Studies on the phonological word

2-4 14195 Berlin Germany rrenate@zedat.fu-berlin.de Kevin Russell Linguistics Department University of Manitoba Winnipeg, Manitoba R3T 2N2 Canada krussll@cc.UManitoba.ca Wendy Sandler De

STUDIES ON THE PHONOLOGICAL WORD

Volume 174

T Alan Hall and Ursula Kleinhenz (eds.)

Studies on the Phonological Word

CURRENT ISSUES IN LINGUISTIC THEORY

AMSTERDAM STUDIES IN THE THEORY AND HISTORY

OF LINGUISTIC SCIENCE – Series IV

General EditorE.F.K KOERNERZentrum für Allgemeine Sprachwissenschaft, Typologie

und Universalienforschung, Berlinefk.koerner@rz.hu-berlin.de

Current Issues in Linguistic Theory (CILT) is a theory-oriented series which welcomes

contributions from scholars who have significant proposals to make towards the advancement

of our understanding of language, its structure, functioning and development CILT has been established in order to provide a forum for the presentation and discussion of linguistic opinions of scholars who do not necessarily accept the prevailing mode of thought in linguistic science It offers an outlet for meaningful contributions to the current linguistic debate, and furnishes the diversity of opinion which a healthy discipline must have.

A complete list of titles in this series can be found on http://benjamins.com/catalog/cilt

Advisory Editorial Board

Raimo Anttila (Los Angeles) Lyle Campbell (Christchurch, N.Z.) John E Joseph (Edinburgh) Manfred Krifka (Austin, Tex.) Hans-Heinrich Lieb (Berlin) Ernst Pulgram (Ann Arbor, Mich.) E.Wyn Roberts (Vancouver, B.C.) Hans-Jürgen Sasse (Köln)

STUDIES ON THE PHONOLOGICAL WORD

Edited by

T ALAN HALL URSULA KLEINHENZ

Zentrum für Allgemeine Sprachwissenschaft,

Typologie und Universalienforschung, Berlin

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA

Library of Congress Cataloging-in-Publication Data

Conference on the Phonological Word (1997 : Berlin, Germany)

Studies on the phonological word / edited by T Alan Hall and Ursula Kleinhenz.

p cm (Amsterdam studies in the theory and history of linguistic science Series IV,

Current Issues in Linguistic Theory, issn 0304-0763 ; v 174)

Includes bibliographical references and index.

1 Word (Linguistics) Congresses 2 Prosodic analysis (Linguistics) Congresses I. Hall,

T Alan II Kleinhenz, Ursula III Title IV Series.

8TM The paper used in this publication meets the minimum requirements of

the American National Standard for Information Sciences – Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.

TABLE OF CONTENTS

List of Contributors vi The Phonological Word: A Review 1

Diagnostics for Prosodie Words Revisited: The Case of 133

Historically Prefixed Words in English

Vrije Universiteit Amsterdam

Fakulteit der Letteren

Goßlerstr 2-4

14195 Berlin Germany

rrenate@zedat.fu-berlin.de

Kevin Russell Linguistics Department University of Manitoba Winnipeg, Manitoba R3T 2N2

Canada

krussll@cc.UManitoba.ca

Wendy Sandler Department of English The University of Haifa

and

Sign Language, Linguistics, and Cognition Research Laboratory The University of Haifa

31905 Haifa Israel

wsandler@research.haifa.ac.il

Marina Vigário Departamento de Português, ILCH Universidade do Minho

4710 Braga Portugal

marina.vigario@bigfoot.com

THE PHONOLOGICAL WORD: A REVIEW*

T.ALAN HALL

Zentrum für Allgemeine Sprachwissenschaft

The present volume consists of eight articles dealing with the role of the prosodic constituent 'phonological word' in various typologically diverse lan­guages These languages and their respective families subsume Indo-European (Dutch, German, English, European Portuguese), Bantu (SiSwati, KiNande), Algonquian (Cree), Siouan (Dakota), and Salishan (Lushootseed) One contri­bution examines the phonological word in a sign language

1

The goal of this introductory article is to summarize some of the research that has been done to date on the phonological word, to show where the articles in the present volume fit in to this line of research, and to outline several questions for further study

This article is organized as follows Section 1 consists of introductory remarks In section 2 I present examples from various languages of the kinds of phenomena that require reference to the phonological word Section 3 concerns itself with the relationship between the phonological word and other prosodic constituents, e.g syllable, foot, phonological phrase, and section 4 outlines the connection between the phonological word and morphosyntactic structure Section 5 deals with the kinds of conflicts among diagnostics for phonological words one encounters in individual languages A short conclusion is provided

2 T A HALL

smaller than the phonological phrase Since this constituent can be shown to be roughly the size of a grammatical word, it is often referred to in the literature as the 'phonological word', or alternatively as the 'prosodic word' The authors in this volume use both terms; in the present article I refer to the phonological/ prosodic word simply as the 'pword.'2

An important point that has been made repeatedly in the literature is that the pword is an entity that is distinct from the grammatical word Thus, many languages are attested in which a single grammatical word consists of two or more pwords (e.g each part of a compound word, or both the prefix and following stem) In other languages a pword has been argued by some authors

to consist of two grammatical words (e.g in a sequence of lexical word+clitic) There is also near unanimity in the literature that pword boundaries — unlike those of syllables and feet — must align with morpho(syntactic) boundaries Thus, if a single grammatical word consists of two or more pwords, then the pwords correspond directly to morphemes This view of the pword implies that

a monomorphemic word cannot be parsed as two or more pwords

In the present volume all of the authors show that the pword is isomorphic with the grammatical word In addition, all of the authors in the present volume agree that there is a direct connection between pword boun­daries and morpho(syntactic) boundaries, as described in the preceding para­graph

non-All of the authors of the present volume argue that the pword is an pensible prosodie constituent because it forms the domain for various phonological generalizations (see section 2) Almost all of the contributors operate under the assumption that the phonological generalizations that refer to the pword are 'concrete' in the sense that they refer to the surface level of representation and not to an abstract stage in a derivation

indis-2 Historically the term 'phonological word' was employed by Dixon (1977a, b) and later adopted by other writers (e.g Booij 1983, van der Hulst 1984, Nespor 1986, Nespor & Vogel

1986, Hannahs 1995a, b) One of the first linguists to employ the locution 'prosodie word' was Selkirk (1978, 1980a, b); this term has subsequently gained currency in the literature on Prosodie Morphology and Optimality Theory (e.g McCarthy & Prince 1986, 1993, 1994, Prince & Smolensky 1993, McCarthy 1993, Peperkamp 1997) Some authors (e.g Liberman & Prince 1977, Booij & Rubach 1984) refer to the pword as the 'mot' A survey of the literature

on pwords, including numerous historical notes, is contained in Smith (1996)

Pregenerative studies sometimes recognized a unit larger than a syllable but smaller than a grammatical word that bears a strong affinity to the 'phonological' or 'prosodie' word referred

to by all of the authors cited in the previous paragraph For example, Aoki (1966) writes that vowel harmony in Nez Perce occurs within a 'harmonic sequence', which is a group of seg­ ments not always identical to a grammatical word

THE PHONOLOGICAL WORD: A REVIEW 3

2 Evidence for the pword

The pword derives motivation as the constituent that defines the domain for various phonological generalizations These generalizations can be reduced to three types:

