň změň dlaň jabloň člun /K/ smyk vlek lak mlok kluk /X/ zdvih mech strach hloh pluh /j/ pij hřej kraj kroj sluj /ř/ miř měř lékař příkoř nekuř /r/ sekyr per par mor chmur /l/ myl měl kal kol mul

Figure V: Examples of words where a given ‘VC’ combination with a short vowel is attested.

/ī/ /ē/ /ā/ /ō/ /ū/ /P/ cíp chléb dráp rób drůbky /F/ dřív chlév páv gróf sův /m/ vím krém rám dóm dům /T/ mít portrét vát kód úd /S/ peníz fréz vás próz vůz /n/ vín fén pán tón trůn /Ť/ píď takéť záď /Š/ výš též náš čikóš nůž /ň/ skříň stráň tůň /K/ vík lék mák jóg /X/ líh hrách bůh /j/ šíj ráj gój svůj /ř/ uhlíř bankéř lhář hůř /r/ výr pér pár pór dvůr /l/ víl podél sál pól půl

Figure VI: Examples of words where a given ‘VC’ combination with a long vowel is attested.

Finally, Figure VIII shows the ability of the nuclear semiconsonants to combine with post-nuclear phonemes. They cannot by followed by /ř/ and /j/ due to restriction RN10a. The combinations /rr/ and /ll/ are impossible due to restriction RG1a. The absence of the other combinations is accidental. On the word umrlce, see Section 7.2.

224 /ë/ /ä/ /ö/ /P/ šturmhaub sloup /F/ lauf houf /m/ (Rosenbaum) /T/ terapeut aut pout /S/ aplaus vous /n/ klaun letoun /Ť/ pouť /Š/ flauš souš 10 /ň/ letouň /K/ brouk /X/ šlauch rouch /j/ – – – /ř/ kouř /r/ eur aur kocour /l/ aul kotoul

Figure VII: Examples of words where ‘VC’ combinations with a diphthongal vowel are attested.

/r/ /l/ /P/ vrb blb /F/ ponejprv /m/ hrm /T/ chrt hlt /S/ prs slz /n/ trn vln /Ť/ vrť pohlť /Š/ strž plž /ň/ vrň plň /K/ krk vlk /X/ vrch plch /j/ – – /ř/ – – /r/ – /l/ umrlce –

Figure VIII: Examples of words where ‘VC’ combinations with a semiconsonant are attested.

10 A Czech name of the order Chiroptera (bats); see Havlová (2010: 100). Also in the place name Hostouň.

225 13.3 Combinability of nuclear phonemes with peripheral combinations

If the combinability of nuclear phonemes with single peripheral phonemes could be relatively easily described, the situation is different with peripheral combinations. It is difficult to examine the combinability in the same manner, that is, to find out whether each and every peripheral combination can co-occur with each and every nuclear phoneme because there are as much as 430 pre-nuclear combinations and 101 post-nuclear combinations (marginal combinations included). Some are attested only before one or two nuclear phonemes like /FSkvj/ attested only before /e/ in vzkvět, /lStn/ only before /ī/ in lstný, /Kt/ only before /e/ in který; the same is true for post-nuclear combinations, e.g. /PST/ is attested only after /ā/ in zábst. Tables demonstrating this type of combinability would not only run through many pages, but would also contain many gaps. Such absences must be accidental in many cases, though they are sometimes governed by regular collocational restrictions. As argued in Section 3.4, in order to determine whether a certain non-occurrence is regular, we must find out whether it pertains to some well-defined class of entities or features. To begin with, we have failed to discern any restriction affecting the co-occurrence of the short vowels with pre-nuclear combinations, by which we conclude that any short vowel has the capacity to be preceded by any pre-nuclear combination. It is supported by two conclusions. The first of them is the fact that any short vowel can be preceded by any single peripheral phoneme (see Section 13.1), and the second is one we arrived at in Chapter 7, namely that short vowels can be preceded by any number of peripheral phonemes (but, of course, only up to five of them). On the other hand, the combinability of the long vowels with pre-nuclear combinations is limited by several collocational restrictions. First, it follows from restriction RN3 introduced in Section 13.1 that the long mid vowels /ē/, /ō/ cannot be preceded by any pre-nuclear combination ending in a palatal occlusive or the palatal nasal (e.g. by /Křť/, /FSď/, /šň/). We can introduce another restriction limiting the combinability of the palatals with the long vowels, but also with the diphthongal vowels: A long and a diphthongal vowel cannot be preceded by a combination of type ‘CnCC1’ where ‘C1’ is a palatal or /j/, and ‘Cn’ is any number of peripheral phonemes (= RN4a and RN4b), i.e.

226 they cannot be preceded by combinations like /FSď/, /Spj/, /zMň/, /FSkvj/; the short vowels have this capacity. The combinability of the long vowels is further restricted by a fact we already mentioned in Chapter 7: A long vowel cannot be preceded by more than four peripheral phonemes (= RN2). This restriction rests first of all on there being no such combination, but it is also a consequence of other facts. If there is a pre-nuclear combination of five phonemes, ‘CCCCC’, the rightmost peripheral phoneme can only be from pos ‘e1’ (i.e. /m/, /n/, /ň/, /j/ /ř/, /r/ or /l/). In effect, this phoneme can only be /j/ because the nasals, the semiconsonants and /ř/ cannot be preceded by four peripheral phonemes in the pre-nuclear context (see Chapter 8). However, the approximant /j/ is also ruled out because as we have just said, the long and diphthongal vowels cannot be preceded by /j/ if that is preceded by two or more peripheral phonemes. The combinability of the diphthongal vowels with pre-nuclear combinations is also limited, namely by restrictions RN4a–b (introduced in the previous paragraph), by RN5 (introduced in Section 7.1 according to which a diphthongal vowel cannot be preceded by more than three peripheral phonemes). These are the only restrictions we find appropriate to introduce. In comparison with the vowels, the ability of nuclear /r/, /l/ to be preceded by pre- nuclear combinations is very limited. They cannot be preceded by five pre-nuclear phonemes, and this is already encoded in the distributional unit. Many other restrictions can be deduced from the distributional unit, and also from the restrictions which hold for /r/, /l/ in general. In Section 7.2 we mentioned that when occurring at the end of a phonological word, the nuclear semiconsonants cannot be preceded by three or more peripheral phonemes, though they have this capacity when not final (cf. /TŠtvrT/ čtvrt); this is restriction RN7b. In Chapter 8 we also introduced several restrictions pertaining to the nuclear semiconsonants, namely RN9a, RN9b, RN9c, RN9g, RN9h and RN9i (see Ap- pendix A). They are also limited by restriction RG17 introduced above in Section 8.1: If a semiconsonant occurs in the same section with /j/, the semiconsonant occurs in a more peripheral position, which must be ‘pre2’. Several other restrictions can be introduced once we return to properties of peripheral combinations discussed in Chapter 10. We mentioned there which types of peripheral combinations are attested to occur before or after a nuclear semiconsonant, but we did

227 not discuss them in any detail. Accordingly, the following collocational restrictions may be proposed: Nuclear semiconsonants cannot stand in close proximity with a combination of or including type ‘OO’ (= RN8) where ‘O’ is an occlusive, as compared to vowels, e.g. /ā/ in /TkāT/ tkát or /i/ in /SkriPT/ skript. They cannot also be preceded by a combination of or including type ‘ON’ (= RN9d) where ‘N’ is a nasal, as opposed to vowels, e.g. /a/ in /tma/ tma. They can be preceded by a combination of type ‘FN’ where ‘F’ is a fricative, as in /sMrT/ smrt, but they cannot be preceded by a combina- 11 tion ‘CnFN’ where ‘Cn’ is any number of peripheral phonemes (= RN9e). Finally, they cannot also be preceded by a combination of or including types ‘LP’ or ‘KP’ or ‘KA’ (= RN9f) where ‘L’ is a labial, ‘P’ is a palatal, ‘K’ is a velar, and ‘A’ is an alveolar, as opposed to vowels, e.g. /e/ in /Fše/ vše, /i/ in /KšilT/ kšilt and /i/ in /Ksilofon/ xylofon. On the other hand, the nuclear semiconsonants are preceded by a labial which is preceded by a velar, but only if the velar is further preceded by another peripheral phoneme, that is, we have /Skvrn/ skvrn but not something like /kvrn/. Let us now move to the ability of the nuclear phonemes to co-occur with post- nuclear combinations. We have not been able to discern any restriction on such ability in the case of the short and long vowels. This conclusion must be confronted with the claims made by Trnka (1961), who is one of the few (if not the only one) to consider the distribution of vowel length in Czech.12 On examining occurrences and co-occurrences of the long vowels in Czech words, he (op. cit.: 12) made the following observations (Trnka’s terminology and transcription are retained):

(α) Long vowels do not occur before homomorphemic consonant clusters other than st, sť, sk, zd, zď, šť, žď, tr, tř, dr, dř, rt, rť, mň. (β) Long vowels are non-occurrent in the first syllable of disyllabic morphemes. (γ) When in homomorphemic hiatus, vowels are only short. (δ) The opposition i|í and u|ú is neutralized before j and before the tautosyllabic m.

11 It does not hold for dialectal words such as čmrkat “to scribble” containing /TšM/. 12 We have also discussed this problem in Bičan (2008a, Section 5.6).

228 Though Trnka’s claims may be of some interest and are confirmed by examples he gave, they are all dubious except for the claim (α). The oppositions /i/ ~ /ī/ and /u/ ~ /ū/ can indeed be neutralized in the mentioned environments (Krčmová 2008: 135, 137, 200), but this phenomenon is not regarded as orthoepic, that is, the length is required to be retained in Standard Czech (Palková 1997: 264, 323, Zeman 2008: 36). The dubiousness of Trnka’s other claims stems from his allusions to homomorphemic clusters, disyllabic morphemes and homomorphemic hiatus. Whether something belongs to one or to two morphemes must and can only be decided on morphological criteria, not on phonological ones (see Section 13.1). This is not to say that morphological boundaries cannot be phonologically signaled, but that it is morphological analysis that tells us what morphological units the phonological signals mark or whether what they mark are boundaries of morphological units at all. But even if we employed morphological criteria, we would still have make recourse to the history of Czech to find out whether some words contain a morphological boundary and others not. Synchronically, it is hardly possible to say that základ “basis” contains two morphemes zá and klad or that vážka “dragon-fly” is built of váž and ka13; words like these thus contradict the claim (α). It is also contradicted by úct, gen. pl. of úcta or by půjč, imper. sg. of půjčit. Even if the list of the exceptional combinations mentioned under (α) were extended to contain some more such as kl, ct or jč, it would still be of no avail. The rule is simply not valid in current Czech, at least in this form.14 Similarly, it is hardly possible to claim that důvod “reason” or kámen “stone” contain synchronically two morphemes. There- fore, the claim (β) cannot be valid either. Finally, the claim (γ) can only be valid provided that we know where a morphological boundary falls (i.e. that there is a morphological boundary in abbéové “abbots” or tříetapový “three-period”) or provided that the words příušnice “mumps” and náušnice “earrings” can be synchronically proven to contain a morphological boundary between pří and ušnice and ná and ušnice; that both words are associated with and were originally derived from uši “ears” does not mean they contain, synchronically, any such morpheme.

13 vážka is ultimately derived from váž-, váha “weight”, but this origin is utterly obscured in Modern Czech. See Machek (1968) for the etymology of this and the other words. 14 It is noteworthy that in Trnka (1982: 188), where Trnka (1961) is reprinted, the list reproduced here under (α) does not include šť and žď.

229 The bottom line is that the claims (α), (β) and (γ) have no validity in our description. Ergo, the combinability of the long vowels is not limited in any obvious way in Czech. But the combinability of the diphthongal vowels is limited, namely by restriction RN6 introduced in Section 13.2: As they cannot be followed by /j/, they cannot also be followed any post-nuclear combination where /j/ stands in close proximity with them. Aside from this, however, we have not been able to discern any other restriction. On the other hand, the ability of nuclear /r/, /l/ to be followed by post-nuclear combinations is subject to many restrictions. Like the diphthongal vowels, they cannot be followed by /j/ (cf. RN10a), so any post-nuclear combination with /j/ is ruled out. They cannot be followed by two occlusives as mentioned earlier in this section (cf. RN9b), and they are also limited by restrictions RN10c and RN10d introduced in Chapter 9. Fi- nally, we can also introduce a new restriction: A nuclear semiconsonant cannot be followed by a combination of or including type ‘NF’ (= RN10b); vowels can, cf. /i/ in /nimF/ nymf.

13.4 ‘CVC’ combinations

In this final section we explore ‘C1VC2’ combinations within single phonotagms in order to find out whether the selection of a peripheral phoneme or a peripheral combination in the ‘C1’ place affects in any way the selection of a peripheral phoneme or a peripheral combination in the ‘C2’ place, and vice versa. Thus, strictly speaking, we will not deal with combinations of peripheral and nuclear phonemes, but with an interaction between pre-nuclear and post-nuclear phonemes and their combinations. Until now we have silently assumed that the pre-nuclear context and the post-nuclear context are mutually independent domains of distributions. Even though this assumption has obvious advantages, it also has one serious drawback. On the one hand, it is argua- bly much easier to deal with the distribution and combinations of peripheral phonemes separately in the context before the nucleus and in the context after it (which has in fact been a practice of many other linguists). On the other hand, such a treatment must assume that the occurrence of phonemes in the pre-nuclear context is not conditioned the

230 occurrence of phonemes in the post-nuclear context, and vice versa. But this need not be necessarily so. One conspicuous piece of counterevidence has been cited for Kamali Arabic where the occurrence of a consonantal combination in either the pre-nuclear or post-nuclear context precludes the occurrence of a consonantal combination in the other context. So, whereas the phonotagms /nmal/ and /malx/15 are well-formed in this language, a phonotagm like /nmalx/ is not because a vowel cannot be both preceded and followed by a consonantal combination. Similar examples could no doubt be found in other languages, though there are sometimes rather tendencies to avoid certain combinations than strict restrictions (cf. also Davis – Baertsch 2011). Cairns (1988) discusses several pan-syllabic constraints (as he calls them) limiting the co-occurrence of pre- nuclear combinations and post-nuclear combinations in English16. Earlier, Fudge (1969: 279ff.) mentioned that English avoids syllables both beginning and ending with a combinations containing /l/ or /r/ (the first violated by flail), and the very same tendency has been cited for Norwegian (Vogt 1942: 22). Similarly, German and Swedish tend to avoid syllables both beginning and ending with /r/, /l/, /m/ or /n/ (Twaddell 1940–41: 41, Sigurd 1958: 47). To the best of our knowledge, the same thing has not been examined for Czech, and it is the purpose of this section to pay the debt—at least partly because we will not be primarily interested in whether there is a tendency in Czech words to avoid a certain ‘CVC’ combination, that is, whether some combinations are more frequent than the others, but whether some or all combinations are possible at all. Figures IX and X provide examples of words in which a given ‘CVC’ combination has been found. Note that the choice of the nuclear phoneme is irrelevant here except for the lines with the nasal archiphoneme /M/ which can occur only before nuclear semiconsonants. Consequently, the combinations /M_j/, /M_ř/, /M_r/ and /M_j/ are impossible, as explained in the previous section. When compiling the tables, we have first attempted to find a word whose phonological form corresponds exactly to a given ‘CVC’ combination; a vast majority of such words are found. If none has been found,

15 These are phonological forms of the signs with the meaning “to be fed-up with” and “to cause damage to”, respectively. The examples are from Hervey (1978: 58). 16 One is that syllables of the structure /sC1(C)VC1/, where ‘C1’ is a particular consonant and ‘(C)’ is optional, are not possible, i.e. the words like spup, skack, snon, sman, splup and scrack are not possible, but words like pup, crack, none, skulk or smarm are possible (p. 229).

