256 3 9MB
English Pages [125] Year 1997
ProsodicWords
Academisch
Proefschrift
ter verkrijging van de graad van doctor aan de Universiteit van Amsterdam op gezag van de Rector Magnificus prof.cir ]J.M. Franse ten overstaan van een door het college van dekanen ingestelde commissie in het openbaar te verdedigen in de Aula der Universiteit
op woensdag 10 december 1997 te 15.00 uur
door Sharon
Andrea
Peperkamp
geboren te Naarden
l
. '
i
FACULTEITDER LETTEREN UNNERSITEIT VAN AMSTERDAM
Preface
promotor: Pro£dr G.E. Booij, Vrije Universiteit co-promotor: Dr M. Nespor, Universiteit van Amsterdam Overige !eden promotiecommissie: Profdr P. M_uysken,Universiteit van Amsterdam Profdr N. V1ncens University of Manchester Profdr W. Zonneveld, Rijksuniversiteit Utrecht Dr H. va~ der Hulss Rijksuniversiteit Leiden Dr N. Smith, Universiteit van Amsterdam
Peperkamp, Sharon Prosodic Words / Sharon Peperkamp - Den Haag· Holland A d . Graphics._- (HIL dissertations ; 34) . . ca em1c Proefschnft Universiteit van Amsterdam - Met lit. opg. - Met samenvatting in het Nederlands ISBN 90-5569-035-x NUG! 941 Trefw.: phonology; morphology; generative grammar
ISBN 90-5569-035-x
©_1997_by Sharon Peperkamp. All rights reserved Prmted m The Netherlands. ·
Omslag: Piet Mondriaan C · • . .. , omposittezn Lyn, eerstestaat, 1916/l91 7.
That I ended up writing a dissertation in linguistics is to a large extent the credit of Marina Nespor. Ever since I took my first undergraduate class in phonology, Marina has been an inspiring teacher and advisor to me. During my years as a graduate student, she was my guide in all academic skills, from doing research and writing papers to preparing talks and teaching classes. She accurately read and commented all my work without delay, and I very much enjoyed our frequent and informal contact. I am also grateful for her constant moral support. Geert Booij was actively involved as a supervisor from the beginning of my research project at HIL, and never failed to provide me with encouraging feedback. As this thesis neared completion, his comments and suggestions became increasingly important, and I am particularly grateful for his careful reading of a previous version of the entire manuscript. Farther away from home, I also met several people to whom I am indebted. First of all, I benefited a lot from the 1993 LSA Linguistic Institute at OSU, especially from the classestaught by Anne Cutler, Sharon Inkelas, and Arnold Zwicky. Second, I would like to thank Pier Marco Bertinetto and Sergio Scalise,who offered hospitality at the ScuolaNormale Superiore in Pisa and the University of Ferrara, respectively, between November 1992 and December 1993. Third, the semester I spent at the University of Massachusetts at Amherst in 1994 was both fruitful and enjoyable. I am grateful to John McCarthy and Lisa Selkirk, whom I came to know as very generous with their time. The regular meetings I had with them had a significant impact on my thinking. Furthermore, I greatly appreciated discussions with Jim Harris and Michael Kenstowicz, both during several visits at MIT and via e-mail.
viii I gratefully acknowledge financial support from the Netherlands Organization for Scientific Research and the Italian Ministry of Foreign Affairs for my visits abroad. Many other people contributed to the completion of this thesis in one way or another. In particular, I am indebted to Ana Arregui, Laura Bafile, Jose Benki, Anna Cardinaletti, Joao Costa, Sonia Frota, Jorge Guitart, T.A. Hall, Michael Hammond, Harry van der Hulst, Ursula Kleinhenz, Frank van der Leeuw, Giulio Lepschy, Rochelle Lieber, Paola Monachesi, Alan Prince, Angela Ralli, Grazyna Rowicka, Norval Smith, Michal Starke, Anna Thornton, Hubert Truckenbrodt, and Ann Wennerstrom. Special thanks are due to all the friends, colleagues, and other people who served as informants. Too numerous to be mentioned individually, they are the ones who ultimately made this thesis possible. Finally, my warmest thanks to Roos Vogel, with whom I shared so much more than just an office, as well as to Qyirine Diesbergen, Magdeleen Matter, and Marian van der Meer for invaluable friendship and support throughout the years. When I first entered the department of Italian as an undergraduate student, I could not foresee the consequences of this step for my career, nor could I realize how several visits to Italy would enrich my life. With great pleasure, I dedicate this thesis to my friends in Ferrara, where I found a second home.
Table of Contents
Abbreviationsand Symbols................ :...................................................... xiii