(1) a The domain of phonological rules

b The domain of phonotactic generalizations

 The domain for minimality constraints

The term 'rule' in (la) is to be understood in the broadest sense of the word, subsuming segmental processes like assimilations, processes manipulating tones, as well as prosodic rules that assign syllables, and stress While not recognized by some authors (e.g Nespor & Vogel 1986), phonotactic constraints (i.e (lb)) have been proposed for various languages that refer crucially to the pword That the domain of minimality constraints is the pword has been recognized by Dixon (1977a, b) and more recently by McCarthy & Prince (1986) In the following subsections examples of (la)-(lc) will be provided

2.1 The pword as the domain for phonological rules

Many of the examples discussed in the literature that argue for the pword involve segmental phonological rules that make crucial reference to this constituent This point can be illustrated with Hungarian vowel harmony, which has been argued to apply only when its trigger and target belong to the same pword (Booij 1984, Nespor & Vogel 1986) The examples in (2a) show that all vowels within a sequence of stem+suffix agree in backness (In (2) and below the symbol ' ω ' represents the pword) By contrast, the individual parts

of a compound, as in (4b), and prefixes as well as the stems to which they attach, as in (4c), constitute separate vowel harmony domains Booij (1984) and Nespor & Vogel (1986) capture these facts by positing that the sequence stem+suffix constitutes a single pword (cf the prosodic structure indicated in (2a)) Each part of a compound is analyzed as a separate pword, as in (2b), and prefixes and the stems to which they attach are separate pwords, as in (2c) The rule of vowel harmony can now be stated as in (2d)

4 T A HALL

(2) a stem+suffix: (ölelés+nek) ω 'embracement' (dat sg suffix)

stem+suffix: (hajó+nak) ω 'ship' (dat sg suffix)

b stem+stem: (könyv) ω (târ) ω 'library'

 prefix+stem: (oda) ω (menni) ω 'to go there'

d V → [aback]/ ( [αback] C 0 _ ) ω

Booij (1984) and Nespor & Vogel (1986) note that the prosodic constituency indicated in the words in (2a)-(2c), which derives motivation by vowel harmony in (2d), is further supported by a rule of palatalization that holds only

in the (2a) contexts but not between the two parts of a compound, as in (2b) or between the prefix plus stem juncture, as in (2c)

Significantly, in the example in (2) the domain of vowel harmony cannot

be captured with another prosodic constituent, e.g syllable, foot An equally important point is that the pword in (2) is not isomorphic with a morphological unit, e.g the morpheme, the grammatical word.3

Other examples of segmental rules that have been argued to refer to the pword have been proposed by Dixon (1977a, b) for Yidiny, Selkirk (1980b) for Sanskrit, Nespor & Vogel (1986) for Greek, Latin, Turkish, and Italian, Booij

& Rubach (1987) for Polish, Kang (1991) for Korean, Zsiga (1992) for Igbo, Tortora (1993) for Dakota, Raffelsiefen (1993) for English, Hannahs (1995a, b) for French, Booij (1995) for Dutch, and Yu (1992) and Wiese (1996) for German Rules altering underlying tones have also been argued to make crucial reference to the pword, e.g Myers (1995) for Shona, and Leben & Ahoua (1997) for Baule

In the present volume Downing, Russell, and Vigário propose that segmental rules in KiNande, Dakota and European Portuguese respectively can

be used as diagnostics to define the pword in these three languages Downing also argues that a rule altering the underlying tonal structure of SiSwati refers

to the pword

A number of writers contend that the pword is necessary not only as a constituent that defines the domain of segmental rules but also of prosodie ones like stress assignment and syllabification For example, Dixon (1977a: 88-90; 1977b: 25ff) argues that stress assignment in the Australian language Yidiny

3 In Lexical Phonology (Kiparsky 1982, Kaisse & Shaw 1985, Halle & Mohanan 1985) one might account for the Hungarian examples in (2) not with the pword, but instead by ordering the relevant morphological operations into distinct levels For example, level 1 might consist of suffixation and level 2 of prefixation and compounding Given this ordering, the rule of vowel harmony would apply at level 1 only For a comparison of Lexical Phonology and Prosodie Phonology see Vogel (1991) and Tortora (1993), who examine Italian and Dakota respec­ tively

THE PHONOLOGICAL WORD: A REVIEW 5

only applies within but not across pwords Nespor & Vogel (1986) similarly cite stress assignment in Latin and Turkish and syllabification in Dutch as prosodic processes that operate within but not across pwords Similar prosodic rules have also been posited for Indonesian (Cohn 1989), German (Booij 1985, Wiese 1996), and Greek (Nespor & Ralli 1996) In the present volume Russell and Raffelsiefen argue that stress patterns are reliable diagnostics for pwords in Cree and English respectively

An often cited example of a syllabification rule that operates within but not

across pwords involves German examples like the ones in (3) In the word liebe

the stem-final /b/, which is syllable-initial when a vowel-initial suffix follows,

surfaces as [b] However, in the word lieblich the same segment undergoes

Final Devoicing to [p], because it is syllable-final The second example in (3)

is important because it illustrates that Mich serves as a 'barrier' to onset maxi­ mization (German permits [bl] onsets, e.g Blitz [blits] 'lightning', bibl+isch

[bi.bhʃ] 'biblical')

(3) lieb+e [:.e] 'love'(first person sg.)

lieb+lich [li:p.liç] 'dearly'

Based on earlier work by Booij (1985), Wiese (1996) argues that the German rule of syllabification (i.e onset maximization) only applies when the segments being syllabified belong to the same pword Restricting our discussion to data like the ones in (3), Wiese defines the pword in German as (i) a bare stem plus

vowel-initial suffix like +e, and (ii) consonant-initial suffixes like Mich This

analysis therefore requires that the two words in (3) be parsed as (lieb+e) and (lieb)ω(lich)ω respectively.4

Another rule type that has been argued to be sensitive to pword structure is the phenomenon sometimes referred to as 'coordinate structure deletion' in German and Dutch, which is illustrated with the German examples in (4) This rule says that the second part of a compound deletes in a coordinate structure if

it is identical to the second half of the following compound, e.g Ebenen in the

4 It is possible to analyze the German data in (3) in Optimality Theory in such a way that reference to pwords is unnecessary Given the familiar constraint ONSET, which ensures that all syllables begin with a consonant, the syllabification [li:.be] is better than [:.e] if ONSET dominates ALIGN-R (stem, right, syllable, right) Given ALIGN-R, the syllabification [li:p.liç]

is correctly predicted to be better than [li:.bliç] (or [li:.pliç]), which both incur an ALIGN-R violation The important point is that the German syllabification facts can be accounted for without reference to the pword if ALIGN-R aligns the right edge of a stem with the right edge

of a syllable However, the pword is motivated as a prosodie constituent for German for

independent reasons (see Hall this volume)

6 T A HALL

first example in (4a) The second example in (4a) demonstrates that

consonant-initial suffixes like +lich can delete in a coordinate structure if the second word

in the phrase ends in the same suffix By contrast, vowel-initial suffixes like

+ig in (4b) do not delete Wiese (1996) (who bases his analysis on the earlier

claim of Booij 1985) argues that the element that deletes in the German exam­

ples in (4a) is the pword This proposal requires that suffixes like +lich but not ones like +ig to be separate pwords; e.g mütterlich and winzig would be repre­

sented prosodically as (mütterr)ω(lich)ω and (winzig)ω respectively Note that the distinction between consonant-intial and vowel-initial suffixes derives addi­tional support from the syllabification facts in (3) above