231 we have looked for words where a given combination occurs, but where one of the peripheral phonemes is further preceded or followed by another peripheral phoneme. Thus, /švāP/ šváb attests the combinability of the pre-nuclear /v/ and the post-nuclear /P/ because there is no reason to think a word with the phonological form /vāP/ could not be possible in Czech. All these words have a mono-monotagmic phonological form. However, some ‘CVC’ combinations have only been found in words with pluri- phonotagmic phonological forms. For example, the combination of the pre-nuclear /f/ and the post-nuclear /F/ is found in šéfův.17 If its phonological form /šēfūF/ is to be parsed to individual phonotagms, the latter must be /fūF/. The same holds for several other combinations: xenofob containing /foP/, doufáš containing /fāŠ/, pravděpodobně containing /ďeP/, marketingář (SN1) containing /gāř/, cukrář containing /rāř/, and others. The phonological word /přiřTšeno/ (přiřčeno) attests /řiř/ because it is not possible to parse the form to /při–řTšeno/, /řTš/ not being a possible pre-nuclear combination (/ř/ would have to occur here in ‘pre2’, not in ‘pre1’ where it belongs). Several combinations occur probably only in proper names such as Machek attesting /xeK/, Jób attesting /joP/, Sas attesting /saS/18, Lstěň attesting /ťeň/, and Kníř attesting /ňīř/. Although these words can only serve as indirect evidence, there is no reason to suppose that the given ‘CVC’ combinations are not possible in “normal” vocabulary of Czech. Empty cells in Figures IX and X indicate that no such ‘CVC’ combination has been found. They are: /ď_Ť/, /ď_ň/, /ň_ň/, /j_ň/, /j_ř/, /ř_r/, /r_r/ and /M_P/ (the last refers to the situation when a nuclear semiconsonant is preceded by /M/ as the only nasal which can precede it). The question we should now ask is whether their absence can be regularized or whether they are missing by accidence. Having found no such regularity, we opt for the latter. Moreover, /ď_Ť/ could be attested in /ďīťe/ if that is parsed to /ďīŤ– ťe/ with functional amalgamation (see Section 3.5). Similarly, /ď_ň/ could occur in /hoďiňe/ hodině if we assume the division /hoŤ–ďiň–ňě/; /ň_ň/ in /koňiňe/ konině (i.e. /koň–ňiň–ňe/); /j_ň/ in /kojeňī/ kojení (i.e. /koj–jeň–ňī/); /j_ř/ in zbrojíř (i.e. /Sbroj–jīř/); /ř_r/ in /přīroda/ příroda (i.e. /přīr–roT–da/); /r_r/ in /erār/ erár (i.e. /er–rār/). Admit-

17 There is also a surname Foff. 18 Unless we accept ses, a variant of se with -s shortened from jsi.

232 tedly, such divisions may be doubtful, as there is no evidence for them, but there is likewise no evidence they could not be possible. Thus, to sum it up, there is no restriction on the combinability of pre-nuclear and post-nuclear phonemes within a phonotagm. That is, in the combination ‘C1VC2’ the selection of ‘C1’ does not limit or determine the selection of ‘C2’ and vice versa: the selection of ‘C2’ does not limit or determine the selection of ‘C1’. A different matter is preferences. It is obvious that Czech prefers to avoid the combinations /ď_Ť/, /ď_ň/, /ň_ň/, /j_ň/, /j_ř/, /ř_r/, /r_r/, which have not been found, but also the combinations /ť_ň/, /ň_n/, /ň_ř/, /x_K/, /g_P/, /j_P/, which have been found in symbols (such as proper or place names) only. Some other combinations are furthermore little frequent, especially those involving /g/ and /f/ or combinations involving the same phonemes (e.g. /n_n/, /l_l/). The very last thing to look at is the ability of pre-nuclear combinations to occur with post-nuclear combinations within a single phonotagm. To begin with, it is obvious that if e.g. the combination /j_ř/ has not been found, then there is no phonotagm which contains a pre-nuclear combination ending in /j/ (such as /Skvj/) and a post-nuclear combination beginning with /ř/ (such as /řT/), that is, no phonotagm like /SkvjeřT/. However, this does not necessarily mean it is not possible. The point is to find out whether certain types of pre-nuclear combinations cannot occur with certain types of post-nuclear combinations. One such regularity was noted in Section 6.1: No phonotagm in Czech can contain more than six peripheral phonemes. Thus, if a nuclear phoneme is preceded by four peripheral phonemes, it cannot be followed by three peripheral phonemes because the phonotagm type ‘CCCCVCCC’ is impossible; likewise, if a nuclear phoneme is preceded by five peripheral phonemes, it cannot be followed by two or more peripheral phonemes because the phonotagm types ‘CCCCCVCC’ and ‘CCCCCVCCC’ are also impossible. It remains to be seen whether any other restrictions can be introduced.

233 /P/ /F/ /T/ /S/ /Ť/ /Š/ /K/ /X/ /p/ pop páv pod pas pouť piš puk puch /b/ bab byv bot bos buď buš bok bůh /f/ xenofob šéfův fit fis foť doufáš fík foch /v/ šváb lvův vod vaz voď výš vak vah /m/ map domov mít maz meť myš mák much /t/ top niv tát tas teď tuž tak tah /d/ dob dav dát dóz vždyť dáš dík duh /s/ sob sov sít (Sas) síť suš sok sáh /z/ zub název zad zas záď vozíš vozík vrazích /n/ neb nov nad nás nať nůž nok nach /ť/ vtip návštěv tít tis štiť tíž tik tíh 19 /ď/ ) div dít děs díž dík zdích /š/ šíp šéf šat šos šeď koušeš šik šach /ž/ žup živ žít užas složitě lžeš žok žeh /ň/ hňup hněv nit kněz niť níž vznik nich /k/ kup kov kat kaz káď kuš skok řekách /g/ (Egypt) kolegův fregat vagus fregatě langoš pedagog ligách /x/ chyb chov chod chas choď chýš (Machek) plochých /h/ hub háv had has hať jehož hák hoch /j/ (Jób) jev jít jas jeď již jak jich /ř/ řep řev řád říz řiť řež řek hřích /r/ ryb rév rád ráz raď růž rak roh /l/ líp lev let les leť lež lok líh /M/ mrv smrdutý mls mrť mlž mrkat mlh

Figure IX: Examples of words where ‘CVC’ combinations are attested. The nuclear phoneme is irrelevant. Continued in Figure X.

19 Attested only in pravděpodobně.

234 /m/ /n/ /ň/ /j/ /ř/ /r/ /l/ /p/ pum pán peň pyj spoř par pel /b/ bomb bon báň boj boř bor bol /f/ fám fén fuň trofej šafář fór faul /v/ vám ven voň voj vař vor val /m/ mám min míň máj miř mor mír /t/ tom ten tůň taj čtyř tur týl /d/ dam den daň dej dař dar dál /s/ sám sen saň sej tesař sýr sel /z/ zem zón vězeň zuj zář vzor zel /n/ nám nán naň hnůj snář nor nul /ť/ tím stín (Lštěň) chtěj rytíř štír těl 20 /ď/ dím hodin děj hmoždíř děr děl 21 /š/ šum výšin vášeň šíj šíř šár šel /ž/ lžím žen žeň žij lyžař žer žal 22 /ň/ ním (Knín) něj (Kníř) knír snil /k/ kým kin kůň kyj keř kur kal 23 /g/ gum gen kolegyň reaguj ) figur gel 24 /x/ uchem tchán kuchyň nechej tchoř chýr chyl /h/ jhem han hoň háj hoř her hol /j/ jím jen jej jer jel /ř/ třem křen dřeň hřej přiřčeno přál /r/ rám ran roň ráj cukrář ryl /l/ lom lan laň lej sklář lir lil 25 /M/ mrmlat mln mrně – – – –

Figure X: Continuation of Figure IX.

20 Also in the place name Bdín, /Pďīn/. 21 Also in the place name Všeň, /Fšeň/. 22 Also in zvěřejněn provided that it is parsed to /zveř–řej–ňen/. 23 Attested only in marketingář (SN1). 24 Also in the vulgarism chuj, /xuj/. 25 Gen. sg. of mlna (SSJČ).

235 SUMMARY AND CONCLUSION

The phonotactics of the phonemes of Czech, whose inventory was discussed in Chapter 2, was accounted for by the distributional unit, a notion introduced in Chapter 4 and applied to Czech in Chapter 5. It is an abstract network of positions, and positions are divisions within a phonotactic construction (defined with related notions in Chapter 3) corresponding to immediate constituents (phonemes and archiphonemes) between which various functional and occurrence dependencies hold. A position may also be empty if a phoneme occurring there can be omitted without violating the well- formedness (but not the identity) of the construction. A phoneme functionally superior and not dependent on the occurrence of any other one (and not omissible) is a nuclear entity standing in nuclear position. A phoneme dependent on another entity for its phonotactic function and for its occurrence is a peripheral, non-nuclear entity standing in peripheral, non-nuclear position. There is generally only one nuclear position within a distributional unit, at least in Czech. An instance of a distributional unit where all positions are either filled with a phoneme (one at a time) or left empty is a phonotagm. In the course of our analysis it proved necessary to operate with two types of distributional unit: the main distributional unit and the semi-distributional unit; the reasons were given in Chapter 6. Moreover, it was necessary in Chapter 8 to recognize several accidental appendices, these being phonemes with peculiar and in a way accidental distribution falling outside the distributional unit. The overall model of the phonotactics of Czech consists, as schematized in Figure I, of the main distributional unit which is essential, of the semi-distributional unit which may be optionally prefixed to the main distributional unit, and of an accidental appendix which may also be optionally prefixed either to the main distributional unit or to the semi-distributional unit. Examples of phonological forms where an accidental appendix is prefixed to the semi-distributional unit have not been found, but they are possible. The phonological form /Strom/ strom is an instance of the main distributional unit without any prefix; /obMňen/ obměn is an in-

236 stance of the main distributional unit corresponding to /Mňen/ together with a semi- distributional unit corresponding to /ob/; /StřMen/ střmen, which is a variant of třmen, is then an instance of the main distributional unit prefixed with an accidental appendix /S/. The other accidental appendix is /z/ at the beginning of /zMdlelī/ zmdlelý, which is a variant of zemdlelý. /S/ and /z/ occur, as it were, accidentally in these words.

( Accidental appendix ) ( Semi-distributional unit ) Main distributional unit

Figure I: The overall model of the phonotactics of Czech consisting of the obligatory main distributional unit and an optional semi-distributional unit and an optional accidental appendix.

The main distributional unit accounts for the structure of self-contained phonotactic entities, i.e. of major-type phonotagms. A major-type phonotagm is either a single phoneme in phonological forms like /a/ a or a combination of phonemes; in either case a phonotagm is a phonotactic entity which is autonomous and not dependent on any other phonotactic entity; as a rule, it corresponds to phonotactic bases of single diaereme groups (“phonological words”). The main distributional unit comprise nine positions; one of them is nuclear (marked ‘n’), five are pre-nuclear (‘e1’, ‘e2’, ‘e3’, ‘pre1’, ‘pre2’) and three post-nuclear (‘i1’, ‘i2’, ‘i3’). The nuclear position must be obligatorily occupied by a phoneme (a vowel), whereas the non-nuclear positions can be optionally occupied with a phoneme (a consonant), i.e. they can be empty. It was explained in Chap- ter 5 that under certain circumstances the phonotactic difference between some of these positions becomes redundant resulting in four archi-positions, two for the pre-nuclear context (‘E2’, ‘E3’), one for the post-nuclear context (‘I’), and one stretching over the pre-nuclear and nuclear context (‘N’). Archi-positions are always occupied by a phoneme. Archi-position ‘N’ is occupied by the semiconsonants /r/, /l/ when they function as nuclear entities; in that case they must be obligatorily preceded by a least one consonant. Two types of pre-nuclear positions have been recognized: explosive (‘e1’, ‘e2’, ‘e3’, ‘E2’, ‘E3’) and pre-explosive (‘pre1’, ‘pre2’). The latter are occupied by a phoneme if and only if at least one of the explosive positions is occupied by a phoneme; otherwise they are empty. In the majority of cases the pre-explosive positions are filled

237 with voicing archiphonemes. The occurrence of phonemes in particular positions and archi-positions was discussed in Chapters 7, 8 and 9. The set of phonemes capable of occurring in the same position constitutes a position class; the members of the same position classes share the same phonotactic properties, i.e. the same distributional and combinational potentials. The individual position classes of the main distributional unit are given in Figure II. The cells stretching over two or three positions are position classes correspond to the archi-positions which are intersections of two or more adjacent positions. The symbol ‘∅’ means a position can be empty. All pre-nuclear and post- nuclear positions, except for the archi-positions which are never empty, can be empty as in the phonotagm /a/ a. This phonotagm is formally equal to /∅∅∅∅∅a∅∅∅/ where every ∅ is potentially replaceable by a phoneme. All pre-nuclear positions can be filled with a phoneme as in /FSkvjeT/ vzkvět, and all post-nuclear positions can be filled with a phoneme as in /vojSK/ vojsk. However, we concluded in Chapter 6 that it is impossible that all pre-nuclear and post-nuclear positions be filled with a phoneme at the same time because no phonotagm in Czech can contain more than six peripheral phonemes. The idea of empty positions, though not unique in linguistic descriptions, allows for a novel analysis of major-type phonotagms. A more traditional analysis similar to the one previously offered by Kučera (1961) is given in Appendix E.

Figure II: Main distributional unit and its position classes. Phonemes stretching over several positions belong to an archi-position. ∅ means a position can be empty.

As explained in Chapter 6, certain phonotagm-like combinations of phonemes, col- lectively called minor-type phonotagms, must be described upon a special type of the distributional unit called the semi-distributional unit. Minor-type phonotagms are dependent on major-type phonotagms (i.e. they cannot occur separately). However, except for this occurrence dependency, they look like major-type phonotagms. Their structure

238 is exhaustively accounted for by the semi-distributional unit which consists of seven positions. Six of them were identified with those of the main distributional unit: pre- nuclear ‘e1’, ‘e2’, ‘e3’, ‘pre1’, ‘pre2’, and nuclear ‘n’ (the difference between these positions can again be canceled to result in archi-positions ‘E3’, ‘E2’ and ‘N’). The sev- enth position is specific to the semi-distributional unit; it is one and the only post- nuclear position ‘im’. If a semi-distributional unit (underlying all minor-type phonotagms) is attached to the main distributional unit (underlying all major-type phonotagms), only positions ‘e2’ or ‘e1’ are occupied by a phoneme in the main distributional unit, the other pre-nuclear positions being empty. This is a consequence of there being no neutralization of voicing before a phoneme occurring in ‘e2’ or ‘e1’. The absence of neutralization of voicing is a feature specific and characterizing position ‘im’ of the semi-distributional unit because it can be occupied by both voiceless and voiced consonants, whereas the post-nuclear positions of the main distributional unit can only be occupied by voicing archiphonemes, i.e. by consonants which are neither voiceless nor voiced. The structure of the semi-distributional unit and its position classes is given in Figure III.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘n’ ‘im’ k g x h P T K v r l t d F S Š X T M p b f s z š ž m n ň M j r l S Š ∅ i e a o u t d v t h ř ∅ j ř ť ď r l ∅ ī ē ā ō ū (b) ∅ p b f ë ä ö k g h x ∅ ť ď

Figure III: Semi-distributional unit and its position classes. Phonemes stretching over several positions belong to an archi-position. ∅ means a position can be empty.

It is obvious that certain phonemes can occur in more than one position. Vowels occur only in the nuclear position of either the main distributional unit or the semi- distributional unit. The consonants /š/, /ž/ occur only in one position, namely in ‘e3’. The other consonants occur in two or more positions, including the post-nuclear position ‘im’ of the semi-distributional unit. The semiconsonants /r/, /l/ also occur in more positions. The occurrence of peripheral phonemes in particular peripheral positions is given in Figure IV. Nuclear phonemes are not included, as the vowels occur only in po-

239 sition ‘n’, and the nuclear semiconsonants only in archi-position ‘N’. Cells which stretch horizontally over two or more positions stand for an archi-position; thus, /r/ can occur in post-nuclear archi-position ‘I’ which corresponds to the intersection of ‘i2’ and ‘i3’. Cells which stretch vertically over two or more phonemes stand for archiphonemes. For example, /S/ corresponding to the intersection of /s/ and /z/ occurs in position ‘pre1’; neither /s/ nor /z/ occurs here. Note that the figure does not show the occurrence of /t/, /h/, /v/ and /b/ in ‘pre2’ on which see Section 8.7.

‘pre2’ ‘pre1’ ‘e3’ ‘e2’ ‘e1’ ‘i1’ ‘i2’ ‘i3’ ‘im’ /ř/ + + + /j/ + + + /r/ + + + + /l/ + + + + /m/ + + + /n/ + + + + + /ň/ + + + /x/ + + + + /h/ + + /š/ + + + + + /ž/ + /s/ + + + + + /z/ + /t/ + + + + + + + /d/ + + /k/ + + + + + + /g/ + + /p/ + + + + + + /b/ + + /f/ + + + + /v/ + /ť/ + + + /ď/ + +

Figure IV: Occurrence of peripheral phonemes in peripheral positions. Shading and the plus sign mean a phoneme occurs in a given position. Cells stretching horizontally over two or more positions stand for an archi-position. Cells stretching vertically over two or more phonemes stand for archiphonemes.