1 INTRODUCTION...................... ,......................................................... l
2
7 PRELIMINARIES ...........................................................................
;:~ ~~i.~~~:~E:if i~i::::::::::::::::::::::::::::::::::::::::::::::::::::: ....... ::::::::::::::::::::::::::: i 1 2.2.2 The Prosodic Word ........................................................................ . ............................. 11 2 .2.2. 1 The Domain of the Prosodic Word · 16 . 2 2 2 2 The Prosodic Word in the Lexicon............................ . . . · 19 2.2.3 Syllabification across Prosodic Words ........:···:···························· 2.2.3.1 Extraprosodicity and Ambisyllabicity........................ . 20 2.2.3.2 Resyllabification and Strict Layering.......................... 27 31 2.3 Optimality Theory ....................................................................................... 2.3.1 Constraint Families ......................................................................... 32 2.3.2 Correspondence ................................................................................ 37 2.3.2.1 Paradigmatic Phonology ................................................ 39 2.3.2.2 Problems ........................................................................... 42 52 2.4 Conclusion .................................. ---············---························
X
XI
3 DERNATION ........................................................................................ 55 3.1 Introduction ................................................................................................... 55 3.2 Suffixation ....................................................................................................... 57 3.2.1 The Base ofDerivation .................................................................... 58 3.2.2 Italian ................................................................................................. 61 3.2.2.1 Hiatus ................................................................................. 62 3.2.2.2 Intervocalic s-Voicing ......................................................65 3.3 Prefixation ....................................................................................................... 69 3.3.1 Italian .................................................................................................. 69 3.3.1.1 Stress ................................................................................... 70 3.3.1.2 Intervocalic s-Voicing......................................................74 3.3.1.3 Voice Assimilation ...........................................................82 3.3.1.4 Total Nasal Assimilation ................................................84 3.3.2 Spanish ............................................................................................... 87 3.3.2.1 e-Epenthesis ....................................................................... 87 3.3.2.2 r-Strengthening ................................................................. 94 3.3.2.3 s-Aspiration and n-Velarization ................................. 100 3.4 Conclusion ................................................................................................... 104 4
COMPOUNDING........................................................................... 107
4.1 Introduction .........................·......................................................................107 4.2 Background .................................................................................................. 109 4.2.1 Head Theories in Morphology ....................................................109 4.2.2 Compounding versusDerivation ...............................................114 4.3 Compounding in Italian ...........................................................................117 4.3.1 Word+Word Compounds ............................................................118 4.3.1.1 Morphological Structure .............................................118 4.3.1.2 Prosodic Structure .........................................................123 4 .3.2 Root+ Root Compounds ..............................................................129 4.3.2.1 Morphological Structure .............................................129 4.3.2.2 Prosodic Structure .........................................................132 4.3.3 Root+Word Compounds .............................................................134 4.3.3.1 Non-native Roots ..........................................................135 4.3.3.2 Native Roots ..................................................................139 4.3.3.3 Prosodic Structure .........................................................144 4.3.4 The Absence ofWord+Root Compounds ................................146 4.3.4.1 No Leftheaded Word+Root Compounds ................ 146 4.3.4.2 No Rightheaded Word+Root Compounds .............. 148 4.4 Conclusion ................................................................................................... 153
5 CLITICIZATION············································································ 157
Introduction ................................................................................................ 157 ;:~ The Clitic Group in Prosodic Phonology .............................................1;8
5.3
5.4
55 · 5.6
5.2.1 English····························································································:~6; 5.2.2 Greek........................................... 66 5.2.3 Hungarian ....................................................................................... 1 Problems for the Clitic Group ......,..........................................................167 5.3.1 Review of Arguments ....................................................................16~ 5.3.2 Asymmetries .................................................................................... 1; 5.3.2.1 English ............................................................................1 1 5.3.2.2 Dutch ............................................................................... l 74 · · Vanauon · · .................................................... 176 Enclisis and Crosslingmst1c 5.4.1 Neapolitan ....................................................................................... 1;: 5.4.1.1 Previous Analyses ..........................................................~ 5.4.1.2 A Prosodic Account... ................................................... 84 5.4.2 Lucanian .......................................................................................... l91 5.4.3 Standard Italian ..............................................................................193 More on Cliticization in Italian ........:····.................................................199 5.5.1 Truncation ...................................................................................... l 99 5.5.2 Consonant Doubling .................................................................... ~~~ Conclusion .................................................................................................. .
6 CONCLUSION............................................................................... 2 13
~l:
References ................................... :······· ................................................................ Samenvatting_(Summaryzn Dutch).................................................... 2 Curriculum Vitae....................................................................................... 233
n
!
Abbreviationsand Symbols
A
adjective
ACC
accusative
attr. C
attributive consonant
cl.
clitic
CG
clitic group consonantal
cons. cont DAT DEF FEM
continuant
IND
dative definite feminine foot imperative indicative
L
left
MASC
masculine noun plural phonological phrase prosodic word right reflexive syllable singular Strict Layer Hypothesis sonorant The Sound Pattern of English (Chomsky & Halle 1968)
Ft IMP
N PL PPh PW R REFL
cr SG
SLH son SPE
XIV
V
verb vowel
vce Xo
voice
XP
+ #
* *!
..,,
1
Introduction
lexical word phrase primary word stress secondary word stress length syllable boundary morpheme boundary word boundary constraint violation fatal constraint violation winning candidate Derivation, compounding and cliticization are common processes which center around the level of the word. This thesis deals with phonological aspects of these three processes, in particular with regard to the definition of the prosodic word. The languages that contribute most to the empirical basis of this thesis are Italian and, tci a lesser extent, Spanish. Other languages that will be examined include Dutch, English, and two more languages spoken in Italy, Neapolitan and Lucanian.1 Prosodic words are typically characterized as being the domain of word stress, phonotactics and segmental word-level rules. Crucially, they are not necessarily isomorphic to morphological words, i.e. syntactic atoms. For instance, derivatives and compounds are morphological words which can be composed of more than one prosodic word, whereas clitics are morphological words which generally do not constitute prosodic words. The prosodic word is but one element in a series of hierarchically ordered phonological constituents known as the prosodic hierarchy (Selkirk 1980, 1981, 1984, 1986; Nespor & Vogel 1982, 1986).The category of the prosodic word straddles the boundary between the lexical part and the postlexical part of the hierarchy. That is, the categories at the lower end of the hierarchy, including the prosodic word, are constructed lexically, whereas the remaining, phrasal, categories are constructed postlexically. A central feature of the theory involving the prosodic hierarchy is the Strict Layer Hypothesis (SLH) of Selkirk (1984) and Nespor & Vogel (1986), which consists of two clauses. The first clause concerns prosodic domination; it requires each prosodic constituent to directly dominate constituents of the immediately lower category only. The second clause 1 Neapolitan and Lucanian are commonly referred to as dialects, a term which I avoid becauseof its lack of theoreticalstatus in formal linguistics.