(4) a Tiefebenen und Hochebenen 'high plains and low plains'

Tief- und Hochebenen

mütterlich und väterlich 'motherly and fatherly'

mütter- und väterlich

b winzig oder riesig 'tiny or huge'

*winz- oder riesig

I will return to the German data in (4) in sections 2.3 and 5 below

2.2 The pword as the domain for phonotactic constraints

An important and often discussed topic in phonology involves the domain for phonotactic constraints For example, much debate has centered on the question

of whether or not such constraints refer to the morpheme, or if 'morpheme structure conditions' can be reanalyzed as constraints that are syllable-based (e.g Hooper 1972, 1976, Davis 1991, Paradis & Prunet 1993, Raffelsiefen 1993)

One question few linguists have addressed is whether or not the pword can serve as the domain for phonotactic constraints While one could easily imag­ine a language in which phonotactic constraints barring some sound or sequence of sounds only hold at the beginning or at the end of a pword, explicit examples are rare in the literature.5

Booij (this volume) and Hall (this volume) posit a series of phonotactic constraints for Dutch and German respectively that they argue refer crucially to the pword As Hall points out, most of these constraints have either been ignored in the literature on German phonology or have been misanalyzed as

5 As noted by Booij (1988: 517), none of Nespor & Vogel's (1986) examples illustrating the necessity of the pword involves a phonotactic constraint

THE PHONOLOGICAL WORD: A REVIEW 7

syllable-based For example, Hall argues that the constraint barring short lax nonlow vowels, i.e [I Y e œ u ] in 'final'position holds at the end of the pword, as in (5), and not at the end of the syllable

(5) *[I Y Ε œ u ]) ω

Booij argues that the Syllable Contact Law (Hooper 1976, Vennemann 1988) is active in Dutch phonology This constraint says that in a sequence of adjacent heterosyllabic consonants, e.g VC.CV, the first consonant is more sonorous

than the second Thus, monomorphemic pwords like album [album] show

optimal syllable contacts Systematic exceptions to the Syllable Contact Law

are compound words, e.g doop+mal [do:p.mal] 'christening feast', which can

have 'bad' contacts like Vp.mV However, these examples do not violate the Syllable Contact Law if the two adjacent heterosyllabic C's in its structural description belong to the same pword (i.e *(VC.CV)ω) and if each part of compounds are parsed as separate pwords

Clearly, one area of further research on the pword regards the extent to which phonotactics provides evidence for this constituent

2.3 Minimality constraints

Many languages impose restrictions on the minimal size a word can have The generalization that governs this kind of phenomenon is therefore usually referred to in the literature as a 'minimal word' constraint For example, in many Australian languages words can typically be no smaller than two sylla­bles (Dixon 1980: 127ff), so these languages are usually analyzed as having a bisyllabic minimum Minimality restrictions are discussed in Dixon (1977a, b) and in McCarthy & Prince (1986, 1990) and in subsequent work in the field of Prosodic Morphology In the present volume a number of authors argue that minimality constraints are at work in the phonology of individual languages (Booij for Dutch, Downing for SiSwati and KiNande, Hall for German, Raffel-siefen for English, Russell for Cree)

The minimal size of 'words' as described in the preceding paragraph is generally recognized as a property of the pword as opposed to the grammatical word Perhaps the most convincing argument that the domain of minimality constraints is the pword is that in some languages certain strings of segments that do not correspond to grammatical words (e.g prefixes, suffixes) exhibit minimality effects For example, Dixon (1977a) argues that in Yidiny certain suffixes (the 'noncohering' ones in his terminology) do not belong to the same pword as the stem to which they attach He concludes that noncohering affixes

8 T A HALL

are parsed as separate pwords because they obey the same kinds of phonological generalizations as other pwords For example, in Yidiny there is a bisyllabic minimum for all pwords, a generalization that also governs nonco-hering affixes

In many languages with a minimal word requirement there are rules that conspire to maintain the minimality restriction One example is discussed by Downing (this volume), who argues that SiSwati, like Yidiny, has a bisyllabic minimum word requirement In the former language one sees minimality in the imperative formation In SiSwati the imperative form consists of the stem minus the prefix if the stem contains at least two syllables, e.g the imperative

of /+bóna/ 'to see' is [bona] However, monosyllabic verb stems are

augmented in the imperative with the suffix +ni, e.g the imperative of /kû+dlá/

'to eat' is [dláni] and not simply [dlá]

In other languages the restriction governing minimal word size involves not the syllable, but the mora For example, in the Cushitic language Iraqw the smallest pword is bimoraic, i.e pwords of the form V: and VC are allowed because these sequences contain two morae, whereas those consisting of a single short vowel like CV are not (Mous 1993: 26)

Hall (this volume) argues that German pwords are minimally bimoraic and that this requirement governing the size of pwords accounts for the fact that the element that remains after deletion in (4) above is minimally a pword, e.g

Tief- and mütter- in the sentences in (4a) This analysis also explains why

German prefixes can only function as 'remnants' if they satisfy the bimoraic

minimality condition For example, the phrase beladen und entladen 'loaded and unloaded' can surface as be- und entladen only when the prefix be- is

pronounced with a long vowel, i.e [be:] By contrast, the pronunciation [e],

which is correct when be- surfaces as a bound morpheme in examples like

beladen, is ungrammatical when the stem is absent

It should be noted that not all languages have minimal word restrictions For example, in Irish many pwords consist of a single mora (Green 1997) One might be tempted to argue that only languages with length distinctions can have a bimoraic minimum, but this requirement fails for Irish, which contrasts short and long vowels

3 The pword and Prosodic Phonology

The pword is usually recognized as simply one of several prosodic constituents

In section 3.1 and 3.2 I discuss the hierarchical relationship that relates these constituents as well the principles constraining this hierarchy

THE PHONOLOGICAL WORD: A REVIEW 9

3.1 The prosodic hierarchy

According to the theory of Prosodic Phonology as espoused by Selkirk (1980a, b), Booij (1983), Nespor & Vogel (1986), and others, universal grammar speci­fies a series of prosodie constituents, which are arranged in a hierarchy The constituents in this hierarchy differ slightly from author to author, but almost all phonologists recognize at least the six constituents in (6).6

(6) phonological utterance (U)

intonational phrase (IP) phonological phrase () phonological word (ω) foot (F) syllable (a)

According to the hierarchy in (6) the pword is a constituent that dominates the foot, but which is dominated itself by the phonological phrase

Some authors have proposed that the hierarchy in (6) be supplemented by additional constituents as well, e.g the clitic group (Nespor & Vogel 1986, Hayes 1989, Vogel 1990), and the phonological stem ('pstem'; Inkelas 1990) Downing (this volume) argues that SiSwati and KiNande have a pstem in addi­tion to a pword, her evidence being that pstems obey minimality constraints that differ from the ones governing pwords and that pstems constitute the domain for phonological rules that is distinct from the pword.7 The clitic group has been criticized by several authors and has therefore been abandoned by most proponents of Prosodie Phonology (e.g Booij 1988: 527, Inkelas 1990, Zec & Inkelas 1991, Booij 1996, Peperkamp 1996, 1997) In the present