It is possible to redraw the main distributional unit into the two schemes reproduced in Appendix C. The first is for the pre-nuclear context; the second for the post-nuclear context. We have omitted the situation when a semiconsonant is the nucleus of a phono-

240 tagm in which case it occupies archi-position ‘N’. The schemes of this kind has been been called freight yard schemes and it is quite an apt name.1 Imagine they represent maps of a route of a train. Every position can be viewed as a station where a wagon is hitched to the train. Every train must have a machine. This is the nuclear entity occurring in the nuclear position. A train can carry as much as five wagons if it goes through the route in the first scheme, and as much as three wagons if it goes through the route in the second scheme. In other words, major-type phonotagms in Czech can begin with as much as five peripheral phonemes (cf. /FSkvjeT/ vzkvět) and end in as much as three peripheral phonemes (cf. /pomST/ pomst). Now, at every station a wagon is not only hitched to the train, but can also be loaded with some goods, although it can remain empty in some stations. That is, all those five pre-nuclear and all those three post- nuclear positions can be filled with a phoneme, but some positions may remain empty. In both schemes the lines represent routes the train can take. If we start in station ‘n’, the machine can go either to station ‘e1’ or to the station marked ‘E3’, that is, to the first archi-position. In the latter case the train loads /ť/ or /ď/ skipping stations ‘e1’, ‘e2’ and ‘e3’, and going directly to station ‘pre1’. In station ‘E3’ the wagon must always be loaded, i.e. an archi-position is always occupied by a phoneme. But if the train goes to station ‘e1’, the wagon it takes can be loaded with /m/, /n/, /ň/, /j/, /ř/, /r/ or /l/, or it can remain empty. Then it can go either to station ‘e2’ or to station ‘E2’, that is, to the second archi-position. Once again, if choosing station ‘E2’, it skips stations ‘e2’ and ‘e3’, and goes straight to station ‘pre1’. In station ‘E2’ it loads /p/ or /b/ or /f/. If the train goes instead to station ‘e2’, the wagon can be loaded with /v/ or /M/ or remain empty. It continues then to station ‘e3’ where it can take on /t/, /d/, /s/, /z/, /š/, /ž/, /k/, /g/, /x/ or /h/, or the wagon can remain empty. Either way, the train continues to station ‘pre1’ where it could get even if it went through stations ‘E3’ or ‘E2’. At station ‘pre1’ the wagon can be loaded with /T/, /S/, /Š/ or /ř/ or it can remain empty; it can load a phoneme in this station only if it has loaded at least one at the previous stations. Finally, the train continues to station ‘pre2’, the end of its route; the last wagon can be loaded with /P/, /T/, /K/, /F/, /S/, /Š/, /X/, /r/, /l/, /j/ or /M/, and in some cases even with /t/, /h/, /v/ or

1 The term has been adopted from Hockett (1958: 290–1); see also Goldsmith (2009: 21–2) on finite state automata.

241 /b/, but again only if it loaded at least one wagon in the previous stations. The wagon could also remain empty, though. If it the wagon in station ‘pre1’ remains empty, the wagon in station ‘pre2’ cannot be loaded with /T/, /S/ or /Š/ because these phonemes can be primarily loaded in the former, that is, ‘pre1’ is the primary place of their occurrence. The similar procedure applies to the post-nuclear context in the second scheme in Appendix C. The situation is a little more intricate here, but the peculiarities are not in- corporated into the scheme even though it would be possible. Here, the train can take up no more than three wagons. If the wagon is loaded with /P/, /T/, /K/, /m/, /n/, /ň/ in station ‘i1’ or if it remains empty there2, it cannot go to station ‘I’ (the archi-position); it can only go there if and only if the wagon in ‘i1’ is loaded with /j/, /r/ or /l/. Moreover, the conditions under which the wagon in station ‘i2’ can be loaded with /P/, /T/, /K/ are restricted: if the wagon in station ‘i1’ was left empty, the wagon in station ‘i2’ cannot be loaded by these phonemes unless the wagon in station ‘i3’ is loaded with /T/ or /Š/. Similarly, the wagon in station ‘i3’ can be loaded with /S/, /Š/ only if the wagon in station ‘i2’ was not left empty. If the wagons in stations ‘i1’ and ‘i2’ were left empty, the only phoneme that can be loaded in station ‘i3’ is /Ť/. The whole thing is, of course, not so simple. Although every position (“wagon”) can be occupied by one phoneme from a given class of phonemes, it does not mean that the positions are occupied at will. If a phoneme occurs in a certain position, it may happen that another phoneme cannot occur in another position because these two phonemes cannot co-occur. Co-occurrences of nuclear phonemes with peripheral phonemes were discussed particularly in Chapter 13; co-occurrences of various peripheral phonemes were discussed in Chapter 8, 9, 11 and 12. Such limitations, called collocational restrictions, are summarized in Appendix A. They function as filters to the output of the freight-yard schemes sorting out ill-formed combinations of phonemes. Consequently, the structure of the distributional unit and the collocational restrictions ensure that the output of the schemes will not only be well-formed combinations of phonemes, but also all well-formed combinations, that is, all major-type phonotagm in Czech. In particular,

2 That is, if position ‘i1’ is occupied by /P/, /T/, /K/, /m/, /n/, /ň/ or if it is empty; we still hope the reader understands the simile.

242 the output of the schemes is all pre-nuclear and post-nuclear combinations of phonemes of which there are as many as 431 pre-nuclear and as many as 101 post-nuclear combinations (their properties are discussed in Chapter 10, and their inventories are given in Chapters 11 and 12 and in Appendix B). If we run through the freight-yard schemes applying the collocational restrictions, we will get any and all of these combinations. The distributional unit thus provides generalizations about well-formed and ill-formed sequences of phonemes in Czech. It is a formal device that generates only well-formed sequences, and these sequences may be either actually attested as phonological forms of Czech words or potential candidates for phonological form of words (such as /FSdřīT/, /tvjeT/ or /Xřber/). To conclude, our analysis of the phonotactics of Modern Standard Czech has reached all three levels of adequacy (in the sense of Chomsky 1964: 28ff.). It provides an exhaustive account of the constituency of attested phonotactic constructions in terms of occurrences and combinations of phonemes in these constructions, which makes it ob- servationally adequate. It formalizes and generalizes conditions of these occurrences and combinations, which makes it descriptively adequate. And finally, which makes it explanatorily adequate, it provides a principled basis for predicting possible phonotactic constructions and for ruling out the impossible ones.

243

APPENDICES Appendix A COLLOCATIONAL RESTRICTIONS

This appendix summarizes all collocational restrictions introduced in the work. The numbers in square brackets refer to sections where a restriction was first introduced. The restrictions are purposely given in a rather condensed form. If a restriction states that an occlusive, a nasal and a semiconsonant cannot be followed by more than three peripheral phonemes, it does not mean that these phonemes constitute a well-defined set (cf. Section 3.4), but that an occlusive cannot be followed by more than three peripheral phonemes, and a nasal cannot be followed by more than three peripheral phonemes, and a semiconsonant cannot be followed by more than three peripheral phonemes.

Restrictions applicable to the whole phonotagm

No geminate restrictions RG1a Two identical phonemes cannot stand in close proximity. [5.4] RG1b A voiceless and voiced occlusive, or a voiceless and voiced fricative cannot stand in close proximity with a voicing archiphoneme of the same place and manner of articulation. [5.4] Restrictions on the number of peripheral phonemes RG2a No phonotagm can contain more than six peripheral phonemes. [6.1] RG2b No phonotagm can contain more than five peripheral phonemes if /r/ or /l/ is its nucleus. [6.1] Restrictions on the combinability of peripheral phonemes RG3 An alveolar fricative cannot be preceded by a palatal fricative. [10.2] RG4 A palatal occlusive cannot be preceded by an alveolar occlusive. [10.2] RG5 A velar occlusive cannot be preceded by a labial occlusive. [10.2]

245 RG6 Two or more velars cannot stand in close proximity. [10.2] RG7 Three or more occlusives cannot stand in close proximity. [10.1] RG8 A nasal cannot be preceded by /b/. [11.2 sub DE2c] Restrictions on the number of following phonemes RG9 A velar cannot be followed by more than two peripheral phonemes. [8.7] RG10a An occlusive, a nasal and a semiconsonant cannot be followed by more than three peripheral phonemes. [8.7] RG10b A palatal cannot be followed by more than three peripheral phonemes. [8.7] RG11 A semiconsonant and a nasal can be followed by no other three peripheral phonemes than of type ‘FON’. [10.5] RG12 /j/ and /ř/ cannot be followed by more than two peripheral phonemes. [8.6, 8.7] Restrictions on the combinability of /j/ RG13a /j/ cannot be preceded by a phoneme from pos ‘e3’ unless it is an alveolar fricative. [8.1] RG13b If /j/ is preceded by an alveolar fricative, the latter is never preceded by any other peripheral phoneme. [8.1] Restrictions on the combinability of phonemes in the same section of the distributional unit1 RG14a /ř/ cannot occur in the same section with a semiconsonant. [8.1] RG14b /ř/ cannot occur in the same section with a palatal fricative. [11.2 sub DE2b] RG15 The same section cannot contain two identical sonants. [8.1] RG16 The pre-nuclear section cannot contain two instances of a semiconsonant. [8.1] RG17 If a semiconsonant occurs in the same section with /j/, the semiconsonant occurs in a more peripheral position.2 [8.1] Restrictions on types of peripheral combinations3 RG18a A peripheral combination of or including type ‘ONR’ is impossible. [10.4] RG18b A peripheral combination of or including type ‘NRC’ is impossible. [10.4] RG18c A peripheral combination of or including type ‘FRF’ is impossible. [10.4]

1 The same section refers here either to the pre-nuclear or to the post-nuclear context. Archi- position ‘N’ belongs both to the pre-nuclear context and to the nuclear context. 2 The more peripheral position is the one more distant from the nucleus. 3 The symbol ‘C’ without any subscript is any peripheral phoneme in these restrictions.

246 RG18d A peripheral combination of or including type ‘NFR’ is impossible. [10.4] RG18e A peripheral combination of or including type ‘RON’ is impossible. [10.4] RG18f A peripheral combination of or including type ‘NCN’ is impossible. [10.4] 4 RG18g A peripheral combination of or including type ‘AOLFC1’ where ‘C1’ is a nasal or /ř/ or a semiconsonant is impossible. [11.3 sub DE3a]

RG18h A peripheral combination of or including type ‘KFC1’ where ‘C1’ is a nasal or /ř/ or a semiconsonant is impossible. [11.3 sub DE3a]

RG18i A peripheral combination of and including type ‘C1FN’ where ‘C1’ is a semiconsonant or /ř/ is impossible. [11.3 sub DE3b] RG19a A peripheral combination of or including type ‘CFFN’ is impossible. [10.5] RG19b A peripheral combination of or including type ‘CFNR’ is impossible. [10.5] RG19c A peripheral combination of or including type ‘OFFC’ is impossible. [10.5] RG19d A peripheral combination of or including type ‘COFN’ or ‘OFNC’ is impossible. [10.5] RG19e A peripheral combination of or including type ‘OFFF’ is impossible. [10.5]

RG20 Two occlusives cannot be followed by two peripheral phonemes ‘C2C1’ where

‘C2’ agrees in the manner of articulation with ‘C1’. [10.5]

Restrictions on the co-occurrence of peripheral and nuclear phonemes

Restriction applicable to all nuclear phonemes RN1 A nuclear phoneme cannot be preceded by a peripheral combination of type ‘NO’ irrespective of whether these phonemes are further preceded by other peripheral phonemes. [10.3] Restrictions applicable to vowels RN2 A long vowel cannot be preceded by more than four peripheral phonemes. [7.1] RN3 A long mid vowel cannot be preceded by a palatal occlusive or the palatal nasal. [13.1]

4 ‘AO’ is an alveolar occlusive, ‘LF’ is a labial fricative.

247 RN4a A long and a diphthongal vowel cannot be preceded by a combination of type 5 ‘CnCC1’ where ‘C1’ is a palatal. [13.3] RN4b A long and a diphthongal vowel cannot be preceded by a combination of type

‘CnCC1’ where ‘C1’ is /j/. [13.3] RN5 A diphthongal vowel cannot be preceded by more than three peripheral phonemes. [7.1] RN6 A diphthongal vowel cannot be followed by /j/. [13.2] Restrictions applicable to nuclear semiconsonants RN7a A nuclear semiconsonant must be preceded by at least one consonant. [7.2] RN7b If a nuclear semiconsonant occurs at the end of a phonological word, they cannot be preceded by three or more peripheral phonemes. [7.3] RN8 A nuclear semiconsonant cannot stand in close proximity with a combination of or including type ‘OO’. [13.3]

RN9a A nuclear semiconsonant cannot be preceded by a combination of type ‘NCn’. [8.7] RN9b A nuclear semiconsonant cannot be preceded by a combination of type 6 ‘CCAF’. [8.3] RN9c A nuclear semiconsonant cannot be preceded by a combination of type ‘CCK’. [8.3] RN9d A nuclear semiconsonant cannot be preceded by a combination of or including type ‘ON’. [13.3]

RN9e A nuclear semiconsonant cannot be preceded by a combination of type ‘CnFN’. [13.3] RN9f A nuclear semiconsonant cannot be preceded by a combination of or including type ‘LP’ or ‘KP’ or ‘KA’. [13.3] RN9g A nuclear semiconsonant cannot be preceded by a combination of type

‘CCAF’. [8.3] RN9h A nuclear semiconsonant cannot be preceded by a combination of type ‘CCK’. [8.3]

5 ‘Cn’ is any number of peripheral phonemes. 6 ‘AF’ is an alveolar fricative.

248 RN9i A nuclear semiconsonant cannot be preceded by a combination of type or in-

cluding type ‘OC1’ where ‘C1’ is from pos ‘E2’. [8.6] RN10a A nuclear semiconsonant cannot be followed by /j/ or /ř/. [9.1, 9.2] RN10b A nuclear semiconsonant cannot be followed by a combination of or including type ‘NF’. [13.3] RN10c A nuclear semiconsonant cannot be followed by a combination of including types ‘LC’ or ‘KC’. [9.2] RN10d A nuclear semiconsonant cannot be followed by a combination of type ‘CCO’. [9.3]

Restrictions applicable to the pre-nuclear context only

Restrictions on the combinability of pre-nuclear phonemes RE1 A labial cannot be preceded by a nasal. [10.2] RE2 A labial fricative cannot be preceded by another labial. [10.2] RE3 A fricative other than labial cannot be preceded by a velar fricative. [10.2] RE4 An alveolar occlusive cannot be preceded by a velar fricative. [10.2] RE5 A sonant cannot stand in close proximity with /ř/. [8.1] RE6 A son ant and a nasal cannot be followed by two occlusives or by two fricatives. [10.4] Restrictions pertaining to pre-nuclear position classes RE7a A non-nuclear semiconsonant7 and /ř/ cannot be preceded by more than three peripheral phonemes. [8.1] RE7b A non-nuclear semiconsonant and /ř/ cannot be preceded by a combination of

type ‘CCC1’ where ‘C1’ is from pos ‘e2’. [8.1]

RE8 An alveolar fricative cannot be followed by a combination of type ‘C2C1’

where ‘C2’ is from pos ‘e2’ and ‘C1’ from pos ‘e1’ and at the same time preceded by a phoneme from ‘pre1’. [8.3]

7 I.e. when occurring in ‘e1’ as opposed to when occurring in archi-position ‘N’.

249 RE9 A palatal fricative cannot be followed by a combination of type ‘C2C1’ where

‘C2’ is from pos ‘e2’ and ‘C1’ is from pos ‘e1’. [8.3]

RE10a If /ř/ is followed by a combination of type ‘C1C’, ‘C1’ can be no other phoneme than one from pos ‘e2’. [8.6] RE10b /ř/ cannot be followed by a fricative other than labial (i.e. it can only be followed by a fricative from pos ‘e2’). [11.2 sub DE2g] 8 RE11a A nasal and a sonant cannot be followed by a combination of type ‘CC1(C)’

where ‘C1’ is from pos ‘E2’ or from pos ‘e2’. [8.7]

RE11b A nasal and a sonant cannot be followed by a combination of type ‘CCC1(C)’

where ‘C1’ is from pos ‘e2’. [8.7]

RE12 An occlusive cannot be followed by a combination of type ‘C2C1’ where ‘C2’ is

from pos ‘E2’ or from pos ‘e2’, and ‘C1’ is from pos ‘e1’. [8.6]

RE13a A velar cannot be followed by a combination of type ‘C1C’ where ‘C1’ is either from pos ‘e3’ or from pos ‘e2’ or from pos ‘E2’. [8.7]

RE13b A labial occlusive cannot be followed by a combination of type ‘C1C’ where

‘C1’ is from pos ‘e3’ or pos ‘e2’ or from pos ‘E2’. [8.7] RE14 A non-nuclear semiconsonant9 cannot be followed by a voiceless fricative. [10.3] RE15 /m/ cannot be preceded by a phoneme from pos ‘E2’. [8.7] RE16 /j/ cannot be followed by a phoneme from pos ‘E2’. [8.7]

Restrictions applicable to the post-nuclear context only

Restrictions on the combinability of post-nuclear phonemes RI1 An occlusive, a fricative and a nasal cannot be followed by a semiconsonant or by a nasal. [9.1] RI2a An occlusive cannot be followed by an occlusive other than alveolar. [10.2]

8 ‘(C)’ is an optional peripheral phoneme. 9 I.e. when occurring in ‘pre2’.