2
CHAPTER 1
concerns the formation of well-formed prosodic trees, in that it demands each string to be parsed exhaustively into non-overlapping domains. In this thesis, I will examine aspects involving both clauses of the SLH with respect to the formation of prosodic words. First, the prosodization of affixes, clitics and compound members can induce violations of the requirements on prosodic domination. Specifically, some of these elements neither incorporate into an adjacent pr?s_odic word _nor form an independent prosodic word. Rather, they adJom an adJacent prosodic word, or, in the case of clitics, incorporate pros?d1cally at a phrasal level. In the former case, a prosodic word dommates another prosodic word, whereas in the latter case, a phonological phrase has a branch which immediately dominates a syllable. In other words, prosodic recursion and the skipping of levels should be allowed for. In this thesis, I will offer a constrained account of when and how recursion and the skipping of levels occur in derivation, compounding and cliticization. To this aim, I will make use of the decomposition of the ?rst clause of the SLH into separate, violable, constramts, as proposed m Selkirk (1995). Second, postlexical resyllabification constitutes a potential problem for the s~cond clause of the SLH, which regards the nesting of prosodic c?nst1tuents; that is, prosodic constituents should be properly nested withm the constituent that dominates them. Syllabification, therefore cannot cross prosodic word boundaries. In many languages, howeve;, phrasal resyllabification applies across prosodic words. I will argue that resyllabification induces the adjustment of the prosodic word. Thus, postlexical resyllabification results in the formation of postlexical prosodic words, which differ minimally from the lexically built prosodic words from which they derive. As a consequence, syllables are properly nested w1thm prosodic words, both lexically and postlexically, and no ill-formed prosodic trees result. In the remaining part of this introduction, I will lay out the o~ganization of the thesis and the main proposals made in it. In chapter 2, I Introduce the theoretical framework and discuss several issues that are relevant to the rest of the thesis. The chapter consists of two parts. In the first part, I give an overview of prosodic phonology, a theory about the mterface between morphosyntax and phonology (Selkirk 1980, 1981, 1984, 1986; Nespor &_Vogel1982, 1986). The key idea of prosodic phonology is that _phonological_ rule domains are derived from morphosyntactic constituents, to which they are not necessarily isomorphic. The categories that define the phonological rule domains constitute the prosodic hierarchy. The topic of this thesis being the category of the prosodic word I examine both its derivation from morphosyntactic structure and its rol;
:o
INTRODUCTION
3
in the lexicon. Furthermore, I discuss resyllabification across p_rosodic words as it occurs in many languages. In particular, I examme the interaction of resyllabification with segmental ~ules, argu1':g that extraprosodicity and ambisyllabicity are not feasible alternatives to resyllabification. . . . . The phonological phenomena considered m this theS!S are analyzed mainly in terms of output constraints rather than as input-t-output rules. In particular, I use the relatively recent framework of optimal~ty theory (Prince & Smolensky 1993). In the second part of chapter 2, I Introduce this theory and explore some potential limits. Optimality theory has proven to be successful in various areas, including prosodic morphology (cf. McCarthy & Prince 1993ab, 1994, 1995, am_ongo_thers).A n?vel feature of the theory consists in the use of paradigmatic constramts. These constraints compare candidate surface forms not to an underlymg form but to the surface form of paradigmatically related words. The part of optimality theory that deals with these so-called ou_tpuHo-output constraints is called correspondence theory (McCarthy & Pnnce 1995). . Concerning the general research question as to whether a constramtbased theory is to be preferred to rule-based theories in the realm .of prosodic phonology, the aim of this thesis is twofold. On the one hand, I will argue in favor of optimality theory as opposed_:o rule-based :heones, by providing supporting evidence for the decomposltlon _ofthe Stnct Layer Hypothesis on prosodic constituency into ranked and v10lable constramts (Selkirk 1995). On the other hand, I will argue against the use of output-tooutput constraints. Crucially, I will attempt to show that the correspondence model fails to account for vario~s. effects of ~honological under- and overapplication in derivation and cht1C1zat1on. I will thus argue in favor of a version of optimality theory in which at least a lexical and a postlexical level are recognized. . . The body of this thesis consists of chapters 3, 4 and 5, which deal with derivation, compounding and cliticization, respectively. _Each of these chapters has a high degree of autonomy and can be read mdependently. Several phenomena, however, are taken up in more than one chapter. For the reader's convenience, I have indicated as many cross-references to the various chapters as possible. . . In chapter 3, I examine derivation in Italian and Spamsh. Fu~t of all,_I compare two proposals regarding the nature of the base ?f denvat1on m these morphologically similar languages. Accordmg to Scalise (1983, _1984), derivation in Italian is based on fully inflected words, whereas accordmg to Harris (1983, 1991), derivation in Spanish is based on stems. I will argue that the latter proposal should be extended to derivation in Italian.