6 As Booij (1983) points out, the idea of a hierarchy consisting of prosodie constituents was often assumed by pregenerative linguists as well (e.g Hocket 1955, Haugen 1956)

Many authors consider the mora to be a prosodie constituent that is immediately dominated

by the syllable

7 In her analysis of the Northern Athabaskan language Slave, Rice (1993) argues that a single grammatical word (i.e a verb) is parsed as a phonological phrase, which in turn consists of 'words' and 'small words' The latter two units are prosodie entities that are the domains for phonological rules

(8) ( ) IP

( )( ) 

( ) ( ) ( ) ω ( ) ( ) ( ) ( ) ( ) ( ) F

Two examples of prosodie representations that are ill-formed on the basis of the principle in (7) are presented in (9)

(9) a ω b ω

F ω

a a

(9a) involves the 'skipping' of a level and (9b) illustrates recursivity

Much current work in Prosodie Phonology has argued that two aspects of the Strict Layer Hypothesis, namely, the skipping of levels in (9a), and recur­sivity in (9b), should be allowed (Selkirk 1995, Booij 1996, Peperkamp 1996, 1997).8 Operating within the Optimality Theoretic framework, Selkirk therefore

8 See also Cohn (1989: 200), who notes that prefixes in Indonesian cannot belong to the pword

of the stem to which they are attached Cohn concludes that prefixes cannot themselves be independent pwords and therefore proposes that prefix+stem sequences in Indonesian have a representation similar to the one in (9a), in which a prosodie level has been skipped (Cohn assumes that the syllable of the prefix is linked to a clitic group and not to the pword)

THE PHONOLOGICAL WORD: A REVIEW 11

replaces the Strict Layer Hypothesis in (7) with a series of four separate constraints in (10)

(10) a LAYEREDNESS: No C i dominates a Cj, j > i

b HEADEDNESS: Any C i must dominate a C i-1

 NONRECURSIVITY: No C i dominates another C i

d EXHAUSTIVITY: No C i immediately dominates   k  < i-1

The structures in (9a) and (9b) violate EXHAUSTIVITY and NONRECURSIVITY respectively Selkirk concludes that these two constraints are violable and that languages that allow them rank the relevant constraint(s) low

In several of the articles in the present volume the authors assume that the Strict Layer Hypothesis in (7) is violable (e.g Beck, Booij, Hall, Raffelsiefen, Vigário) and consequently propose prosodic representations like the ones in (9a) and/or (9b) The details of these analyses will be outlined in the following section

3.2 The prosodic representation of clitics

In many languages with minimal word constraints 'function' words such as pronouns, determiners and auxiliary verbs can be smaller than the predicted minimum For example, in the Cushitic language Iraqw referred to above, a language with a bimoraic minimality requirement, there are some function words of the shape CV, e.g certain forms of the verb 'to be' (Mous 1993) On the basis of these kinds of examples most phonologists conclude that weak forms of functions words, in contrast to lexical words, are not independent pwords (see, for example, Selkirk 1995, Booij 1995, 1996, Peperkamp 1996, 1997)

Several authors in the present volume examine the phonology of function words For example, Hall (this volume) and Vigário (this volume) demonstrate that the weak forms of function words in German and European Portuguese respectively are exempt from the bimoraic minimum that holds for pwords in both of these languages The conclusion these authors draw is that the weak forms of function words cannot be considered to be independent pwords According to Beck (this volume) there are two word classes in Lushootseed, predicative and nonpredicative, where, generally speaking, only the former function in the phonology as pwords This distinction between these two word classes is roughly the same as the one usually drawn between lexical words and function words respectively, but as Beck points out the distinction is not entirely clear cut One complication is that certain predicative words (e.g the

(11) Bob's swimming

Note that this example also illustrates the non-isomorphy between syntactic and prosodic structure The enclitic [z] fuses phonologically with the subject of

the sentence Bob (in a way to be made explicit below) and consequently under­

goes voicing assimilation, but [z] forms a syntactic unit with the following

word, swimming

One question that has been discussed at great length in recent years is the prosodie representation of clitics Thus, given a host+enclitic or proclitic+host sequence, how does one capture formally the prosodie constituency, as, for

instance, the sequence Bob and s in (11)? In this particular example the [z] clearly belongs to the same syllable (and pword) as the host Bob, but when the

clitic contains a vowel the prosodie representation of host+clitic and clitic+host sequences is far from obvious A number of competing proposals have been made, most of which refer crucially to the pword For example, several authors have argued that host+enclitic sequences in various languages constitute a single pword, as in (12a) (e.g Lahiri et al 1990 and Booij 1995, 1996 for Dutch, Peperkamp 1997 for the Lucanian dialect of Italian).9

(12) a (host enclitic) ω b ((host) ω enclitic) φ  ((host) ω enclitic) ω

Sandler (this volume) discusses the prosodie mergings that occur when a content word combines with a following function word in Israeli Sign Lan­guage These mergings are evident in handshape assimilation and another pro­cess Sandler dubs 'coalescence' She argues that both processes suggest that a host+enclitic sequence form a single pword, as in (12a)

Other authors contend that the enclitic does not belong to the same pword

as the host, but that it is attached to a constituent higher up in the hierarchy One possible representation along these lines is presented in (12b), where the

9 See also Selkirk (1986: 387, 402, note 9), who writes that the pword can contain a sequence

of lexical word plus clitic

THE PHONOLOGICAL WORD: A REVIEW 13

enclitic and the host attach to the same phonological phrase See, for example, Peperkamp (1997), who proposes the representation in (12b) for Standard Italian Yet another possible representation for enclitics is the one in (12c) Here the pword is taken to be recursive (recall (9b)), and the enclitic is linked

to the higher of the two pwords The recursive structure in (12c) has been argued to be correct for Serbo-Croatian (Zec & Inkelas 1991), English (McCarthy 1993), the Cologne dialect of German (Kleinhenz 1997), and the Neapolitan dialect of Italian (Peperkamp 1997)

The same kinds of representational questions can be posed with respect to proclitics, which could in principle have the mirror image representations of (12) Several studies of the phonology of cliticization in Germanic languages have concluded that the mirror image of (12a) is not correct and that the proclitic cannot belong to the same pword as the host See Selkirk (1995) and Booij (1995, 1996), who hold that proclitic+host sequences in English and Dutch respectively have the mirror image representation of (12c)

Vigário (this volume) argues that pronominal enclitics in European Portu­guese have representations like the ones in (12a) because the sequence host+enclitic behaves phonologically like a single pword One argument for her conclusion comes from a rule that deletes a nonback vowel at the end of a pword, regardless of what follows In host+enclitic sequences like [pedeo] /pede+o/ '(he) asks for it' the /e/ does not delete, suggesting that the host /pede/

in the encliticized construction is not a separate pword Vigârio concludes that the correct prosodic structure for this and similar host+enclitic sequences is (pede+o)ω, thereby correctly predicting that nonback vowel deletion does not apply

Hall (this volume) follows a similar line of reasoning in arguing that (some) host+enclitic sequences in German are represented prosodically as in (12a) For example, in a sequence of host+enclitic where the host ends in a consonant and the enclitic begins with a vowel the correct prosodie structure is

(12a), e.g kommt es [kom.tes] 'is it coming' is parsed (kommt es)ω as opposed

to (kommt)ω es The former representation (in which the foot has been omitted)