250 RI2b An occlusive cannot be followed by a labial fricative or by a velar fricative. [10.2] RI3 Two occlusives or two fricatives cannot be followed by any other peripheral phoneme. [10.4] RI4 A fricative cannot be preceded by a labial fricative or by a palatal fricative. [10.2] RI5a A labial cannot be preceded by another labial unless it is a labial nasal. [10.2] RI5b A labial cannot be preceded by a palatal or by a velar. [10.2] RI6 A labial fricative, a velar fricative and /ř/ cannot be preceded by a peripheral phoneme and at the same time followed by a peripheral phoneme. [9.2] RI7 An alveolar fricative cannot stand in close proximity with any fricative. [10.2] RI8 /ř/ cannot stand in close proximity with a fricative or with a nasal. [10.2] Restriction on types of post-nuclear combinations 10 RI9 A combination of type ‘OPFO’ is impossible. [12.4 sub DI3]

10 ‘O’ is an occlusive and ‘PF’ a palatal fricative

251 Appendix B PERIPHERAL COMBINATIONS AND THEIR PROPERTIES

This appendix provides a complete listing of 431 pre-nuclear and 101 post-nuclear combinations analyzed in this work; they are tabulated in Chapters 11 and 12 and also in Bičan (2011a) where they are arranged according to their length. Every combination is given a series of analytical properties; they are discussed in Chapter 10.

Explanations: COLUMN Com The combinations Parenthesized combinations are attested only in symbols and other marginal words. Italicized combinations are attested also before or after nuclear semiconsonants. MT Manner type O = occlusive F = fricative N = nasal R = sonant PT Place type L = labial A = alveolar P = palatal K = velar I = isolated consonant or semiconsonant DT Distributional type E Expandability L = left-hand expandable R = right-hand expandable L! = left-hand expandable, but not immediately expandable R! = right-hand expandable, but not immediately expandable X = left-hand or right-hand non-expandable (X) = marginally left-hand or right-hand non-expandable - = left-hand or right-hand anti-expandable R Reducibility L = left-hand reducible R = right-hand reducible X = left-hand or right-hand (depending on position) non-reducible - = left-hand or right-hand anti-reducible

252 Rs Resolvability P = partly resolvable C = completely resolvable - = resolvability not applicable P Pairedness (not applicable to post-nuclear combinations) P = paired S = single - = pairedness not applicable V Voicing of the combination (not applicable to post-nuclear combinations) + = voiced combination – = voiced combination M = mixed combination I = indifferent combination M Mirror effect Y = has a mirror counterpart (Y) = has a mirror counterpart in a marginal combination N = does not have a mirror counterpart Example Examples of words where a combination is attested. If attested only in PSJČ or SSJČ, it is mentioned.

Pre-nuclear combinations

Com MT PT DT E R RsP V M Example /Ml/ NR II DE2a L- -- - - I Y mlok /Mn/ NN IA DE2a L- -- - - I N mnout si /Mň/ NN IP DE2a L- -- - - I N město /Mr/ NR II DE2a L- -- - - I Y mrak /Mř/ NR II DE2a L- -- - - I N mříže (/Mj/) NR II DE2a X- -- - - I (Y) Mjachký (surname) /vj/ FR LI DE2a L- -- - P + Y vědět /vl/ FR LI DE2a L- -- - P + Y vlast /vm/ FN LL DE2a XR -- - S + Y vmísit se /vn/ FN LA DE2a LR -- - S + Y vnuk /vň/ FN LP DE2a LR -- - P + N vnější /vr/ FR LI DE2a L- -- - P + Y vrata /vř/ FR LI DE2a L- -- - S + N vřít /dl/ OR AI DE2b L- -- - P + Y dlouho /dm/ ON AL DE2b LR -- - P + Y dmout /dn/ ON AA DE2b XR -- - P + Y dnes /dň/ ON AP DE2b XR -- - P + N dni (pl. of den) /dr/ OR AI DE2b L- -- - P + Y drahý /dř/ OR AI DE2b L- -- - P + Y dřevo /gl/ OR KI DE2b L- -- - P + Y glazúra /gn/ ON KA DE2b XX -- - P + Y gnoseologický /gr/ OR KI DE2b L- -- - P + Y granule /hl/ FR KI DE2b L- -- - P + Y hlava

253 Com MT PT DT E R RsP V M Example /hm/ FN KL DE2b LX -- - P + Y hmat /hn/ FN KA DE2b LX -- - P + N hnát /hň/ FN KP DE2b LX -- - P + N hnít /hr/ FR KI DE2b L- -- - P + Y hrách /hř/ FR KI DE2b L- -- - P + N hřešit /kl/ OR KI DE2b L- -- - P – Y klobouk /km/ ON KL DE2b LX -- - S – N kmet /kn/ ON KA DE2b LX -- - P – Y knot /kň/ ON KP DE2b LX -- - S – Y kniha /kr/ OR KI DE2b L- -- - P – Y krádež /kř/ OR KI DE2b L- -- - S – Y kříž /sj/ FR AI DE2b X- -- - P – Y sjezd /sl/ FR AI DE2b L- -- - P – Y sláma /sm/ FN AL DE2b LR -- - P – Y smát se /šm/ FN PL DE2b LX -- - P – Y šmíra /sn/ FN AA DE2b LR -- - P – Y snář /sň/ FN AP DE2b LR -- - P – N snít /sr/ FR AI DE2b X- -- - P – Y srub /šl/ FR PI DE2b L- -- - P – Y šlépěj /šn/ FN PA DE2b XX -- - P – Y šnek /šň/ FN PP DE2b LX -- - P – N šněrovat /sř/ FR AI DE2b X- -- - P – N sřezat /šr/ FR PI DE2b X- -- - P – Y šramot /tl/ OR AI DE2b L- -- - P – Y tlak /tm/ ON AL DE2b LR -- - P – Y tma /tn/ ON AA DE2b L!R-- - P – Y tnout /tň/ ON AP DE2b L!R-- - P – N tni (imper. sg. of tnout) /tr/ OR AI DE2b L- -- - P – Y tráva /tř/ OR AI DE2b L- -- - P – Y třeba /xl/ FR KI DE2b L- -- - P – Y chlap /xm/ FN KL DE2b LX -- - P – Y chmury /xn/ FN KA DE2b LX -- - P – N from nad-chnout /xň/ FN KP DE2b LX -- - P – N chňapnout /xr/ FR KI DE2b L- -- - P – Y chrápat /xř/ FR KI DE2b L- -- - P – N chřípí /zj/ FR AI DE2b X- -- - P + Y zjev /zl/ FR AI DE2b L- -- - P + Y zlatý /žl/ FR PI DE2b X- -- - P + Y žlutý /zm/ FN AL DE2b LR -- - P + Y zmatek /žm/ FN PL DE2b XX -- - P + Y žmoulat /zn/ FN AA DE2b LR -- - P + Y znak /zň/ FN AP DE2b LR -- - P + N zničit /žn/ FN PA DE2b XX -- - P + Y žnout /žň/ FN PP DE2b XX -- - P + N žně /zr/ FR AI DE2b L- -- - P + Y zrak /zř/ FR AI DE2b L- -- - P + N zřejmě

254 Com MT PT DT E R RsP V M Example /žr/ FR PI DE2b X- -- - P + Y žrát /bj/ OR LI DE2c L- -- - P + Y běloba /bl/ OR LI DE2c L- -- - P + Y bláto /br/ OR LI DE2c L- -- - P + Y bratr /bř/ OR LI DE2c L- -- - P + N bříza /fj/ FR LI DE2c X- -- - P – Y fěrtoch (SSJČ) /fl/ FR LI DE2c L- -- - P – Y flám /fň/ FN LP DE2c X- -- - P – N fňukat /fr/ FR LI DE2c L- -- - P – Y frak /pj/ OR LI DE2c L- -- - P – Y pět /pl/ OR LI DE2c L- -- - P – Y plat /pn/ ON LA DE2c L!- -- - S – N pnout /pň/ ON LP DE2c L!- -- - S – N pni (imper. sg. of pnout) /pr/ OR LI DE2c L- -- - P – Y práce /př/ OR LI DE2c L- -- - P – N příbor /rm/ RN IL DE2e -X -- - - I N rmoutit (/lj/) RR II DE2e ------I Y Ljuba (proper name) (/rj/) RR II DE2e ------I Y Rjazaň (for. place name) /jm/ RN IL DE2e -R -- - - I N jméno /ln/ RN IA DE2e -X -- - - I N lnout /lň/ RN IP DE2e -X -- - - I N lněný /dv/ OF AL DE2f LR -- - P MY dva /gv/ OF KL DE2f XX -- - P MY guajak (SSJČ) /hv/ FF KL DE2f XR -- - P MN hvozd /kv/ OF KL DE2f LR -- - P MY kvést /sM/ FN AI DE2f X- -- - P – N smrt /sv/ FF AL DE2f LR -- - P MN svatba, svrbět /šM/ FN PI DE2f X- -- - S – N šmrnc /šv/ FF PL DE2f LR -- - P MN švagr, Švrček (surname) /tv/ OF AL DE2f LX -- - P MY tvůj, tvrz /xv/ FF KL DE2f LR -- - P MN chvíle /zM/ FN AI DE2f X- -- - P + N zmrznout /zv/ FF AL DE2f LR -- - P MN zvát, zvrhlík /žv/ FF PL DE2f XX -- - P MN žvanit /řv/ RF IL DE2g (X)R-- - S + N řvát /rv/ RF IL DE2h -R -- - S + N rvát /lv/ RF IL DE2h -X -- - S + N lvi (pl. of lev) /vM/ FN LI DE2h -- -- - S + Y vmrštit /řk/ RO IK DE2i XX -- - S – Y řka (archaic form of říct) /Sd/ FO AA DE2i LR -- - P + Y zdát, zdržet /Sg/ FO AK DE2i XR -- - P + Y zgalvanizovat /Sh/ FF AK DE2i LR -- - P + N shon, shrnout /Sk/ FO AK DE2i LR -- - P – Y skákat, skrz /Sš/ FF AP DE2i XX -- - P – N zšednout /St/ FO AA DE2i LR -- - P – Y stát, strmý /Sx/ FF AK DE2i LR -- - P – N schod, zchrtlý (PSJČ)

255 Com MT PT DT E R RsP V M Example /Sž/ FF AP DE2i XX -- - P + N zženštilý /Šk/ FO PK DE2i LR -- - S – Y škola, škrt /Št/ FO PA DE2i LR -- - P – Y štáb, štrbské /Šd/ FO PA DE2i LX -- - P + Y ždáti (SSJČ) /Šh/ FF PK DE2i XR -- - S + Y žhář /Th/ OF AK DE2i XX -- - P + N dharma (SSJČ) /Tk/ OO AK DE2i LR -- - S + Y tkát /Ts/ OF AA DE2i LR -- - P – Y cena, crčet /Tš/ OF AP DE2i LR -- - P – N čelo, črta /Tx/ OF AK DE2i XX -- - P – N tchoř /Tz/ OF AA DE2i XX -- - P + Y dzinkati (SSJČ) /Tž/ OF AP DE2i LX -- - P + Y džus /Fd/ FO LA DE2j -R -- - P + N vdát se /Fg/ FO LK DE2j -X -- - P + N vgalvanisovati (PSJČ) /Fh/ FF LK DE2j -R -- - P + N vhodit, vhrnout /Fk/ FO LK DE2j -R -- - P – N vkus /Fs/ FF LA DE2j -R -- - P – N vsadit, vsrknout /Fš/ FF LP DE2j -R -- - P – N všechno /Ft/ FO LA DE2j -R -- - P – N vtom, vtrhnout /Fx/ FF LK DE2j -R -- - P – N vcházet, vchrstnout /Fz/ FF LA DE2j -R -- - P + N vzít /Fž/ FF LP DE2j -R -- - P + N vžít se /jd/ RO IA DE2j -X -- - S + N jde (3rd person sg. of jít) /jh/ RF IK DE2j -X -- - S + N jho /js/ RF IA DE2j -R -- - S – N jsi (2nd person sg. of být) /Kd/ OO KA DE2j -X -- - P – N kdo /Ks/ OF KA DE2j -X -- - S – Y xylofon /Kš/ OF KP DE2j -R -- - S – N kšandy /Kt/ OO KA DE2j -X -- - P + N který /lh/ RF IK DE2j -X -- - S + N lhostejný /lk/ RO IK DE2j -X -- - S – N lkát /lz/ RF IA DE2j -X -- - S + N lze /lž/ RF IP DE2j -X -- - S + N lži (pl. of lež) /Mš/ NF IP DE2j -X -- - P – N mše /Mz/ NF IA DE2j -X -- - S + N mze (SSJČ) /Mž/ NF IP DE2j -X -- - P + N mžitky /Ps/ OF LA DE2j -R! -- - P – Y psát /Pš/ OF LP DE2j -R! -- - P – N pšenice /Pt/ OO LA DE2j -X -- - S + N pták /Px/ OF LK DE2j -X -- - S – N pcháč /Pz/ OF LA DE2j -X -- - P + Y bzukot /Pž/ OF LP DE2j -X -- - P + N bžunda /rd/ RO IA DE2j -X -- - P + N rdousit /rt/ RO IA DE2j -R -- - P – N rty (pl. of ret) /rz/ RF IA DE2j -X -- - S + N rzi (pl. of rez) /rž/ RF IP DE2j -X -- - S + N ržát

256 Com MT PT DT E R RsP V M Example /Mh/ NF IK DE2j -X -- - S + N mhouřit /řb/ RO IL DE2k LX -- - S + Y řbuchan1 /Sb/ FO AL DE2k LR -- - P + Y sbírat, zbrklý /Sf/ FF AL DE2k XR -- - S – N sfouknout /Sp/ FO AL DE2k LR -- - P – Y spát, sprcha /Šb/ FO PL DE2k LR -- - P + (Y) žbublati, žbrdolek2 /Šp/ FO PL DE2k LR -- - P – (Y) špatný, šplhat /Tb/ OO AL DE2k XX -- - S – Y dbát /Fb/ FO LL DE2l -R -- - P + N vbodnout, vbrzku /Fp/ FO LL DE2l -R -- - P – N vpadnout, vpršeti (PSJČ) (/Kf/) OF KL DE2l -X -- - S – Y Kfely (Czech place name) /Kb/ OO KL DE2l -X -- - S – N kbelík /lb/ RO IL DE2l -X -- - P + N lbový (SSJČ) /lp/ RO IL DE2l -R -- - P – N lpí (3rd pers. sg. of lpět) /Xb/ FO KL DE2l -X -- - S + N hbitý /Sť/ FO AP DE2m L- -- - P – N stín /Sď/ FO AP DE2m L- -- - P + N zdědit /Šť/ FO PP DE2m L- -- - P – N šťavnatý /Šď/ FO PP DE2m X- -- - P + N ždímat /Pď/ OO LP DE2n -- -- - (P)– N bdít (/Pť/) OO LP DE2n -- -- - P + N Ptice (Czech place name) (/Kť/) OO KP DE2n -- -- - S + N Ktiš (Czech place name) /Fť/ FO LP DE2n -- -- - P – N vtělený /Fď/ FO LP DE2n -- -- - P + N vděčný /jď/ RO IP DE2n -- -- - S + N jdi (imper. sg. of jít) /rď/ RO IP DE2n -- -- - P + N rdít se /rť/ RO IP DE2n -- -- - P – N rtěnka /Xť/ FO KP DE2n -- -- - S – N chtít /dMň/ ONN AIP DE3a X- LR C P + N dměte (imp. pl. of dmout) /dvj/ OFR ALI DE3a X- LR C S M N dvě (fem. of dva) /hvj/ FFR KLI DE3a L- LR C P M N hvězda /kvj/ OFR KLI DE3a L- LR C S M N květ /sMl/ FNR AII DE3a X- LR C P – N smlouva /sMň/ FNN AIP DE3a L- LR C P – N směna /sMr/ FNR AII DE3a X- LR C P – N smrad /svj/ FFR ALI DE3a X- LRC P M N svět /svl/ FFR ALI DE3a X- LRC P M N svlak /svr/ FFR ALI DE3a X- LRC P M N svrab /tMň/ ONN AIP DE3a L- LR C P – N tmě (loc. sg. of tma) /xvj/ FFR KLI DE3a L- LR C P M N chvět /zMl/ FNR AII DE3a X- LR C P + N zmlátit /zMn/ FNN AIA DE3a X- LR C S + N zmnohonásobit /zMň/ FNN AIP DE3a L- LR C P + N změna