4
CHAPTER 1
In the remaining part of the chapter, I focus on phonological asymmetries between prefixation and suffixation, drawing on a variety of phonological rules and their interaction with syllable structure. Analogously to what has been reported for many other languages, prefixes will be shown to be less coherent to their base than suffixes in both Italian and Spanis?. Elaborating on proposals by Nespor & Vogel (1986), among others, I will argue that whereas suffixes form a prosodic word together with the stem, prefixes are not incorporated into the base prosodic word. Rather, they either form independent prosodic words, or, if they do not satisfy the minimal conditions for constituting a prosodic word, they ad1om t~ the base prosodic word, giving rise to recursion of this category. As a desirable result, they are outside the prosodic word to which they attach without forming a prosodic word themselves. Chapter 4 is dedicated to compounding. This chapter is more morphologically oriented than those on derivation and cliticization. In fact, a great deal of attention is paid to the question of the position of the head in compou?ds. Romance languages are known for the fact that they have compoundhke structures, or juxtapositions, which are inflected at the e".d of the first ekrnent. For Italian, I will argue, following the analysis of D1 Scmllo & Williams (1987) for French, that these structures are listed phrases. I will then distinguish three structural types of true compounding m Italian, 1.e. root+root compounding, root+word compounding and word+word compounding. Roots are generally elements of Greek or Latin origin, e.g. pseudo and graft. However, I will show that roots can also be derived from native Italian words by clipping and suffix truncation; these roots, then, also participate in root+word compounding. Crucially productivecompounding in Italian will be shown to be either exocentric o; rightheaded. Moreover, I will account for the absence of the fourth logically possible type, word+root compounding. That is, the absence of leftheaded word+root compounds follows straightforwardly from the hypothesis that endocentric compounding is uniformly rightheaded. Rightheaded word+root compounds, by contrast, are ruled out because roots cannot be morphological heads. On the basis of this structural typology, I will examine the prosodic structure of Italian compounds. Specifically, I will show that root+root compounds form single prosodic words. Root+word and word+word compounds ~re generally mapped onto two adjacent prosodic words; howe~er, mdiv1~ual compounds of these types can alternatively form recursive prosodic word structures, with the first element adjoined to the second one.
In chapter 5, I consider the prosodic structure of clitics and their hosts. The aim of this chapter is to provide evidence that the category of the
INTRODUCTION
5
clitic group, placed between the prosodic word and the phonological phrase in the prosodic hierarchy (Hayes 1989; Nespor & _Vogel 1986), IS redundant. I will argue, first of all, that the cases advanced i". the hterat~re to show the need for the clitic group should be reanalyzed witho~t- makmg reference to this constituent. Furthermore, I will argue that a clmc group account fails to accommodate data regarding phonological asymmetries between proclisis and enclisis. A novel argume".t agamst the chuc group will be developed in a detailed study of the mteract1on ~f stress ~nd enclisis in three closely related languages spoken in Italy. Crucially'. vary1".g stress behavior will be shown to be a consequence of vananc~ m representations. That is, under the assumption that clitics fail to constitute independent prosodic words, I will propose_ that each of these languages adopt a different strategy for the prosodizauon of encht1cs. !n. particular, enclitics can incorporate into the base prosodic word, ad1om to 1t, ?r incorporate at the level of the phonological phrase. The propose~ analysis will be shown to have more explanatory power than a chuc group approach. In the latter, in fact, a)l represe".tations are the same, hence no principled account of the stress differences is available. . . Finally, chapter 6 summarizes the main results of th!S thes!S and contains concluding remarks.
2
Preliminaries
2.1 Introduction This thesis is couched within a framework which combines principles from two theories, prosodic phonology (Selkirk 1984, 1986; Nespor & Vogel 1986) and optimality theory (Prince & Smolensky 1993). The former is concerned with the definition of purely phonological rule domains; the latter provides a general, representational, model of the relation between underlying and surface form in phonology. Thus, prosodic phonology and optimality theory are not competing theories. Most work in prosodic phonology is based on derivational models of phonology. By contrast, I will adopt an optimality theoretic approach (cf., among others, McCarthy 1993; Peperkamp 1995ab, 1996, to appear a; Selkirk 1995; Truckenbrodt 1995). In this chapter, I will provide introductions to both prosodic phonology (section 2.2) and optimality theory (section 2.3). Anticipating data to be discussed in the rest of this thesis, I will make several proposals throughout the chapter. A summary of these proposals is given in section 2.4.
2.2 Prosodic Phonology Prosodic phonology is concerned with purely phonological rules, i.e. rules which do not refer to specific morphemes, words, or syntactic phrases, and instead apply 'across the board' within certain prosodic domains. These domains are hierarchically arranged units ranging from the syllable1 to the utterance. Traditionally, word-internal phonological rules - applying across 1 Sometimes the mora is taken to be the smallest constituent of the prosodic hierarchy. See, for instance, McCarthy & Prince (1986, 1993a) and Zee (1988).