14 T A HALL

In contrast to Vigário (and almost all other authors who have discussed the prosodic representation of clitics), Hall argues that a single language or dialect need not have a single prosodic representation for enclitics For instance, the example discussed in the previous paragraph requires the prosodie structure in (13) (i.e (12a)), but certain consonant-initial enclitics require a representation

along the lines of (12b) Thus, in kann sie [kanzi] 'can she' both (12a) and

(12c) are incorrect because [I] cannot stand at the end of a pword, as expressed

by the exceptionless constraint in (5)

The reason (13) — as opposed to the incorrect parsing (kommt)ω es — is correct is a consequence of the fact that the pword dominates the syllable in the prosodie hierarchy (recall (6)) What this dominance relationship means is that

if a (nonambisyllabic) consonant is in syllable-initial position, then this same consonant cannot be final in the pword This generalization is stated in (14)

(14) If C a is nonambisyllabic and syllable-initial, then C a is not in pword-final position

Aronoff & Sridhar (1983) and Hannahs (1995a) propose prosodie structures for English and French respectively, in which (14) is implicitly violated I comment on this in the following section

The generalization in (14) brings up the question what happens to pwords

in languages with resyllabification across word boundaries (e.g VC.#.VC → V.CVC) In some Romance languages like French, Spanish and Italian the final consonant of one word resyllabifies into the onset of a following vowel-initial word An example of this resyllabification is illustrated with the Italian phrase

bar aperto 'open bar' in (15) Peperkamp (1997) argues that the lexically

assigned prosodie structure in this example is as in (15a), where both bar and

aperto belong to separate syllables and separate pwords

THE PHONOLOGICAL WORD: A REVIEW 15

pword in (15b) is not bar, but instead ba, and the second one is not aperto, but instead raperto.10

The restructuring of pwords as described in the preceding paragraph brings

up two very general unresolved questions First, as I noted in section 1 above, it

is generally either explicit or implicit that the segments that make up the pword belong exclusively to a morpheme or to a sequence of more than one morpheme For example, Nespor & Vogel's (1986) algorithms that map morphological structure into prosodic structure (see, for example, (16) below) apply to morphemes and not to arbitrary sequences of sounds Note that the

pword structure in (15b) challenges this claim, since neither ba, nor raperto are

meaningful elements.11 Second, the first pword in (15b), namely ba, violates

the bisyllabic minimality requirement that Peperkamp proposes for Italian pwords One might respond that both of these conditions for pwords only hold lexically and that they 'turn off postlexically This analysis can be tested by determining whether or not other postlexical phonemena in Italian require reference to the minimality condition or to the requirement that all pwords correspond to one or more morphemes

4 The connnection between pwords and morphological structure

Many authors see the pword as a prosodic constituent that represents the interaction between the phonological and morphological components of the grammar (e.g Booij & Rubach 1984, Nespor & Vogel 1986, Szpyra 1989, Booij & Lieber 1993, Hannahs 1995a, b) For example, Nespor & Vogel (1986: 107) define the pword as 'the lowest constituent of the prosodie hierarchy which is constructed on the basis of mapping rules that make substantial use of nonphonological notions' This statement is intended to mean that in any given language the pword (as opposed to the syllable and the foot) consistently correlates with morphological boundaries Thus, in their 1986 book, Nespor & Vogel posit specific algorithms for individual languages that map morphologi­cal structure into pwords An example of an algorithm along these lines for Hungarian is presented in (17)

10 A possible alternative to Peperkamp's treatment of Italian resyllabification is the analysis of very similar data in French by Booij (1983: 270) Booij concludes that phrasal resyllabification

in that language implies that each phonological phrase consists of a single pword

111 On occasion one encounters proposals in which pwords can consist of an 'arbitrary' se­

quence of sounds that do not correspond to one or more morphemes For example, Wenner­

strom (1993) argues that the string 'ternal' in words like internal and external is a separate

pword Inkelas (1993) parses English words with antepenultimate stress in such a way that the final syllable is 'stray', e.g () ω 1 See also Booij (this volume), who treats certain mono- morphemic Dutch words, e.g [a:lmu:s] 'alms' as 'prosodie compunds', e.g (:1) ω (mu:s) ω

16 T A HALL

b stem+suffix → (stem+suffix) ω

 prefix+stem → (prefix) ω (stem) ω

d stem+stem → (stem) ω (stem) ω

While none of the authors in the present volume posit explicit algorithms like the one in (16), all implicitly assume that there is a direct connection between morphological boundaries and pword boundaries

Nespor & Vogel (1986) see the pword as a prosodic constituent that is always 'word-internal' in the sense that it cannot span two or more grammati­cal words Other authors contend that the pword can be composed of more than one grammatical word One example of this situation is a representation like the one in (12a), in which a host+enclitic count as a single pword Other authors argue that two adjacent lexical words can under some syntactically defined conditions form one pword (see Myers 1995 for Shona)

One way languages can differ is in the algorithm that maps tic structure into pwords For example, many languages (e.g Hungarian) parse

morphosyntac-a sequence of stem plus suffix morphosyntac-as morphosyntac-a single pword, morphosyntac-as in (17morphosyntac-a), wheremorphosyntac-as other languages apparently do not parse (some) suffixes with the pword of the stem

to which they attach, as in (17b) Examples of languages of the latter type are Yidiny (Dixon 1977a, b), English (Aronoff & Sridhar 1983, Inkelas 1990), and German (Booij 1985, Wiese 1996) Following Dixon (1977a, b) and subse­quent writers, I refer to suffixes of the (17a) type as 'cohering' and those of the (17b) type as 'noncohering'.12

(17) a (stem suffix) ω b (stem) ω suffix

 (prefix stem) ω d prefix (stem) ω

The mirror image representations for prefix+stem sequences are displayed in (17c, d) respectively

An intriguing question is whether or not one can predict in any way under what conditions an affix be considered cohering, as in (17a), or noncohering, as

in (17b) Some phonologists have attempted to answer this question by corre­lating pwords with the two 'types' of affixation For example, Siegel (1974) and many subsequent linguists have argued that English requires a distinction

12 One question I have not considered here is whether or not the noncohering suffix in (17b) has its own pword or if it is attached to a constituent higher up in the prosodic hierarchy, as in (12b) I refer to the suffix in both configurations as 'noncohering'; hence, this question is not relevant for purposes of this paper

THE PHONOLOGICAL WORD: A REVIEW 17

between affixes of 'Class  (e.g +ity, in+) and 'Class IF (e.g +ness, un+)

Szpyra (1989) proposes that Class I affixes are incorporated into the pword of the stem to which they attach, as in (17a, c), whereas Class II affixes are not, as

in (17b, d) Two examples are provided in (18)

un+believable → (un) ω (believable) ω

The contrastive parsings in (18) are commonly assumed to derive support from

nasal place assimilation Thus, words like impolite would have to be parsed

(impolite)ω on this view because the rule that assimilates /n/ to the place of articulation of a following consonant only applies when these two segments belong to the same pword In contrast, there is no obligatory assimilation of the

/n/ in the un+ prefix, e.g unbelievable I return to the examples in (18) and the

English rule of nasal place assimilation in the following section.13

5 Conflicts among diagnostics for pwords

An interesting and often neglected question involves 'conflicting' criteria for pwords The general difficulty can be illustrated by examining Aronoff & Sridhar's (1983) treatment of English Aronoff & Sridhar argue that stress patterns are the main diagnostic for pword structure in English Specifically, they argue that English suffixes that are stress-shifting are cohering, e.g (feminin+ity)ω, whereas those that are stress neutral are noncohering, e.g (feminine)ωness Thus, a word like alphabetize with the stress-neutral suffix