1 Machek (1957: 256). 2 Both attested in SSJČ.

257 Com MT PT DT E R RsP V M Example /zMr/ FNR AII DE3a X- LR C P + N zmrazit /zMř/ FNR AII DE3a X- LR C S + N zmřížiti (PSJČ) /zvj/ FFR ALI DE3a L- LR C P M N zvěd /zvl/ FFR ALI DE3a X- LRC P M N zvlášť /zvn/ FFN ALA DE3a X- LR C S M N zvnaditi (PSJČ) /zvň/ FFN ALP DE3a X- LR C S M N zvnějšku /zvr/ FFR ALI DE3a X- LRC P M N zvracet /zvř/ FFR ALI DE3a X- LRC S M N zvředovatět /řvj/ RFR ILI DE3b X- LRC S + N řvěte (imper. pl. of řvát) /vMň/ FNN LIP DE3c -- LR C S M N vměstnat se /vMl/ FNR LII DE3c -- LR C S M N vmlátiti (SSJČ) /rvj/ RFR ILI DE3c -- LR C S + N rvěte se3 /jMň/ RNN IIP DE3c -- LR C - I N jmění /Sdl/ FOR AAI DE3d X- LR C P + N zdlouhavý /Sdm/ FON AAL DE3d LX LR C P + N sdmýchnouti (PSJČ) /Sdr/ FOR AAI DE3d X- LR C P + Y zdraví /Sdř/ FOR AAI DE3d X- LR C P + N zdřímnout si /Sgl/ FOR AKI DE3d X- LR C P + N zglajšaltovat /Sgr/ FOR AKI DE3d X- LR C P + Y zgranulovat /Shl/ FFR AKI DE3d L- LR C P + N zhloupnout /Shm/ FFN AKL DE3d XX LR C P + N zhmotnit se /Shn/ FFN AKA DE3d XX LRC P + N zhnusit /Shň/ FFN AKP DE3d XX LRC P + N zhnít /Shr/ FFR AKI DE3d X- LR C P + N zhroutit se /Shř/ FFR AKI DE3d L- LR C P + N zhřešit /Skl/ FOR AKI DE3d L- LR C P – N sklon /Skm/ FON AKL DE3d XX LR C S – N skmotřiti se (SSJČ) /Skn/ FON AKA DE3d XX LR C S – Y sknotiti se (SSJČ) /Skň/ FON AKP DE3d XX LR C S – N zknižnění (PSJČ) /Skr/ FOR AKI DE3d X- LR C P – Y zkrachovat /Skř/ FOR AKI DE3d L- LR C S – N zkřížit /Sšl/ FFR API DE3d X- LRC S – N zšlechtiti (PSJČ) /Stl/ FOR AAI DE3d L- LR C P – Y stlačit /Stm/ FON AAL DE3d LR LR C P – N stmívat se /Str/ FOR AAI DE3d L- LR C P – Y ztratit se /Stř/ FOR AAI DE3d L- LR C P – N střádat /Sxl/ FFR AKI DE3d L- LR C P – N schlíplý /Sxm/ FFN AKL DE3d XX LR C P – N zchmuřovat /Sxn/ FFN AKA DE3d XX LRC P – N schnout /Sxň/ FFN AKP DE3d XX LRC P – N schni4 /Sxr/ FFR AKI DE3d X- LR C P – N schránka

3 Imper. pl. of rvát se. 4 Imper. sg. of schnout.

258 Com MT PT DT E R RsP V M Example /Sxř/ FFR AKI DE3d X- LR C P – N zchřadnout /Štr/ FOR PAI DE3d L- LR C S – Y štrachat /Škr/ FOR PKI DE3d X- LR C S – N škrábat /Škl/ FOR PKI DE3d X- LR C S – N šklebit se /Šhn/ FFN PKA DE3d XX LRC S + N žhnout /Šhň/ FFN PKP DE3d XX LRC S + N žhni (imp. pl. of žhnout) /Tkl/ OOR AKI DE3d L- LR C S – N tklivý /Tkn/ OON AKA DE3d LX LR C S – Y tknout /Tkň/ OON AKP DE3d LX LR C S – N tkni (imper. sg. of tknout) /Tsl/ OFR AAI DE3d X- LR C S – Y clona /Tšl/ OFR API DE3d L- LR C S – N člověk /Tsm/ OFN AAL DE3d XX LR C S – Y cmunda /Tšm/ OFN APL DE3d XX LR C S – Y čmárat /Tšň/ OFN APP DE3d XX LR C S – N čnět /Txn/ OFN AKA DE3d XX LR C S – N dchnouti (SSJČ) /Txň/ OFN AKP DE3d XX LR C S – N dchni5 /Fdl/ FOR LAI DE3e -- LR C P + N vdláždit /Fdm/ FON LAL DE3e -X LR C P + N vdmýchnouti (SSJČ) /Fdr/ FOR LAI DE3e -- LR C P + N vdrápati se (SSJČ) /Fhl/ FFR LKI DE3e -- LR C P + N vhled /Fhm/ FFN LKL DE3e -X LR C S + N vhmatávati se (PSJČ) /Fhň/ FFN LKP DE3e -X LR C S + N vhníst /Fhr/ FFR LKI DE3e -- LR C S + N vhrabat se /Fhř/ FFR LKI DE3e -- LR C S + N vhřížiti (SSJČ) /Fkl/ FOR LKI DE3e -- LR C S – N vklouznout /Fkr/ FOR LKI DE3e -- LR C S – N vkrádat se /Fkř/ FOR LKI DE3e -- LR C S – N vkřiknouti (SSJČ) /Fsl/ FFR LAI DE3e -- LR C P – N vsled (SSJČ) /Fšl/ FFR LPI DE3e -- LR C S – N všlehati (SSJČ) /Fsm/ FFN LAL DE3e -R LR C P – N vsmýknouti (SSJČ) /Fsň/ FFN LAP DE3e -R LR C P – N vsníti (PSJČ) /Fšr/ FFR LPI DE3e -- LR C P – N všroubovat /Ftl/ FOR LAI DE3e -- LR C P – N vtlačit /Ftm/ FON LAL DE3e -X LR C P – N vtmelit /Ftr/ FOR LAI DE3e -- LR C P – N vtrousit /Ftř/ FOR LAI DE3e -- LR C S – N vtřískat /Fxl/ FFR LKI DE3e -- LR C P – N vchlípenina /Fzl/ FFR LAI DE3e -- LR C P + N vzlétnout /Fzm/ FFN LAL DE3e -R LR C P + N vzmužit se /Fzn/ FFN LAA DE3e -R LR C S + N vznášet se /Fzň/ FFN LAP DE3e -R LR C P + N vznik /Fzr/ FFR LAI DE3e -- LR C P + N vzruch /Fzř/ FFR LAI DE3e -- LR C S + N vzření /Fžr/ FFR LPI DE3e -- LR C P + N vžráti se (SSJČ)

5 Imper. sg. of dchnouti (SSJČ).

259 Com MT PT DT E R RsP V M Example /jsm/ RFN IAL DE3e -X LR C S – N jsme (1st pers. pl. of být) /Mdl/ NOR IAI DE3e -- LX P S + N mdlý /Sbj/ FOR ALI DE3f L- LR C P + N sběr /Sbl/ FOR ALI DE3f X- LR C P + N zbláznit se /Šbl/ FOR PLI DE3f X- LR C P + N žbluňkat /Sbr/ FOR ALI DE3f X- LR C P + N zbrázděný /Sbř/ FOR ALI DE3f L- LR C P + N zbřidličnatět /Sfl/ FFR ALI DE3f X- LRC S – N zflákati (SSJČ) /Sfr/ FFR ALI DE3f X- LRC S – N zfruktizovati (SSJČ) /Spj/ FOR ALI DE3f L- LR C P – N spěchat /Spl/ FOR ALI DE3f L- LR C P – N splav /Spr/ FOR ALI DE3f L- LR C P – N správa /Spř/ FOR ALI DE3f L- LR C P – N spřízněný /Špr/ FOR PLI DE3f X- LR C P – N šproch /Šbr/ FOR PLI DE3f X- LR C P + N žbrunda /Špl/ FOR PLI DE3f X- LR C P – N šplouchat /Fbj/ FOR LLI DE3g -- LR C P + N vběhnout /Fbl/ FOR LLI DE3g -- LR C P + N vblízku (SSJČ) /Fbr/ FOR LLI DE3g -- LR C P + N vbruslit /Fbř/ FOR LLI DE3g -- LR C P + N vbřednouti (SSJČ) /Fpj/ FOR LLI DE3g -- LR C P – N vpěchovat /Fpl/ FOR LLI DE3g -- LR C P – N vplout /Fpr/ FOR LLI DE3g -- LR C P – N vpravit /Fpř/ FOR LLI DE3g -- LR C P – N vpřed /lpj/ ROR ILI DE3g -- LR C S – N lpět /Sdv/ FOF AAL DE3i XX LRC P M N zdvojit /Skv/ FOF AKL DE3i XR LRC S M N skvost, skrvna /Stv/ FOF AAL DE3i XX LRC P M N stvořit, stvrdit /Sxv/ FFF AKL DE3i LR LRC S M N schválit /Škv/ FOF PKL DE3i XX LRC S M N škvařit se, škvrně /Štv/ FOF PAL DE3i LX LRC S M N štvát /Tkv/ OOF AKL DE3i LR LR C S M N tkví (3rd pers. sg. of tkvět) /Tšv/ OFF APL DE3i XX LRC S M N čvachtat /Tsv/ OFF AAL DE3i LX LRC S M N cválat, cvrnkat /Ftv/ FOF LAL DE3j -X LR C S M N vtvořiti (SSJČ) /Fzv/ FFF LAL DE3j -R LR C S M N vzvolati (SSJČ) /Fkv/ FOF LKL DE3j -X LR C S M N vkvapiti (PSJČ) (/břv/) ORF LIL DE3k -X LR C S M N Břve (Czech place name) /třM/ ORN AII DE3k -R XR P S – N třmen /hřM/ FRN KII DE3k -R XR P S + N hřmí (3rd p. sg. of hřmět) /Fřt/ FRO LIA DE3l -X XRP S – N from ote-vřte /FSd/ FFO LAA DE3l -R LR C P + N vzdálit se /FŠd/ FFO LPA DE3l -R LR C S + N vždy /FSh/ FFF LAK DE3l -R LR C P + N vzhůru /FSk/ FFO LAK DE3l -R LR C S – N vzkázat /FŠk/ FFO LPK DE3l -R LR C S – N vškatulkovati (SSJČ)

260 Com MT PT DT E R RsP V M Example /FSš/ FFF LAP DE3l -R LR C S – N vzšuměti (SSJČ) /FSt/ FFO LAA DE3l -R LR C P – N vstát, vstrčit /FSx/ FFF LAK DE3l -R LR C P – N vzchopit se /FTk/ FOO LAK DE3l -X LR C S – N vtkát /FTs/ FOF LAA DE3l -X LR C S – N vcelku /FTš/ FOF LAP DE3l -R LR C S – N včera, včrtati (PSJČ) /jSt/ RFO IAA DE3l -X LRC S – N jste (2nd pers. pl. of být) /Křt/ ORO KIA DE3l -X XR P S – N křtu (gen. sg. of křest) (/TŠk/) OFO APK DE3l -X LR C S – N Čkyně (Cz. place name) /MSd/ NFO IAA DE3l -X LX P P + N mzda /MSt/ NFO IAA DE3l -R LX P P – N msta (/MTs/) NOF IAA DE3l -X LX P S – N Mcely (Cz. place name) /rTs/ ROF IAA DE3l -X LR C S – N rci (archaic form of říct) /rTš/ ROF IAP DE3l -X LR C S – N rčení (/lTš/) ROF IAP DE3l -X LX P S – N Lčovice (Cz. place name) /STk/ FOO AAK DE3l -R LR C S – N stkát /STs/ FOF AAA DE3l -R LR C S – N scestný /STš/ FOF AAP DE3l -X LR C P – N zčeřit /STž/ FOF AAP DE3l -X LR C P + N zdžezovati (SSJČ) /ŠTk/ FOO PAK DE3l -R LR C S – N štkát /TŠt/ OFO APA DE3l -R LR C S – N čte6, čtrnáct /Xřt/ FRO KIA DE3l -X XR P S – N chřtán /XTs/ FOF KAA DE3l -X LX P S – N chce (3rd p. sg. of chtít) /XTš/ FOF KAP DE3l -X LX P S – N chčije7 /FSb/ FFO LAL DE3m -R LR C P + N vzbudit /FSp/ FFO LAL DE3m -R LR C P – N vzpažit, vzprchnouti8 /Třp/ ORO AIL DE3m -X XR P S – N třpit /TŠb/ OFO APL DE3m -X LR C P + N džbán /TSp/ OFO AAL DE3m -X LR C (P)– Y cpát se (/TSb/) OFO AAL DE3m -X LR C P + Y Dzbel (Cz. place name) /TŠp/ OFO APL DE3m -R LR C P – N čpí (3rd pers. sg. of čpět) /Xřb/ FRO KIL DE3m -X LR C S + N hřbitov /PSď/ OFO LAP DE3n -- LR C S + N bzdíti (SSJČ) (/PŠť/) OFO LPP DE3n -- LR C S – N Pština (Cz. place name) /FSď/ FFO LAP DE3n -- LR C P + N vzdělaný /FSť/ FFO LAP DE3n -- LR C P – N vstěhovati (SSJČ) /FŠť/ FFO LPP DE3n -- LR C S – N vštípit /Křť/ ORO KIP DE3n -- XR P S – N křtiny /KŠť/ OFO KPP DE3n -- LR C S – N kštice /lSť/ RFO IAP DE3n -- LX P S – N lstivý

6 3rd person sg. of číst. 7 3rd person sg. of chcát. 8 Attested in PSJČ.

261 Com MT PT DT E R RsP V M Example (/lŠť/) RFO IPP DE3n -- LX P S – N Lštěň (Cz. place name) /MSď/ NFO IAP DE3n -- LX P P + N mzdě (loc. sg. of mzda) /MSť/ NFO IAP DE3n -- LX P P – N mstít se /Třť/ ORO AIP DE3n -- XR P S – N třtina /TSť/ OFO AAP DE3n -- LR C S – N ctí (3rd pers. sg. of ctít) /TŠť/ OFO APP DE3n -- LR C S – N čti (imper. sg. of číst) /Tkvj/ OOFR AKLI DE4a X- LR C S M N tkvět /StMň/ FONN AAIP DE4a X- LR C S – N stmělý (SSJČ) /Skvj/ FOFR AKLI DE4a L- LR C S M N skvělý /Shvj/ FFFR AKLI DE4a X- LX P P M N zhvězditi (SSJČ) /Sxvj/ FFFR AKLI DE4a X- LR C P M N schvělý (SSJČ) /FsMň/ FFNN LAIP DE4b -- LR C P – N vsměstnati (PSJČ) /Fzvj/ FFFR LALI DE4b -- LR C S M N vzvěděti (PSJČ) /FzMň/ FFNN LAIP DE4b -- LR C P + N vzměť (PSJČ) /třMň/ ORNN AIIP DE4c -- XR P S – N třměti (SSJČ) /hřMň/ FRNN KIIP DE4c -- XR P S + N hřmět /FSdm/ FFON LAAL DE4d -X LR C S + N vzdmouti se (PSJČ) /FShl/ FFFR LAKI DE4d -- LR C P + N vzhlížet /FShř/ FFFR LAKI DE4d -- LR C S + N vzhřížiti (PSJČ) /FSkl/ FFOR LAKI DE4d -- LR C S – N vzklíčit /FSkř/ FFOR LAKI DE4d -- LR C S – N vzkřísit /FŠkr/ FFOR LPKI DE4d -- LR C S – N vškrábat se /FStl/ FFOR LAAI DE4d -- LR C S – N vztlak /FStř/ FFOR LAAI DE4d -- LR C S – N vstřícný /FSxl/ FFFR LAKI DE4d -- LR C P – N vzchlípiti (PSJČ) /FTšl/ FOFR LAPI DE4d -- LR C S – N včlenit /lStm/ RFON IAAL DE4d -X LX P S – N lstmi (instr. pl. of lest) /lStn/ RFON IAAA DE4d -X XX P S – N lstný /lStň/ RFON IAAP DE4d -X XX P S – N lstně /MStn/ NFON IAAA DE4d -X XR P S – N mstný (SSJČ) /MStň/ NFON IAAP DE4d -X XR P S – N mstně (PSJČ) /PStr/ OFOR LAAI DE4d -- LX P S – N pstruh /PStř/ OFOR LAAI DE4d -- LX P S – N pstřeň (PSJČ) /PŠtr/ OFOR LPAI DE4d -- LX P S – N pštros /STkl/ FOOR AAKI DE4d -- LR C S – N ztklivět /ŠTkn/ FOON PAKA DE4d -X LR C S – N štknout /ŠTkň/ FOON PAKP DE4d -X LR C S – N štkni (imp. sg. of štknout) /TStm/ OFON AAAL DE4d -- LX P S – N ctmi (instr. pl. of čest) /TStn/ OFON AAAA DE4d -- LX P S – N ctný (SSJČ) /TStň/ OFON AAAP DE4d -- LX P S – N ctně /FSbj/ FFOR LALI DE4e -- LR C P + N vzběhnouti (SSJČ) /FSbř/ FFOR LALI DE4e -- LR C P + N vzbředlý (PSJČ) /FSpj/ FFOR LALI DE4e -- LR C P – N vzpěrač /FSpl/ FFOR LALI DE4e -- LR C S – N vzplanout