CHAPTER 2
8
morphemes - are referred to as internal sandhi, and phrasal rules - applying across words - as external sandhi. In this thesis, I will be concerned mainly with phonology at the border of internal and external sandhi, i.e. with processes applying either within roughly a word-size domain or between words and clitics. In the following sections, I lay out the basics of prosodic phonology (2.2.1), I discuss several issues concerning the category of the prosodic word (2.2.2), and I deal with syllabification across prosodic word boundaries (2.2.3).
2.2.1 Basic Tenets Apart from prosodic phonology, two approaches have been taken to the interface between morphosyntax and phonology. First, in early generative theory, it was proposed that phonological rules refer to boundary symbols which are placed at the edges of morphosyntactic constituents (Chomsky & Halle 1968; Selkirk 1972). In this approach, the strength of junctures is expressed by the number of boundary symbols, and phonological rules refer to these symbols in one of two ways; they either require a certain number of boundary symbols in order to apply, or, conversely, they are blocked by the presence of a certain number of boundary symbols. Selkirk (1980) and Hayes (1989) convincingly argue against such a boundary approach, given its overly powerful character. Second, in the direct syntax approach advocated, for instance, by Kaisse (1985), phrasal phonological rules make crucial reference to syntactic notions such as c-command. 2 As extensively argued by Selkirk (1980, 1981), Nespor & Vogel (1986) and Hayes (1989), there are reasons not to adopt a direct syntax approach either. First, phonological rules make use of an impoverished amount of morphosyntactic information. For instance, they do not distinguish morphosyntactic categories from one another (Selkirk 1972, 1986; Nespor & Vogel 1986),3 and they are insensitive to syntactically empty elements such as wh-traces (Berendsen 1985; Nespor & Scorretti 2
Phonological rules that make reference to syntactic notions can also be found in earlier
work, ;.g. Clements (1978), Rotenberg (1978), and Napoli & Nespor (1979). Most of the rules discussed by these authors, however, have been reanalyz.edwithout direct reference to syntax in later work, such as Selkirk (1981, 1986) Nespor & Vogel (1982 1986) and Hayes (1989). ' . ' ' R~cent ~irect synt~x accounts of phra~al,Phonol~gical rules include Odden (1987),
PRELIMINARIES
1985; Nespor & Vogel 1986). In a direct syntax approa~h, these generalizations must be stipulated. Second, there are frequent mismatches between syntactic and phonological constituency that cannot be accounted for. Third, postlexical rules of an individual language either refer to the same domain, or have different domains which crucially do not overlap.' [n a direct syntax approach, a domain which is referred to by more than one rule must be defined separately for each of those rules, and the nesting of different domains remains unaccounted for. The theory of prosodic phonology, as developed by Selkirk (1980, 1981, 1986) and Nespor & Vogel (1982, 1986), among others, differs from the direct syntax approach in that the domains of prosodic rules are derived from but not isomorphic to morphosyntactic constituents. In fact, prosodic phonology has two components. The so-called mapping rules derive prosodic domains from morphosyntactic constituency. The phonological rules proper then apply with reference to these domains. Rule domains are thus defined only once, but can be referred to by more than one rule. The mapping rules treat morphosyntactic category labels in a uniform manner. Moreover, they refer to surface syntactic structure, hence they are insensitive to syntactically' empty elements. The rule domains, or prosodic constituents, are joined together into a hierarchical structure, called the prosodic hierarchy. Different versions of the prosodic hierarchy have been proposed (cf., among others, Selkirk 1980, 1981, 1986; Nespor & Vogel 1986; Selkirk & Tateishi 1988, 1991). All of these contain the categories of the syllable, the foot, the prosodic word,' and one or more phrasal categories. By contrast, the clitic group, introduced by Nespor & Vogel (1986) and Hayes (1989) as the constituent immediately dominating the prosodic word, is highly controversial. Following opponents of this category, e.g. Zee (1988, 1993), Booij (19886, 1995, 19966), Inkelas (1989), Zee & Inkelas (1991) and Selkirk (1995), I will argue in chapter 5 that the clitic group is not a necessary constituent of the prosodic hierarchy, and that phonological cliticization is treated best without this constituent. The nesting of prosodic constituents in the prosodic hierarchy is guaranteed by one of the fundamental tenets of prosodic phonology, known as the Strict Layer Hypothesis (SLH) (Selkirk 1984; Nespor & Vogel 1986). The definition from Nespor & Vogel (1986: 7) is given in (1).
Cardm~ettI (1,993), Cmque (1?93), and ~.IZZI &, Savo1a (1993). A reanalysis in terms of prosod1c constituents of the Kimatuumbt data discussed by Odden (1987) is provided in
Truckcnbrodt (1995). 3 See, however, ~ayes (1,990) and Nespor (1990) for phrasal rules
that apply only in certain morphosyntact1c constituents. One such rule, Italian troncamento, will be discussed in chapter 5.
9
4 See Hyman, Katamba& Walusimbi (1987),however,for a potential counterexample. 5 This constituent is alternativelyreferredto as the phonological word.