+ize has the noncohering prosodic representation in (19)

(19) (alphabet) ω ize

Inkelas (1990, 1993) adopts a representation for English words with antepen­

ultimate stress, e.g Pamela, recall note 11, along the lines of the one in (19)

13 Nespor & Vogel (1986: 136ff.) write that in certain languages a pword can be a morpheme that is marked with an arbitrary 'diactitic feature' — a treatment that implies that the pword in such languages is not predictable from morphosyntactic structure However, the only example they give of a 'nonpredictable' pword is the Dutch morpheme 'achtig', which they assume, following van der Hulst (1984), is a suffix Several convincing arguments suggest that, in

contrast to 'true' vowel-initial suffixes, the prosodic structure of words containing achtig are noncohering and that achtig is dominated by its own pword, e.g a word like rood+achtig 'red-

like' is prosodically (rood) ω (achtig) ω Interestingly, in the literature on German phonology many researchers consider the German cognate 'artig' to be 'predictably' a separate pword (see Olson 1988, Hall 1992, Wiese 1996) The reason is that the latter linguists do not treat 'artig' as a suffix, but instead as the second part of a compound

18 T A HALL

As Raffelsiefen (to appear) points out, the problem with the representation

in (19) is that it violates the exceptionless generalization stated in (14): The /t/

in (19) is syllable-initial, but pword-final, contrary to what one would expect

if the pword dominates the syllable in the prosodiω hierarchy Since the condi­tion in (14) is universal and inviolable, one would probably conclude from this example that stress cannot be the correct diagnostic for English pwords However, in other languages the choice of which criterion 'wins' is not so clear

Hall (this volume) notes that the Coordinate Structure Deletion data in (4) also involve a set of two conflicting criteria for pwords As mentioned above, Booij (1985) and Wiese (1996) argue that the element that deletes in coordinate structures like the ones in (4a) is a pword Consistency would force one to

regard the diminutive suffix +chen as a pword as well, since this suffix can also

delete, as illustrated in (20)

(20) Brüderchen und Schwesterchen 'little brother and little sister'

Brüder- und Schwesterchen

The problem with analyzing +chen as a pword is that this suffix contains schwa

as its only nuclear vowel Were +chen a pword, then it would violate the

otherwise exceptional generalization that lexical words (i.e pwords) must contain at least one full vowel

Raffelsiefen (this volume) argues that the only diagnostics that reliably predict pwords for English are those pertaining to the prosodic hierarchy, i.e mora, syllable and foot Thus, she demonstrates that various stress patterns, minimal word requirements and syllabification conditions all conspire to predict under what conditions prefixes and suffixes are cohering or noncoher-

ing For example, Raffelsiefen argues that the English prefixes in+ and un+ are

parsed as separate pwords from the base to which they attach Two representa­tive examples are provided in (21)

(21) im+polite: [,Impe'lait] → (im) ω (pelait) ω

un+aware: [une'wer] → (un) ω (ewεr) ω

The hearer knows on the basis of the stress patterns that the words like impolite and unaware are parsed as in (21) Thus, evidence for the parsing in (21) as opposed to the one in (18) is that the stress pattern of words with in+ and those with un+ are identical and that the parsing of a word like impolite as a single

THE PHONOLOGICAL WORD: A REVIEW 19

pword would require a stress pattern that is otherwise nonattested within an English pword

Raffelsiefen also shows that rules/constraints referring to prosodic proper­ties like stress, syllabification and minimal word requirements also correlate with semantic compositionahty and the wordhood of the base Thus, examples like the ones in (21) above have 'predictable' meanings, i.e compositional semantics, and they consist of a prefix plus a base that is always an existing word.14

In contrast to the prosodic properties listed above, Raffelsiefen argues that segmental evidence, e.g assimilation rules, cannot be used as diagnostics for pword structure in English because the pwords that would be required for these rules do not correlate with the pwords predicted by the prosodie evidence Raffelsiefen concludes that the English rule of nasal place assimilation discussed in the previous section not a reliable diagnostic for pword structure Note that Raffelsiefen's claim is clearly at odds with the majority of the linguists cited in §2.1, who see segmental rules like vowel harmony in Hun­garian as reliable diagnostics for pword structure

An interesting question for future work is to determine whether or not the diagnostics for pwords proposed for individual languages truly correlate to consistently predict pword structure

or if there are conflicts like the ones described above A second question is whether or not segmental rules like assimilations in languages other than English can be shown to be reliable diagnostics for pwords, as is generally assumed in the literature

6 Conclusion

To summarize, all of the authors in the present volume argue that the pword is

an indispensible prosodie constituent, that the pword is not always isomorphic with the grammatical word, and that there is a direct connection between morphosyntactic boundaries and pword boundaries Most of the authors reject

Wennerstrom (1993) also argues that English words with in+ and un+ are represented

prosodically as (21), but in contrast to Raffelsiefen, her main diagnostic is whether or not the prefix can be focussed: In her model, a prefix is parsed as a separate pword iff it can be focussed Wennerstrom also claims that focussability correlates with what she refers to as 'semantic analyzability' However, Wennerstrom's definition of 'semantically analyzable' is

unclear because she treats English words like internal and external as in (21), e.g

(in) ω (ternal) ω , even though the string 'ternal' is meaningless See Raffelsiefen (this volume), who points out that Wennerstrom's 'focus' diagnostic does not correlate with other criteria for pwords

References

Aoki, H 1966 "Nez Perce Vowel Harmony and Proto-Sahaptian Vowels" Language 42

759-767

Aronoff, Mark & S N Sridhar 1983 "Morphological Levels in English and Kannada; or,

Atarizing Reagan" Papers from the Paresession on Phonology, Morphology, and Syntax Chicago 22.-23 April 1983 ed by John F Richardson, Mitchell Marks & Amy Chuker-

Booij, Geert 1985 "Coordination Reduction in Complex Words: A case for prosodie

phonology" Advances in Nonlinear Phonology ed by Harry van der Hulst & Norval

Smith, 143-160 Dordrecht: Foris

Booij, Geert 1988 Review of Nespor & Vogel 1986 Journal of Linguistics 24 515-525 Booij, Geert 1995 The Phonology of Dutch Oxford: Clarendon Press

Booij, Geert 1996 "Cliticization as Prosodie Integration: The case of Dutch" The Linguistic Review 13 219-242

Booij, Geert & Rochelle Lieber 1993 "On the Simulaneity of Morphological and Prosodie

Structure" Studies in Lexical Phonology ed by Sharon Hargus & Ellen Kaisse, 23-44

San Diego: Academic Press

Booij, Geert & Jerzy Rubach 1984 "Morphological and Prosodie Domains in Lexical

Phonology" Phonology Yearbook 1.1-27

Booij, Geert & Jerzy Rubach 1987 "Postcyclic Versus Postlexical Rules in Lexical

Phonology" Linguistic Inquiry 18.1-44

Cohn, Abigail  1989 "Stress in Indonesian and Bracketing Paradoxes" Natural Language and Linguistic Theory! 167-216

Davis, Stuart 1991 "Coronals and the Phonotactics of Nonadjacent Consonants in English"