262 Com MT PT DT E R RsP V M Example /FSpň/ FFON LALP DE4e -- XR P S – N vzpni se9 /FSpr/ FFOR LALI DE4e -- LR C S – N vzpruha /FSpř/ FFOR LALI DE4e -- LR C P – N vzpřímený /TŠpj/ OFOR APLI DE4e -- XR P S – N čpět /FSkv/ FFOF LAKL DE4f -R LR C S M N vzkvétat /FSxv/ FFFF LAKL DE4f -X LR C S M N vzchvívati se (PSJČ) /TŠtv/ OFOF APAL DE4f -X LR C S M N čtvero, čtvrtek /STsv/ FOFF AAAL DE4f -X LR C S M N scvaknout, scvrknout /STkv/ FOOF AAKL DE4f -X LR C S M N stkví10 /FSkvj/ FFOFR LAKLI DE5 -- LR C S M N vzkvět /STkvj/ FOOFR AAKLI DE5 -- LR C S M N stkvěl11

Post-nuclear combinations

Com MT PT DT E R RsM Example /jF/ RF IL DI2a -X -- - Y sejf /jK/ RO IK DI2a -X -- - N laik /jP/ RO IL DI2a -X -- - Y knajp (gen. pl. of knajpa (SSJČ)) /jS/ RF IA DI2a -R -- - Y rorejs /jŠ/ RF IP DI2a -X -- - N jejž /jT/ RO IA DI2a -R -- - N prejt /jX/ RF IK DI2a -X -- - N cejch (/jř/) RR II DI2a -X -- - N Kejř (surname) /lF/ RF IL DI2a -X -- - Y salv (gen. pl. of salva) /lK/ RO IK DI2a -X -- - Y kalk /lP/ RO IL DI2a -X -- - Y alb (gen. pl. of album) /lS/ RF IA DI2a -R -- - Y pulz /lT/ RO IA DI2a -R -- - Y kolt /lX/ RF IK DI2a -X -- - Y valch (gen. pl. of valcha) (/lŠ/) RF IP DI2a -X -- - Y Tylš (surname) /mF/ NF LL DI2a -X -- - Y nymf (gen. pl. of nymfa) /mP/ NO LL DI2a -R -- - N lamp (gen. pl. of lampa) /mS/ NF LA DI2a -R -- - Y říms (gen. pl. of římsa) /mŠ/ NF LP DI2a -R -- - Y jímž /mT/ NO LA DI2a -X -- - Y vikomt (SSJČ) /mX/ NF LK DI2a -X -- - Y střemch (gen. pl. of střemcha) /nF/ NF AL DI2a -X -- - Y tonf (gen. pl. of tonfa (SN1)) /nK/ NO AK DI2a -R -- - Y tank, from cvrnk-at /nS/ NF AA DI2a -R -- - Y tranz

9 Imper. sg. of vzpnouti si (SSJČ). 10 3nd person sg. of stkvíti se (SSJČ). 11 3rd person sg. of stkvíti se, past tense (SSJČ).

263 Com MT PT DT E R RsM Example /nŠ/ NF AP DI2a -X -- - Y jenž /nT/ NO AA DI2a -R -- - Y moment /ňK/ NO PK DI2a -X -- - Y šizuňk (SSJČ) /rF/ RF IL DI2a -X -- - Y harf (gen. pl. of harfa) /rK/ RO IK DI2a -R -- - Y park /rP/ RO IL DI2a -R -- - Y orb (gen. pl. of orba) /rS/ RF IA DI2a -R -- - Y kurz /rŠ/ RF IP DI2a -X -- - Y verš /rT/ RO IA DI2a -R -- - Y nárt /rX/ RF IK DI2a -X -- - Y arch /Př/ OR LI DI2a X- -- - Y pepř /PS/ OF LA DI2a XR -- - Y zips /Tř/ OR AI DI2a -X -- - Y dovnitř /TS/ OF AA DI2a LR -- - Y pec, mudrc /Kř/ OR KI DI2a X- -- - Y mokř (SSJČ) /KS/ OF KA DI2a LR -- - Y koks /jŤ/ RO IP DI2b -- -- - N pojď (imper. sg. of jít) (/lŤ/) RO IP DI2b -- -- - N bylť (archaic form) (/mŤ/) NO LP DI2b -- -- - N jsemť (archaic form) (/nŤ/) NO AP DI2b -- -- - N onť (archaic form) /rŤ/ RO IP DI2b -- -- - N žerď /jl/ RR II DI2c -- -- - (Y) koktejl /jm/ RN IL DI2c -- -- - (Y) Sejm (SSJČ), tajm (SN1) /jn/ RN IA DI2c -- -- - N hejn (gen. pl. of hejno) /jr/ RR II DI2c -- -- - (Y) chejr (SSJČ) /lm/ RN IL DI2c -- -- - Y jilm (/ln/) RN IA DI2c -- -- - Y Lincoln (surname) /rl/ RR II DI2c -- -- - N stárl (see Section 7.2) /rm/ RN IL DI2c -- -- - Y šarm /rn/ RN IA DI2c -- -- - Y koncern /rň/ RN IP DI2c -- -- - Y čerň /FK/ FO LK DI2d X- -- - Y tomahavk /FT/ FO LA DI2d X- -- - Y pravd (gen. pl. of pravda) /PT/ OO LA DI2d LX -- - Y skript (gen. pl. of skriptum) (/PŠ/) OF LP DI2d XX -- - Y Hybš (surname) /TŠ/ OF AP DI2d L- -- - Y meč, mlč /KŠ/ OF KP DI2d XX -- - Y jakžtakž /KT/ OO KA DI2d LX -- - Y akt /řK/ RO IK DI2d X- -- - Y ouřk (SSJČ), Pyřk (surname) /řT/ RO IA DI2d X- -- - Y buřt /SK/ FO AK DI2d L- -- - Y lesk, prsk (SSJČ) /ST/ FO AA DI2d L- -- - Y dost, prst /SŤ/ FO AP DI2d X- -- - N pusť (imper. sg. of pustit), prsť (/SP/) FO AL DI2d L- -- - Y výsp (gen. pl. of výspa) /ŠT/ FO PA DI2d L- -- - Y mošt, obršt (SSJČ) /ŠŤ/ FO PP DI2d X- -- - N poušť, smršť /XŠ/ FF KP DI2d X- -- - Y jejichž /XT/ FO KA DI2d X- -- - Y ksicht (SSJČ)

264 Com MT PT DT E R RsM Example /XŤ/ FO KP DI2d X- -- - N nechť /jSK/ RFO IAK DI3 -- LR C N vojsk (gen. pl. of vojsko) /jST/ RFO IAA DI3 -- LR C N gajst (SSJČ) /jTS/ ROF IAA DI3 -- LR C N AIDS12 /jTŠ/ ROF IAP DI3 -- LR C N půjč (imper. sg. of půjčit) /KST/ OFO KAA DI3 -- LR C N text /lTS/ ROF IAA DI3 -- LR C Y sulc /mST/ NFO LAA DI3 -- LR C Y pomst (gen. pl. of pomsta) /mŠT/ NFO LPA DI3 -- LR C Y kumšt /nKT/ NOO AKA DI3 -- LR C Y adjunkt /nTS/ NOF AAA DI3 -- LR C N princ, šmrnc /nTŠ/ NOF AAP DI3 -- LR C N pomeranč /PST/ OFO LAA DI3 -- LR C Y zábst /rKT/ ROO IKA DI3 -- LR C N infarkt /rPT/ ROO ILA DI3 -- LR C N excerpt (SSJČ) /rST/ RFO IAA DI3 -- LR C N verst (gen. pl. of versta) /rTS/ ROF IAA DI3 -- LR C Y herz /rTŠ/ ROF IAP DI3 -- LR C Y terč /TSP/ OFO AAP DI3 -- LR C N zácp (gen. pl. of zácpa) /TST/ OFO AAA DI3 -- LR C N péct (/lSK/) RFO IAK DI3 -- LR C N Tobolsk (foreign place name) (/lST/) RFO IAA DI3 -- LR C Y Holst (surname) (/lTŠ/) ROF IAP DI3 -- LR C N Telč (Czech place name) (/nKS/) NOF AKA DI3 -- LR C Y Sfinx (brand name) (/nSK/) NFO AAK DI3 -- LR C N Smolensk (foreign place name) (/PSK/) OFO LAK DI3 -- LR C N Vitebsk (foreign place name) (/rKS/) ROF IKA DI3 -- LR C Y Marx (surname) (/rSK/) RFO IAK DI3 -- LR C N Magnitogorsk (for. place name) (/TSK/) OFO AAK DI3 -- LR C N Kuzněck (foreign place name)

12 Pronounced as [ajʦ].

265 Appendix C FREIGHT-YARD SCHEMES

n i e a o u ī ē ā ō ū ë ä ö

e1 E3 m n ň ř j r l ∅

ť ď e2 E2 v M ∅

p b f

e3 t d s z š ž k g x h ∅

pre1 T S Š ř ∅

pre2 P T K F S Š X M j r l t h v (b) ∅

Scheme 1: Pre-nuclear context of the main distributional unit

266

n i e a o u ī ē ā ō ū ë ä ö

i1 P T K m n ň j r l ∅

i2 I P T K S Š F X ř ∅

m n ň r l

i3 P T K S Š Ť ∅

Scheme 2: Post-nuclear context of the main distributional unit

267 Appendix D DISTRIBUTION OF NUCLEAR PHONEMES

This appendix provides examples of words in which phonotactic properties of nuclear phonemes are attested; they are discussed in Chapter 7. Phonological representation is given only when necessary.

/i/ F0 F1 F2 F3 P0 i Ir Ind P1 ti /miŠ/ myš /ňiXŠ/ nichž (Minsk) 1 P2 zdi /pliŠ/ plyš /KsiXT/ ksicht /StinKT/ P3 vždy vznik /FStiTŠ/ vztyč P4 vstřik P5

/e/ F0 F1 F2 F3 P0 es erb P1 ne jet jenž verst P2 vře /vjeT/ vět /xřeST/ chřest P3 jsme /kvjeT/ květ /StřemX/ střemch P4 /lStňe/ lstně vzhled P5 /STkvjel/ stkvěl

/a/ F0 F1 F2 F3 P0 a až akt P1 na pak tank /farKT/2 P2 kra mrak start 3 P3 skla /StraX/ strach /StlaTŠ/ stlač /SxramST/ P4 vztlak P5

1 Extracted from instinkt. 2 Extracted from infarkt. 3 Extracted from schramstnout.

268 /o/ F0 F1 F2 F3 P0 o on orb P1 to tok most vojsk P2 sto /StoX/ stoh sport P3 sklo vzdor skvost P4 pštros /TStnoST/ ctnost P5

/u/ F0 F1 F2 F3 P0 u už uzd P1 tu tuž kurt kumšt P2 mnu stud špunt P3 mzdu vzmuž (se) /StluTŠ/ stluč P4 pstruh P5

/ī/ F0 F1 F2 F3 P0 P1 jí jít míst /řīTST/ říct P2 spí znít sníst /zřīTST/ zříct P3 tkví /SkrīT/ skrýt zhníst P4 stkví /lStnīX/ lstných /FStřīTS/ vstříc P5

/ē/ F0 F1 F2 F3 P0 ér P1 té fén lézt /pēTST/ péct P2 své sfér plést /StēTST/ stéct P3 mdlé mdlém vznést /svlēTST/ svlést P4 /TStnē/ ctné /TStnēm/ ctném vzkvést P5

/ā/ F0 F1 F2 F3 P0 ár P1 já náš nárt zábst P2 svá zdát zmást P3 mdlá /Xřtān/ chřtán střást P4 /TStnā/ ctná vzplát P5

/ō/ F0 F1 F2 F3 P0 ód P1 /tō/ fr. šapitó gól P2 /grō/ gros /xlōr/ chlór P3 P4 P5

269 /ū/ F0 F1 F2 F3 P0 úl úst /ūTST/ úct P1 kůl půst P2 rtů stůl vrůst P3 křtů stvůr vzrůst P4 P5

/ë/ F0 F1 F2 F3 P0 /ër/ eur P1 /lë/ leu4 /zëK/ fr. zeugma P2 P3 P4 P5

/ä/ F0 F1 F2 F3 P0 /äT/ aut P1 /tä/ tau /räT/ raut /gäTŠ/ gauč P2 /Tšä/ čau /SkäT/ skaut P3 P4 P5

/ö/ F0 F1 F2 F3 P0 P1 /mö/ mou /pöŤ/ pouť /pöŠŤ/ poušť P2 /dvö/ dvou /xröST/ chroust /tlöTST/ tlouct P3 /Mdlö/ mdlou /FplöT/ plout /StlöTST/ stlouct P4 /TStnö/ ctnout (/Škrön/)5 P5

/r/ F0 F1 F2 F3 P1 hr vrt prst P2 /Str/ from mistr šprt smršť /šMrnTS/ šmrnc P3 skvrn /FStrTŠ/ vstrč P4 /TŠtvrT/ čtvrt

/l/ F0 F1 F2 F3 P1 /hl/ from vrhl mlž plst P2 /Sxl/ from uschl šplh /sMlTŠ/ smlč P3 P4

4 The name of the Romanian currency. 5 Extracted from the place name Lanškroun.

270 Appendix E ALTERNATIVE ANALYSIS OF MAJOR-TYPE PHONOTAGMS

This appendix provides an alternative description of major-type phonotagms inspired by Haugen (1956b) and especially by the analysis of the Czech syllable by Kučera (1961), which is the first and, besides ours, the only analysis of this kind. The term place (borrowed from Fudge 1969 but used differently) refers here to relative placement of a phoneme within a phonotagm (relative to the nucleus) and has the same reference as Kučera’s position; our use of the latter is different. /d/ in /dva/ and /p/ in /ploT/ occur in the same place, i.e. have the same placement relative to the vowel, but they do not occur in the same position of the distributional unit. Nuclear phonemes occur in nuclear “peak” places; peripheral phonemes occur in pre-nuclear “onset” places and in post- nuclear “coda” places. Every non-nuclear place is indexed with a two-digit number. The first one refers to the length of a peripheral combination. /j/ occurs in ‘O51’ in /FSkvjeT/ vzkvět but in ‘O31’ in /svjeT/ svět. The second digit is the number of the place relative to the nucleus. The membership of the individual places follows on the next page. It means every member of a set can occur in a given place, although there are limitations on the co-occurrence of certain phonemes (see Appendix A).

“Peak” “Onset” places “Coda” places places

O55 O54 O53 O52 O51 C31 C32 C33

O44 O43 O42 O41 C21 C22 PW O33 O32 O31 C11 PR O22 O21

O11

271 Nuclear places:

‘PW’ = {/e/, /i/, /a/, /o/, /u/, /ē/, /ī/, /ā/, /ō/, /ū/, /ë/, /ä/, /ö/}

‘PR’ = {/r/, /l/} Pre-nuclear places:

‘O11’ = {/p/, /b/, /t/, /d/, /ť/, /ď/, /k/, /g/, /f/, /v/, /s/, /z/, /š/, /ž/, /x/, /h/, /m/, /n/, /ň/, /j/, /ř/, /r/, /l/, /M/}

‘O21’ = same as ‘O11’

‘O22’ = {/p/, /b/, /t/, /d/, /k/, /g/, /f/, /v/, /s/, /z/, /š/, /ž/, /x/, /h/, /j/, /ř/, /r/, /l/, /P/, /T/, /K/, /F/, /S/, /Š/, /X/, /M/}

‘O31’ = {/p/, /b/, /t/, /d/, /ť/, /ď/, /k/, /v/, /s/, /š/, /ž/, /x/, /h/, /m/, /n/, /ň/, /j/, /ř/, /r/, /l/, /M/}

‘O32’ = {/p/, /b/, /t/, /d/, /k/, /f/, /v/, /s/, /z/, /š/, /ž/, /x/, /h/, /ř/, /T/, /S/, /Š/, /M/} 1 ‘O33’ = {/b/, /t/, /d/, /k/, /v/, /s/, /z/, /x/, /h/, /j/, /ř/, /r/, /l/, /P/, /T/, /K/, /F/, /S/, /Š/, /X/, /M/}

‘O41’ = {/v/, /m/, /n/, /ň/, /j/, /ř/, /r/, /l/}

‘O42’ = {/p/, /b/, /t/, /d/, /k/, /v/, /s/, /š/, /x/, /h/, /M/}

‘O43’ = {/t/, /k/, /s/, /z/, /x/, /h/, /ř/, /T/, /S/, /Š/}

‘O44’ = {/t/, /h/, /l/, /P/, /T/, /F/, /S/, /Š/, /M/}

‘O51’ = {/j/}

‘O52’ = {/v/}

‘O53’ = {/k/}

‘O54’ = {/T/, /S/}

‘O55’ = {/F/, /S/} Post-nuclear places:

‘C11’ = {/P/, /T/, /Ť/, /K/, /F/, /S/, /Š/, /X/, /m/, /n/, /ň/, /j/, /ř/, /r/, /l/}

‘C21’ = {/P/, /T/, /K/, /F/, /S/, /Š/, /X/, /m/, /n/, /ň/, /j/, /ř/, /r/, /l/}

‘C22’ = {/P/, /T/, /Ť/, /K/, /F/, /S/, /Š/, /X/, /m/, /n/, /ň/, /ř/, /r/, /l/}

‘C31’ = {/P/, /T/, /K/, /m/, /n/, /j/, /r/, /l/}

‘C32’ = {/P/, /T/, /K/, /S/, /Š/}

‘C33’ = {/T/, /K/, /S/, /Š/}

1 /b/ occurs only in /břv/, attested in the place name Břve.