CHAPTER 2
10
(1) Strict LayerHypothesis a. A given non-terminal unit is composed of one or more units of the immediately lower category. b. A unit of a given level is exhaustively contained in the superordinate unit of which it is part. Lacking recursion, prosodic structure is thus flatter than syntactic structure. In this thesis, I will argue, following Ito & Mester (1992), Vogel (1994), Selkirk (1995), Truckenbrodt (1995), and Ladd (1986, 1996), among others, that the Strict Layer Hypothesis cannot be maintained in its strongest form. That is, limited forms of both recursion and the skipping of levels should be permitted under certain circumstances. In particular, I will provide evidence that derivation, compounding and cliticization can all induce recursion of the prosodic word. Recursion in the syntactic sense of one constituent dominating a constituent of a higher category will still be excluded, though. As to the skipping of levels, I will show that affixes can be integrated into prosodic structure without being dominated by a foot, while clitic syllables can skip the levels of both the foot and the prosodic word. Regarding the nature of the mapping rules which derive the prosodic constituency from morphosyntactic structure, Selkirk (1986) proposes that phrasal prosodic constituents, including the prosodic word, comprise the stretch of the surface syntactic structure that is demarcated by the right or left end of a category from the X-bar hierarchy. As a desirable consequence, exhaustive parsing into non-overlapping domains as expressed by the Strict Layer Hypothesis is a derived property of phonological constituency. A formal definition of this edge-based approach, taken from Selkirk & Shen (1990), is given in (2).
(2)
11
PRELIMINARIES
theory, the edge-based approach to the syntax-phonology interface may be implemented in a straightforward manner.
2.2.2 The Prosodic Word In this section, I will review several proposals that have been made concerning the universal nature of the prosodic word (2.2.2.1) and discuss evidence that prosodic words are derived lexically (2.2.2.2).
2.2.2.1 The Domain of the Prosodic Word Booij (1983) argues that prosodic words can be smaller than a lexical word, larger than it, or equal in size. His typology is based on the observation that, given the Strict Layer Hypothesis, syllables (and, for that matter, feet) must be properly contained in prosodic words. Thus, syllabification diagnoses prosodic word boundaries. 6 As to the first option, that of the prosodic word being smaller than a lexical word, Booij notes that in several languages, the members of compounds and certain affixes constitute independent domains of syllabification. This is, for instance, the case in Dutch. I repeat Booij's examples in (3). Here, and in the rest of this thesis, relevant syllable boundaries are indicated by periods. (3) a. b.
morphology
prosody
[[vlees}N[eter}N}N meat eater [[rood]A achtig]A red like
[vles.}pw[etar}rw
. ' 'carnivore
[rot.]PW [axtax]rw 'reddish'
Syntax-Phonology Mapping(Selkirk & Shen 1990: 319) For each category Cn of the prosodic structure of a language there is a two-part parameter of the form C 0 : {Right/Left; Xm} where Xm is a category type in the X-bar hierarchy. A syntactic structure - prosodic structure pair satisfies the set of syntax-phonology parameters for a language iff the Right (or Left) end of each constituent of type Xm in syntactic structure coincides with the edge of constituent(s) of type C 0 in prosodic structure.
Thus, the prosodic constituents of word-size and larger are derived in a uniform manner. In section 2.3.1, it will be shown that within optimality
The final consonants of vleesand roodare not syllabified as onsets into the following syllables. Therefore, Booij concludes that the two members of the compound in (3a), as well as the stem and the suffix in (36), form independent prosodic words.' Selkirk {1986) argues that syllables and feet do not prOperly belong to the prosodic hierarchy,since syllabification and foot construction apply within a constituent which may be either word-internalor phrasal.Rice (1990) also makesa distinction betweenthe syllable and the foot on the one hand, and the remaining categories on the other hand, in that the former are available everywhere,whereas the latter are available only at the point at which theyare constructed. 7 In chapter 3, it will be shown that prefixesconstitute separatedomains for syllabification more often than suffixes. In Dutch, for instance,few suffixes behavelike -achtigin resisting transjuncturalsyllabification,whereasall prefixes conform to this pattern.The example in (36) is thus a markedcase for suffixation(cf. also section 2.3.2.2). 6
CHAPTER 2
PRELIMINARIES
The second option in Booij's typology, that of the prosodic word being larger than a lexical word, is exemplified by French. In this language, syllabification applies within phrases. Therefore, Booij argues that the domain of the prosodic word in French coincides with that of the phonological phrase and can include more than one lexical word. However, it is argued in Nespor & Vogel (1986) and Booij & Rubach (1987) that two types of syllabification must be distinguished in French. First, syllabification initially applies within words. Second, resyllabification applies across words. Evidence for this two-step syllabification comes from the segmental rule of closed syllable adjustment (Schane 1968; Selkirk 1972; Dell 1973). This rule changes /e/ and /a/ into [s] in closed syllables, as exemplified in (4). In the masculine form in (4a), the underlying wordfinal /r/ is not realized, whereas in the feminine form in (4b), it is. Accordingly, closed syllable adjustment applies in the latter example only.
this language end in a vowel, but there are also a number of consonantfinal words, mainly loans or words of foreign origin. If such a word is followed by a vowel-initial word within the same phonological phrase, resyllabification applies (Camilli 1965; Chierchia 1983/1986; Nespor 1986). Consider the example in (6).
12
(4) a. b.
/pramjer/ premier premiere
[pramje] [pramjsr]
'firstMASc' 'firstpEM'
In (5), it is shown that the rule also applies word-finally before a vowelinitial word, despite the fact that on the surface, the syllable containing/ e/ or /a/ is no longer closed. Thus, in both the masculine form in (Sa) and the feminine form in (Sb) does the word-final /r/ surface in an onset. In the latter example, however, closed syllable adjustment has applied.