The Special Status of Coronals Internal and External Evidence ed by Carole Paradis &

Jean-François Prunet, 49-60 New York: Academic Press

Dixon, R M W 1977a A Grammar of Yidin Cambridge: Cambridge University Press

Dixon, R M W 1977b "Some Phonological Rules of Yidin" Linguistic Inquiry 8.1-34 Dixon, R M W 1980 The Languages of Australia Cambridge: Cambridge University Press Green, Antony Dubach 1997 The Prosodie Structure of Irish, Scots Gaelic, and Manx Ph.D

dissertation Cornell University

Hall, T A 1992 Syllable Structure and Syllable Related Processes in German Tübingen:

THE PHONOLOGICAL WORD: A REVIEW 21

Haugen, Einar 1956 "The Syllable in Phonological Description" For Roman Jacobson ed by

Morris Halle et al., 213-221 The Hague: Mouton

Hayes, Bruce 1989 "The Prosodic Hierarchy in Meter" Rhythm and Meter ed by Paul

Kiparsky & G Youmans, 201-260 Orlando, Florida: Academic

Hockett, Charles 1955 A Manual of Phonology Baltimore: Waverly

Hooper, Joan 1972 "The Syllable in Phonological Theory" Language 48 525-540

Hooper, Joan 1976 An Introduction to Natural Generative Phonology New York: Academic

Press

Hulst, Harry van der 1984 Syllable Structure and Stress in Dutch Dordrecht: Foris

Inkelas, Sharon 1990 Prosodic Constituency in the Lexicon New York: Garland

Inkelas, Sharon 1993 "Deriving Cyclicity" Studies in Lexical Phonology ed by Sharon

Hargus & Ellen Kaisse, 75-110 San Diego: Academic Press

Kaisse, Ellen & Patricia Shaw 1985 "On the Theory of Lexical Phonology" Phonology Yearbook! 1-30

Kang, Ongmi 1991 Korean Prosodie Phonology Ph.D Dissertation University of

Washington

Kiparsky, Paul 1982 "Lexical Phonology and Morphology" Linguistics in the Morning Calm

ed by I.-S Yang, 3-91 Seoul: Hanshin

Kleinhenz, Ursula 1997 On Words and Phrases in Phonology A Comparative Study with Focus on German Ph.D dissertation University of Tübingen

Lahiri, Aditi, Allard Jongman & Joan A Sereno 1990 "The Prominimal Clitic [dar] in Dutch:

A theoretical and experimental approach" Yearbook of Morphology, Volume 3 ed by

Geert Booij & Jap van Marie, 115-127 Dordrecht: Kluwer

Leben, William R & Firmin Ahoua 1997 "Prosodie Domains in Baule" Phonology 14

McCarthy, John & Alan Prince 1986 "Prosodie Morphology" Unpublished Ms

McCarthy, John & Alan Prince 1990 "Foot and Word in Prosodie Morphology: The Arabic

broken plural" Natural Language and Linguistic Theory 8 209-284

McCarthy John & Alan Prince 1993 "Generalized Alignment" Yearbook of Morphology Volume 6 ed by Geert Booij & Jap van Marie, 79-153 Dordrecht: Kluwer

McCarthy, John & Alan Prince 1994 "Prosodie Morphology I: Constraint interaction and satisfaction" Ms

Mous, Marten 1993 A Grammar of Iraqw Hamburg: Buske

Myers, Scott 1995 "The Phonological Word in Shona" Bantu Phonology and Morphology ed

by Francis Katamba, 69-92 München: Lincom

Nespor, Marina 1986 "The Phonological Word in Greek and Italian" Sandhi Phenomena in the Languages of Europe ed by Henning Andersen, 65-74 Berlin: Mouton

Nespor, Marina & Irene Vogel 1986 Prosodie Phonology Dordrecht: Foris

Nespor, Marina & Angela Ralli (1996) "Morphology-Phonology Interface: Phonological

domains in Greek compounds" The Linguistic Review 13 357-382

Olson, Susan 1988 "Flickzeug vs abgasarm' Eine Studie zur Analogie in der Wortbildung" Semper idem et novus: Festschrift for Frank Banta ed by Francis G Gentry, 75-97

Göppingen: Kümmerle

Paradis, Carole & Jean-François Prunet 1993 "On the Validity of Morpheme Structure

Constraints" Canadian Journal of Linguistics 38 236-256

22 T A HALL

Peperkamp, Sharon 1996 "On the Prosodic Representation of Clitics" Interfaces in Phonology ed by Ursula Kleinhenz, 102-127 Berlin: Akademie Verlag

Peperkamp, Sharon 1997 Prosodic Words The Hague: Holland Academic Press

Prince, Alan & Paul Smolensky 1993 "Optimality Theory Constraint Interaction in Generative Grammar" Ms

Raffelsiefen, Renate 1993 "Relating Words A Model of Base Recognition Part I" Linguistic Analysis 23 3-159

Raffelsiefen, Renate to appear "Phonological Constraints in English Word Formation"

Yearbook of Morphology ed by Geert Booij & Jap van Marie Dordrecht: Kluwer

Rice, Keren 1993 "The Structure of the Slave (Northern Athabaskan) Verb" Studies in Lexical Phonology ed by Sharon Hargus & Ellen Kaisse, 145-171 San Diego: Academic

Selkirk, Elisabeth O 1978 "On Prosodie Structure and its Relation to Syntactic Structure"

Nordic Prosody II ed by Thorstein Fretheim Trondheim: TAPIR

Selkirk, Elisabeth O 1980a "The Role of Prosodie Categories in English Word Stress"

Linguistic Inquiry 11 563-605

Selkirk, Elisabeth O 1980b "Prosodie Domains in Phonology: Sanskrit revisited" Juncture A Collection of Original Papers ed by Mark Aronoff & Mary-Louise Kean, 107-129

Saratoga: Anma Libri

Selkirk, Elisabeth O 1984 Phonology and Syntax The Relation between Sound and Structure

Cambridge, Mass.: MIT Press

Selkirk, Elisabeth O 1986 "On Derived Domains in Sentence Phonology" Phonology Yearbook 3.371-405

Selkirk, Elisabeth 1995 "The Prosodie Structure of Function Words" Papers in Optimality Theory ed by J N Beekman et a l , 439-469 Amherst

Siegel, Dorothy 1974 Topics in English Morphology Ph.D dissertation MIT

Smith, George 1996 "Das 'phonologische Wort' in der nicht-linearen Phonologie" M.A thesis Free University of Berlin

Szpyra, Jolanta 1989 The Phonology-Morphology Interface London: Routledge

Tortora, Christina M 1993 "Stop Voicing and the Phonological Word in Dakota"

Proceedings of the 9th Eastern States Conference on Linguistics ed by Michael Bernstein,

Vogel, Irene 1991 "Level Ordering in Italian Lexical Phonology?" Certamen Phonologicum II

ed by P M Bertinetto & M Loporcaro, 81-101 Torino: Rosenberg & Sellier

Wennerstrom, Ann 1993 "Focus on the Prefix: Evidence for word-internal prosodie words"

Phonology 10 309-324

Wiese, Richard 1996 The Phonology of German Oxford: Clarendon Press

Yu, Si-Taek 1992 UnterSpezifikation in der Phonologie des Deutschen Tübingen: Niemeyer

Zec, Draga & Sharon Inkelas 1991 "The Place of Clitics in the Prosodie Hierarchy"