272 REFERENCES

Akamatsu, Tsutomu. 1988. The Theory of Neutralization and the Archiphoneme in Functional Phonology. Amsterdam – Philadelphia: John Benjamins Publishing Company. ―. 1992. Essentials of Functional Phonology. Louvain-La-Neuve: Peeters. Appel, Wilhelm. 1957–8. “Energiebasis – Artikulationsbasis”. Wiener slavistisches Jahrbuch 6.73–104. Arnold, Gordon F. 1955–6. “A phonological approach to vowel, consonant and syllable in modern French”. Lingua 5.253–87. ―. 1964. “Vowel and consonant: a phonological definition re-examined”. In Honour of Daniel Jones (eds. David Abercrombie et al.), 16–25. London: Longmans. Awedyk, Wiesław. 1975. The Syllable Theory and Old English Phonology. Wrocław et al.: Wydawnictwo Polskiej akademii nauk. Bagemihl, Bruce. 1991. “Syllable structure in Bella Coola”. Linguistic Inquiry 22:4.589–646. Bárkányi, Zsuzsanna – Kiss, Zoltán. 2010. “A phonetic approach to the phonology of v: A case study from Hungarian and Slovak”. Turbulent Sounds. An Interdisciplinary Guide (eds. Susanne Fuchs – Martine Toda – Margena Żygis), 103–42. Berlin – New York: de Gruyter. Bartoň, Tomáš et al. 2009. Statistiky češtiny. Praha: Nakladatelství Lidové noviny – Ústav Českého národního korpusu. Basbøll, Hans. 1994. “How to derive the sonority syllable from the prototypical peak”. Acta Linguistica Hafniensia 27:1.51–65. Bell, Alan. 1976. “The distributional syllable”. Linguistic Studies Offered to Joseph Greenberg, Vol. 2: Phonology (ed. Alphonse Juilland), 249–262. Saragota, Califor- nia: Anma Libri.

273 ―. 1978. “Syllabic consonants”. Universals of Human Language, Vol. 2: Phonology (ed. Joseph Greenberg), 153–201. Stanford, California: Stanford University Press. Bell, Alan – Hooper, Joan. 1978. “Issues and evidence in syllabic phonology”. Syllables and Segments (eds. Alan Bell – Joan B. Hooper), 3–22. Amsterdam – New York – Oxford: North-Holland Publishing Company. Bell, Alan – Saka, Mohamad M. 1983. “Reversed sonority in Pashto initial clusters”. Journal of Phonetics 11.259–75. Bičan, Aleš. 2008a. Phonematics of Czech: An axiomatic-functionalist view. Masaryk University. Unpublished PhDr. thesis, available online: . ―. 2008b. “Accent and diaereme in Czech”. La Linguistique 44:2.45–66. ―. 2008c. “Ne-volnost přízvuku v češtině”. Varia Slavica. Sborník příspěvků k 80. narozeninám Radoslava Večerky (eds. Ilona Janyšková – Helena Karlíková), 27–39. Praha: Nakladatelství Lidové noviny. ―. 2010a. “Phonematics of Czech”. La Linguistique 46:1.19–41. ―. 2010b. “Čím je Petr zvláštní aneb kombinovatelnost slabičného /r/ v češtině”. Kar- lík a továrna na lingvistiku (eds. Aleš Bičan et al.), 49–68. Brno: Host – Masarykova univerzita. ―. 2011a. “Distribution and combinations of Czech consonants”. Zeitschrift für Slawistik 56:2.153–71. ―. 2011b. “Structure of syllables in Czech”. Formalization of Grammar in Slavic Lan- guages (eds. Peter Kosta – Lilia Schürks), 13–28. Frankfurt am Main: Peter Lang. ―. 2011c. “Distribuční a kombinační vlastnosti /ř/ v současné češtině”. m*OST 2009. Österreichische StudierendenTagung für SlawistInnen. Austriacka Studencka Kon- ferencja Slawistyczna (ed. Dagmar Heeg), 36–42. München – Berlin: Verlag Otto Sagner. Blevins, Juliette. 1995. “The syllable in phonological theory”. The Handbook of Phono- logical Theory (ed. John Goldsmith), 206–44. Oxford: Blackwell. ―. 2003. “Independent nature of phonotactic constraints”. The Syllable in Optimality Theory (eds. Caroline Féry – Ruben van de Vijver), 375–403. Cambridge: Cam- bridge University Press.

274 ―. 2006. “Syllable: typology”. Encyclopedia of Language and Linguistics, Vol. 12 (ed. Keith Brown), 333–7. Amsterdam: Elsevier. Breen, Gavan – Pensalfini, Rob. 1999. “Arrernte: a language with no syllable onsets”. Linguistic Inquiry 30.1–25. Butskhrikidze, Marika. 2002. The Consonant Phonotactics of Georgian. Utrecht: LOT. Butt, Matthias. 1992. “Sonority and the explanation of syllable structure”. Linguistische Berichte 137.45–67. Bybee, Joan. 2001. Phonology and Language Use. Cambridge: Cambridge University Press. Cairns, Charles. 1988. “Phonotactics, markedness and lexical representation”. Phonol- ogy 5.209–36. Cairns, Charles – Raimy, Eric. 2011. “Introduction”. Handbook of the Syllable (eds. Charles Cairns – Eric Raimy), 1–30. Leiden – Boston: Brill. Cholin, Joana. 2011. “Do syllables exist? Psycholinguistic evidence for the retrieval of syllabic units in speech productions”. Handbook of the Syllable (eds. Charles Cairns – Eric Raimy), 225–53. Leiden – Boston: Brill. Chomsky, Noam. 1964. Current Issues in Linguistic Theory. The Hague – Paris: Mou- ton. Clements, George. 1990. “The role of the sonority cycle in core syllabification”. Papers in Laboratory Phonology I: Between the Grammar and Physics of Speech (eds. John Kingston – Mary E. Beckman), 283–333. Cambridge: Cambridge University Press. Clements, George – Keyser, Samuel Jay. 1983. CV Phonology: A generative theory of the syllable. Cambridge, Mass.: MIT Press. Coleman, John. 2001. “The phonetics and phonology of Tashlhiyt Berber syllabic consonants”. Transactions of the Philological Society 99:1.29–64. Český národní korpus. Institute of the Czech National Corpus, Charles University. Available online: . Databáze heslářů. Institute of the Czech Language, Academy of Sciences of the Czech Republic. Available online: . Davidsen-Nielsen, Niels. 1978. Neutralization and Archiphoneme. Two Phonological Concepts and Their History. Copenhagen: Akademisk Forlag.

275 Davis, Stuart – Baertsch, Karen. 2011. “On the relationship between codas and onset clusters”. Handbook of the Syllable (eds. Charles Cairns – Eric Raimy), 71–97. Lei- den – Boston: Brill. Dell, François – Elmedlaoui, Mohamed. 2002. Syllables in Tashlhiyt Berber and in Mo- roccan Arabic. Dordrecht – Boston – London: Kluwer Academic Press. Dickins, James. 1998. Extended Axiomatic Functionalism. Berlin – New York: Mouton de Gruyter. ―. 2007. Sudanese Arabic: Phonematics and syllable structure. Wiesbaden: Otto Har- rassowitz. Dixon, R. M. W. 1980. The Languages of Australia. Cambridge: Cambridge University Press. Duamnu, San. 2008. Syllable Structure: The limits of variation. Oxford: Oxford Univer- sity Press. Duběda, Tomáš. 2002. “Structural and quantitative properties of stress units in Czech and French”. Phonetics and its Applications: Festschrift for Jens-Peter Köster on the Occasion of his 60th Birthday (eds. A. Braun – H. R. Masthoff), 26–9. Stuttgart: Steiner. ―. 2005. Jazyky a jejich zvuky. Praha: Karolinum. Duběda, Tomáš – Votrubec, Jan. 2005. “Acoustic analysis of Czech stress: intonation, duration and intensity revisited”. INTERSPEECH-2005, 1429–32. Lisabon. El-Shakfeh, Fawzi. 1987. The Phonematics, Phonotactics and Para-phonotactics of Southern Standard British English. University of St. Andrews. Unpublished PhD. thesis, available online via the EThOS service: . Fidler, Masako. 2010. “Onomatopoeia as an embryonic word: Sound and submor- phemic properties of Czech onomatopoeic expressions”. Karlík a továrna na lingvistiku (eds. Aleš Bičan et al.), 138–55. Brno: Host – Masarykova univerzita. Firchow, Irwin – Firchow, Jacqueline. 1969. “An abbreviated phoneme inventory”. An- thropological Linguistics 11.9:271–6. Fischer-Jørgensen, Eli. 1952. “On the definition of phoneme categories on a distributional basis”. Acta Linguistica 7.8–39.

276 Frinta, Antonín. 1909. Novočeská výslovnost. Praha: Nákladem České akademie věd císaře Františka Josefa pro vědy, slovesnost a umění. ―. 1916. Fonetická povaha a historický vývoj souhlasky “v” ve slovanštině. Praha: Nákladem České akademie věd císaře Františka Josefa pro vědy, slovesnost a umění. Fudge, Erik. 1969. “Syllables”. Journal of Linguistics 5.253–86. Gabjanda, James D. 1976. An Axiomatic Functionalist Analysis of the Phonology of Yulu. University of St. Andrews. Unpublished PhD. thesis, available online via the EThOS service: . Gardner, Sheena. 1985. Parasyntax and the Sentential Level in Axiomatic Functional- ism. University of St. Andrews. Unpublished PhD. thesis, available online via the EThOS service: . Gardner, Sheena – Hervey, Sándor. 1983. “Structural sentences types”. La Linguistique 19:2.3–19. Garvin, Paul. 1948. “Kutenai I: Phonemics”. International Journal of American Lin- guistics 14:1.37–42. Goldsmith, John. 2009. “The syllable”. Ms, to be published in the second volume of The Handbook of Phonological Theory. Available online: . Greenberg, Joseph. 1962. “Is the vowel–consonant dichotomy universal?”. Word 18.73– 81. ―. 1978. “Some generalizations concerning initial and final consonant clusters”. Uni- versals of Human Language, Vol. 2: Phonology (ed. Joseph Greenberg), 243–79. Stanford, California: Stanford University Press. Grygarová-Rechzieglová, Adela. 1993. “K otázce měkkosti a kombinatoriky českých fonémů”. Slovo a slovesnost 54.255–69. Haas, William. 1957. “Zero in linguistic description”. Studies in Linguistic Analysis (Special volume of the Philological Society), 33–53. Oxford: Basil Blackwell. Haiman, John. 1980. Hua: A Papuan Language of the Eastern Highlands of New Guinea. Amsterdam: John Benjamins. Hála, Bohuslav. 1956. Slabika, její podstata a vývoj. Praha: Nakladatelství Českoslov- enské akademie věd.

277 ―. 1961. “La syllabe, sa nature, son origine et ses transformations”. Orbis 10:1.69–143. ―. 1962. Uvedení do fonetiky češtiny na obecně fonetickém základě. Praha: Nakladatel- ství Československé akademie věd. Hall, T. A. 2002. “Against extrasyllabic consonants in German and English”. Phonol- ogy 19.33–75. Harris, John. 2006. “The phonology of being understood: further arguments against sonority”. Lingua 116.1483–94. Hattala, Martin. 1870. Počátečné skupeniny souhlásek československých. Praha: Tiskem dra. Ed. Grégra. Haugen, Einar. 1956a. “Syllabification in Kutenai” International Journal of American Linguistics 22:3.196–201. ―. 1956b. “The syllable in linguistic description”. For Roman Jakobson (eds. Morris Halle et al.), 213–21. The Hague: Mouton. Havlová, Eva. 2010. České názvy savců. Praha: Nakladatelství Lidové noviny.

Havránek, Bohuslav – Jedlička, Alois. 19886. Česká mluvnice. Praha: Státní peda- gogické nakladatelství. Hjelmslev, Louis. 1936. “On the principles of phonematics”. Proceedings of the Second International Congress of Phonetic Sciences (eds. Daniel Jones – D. B. Fry), 49–54. Cambridge: Cambridge University Press. ―. 1961. Prolegomena to a Theory of Language (trans. by Francis J. Whitfield). Madi- son: The University of Wisconsin Press. Hervey, Sándor. 1972. “Mulder’s ‘axiomatic’ linguistics”. Lingua 28.348–79. ―. 1978. “On the extrapolation of phonological forms”. Lingua 45.37–63. ―. 1982. Semiotic Perspectives. London: George Allen & Unwin. ―. 1990. “Sentences and linguistic data”. La Linguistique 26:1.17–27. ―. 1996. “Functionalism, Axiomatic”. Concise Encyclopedia of Syntactic Theories (eds. Keith Brown – Jim Miller), 110–3. Cambridge: Pergamon. Heselwood, Barry. 2007. “Schwa and the phonotactics of RP English”. Transactions of the Philological Society 105.148–87. Hoard, James. 1966. “Juncture and syllable structure in English”. Phonetica 15.96–109.

278 Hockett, Charles. 1955. A Manual of Phonology. Indiana University Publications in Anthropology and Linguistics. —. 1958. A Course in Modern Linguistics. New York: The Macmillan Company. Hooper, Joan. 1972. “The syllable in phonological theory”. Language 48:3.525–40. Howkins, Douglas William. 1972. The Phonology of San Martín Quechua. University of St. Andrews. Unpublished doctoral thesis, available online via the EThOS service: . Hulst, Harry van der – Ritter, Nancy A. 1999. “Theories of the syllable”. The Syllable. Views and facts (eds. Harry van der Hulst – Nancy A. Ritter), 13–52. Berlin – New York: Mouton de Gruyter. Hůrková, Jiřina. 1995. Česká výslovnostní norma. Praha: Scientia. Hůrková, Jiřina – Hlaváč, Sáva. 1981. “K výzkumu českých souhlásek likvidních”. Slovo a slovesnost 42.269–79. Hyman, Larry M. 2008. “Universals in phonology”. The Linguistic Review 25.83–137. —. 2011. “Does Gokana really have no syllables? Or: what’s so great about being universal?”. Phonology 28.55–85. Jaeger, Jeri J. – Van Valin, Robert D. 1982. “Initial consonant clusters in Yateé Zapo- tec”. International Journal of American Linguistics 48:2.125–38. Internetová jazyková příručka. Institute of the Czech Language, Academy of Sciences of the Czech Republic. Available online: . Jakobson, Roman – Halle, Morris. 1956. Fundamentals of Language. ‘S-Gravenhange: Mouton & Co. Jensen, John T. 2000. “Against ambisyllabicity”. Phonology 17.187–235. Jones, Daniel. 1931. “The ‘word’ as a phonetic entity”. La Maître Phonétique 36.60–65. —. 1956. “The hyphen as a phonetic sign”. Zeitschrift für Phonetik 9:2.99–107. Kahn, Daniel. 1976. Syllable-based Generalizations in English Phonology. Unpublished Ph.D. Thesis, Massachusetts Institute of Technology. Available online: . Key, Harold. 1961. “Phonotactics of Cayuvava”. International Journal of American Linguistics 27:2.143–50.

279 Kohler, K. J. 1966. “Is the syllable a phonological universal?”. Journal of Linguistics 2.207–8. Krakow, Rena. 1999. “Physiological organization of syllables: a review”. Journal of Phonetics 27.23–54. Krámský, Jiří. 1976. Papers in General Linguistics. The Hague – Paris: Mouton.

Krčmová, Marie. 20083. Úvod do fonetiky a fonologie pro bohemisty. Ostrava: Os- travská univerzita v Ostravě. Kučera, Henry. 1961. The Phonology of Czech. ‘s-Gravenhage: Mouton & Co. Kučera, Henry – Monroe, George. 1968. A Comparative Quantitative Phonology of Russian, Czech, and German. New York: American Elsevier Publishing Company. Kuryłowicz, Jerzy. 1948. “Contribution à la théorie de la syllabe”. Bulletin de la Société Polanaise de Linguistique 8.80–114. Lebrun, Y. 1966. “Sur la syllabe, sommet de sonorité”. Phonetica 14.1–15. Lehiste, Ilse. 1960. “An acoustic-phonetic study of internal open juncture”. Phonetica, Supplementum ad Vol. 5. ―. 1965. “Juncture”. Proceedings of the Fifth International Congress of Phonetic Sci- ences (ed. E. Zwirner – W. Bethre), 172–200. Basel: S. Kager. Lekomceva, M. I. 1968. Tipologija struktur sloga v slavjanskix jazykax. Moskva: Nauka. Lin, Yen-Hwei. 1997. “Syllabic and moraic structures in Piro”. Phonology 14.403–36. Lowenstamm, Jean. 1996. “CV as the only syllable type”. Current Trends in Phonol- ogy: Models and Methods (eds. Jacques Durand – Bernard Laks), 419–42. Salford, Manchester: ESRI. Ludvíková, Marie. 1968. “Kombinatorika českých fonémů z kvantitativního hlediska”. Slovo a slovesnost 29.56–65. ―. 1972a. “Quantitative syllable analysis of words in Czech”. Prague Studies in Mathematical Linguistics 3.27–34. ―. 1972b. “Some quantitative aspects of the Czech syllables”. Prague Studies in Mathematical Linguistics 4.141–54. ―. 1976. “On some statistical differences in two spoken texts on the syllabic level”. Prague Studies in Mathematical Linguistics 5.91–104.