(5)
a. b.
premier am1 [pra mje.rami] premiere amie [pramjs.rami]
'firstMASc friendMAsc' 'firstFEMfriendFEM'
Thus, there must be a stage in which single words are syllabified. The rule of closed syllable adjustment applies to the feminine form before resyllabification takes place. Due to resyllabification, the alternation between /e/ and /a/ on the one hand and [s] on the other hand is opaque, in that the target vowel is not in a closed syllable at the surface. By contrast, the masculine form does not have a final [r] when it occurs in isolation. This consonant surfaces only before a following vowel-initial word, in which case it immediately gets syllabified as an onset. The environment for closed syllable adjustment is thus never met. Another argument against Booij's (1983) proposal to consider syllabification across words as a diagnostic for prosodic words comprising more than one lexical word is the following. In some languages, phrases in which syllabification applies across words arguably consist of more than one prosodic word. Such is the case, for instance, in Italian. Most words in
(6)
snob orrendo
[sno.bor:cndo]
13
'terrible snob'
The final /b/ of snob surfaces in the onset position of the first syllable of orrendo. Still, there is evidence that the phrase contains two prosodic words. In Italian, the low mid vowels / o/ and / s/ can only occur in syllables bearing main word stress; in all other positions, they are neutralized to [o] and [e], respectively (Nespor & Vogel 1986). From the fact that both words in (6) contain a low mid vowel, we conclude that they both bear a primary word stress, hence constitute a prosodic word. Nespor & Vogel (1986) make a clear distinction between:syllabification and resyllabification. The prosodic word is the domain of syllabification; this constituent cannot be larger than a lexical word. Resyllabification, by contrast, applies in a phrasal domain, to be defined language-specifically. Two major problems arise with respect to the concept of resyllabification. First, resyllabification is undesirable from a theoretical point of view, given its structure changing nature. In fact, syllabification rules are argued to be structure building only (Steriade 1982, 1988). Second, resyllabification induces violations of the Strict Layer Hypothesis, in that prosodic word boundaries do not coincide with syllable boundaries at the surface. These problems will be taken up in section 2.2.3. Generalizing over proposals they make for various languages, N espor & Vogel (1986) define the domain of the prosodic word universally as in (7). In this definition, Q denotes any terminal element of the syntactic tree. (7)
Domain of PW(Nespor & Vogel 1986: 141)
A. or B.
The domain of PW is Q. I The domain of PW consists of a. a stem;
b. any element identified by specific phonological and/or morphological criteria; c. any element marked with the diacritic [+W] II Any unattached elements within Q form part of the adjacent PW closest to the stem; if no such PW exists, they form a PW on their own.
14
CHAPTER
2
Nespor & Vogel mention several predictions of this definition. First, the prosodic word cannot be larger than a terminal syntactic element. Second, a stem cannot contain more than one prosodic word. Third, in a language in which compounds are mapped onto a single prosodic word, there can be no (sequences of) affixes that form independent prosodic words. These universal characteristics of the prosodic word call for a principled account. Nespor & Vogel's definition, however, fails to provide insight into the nature of the prosodic word. In the edge-based theory of Selkirk (1986), phrasal prosodic constituents including the prosodic word are derived uniformly by alignment with morphosyntactic categories. In particular, Selkirk (1986, 1995) proposes that each lexical terminal element of the syntactic tree be aligned at one edge with a prosodic word. As a consequence, prosodic words are at least as big as a lexical word and can include one or more function words. In Cohn (1989), Kang (1992ab), McCarthy & Prince (1993b) and Peperkamp ( to appear a), among others, proposals are made for edge-based derivations of prosodic constituents smaller than the morphosyntactic word, such as stems and the members of compounds. The edge-based approach to deriving prosodic words is highly general, since no reference is made to morphological units other than Xo and stems. Moreover, it neatly accounts for Nespor & Vogel's predictions regarding the nature of prosodic words. First, the observation that prosodic words cannot be larger than a syntactic terminal is dependent upon Nespor & Vogel's view that clitics form independent prosodic words, which are joined with their host prosodic words into a clitic group. In Selkirk's account, who explicitly argues that function words cannot form prosodic words, this point translates into the observation that a prosodic word cannot comprise more than one lexical word. This follows directly in the edge-based approach, since each lexical word gets aligned with a prosodic word. Second, stems are the smallest morphological constituents that induce alignment. From this it follows directly that a stem cannot contain more than one prosodic word. Third, let us consider the prediction that in a language in which compounds are mapped onto a single prosodic word, there can be no (sequences of) affixes that form independent prosodic words. In chapter 4, it will be shown that languages can have more than one morphological compound type, and that the prosodic structure of compounds accordingly varies (cf. also Nespor & Ralli 1996; Nespor to appear). The observation made by Nespor & Vogel is thus an oversimplification. However, there does exist a basic dichotomy among languages regarding the morphological structure of compounds which relates to their prosodic
15
PRELIMINARIES
structure. Specifically, in one type of languages, compounds are generally made up of words, whereas in the other type, the basic units of compounds are generally stems, consisting of a root possibly followed by a theme vowel. Thus, in the latter type, the inflection of the compound does not necessarily coincide with that of one of its elements. The two basic types of compounding are shown in (8). (8)
a.
word compounding
b.
stem compounding
x,
x,
---------I I
x,
X,
Stem
Stem
I\
Stem
Stem
The languages considered by Nespor & Vogel (1986) in which compounds are typically mapped onto a single prosodic word are Greek and Latin; crucially, these languages have basic stem compounding (cf. Ralli (1992) and Nespor & Ralli (1996) for Greek, and Oniga (1992) for Latin). In languages such as Greek and Latin, then, derived words, whether they contain a prefix or a suffix, also constitute a single prosodic word (Nespor & Vogel 1986). In the edge-based approach, this follows straightforwardly. To see why, first note that in order for stem compounds to be mapped onto a single prosodic word, alignment must be sensitive to Xo's but not to stems. In fact, if alignment were sensitive to either the left or right edge of stems, each element of a compound would form a prosodic word. Consider now the derived structures in (9).
(9)
a.
prefixation
x,
A Stem
Prefix
b.
suffocation
X.
A Stem Suffix
Given that alignment does not regard stems, both the prefixed and the suffixed structure are mapped onto a single prosodic word. Regarding the domain of the prosodic word, there is one more universal characteristic worth consideration. That is, given that prosodic words dominate feet and that feet are branching, prosodic words should be minimally bimoraic or disyllabic, depending on whether feet are moraic or syllabic in a given language. This requirement is known as prosodic
CHAPTER 2
16
PRELIMINARIES
minimality (McCarthy & Prince 1986 et seq.). Thus, lexical words must satisfy a minimality requirement. Moreover, they bear primary stress by virtue of constituting at least one foot. Function words, by contrast, do not have to constitute prosodic words; hence, they can be smaller than two moras or two syllables and do not have to bear primary stress (cf. chapter 5). Thus, prosodic words are characterized by dominating at least one foot, and, consequently, by being stressed. In many languages, prosodic minimality is not only a static constraint on the size of lexical words, but it also imposed actively in morphological processes, as will be shown in the next section.
2.2.2.2 The Prosodic Word ln the Lexicon In early work on the prosodic hierarchy, e.g. Nespor & Vogel (1982, 1986) and Selkirk (1984), all prosodic categories are assumed to be constructed post-syntactically. However, it is argued elsewhere that the categories up to the prosodic word are built lexically, and that morphological and prosodic structure are co-present (Van der Hulst 1984; McCarthy & Prince 1986 et seq.;Booij 1988a; Inkelas 1989; Nespor 1990; Vogel 1991; Booij & Lieber 1993; Cohn & McCarthy 1994). For instance, it is shown by McCarthy & Prince (1986 et seq.) and others, that the categories of the foot and the prosodic word serves as a target in several morphological processes. I will · illustrate this with an example from Italian. Thornton (1996) argues that in Italian, the minimal prosodic word is a disyllabic trochee. In fact, a dictionary search reveals that less than 1O/oof Italian words are monosyllabic,8 and of the disyllabic words only 3.2% are iambic. Furthermore, not more than 3% of the content words end in a consonant, and most of the consonant-final words are loans. Thornton argues that the disyllabic vowel-final trochee functions as a template in the formation of clippings and hypocoristics. Examples are given in (10a) and (106), respectively. (10)
a.
clippings bicicletta frigorifero mitragliatrice
bici frigo mitra
b.
hypocoristic? Federica Fabrizio Valentina
17
Fede Fabri Vale
Thornton shows that truncation of the base after the first two syllables cannot account for the data. For instance, given the stress on the base Fabrizio in (106), truncation would yield iambic '·Fabr~ similarly, we would expect to find * Vdlen, given that the second syllable of the base Valentina ends in /n/. The correct results, then, can only be derived by mapping the base onto the prosodic template. Hence, the morphological processes forming clippings and hypocoristics make reference to the minimal prosodic word, consisting of a disyllabic trochaic foot. 10 Booij & Lieber (1993) argue that under the assumption that morphological and prosodic structure are co-present, bracketing paradoxes receive a straightforward account. Consider, for instance, the English word unhappier.From a morphological point of view, the bracketing must be as in (lla), since the word means 'more unhappy' rather than 'not happier'. Thus, the negative prefix un- must 'be attached before the comparative suffix -er. The latter, though, has a prosodic subcategorization frame, according to which it cannot attach to trisyllabic bases. Hence, from a phonological point of view, the bracketing must be as in (llb), with the suffix attached before the prefix. (11)
a. b.
morphology: phonology:
[[un [happy]] er] [un [[happy] er]]
Following the analysis of Booij & Rubach (1984), Booij & Lieber propose that the prosodic requirement pertain to the prosodic word to which -er attaches, rather than to the morphological word. Under the assumption that the prefix constitutes an independent prosodic word, the prosodic subcategorization can be satisfied if the morphological and the prosodic structure are co-present on separate planes, as illustrated in (12).
'bicycle' 'refrigerator' 'machinegun'
8 Among
the monosyllabic words, we find several content words which contain only one mora, such as tre 'three', blu 'blue',gru 'crane',and re 'king'. A seriesof forms of a few highfrequency irregular verbs such as dare 'to give', essere'to be', and sapere'to know', as well as the names of the letters of the alphabet and the musical notes equally escape prosodic minimalitf:, Finally, St 'yes' and no 'no', which often exhaust an utterance, are also monomora1c.
9 Hypocoristics occasionally do not satisfy prosodic minimality. For instance, Cri
Cristina),Giu (< Giuliano),and Ste(< Stefano)are all monomoraic.
lO For more on
clippings in Italian,see chapter4.
(