280 ―. 1978. “On the occurrence of syllables in different word positions”. Prague Studies in Mathematical Linguistics 6.39–45. Ludvíková, Marie – Kraus, Jiří. 1966. “Kvantitativní vlastnosti soustavy českých fonémů”. Slovo a slovesnost 26.334–44. Machač, Pavel. 2009. “Implications of acoustic variation for the segmentation of the Czech trill /r/”. Cross-Modal Analysis of Speech, Gestures, Gaze and Facial Expres- sions, Lecture Notes in Artificial Intelligence 5641 (eds. Anna Esposito – Robert Vích), 173–81. Berlin – Heidelberg: Springer-Verlag. Machač, Pavel – Skarnitzl, Radek. 2004. “Selected acoustic properties of the Czech palatal plosives”. 13th Czech-German Workshop – Speech Processing (ed. Robert Vích), 29–35. Prague: Ústav radiotechniky a elektroniky AV ČR. ―. 2009. Principles of Phonetic Segmentation. Prague: Epocha Publishing House. Machek, Václav. 1957. Česká a slovenská jména rostlin. Praha: Nakladatelství Československé akademie věd. ―. 1968. Etymologický slovník jazyka českého. Praha: Academia. Malmberg, Bertil. 1964. “Juncture and syllable division”. In Honour of Daniel Jones (eds. David Abercrombie et al.), 116–9. London: Longmans.

―. 19672. Structural Linguistics and Human Communication. Berlin – Heidelberg – New York: Springer-Verlag. Malone, Kemp. 1936. “The phonemic structure of English monosyllables”. American Speech 11:3.205–18. Martinet, André. 1956. La description phonologique avec application au parler franco- provençal d’Hauteville (Savoie). Genève: Librairie Droz. ―. 1965. La linguistique synchronique. Paris: Presses universitaires de France.

―. 19913. Éléments de linguistique générale. Paris: Armand Colin. Mathesius, Vilém. 1929. “La structure phonologique du lexique du tchèque modern”. Travaux du Cercle Linguistique de Prague 1.67–84. ―. 1931a. “O výrazové platnosti některých českých skupin hláskových”. Naše řeč 15.38–40. ―. 1931b. “Několik slov o hiátu v dnešní češtině”. Naše řeč 15.219–21.

281 ―. 1931c. “Zur problem der Belastungs- und Kombinationsfähigkeit der Phoneme“. Travaux du Cercle Linguistique de Prague 4, 148–52. ―. 1932. “Cizí slova se stanoviska synchronického”. Časopis pro moderní filologii 18.231–9. ―. 1947. Čeština a obecný jazykozpyt. Praha: Melantrich. Matteson, Esther. 1965. The Piro (Arawakan) Language. Berkeley: University of Cali- fornia Press. Mazlová, Věra. 1946. “Jak se projevuje zvuková stránka češtiny v hláskových sta- tistikách”. Naše řeč 30.101–11, 146–51. MČ1 = Dokulil, Miloš – Horálek, Karel – Hůrková, Jiřina – Knappová, Miloslava (eds.). 1986. Mluvnice češtiny 1. Praha: Academia. MČ2 = Komárek, Miroslav – Kořenský, Jan – Petr, Jan (eds.). 1986. Mluvnice češtiny 2. Praha: Academia. Meynadier, Yohann. 2001. “La syllabe phonétique et phonologique: une introduction”. Travaux Interdisciplinaires du Laboratoire Parole et Langage 20.91–148. Mills, Elizabeth. 1984. Senoufo Phonology, Discourse to Syllable. Dallas: Summer In- stitute of Linguistics – the University of Texas at Arlington. Available online: . Mulder, Jan W. F. 1968. Sets and Relations in Phonology: An axiomatic approach to the description of speech. Oxford: Clarendon Press. ―. 1987. “Effective methodology and effective phonological description”. La Linguis- tique 23:2.19–42. ―. 1989. Foundations of Axiomatic Linguistics. Berlin – New York: Mouton de Gruyter. ―. 1993. “Negativism as an effective methodology in linguistic description”. Linguis- tics and Philosophy. The controversial interface (eds. Rom Harré – Roy Harris), 179–97. Oxford et al.: Pergamon Press. ―. 1994. “Written and spoken languages as separate semiotic systems”. Semiotica 101:1/2.41–72. ―. 1996. “Methodology and description”. La Linguistique 32:1.17–34.

282 ―. 1998. “Epistemology and linguistics: Anatomy of an approach”. Productivity and Creativity: Studies in general linguistics in honour of E. M. Uhlenbeck (ed. Mark Janse), 115–60. Berlin: Mouton de Gruyter. Mulder, Jan W. F. – Hervey, Sándor G. J. 1972. Theory of the Linguistic Sign. The Hague – Paris: Mouton. ―. 1980. The Strategy of Linguistics. Edinburgh: Scottish Academic Press.

―. 2009. “Postulates for Axiomatic Functionalism”. Linguistica ONLINE, . Mulder, Jan W. F. – Rastall, Paul. 2005. Ontological Questions in Linguistics. Munich: Lincom. MSČ = Cvrček, Václav et al. 2010. Mluvnice současné češtiny 1. Praha: Karolinum. Nechutová, Jana – Šlosar, Dušan – Večerka, Radoslav (eds.). 1982. Čítanka ze slovan- ské jazykovědy v českých zemích. Brno: Univerzita J. E. Purkyně. Novotná, Jiřina. 1962. “K přepisu zvukové podoby některých hláskových skupin v češtině”. Naše řeč 45.82–7. ―. 1972. “Kombinační schopnost českých konsonantických fonémů”. Studia z filologii polskiej i słowiańskiej 12.269–84. Novotná-Hůrková, Jiřina. 1974. “K výslovnosti některých souhláskových skupin a tzv. rázu v češtině”. Slovo a slovesnost 35.113–20. ―. 1980. “Selected problems of consonant clusters”. Phonetica Pragensia 6.89–95. O’Connor, J. D. – Trim, J. L. M. 1953. “Vowel, consonant, and syllable – a phonological definition”. Word 9:2.103–22. Padgett, Jaye. 2002. “Russian voicing assimilation, final devoicing, and the problem of [v]”. Ms. University of California, Santa Cruz. Available online: .

Palková, Zdena. 19972. Fonetika a fonologie češtiny. Praha: Karolinum. ―. 2004. “Přízvukový takt ve struktuře češtiny”. Čeština – univerzália a specifika (eds. Zdeňka Hladká – Petr Karlík), 399–408. Praha: Nakladatelství Lidové noviny. Palková, Zdena – Volín, Jan. 2003. “The role of F0 contours in determining foot boundaries in Czech”. Proceedings of the 15th International Congress of Phonetic Sciences, Vol. 2, 1783–6. Barcelona: Causal Productions Pty Ltd.

283 Pike, Kenneth. 1947. Phonemics: A technique for reducing languages to writing. Ann Arbor: University of Michigan Press.

PMČ = Karlík, Petr – Nekula, Marek – Rusínová, Zdenka (eds.). 19962. Příruční mluvnice češtiny. Praha: Nakladatelství Lidové noviny. Podlipský, Václav Jonáš – Skarnitzl, Radek – Volín, Jan. 2009. “High front vowels in Czech: a contrast in quantity or quality?”. Interspeech 2009, 132–5. Brighton: ISCA. Popper, Karl. 2002. Conjectures and Refutations. London – New York: Routledge Clas- sics. ―. 2003. The Logic of Scientific Discovery. London – New York: Routledge Classics. PSJČ = Příruční slovník jazyka českého I–VIII. Praha, 1935–1957. Available online together with its database: . Pulgram, Ernst. 1970. Syllable, Word, Nexus, Cursus. The Hague – Paris: Mouton. Rastall, Paul. 1993. Empirical Phonology and Cartesian Tables. Lampeter – Lewiston: Mellen. Renský, Miroslav. 1960. “Funkce slabiky v jazykovědném systému”. Slovo a slovesnost 21.86–95. Ridouane, Rachid. 2008. “Syllables without vowels: phonetic and phonological evidence from Tashlhiyt Berber”. Phonology 25.321–359. Romportl, Milan. 1973. Studies in Phonetics. Praha: Academia. Romportl, Milan et al. 1978. Výslovnost spisovné češtiny. Praha: Academia. Romportl, Simeon. 1984. “Úloha rázu a jeho ekvivalentů při signalizování předělu”. Slovo a slovesnost 45.104–120.

Rosetti, Alexandru. 19632. Sur la thérie de la syllabe. The Hague – Mouton. Saporta, Sol – Olson, Donald. 1958. “Classification of intervocalic clusters”. Language 34:2.261–6. de Saussure, Ferdinand. 1972. Cours de linguistique générale (édition critique préparée par Tullio de Mauro). Paris: Payot. Sawicka, Irena. 1974. Struktura grup spółgłoskowych w językach słowiańskich. Wro- cław et al.: Wydawnictwo Polskiej akademii nauk. ―. 1985. “Syllable structure in Slavic languages”. Rocznik slawistyczny 45.3–9.

284 ―. 2009. “Segmental clusters in the Slavic languages”. Die slavischen Sprachen / The Slavic Languages, Part 1 (eds. Sebastian Kempgen et al.), 52–67. Berlin – New York: Mouton de Gruyter. Selkirk, Elizabeth. 1982. “The syllable”. The Structure of Phonological Representa- tions, vol. 2 (eds. Harry van der Hulst – Norval Smith), 337–83. Dordrecht: Foris. Schütz, Albert. 1981. “A reanalysis of the Hawaiian vowel system”. Oceanic Linguis- tics 20:1.1–43. Short, David. 1985. “Some notes on the distribution of /l/ in Czech – with special reference to butterflies”. Phonetica Pragensia 7.35–41. Sigurd, Bengt. 1958. “Tendencies in the combination of prevocal and postvocal consonants in Swedish monosyllables”. Studia Linguistica 12.27–51. ―. 1965. Phonotactic Structures of Swedish. Lund: Uniskol. ―. 1968. “Phonotactic aspects of the linguistic expression”. Manual of Phonetics (ed. Bertil Malmberg), 450–463. Amsterdam: North-Holland. Skaličková, Alena. 1954. “K otázce podstaty slabiky”. Slovo a slovesnost 15.19–24. ―. 1958. “A contribution to the problem of the syllable”. Zeitschrift für Phonetik 11.160–5. Skarnitzl, Radek. 2004. “Acoustic categories of nonmodal phonation in the context of the Czech conjunction ‘a’”. Phonetica Pragensia 10.57–68. SN1 = Nová slova v češtině, Slovník neologizmů 1. Praha, 1998. SN2 = Nová slova v češtině, Slovník neologizmů 2. Praha, 2004. Spang-Hanssen, Hennig. 1958. “Typological and statistical aspects of distribution as a criterion in linguistic analysis”. Proceedings of the Eighth International Congress of Linguists, 182–213. Oslo: Oslo University Press. ―. 1959. Probability and Structural Classification in Language Description. Copenha- gen: Rosenkilde and Bagger

SSČ = Slovník spisovné češtiny. Praha, 20033. SSJČ = Slovník spisovného jazyka českého I–IV. Praha, 1960–1971. Available online: .

285 Steriade, Donca. 1999. “Alternatives to syllable-based accounts of consonantal phonotactics.” Proceedings of LP 1998 (eds. O. Fujimura – B. D. Joseph – B. Palek), 205– 45. Prague: The Karolinum Press. Studenovský, David. 2008. “Změny intenzity v rámci průběhu českých jednoduchých vokálů a diftongů”. Phonetica Pragensia 11.145–53. ―. 2010. “Comparison of Czech diphthongs with the corresponding short simple vowels”. Phonetica Pragensia 12.37–48. Studenovský, David – Trpák, Jindřich. 2004. “The duration of the Czech diphthong [ou] in comparison with vowel-vowel sequences”. Phonetica Pragensia 10.69–77. Tabain, Marija – Breen, Gavan – Butcher, Andrew. 2004. “VC vs. CV syllables: a comparison of Aboriginal languages with English”. Journal of the International Phonetic Association 34:2.175–200. Těšitelová, Marie et al. 1985. Kvantitativní charakteristiky současné češtiny. Praha: Academia. Tolstaja, Cvetlana M. 1968. “Fonologičeskoe rasstojanie i sočetaemosť soglasnyx v slavjanskix jazykax”. Voprosy jazykoznanija 3.66–81. ―. 1974. “K xarakteristike konsonatnyx sočetanij v slavjanskix jazykax (načalo i konec slova). Slavia 43.113–33. Trnka, Bohumil. 1937. Pokus o vědeckou teorii a praktickou reformu těsnopisu. Praha: Filosofická fakulta. ― . 1966. “The distribution of vowel length and its frequency in Czech”. Prague Stud- ies in Mathematical Linguistics 1.11–6. ―. 1972. “On the frequency and distribution of consonant clusters in Czech”. Prague Studies in Mathematical Linguistics 3.9–14. ―. 1982. Selected Papers in Structural Linguistics (ed. Vilém Fried). Berlin – New York – Amsterdam: Mouton Publishers. Trubetzkoy, Nikolai. 1939. Grundzüge der phonologie. = Travaux du Cercle Linguis- tique de Prague 7. Twaddell, W. F. 1939. “Combinations of consonants in stressed syllables in German”. Acta Linguistica 1.189–99.

286 ―. 1940–41. . “Combinations of consonants in stressed syllables in German (continued)”. Acta Linguistica 2.31–50. Uhlenbeck, E. M. 1950. “The structure of the Javanese morpheme”. Lingua 2.239–70. Vachek, Josef. 1932. “Fonologický poměr hlásek i a j v češtině a v slovenštině”. Slavia 11.265–73. ―. 1940. “Poznámky k fonologii českého lexika”. Listy filologické 67.395–402. ―. 1968. Dynamika fonologického systému současné spisovné češtiny. Praha: Acade- mia. Vestergaard, Torben. 1967. “Initial and final consonant combinations in Danish monosyllables”. Studia Linguistica 11.37–66. Vogt, Hans. 1942. “The structure of the Norwegian monosyllables”. Norsk tiddsskrift for sprogvidenskap 12.5–29. ―. 1954. “Phoneme classes and phoneme classification”. Word 10.28–34. ―. 1958. “Structure phonémique du géorgien”. Norsk tiddsskrift for sprogvidenskap 18.5–90. Volín, Jan. 2002. “Čtyři scénáře vývoje české laterály”. Čeština doma a ve světě 1/2002.7–13. Volín, Jan – Churaňová, Eliška. 2010. “Probabilities of consonantal sequences in continuous Czech texts”. Phonetica Pragensia 12.49–62. Volín, Jan – Skarnitzl, Radek. 2005. “Czech voiced labiodental continuant discrimina- tion from basic acoustic data”. Interspeech-2005, 2921–4. Lisbon: ISCA. ―. 2006a. “Fonologická výjimečnost české znělé labiodentály”. Kapitoly z fonetiky a fonologie slovanských jazyků (eds. Zdena Palková – Jana Janoušková), 251–64. Praha: Univerzita Karlova v Praze. ―. 2006b. “K nevyhraněnosti vlastností české labiodentální frikativy”. Tzv. základní výzkum v lingvistice – desideratum, nebo realis? Sborník příspěvků z 5. mezinárodní konference Setkání mladých lingvistů, konané na Filozofické fakultě Univerzity Palackého ve dnech 17.–19. května 2004 (eds. Petr Pořízka – Vladimír P. Polách), 308–14. Olomouc: Univerzita Palackého.

VSČ = Hála, Bohuslav et al. 19672. Výslovnost spisovné češtiny I. Praha: Academia.

287 Zec, Draga. 2007. “The syllable”. The Cambridge Handbook of Phonology (ed. Paul de Lacy), 161–94. Cambridge: Cambridge University Press. Zeman, Jiří. 2008. Základy české ortoepie. Hradec Králová: Gaudeamus. Ziková, Markéta. 2008. Alternace vokálů s nulou v současné češtině – laterální auto- segmentální analýza. Brno: Masarykova univerzity. Unpublished PhD. thesis, available at: .

288