Proceedings of the 10 th West Coast Conference on Formal Linguistics ed by Dawn Bates,

505-519

Zsiga, Elizabeth  1992 "A Mismatch between Morphological and Prosodie Domains:

Evidence from two Igbo rales" Phonology 9 101-135

WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE*

DAVID BECK

University of Toronto

1 Introduction

As with so many of the basic terms in linguistics, the term "word" is one which

is often taken for granted and, although it is characterized differently in differ­ent domains of linguistic research, there is often a tacit assumption that the entity as defined at one level of description is the same entity singled out at another level Thus, in the syntax "word" is often used to mean a "single lexical item" whose linear position and inflectional properties are manipulated by the morphosyntax (Trask 1993), whereas the phonological word is generally defined in terms of the domain of lexical stress assignment or the application of other low-level phonological rules (e.g

Nespor & Vogel

1986) The assump­tion that what is a word for the syntax is equivalent to what is a word for the phonology, however, has been called into question by a good deal of recent work, including a number of the papers in this volume, which seem to indicate that the two types of word do not necessarily match Although there is a good deal of symmetry between the two in the familiar Indo-European languages that have been the focus of the bulk of investigation, the greatest challenge to the idea that the syntactic word is isomorphic with the phonological word comes from languages which are, relatively-speaking, morphologically com-

* Thanks are owed to Jean Balcaen, Trisha Causley, and a number of attentive audiences for listening to more confused earlier incarnations of this paper, to Keren Rice and to Elan Dresher for helpful comments on the current version, and to Elan for convincing me that this stuff was kind of cool The assistance of David Bennett has also been invaluable in sorting out some of the intricacies of the intonational phrase While it is hard to believe that with such able help any errors could possibly remain, those that do can be ascribed to me alone This work has been supported by the Social Sciences and Humanities Research Council of Canada through a Doctoral Fellowship to the author and a SSHRC research grant to Elan Dresher and Keren

24 DAVID BECK

plex (Evans 1986 on Yidiny; Czaykowska-Higgins 1997 on Nxa?amxcin) or which are classified as polysynthetic (Rice 1993 on Slave; Russell (this volume) on Cree and Dakota) Work on the latter group in particular seems to suggest that what is a word for the morphosyntax may actually correspond to a higher-level phonological unit consisting of more than a single phonological word — perhaps even to the phonological phrase

In this article I will examine some data from the Salishan language Lushootseed which shows that, in fact, the phonological word differs markedly from what can reasonably be called a word in the morphosyntax even in languages that are only mildly polysynthetic, and that what is called a word in the syntax may not be a word in the phonology To begin, in section 2, I present the basic patterns of Lushootseed prosody and the constraints that govern phonological phrasing, using an informal definition of the two primary units of the phrasing process, the clitic and the phonological word; following this, in section 3, I discuss the interaction — or, more accurately, the lack thereof — between the processes of phonological phrasing and the syntax of

the language In section 4, I return to the ad hoc notions of clitic and word used

earlier and discuss how these might be defined for the purposes of phonological phrasing I will argue that these units are not amenable to a syntactic definition, but instead depend on a combination of lexical and phonological criteria, the implications of which are discussed in the conclusion to this article

Lushootseed is a language of the Salishan family spoken in the Puget Sound area of Washington State It has the consonantal inventory given in (1)

( 1 ) Lushootseed consonantal inventory

labial alv lateral al-pal velar

unrnd rnd

uvular unrnd rnd

WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE 25

There are three vowels — /i/, /u/, and /a/ — which occasionally show a phone­mic long/short distinction, and /e/, which has only the short variant Word-level stress is largely predictable and all vowels can be stressed, stress falling on the first non-schwa of the morphological root of a major class lexical item (Bianco 1995) or any other lexeme that can be considered a phonological word

My primary source of data is Hess (1995), which consists of a grammar, a reader, and four stories on an accompanying tape; all four texts are traditional legends told by the same consultant, Edward Sam, a speaker of the Snohomish dialect of Northern Lushootseed, recorded in the field in the early 1960s Of the four stories, three have been analyzed for this paper using Signalyze 3.12 voice-analysis software; the examples given below come from one of these three stories — "Little Mink and his Younger Cousin, Tetyika", "Coyote and the Big Stone", and "Bear and Fish-Hawk" — and will be cited by source text and line number Additional data has been drawn from a tape recording of the story of "Pheasant and Raven" as told by Martha Lamont (Hess, to appear), also a speaker of Snohomish; this recording was subjected to waveform and pitch-extraction analysis using WinCECIL software In most cases data is given in phonemic transcription, except where the phonetics is at issue; simi­larly, interlinear glosses are in general as detailed as possible, but in some cases words are left unanalyzed when their composition is not relevant to the discus­sion

2 Phonological Phrasing

The formation of phonological phrases (PhP) in Lushootseed is closely tied to the notion of the phonological word, and the building of phrases in many ways resembles the building of syllables — so much so, that in the following sections I will borrow a great deal of terminology from syllabic phonology as a descriptive convention Like the syllable, the Lushootseed phonological phrase

is built up around a single head or phrasal nucleus, and the ideal or canonical phrase allows for a single initial non-head element — the phrasal onset; on the other hand, phrasing does not allow for any element to follow the head (i.e a phrasal coda) The head of a phonological phrase is a phonological word (W) in the sense that it can stand on its own in a phrase and is a legitimate target for cliticization The head of a phrase bears the unique phrasal stress as marked by amplitude and, usually, vowel length (Barthmaier 1998) As with syllabic nuclei, it is the position of the phrasal heads in the sentence that determines the associations of the various non-heads, or clitics (C), within the next-higher level units Clitics can not be stressed and must be cliticized or affixed to a phonological word, optimally forming a CW phonological phrase

26 DAVID BECK

The domain of phonological phrasing is the intonational phrase (IP), and within the IP lexical elements are parsed exhaustively into PhPs according to their status as phonological clitics or phonological words The IP in Lushoot-seed is marked by pitch-contours that peak on the first full (non-schwa) vowel, whether this vowel is stressed or not; IPs may also be set off by lengthy pauses

at their boundaries, although these are naturally reduced in more rapid speech Between the PhP and the phonological word, there seems to be no evidence for

a more articulated organization of the Lushootseed prosodic hierarchy While the C+W pattern of phonological phrasing described here is somewhat reminis­

cent of the behaviour of clitics in what has been called a clitic group (Nespor &

Vogel 1986; various papers in this volume), there is no evidence that these C+W units constitute an intermediary constituent that might then be recom-bined into PhPs Below the level of the W, the absence of predictable, non-lexicalized secondary stress seems to indicate the absence of metrical feet (or,

at best, argues for left-headed unbounded feet which are isomorphic with the prosodic word); by the same token, the inability of non-words to bear stress makes it difficult to argue for their being parsed into feet at all Thus, at this stage of the game, a Lushootseed sentence appears to have the prosodie struc­ture shown in (2):

(2)

It should be noted, however, that the word-level prosodie phonology of this language awaits detailed investigation, and it seems possible that there are some intermediary domains for segmental alternations that could be used to define prosodie constituents below the word and above the syllable (cf Bianco

1995 on the behaviour of certain suffixes and, in a related language, ska-Higgins 1997) Fortunately, clarification of these issues is a bit beyond the scope of this paper, which is aimed at the processes above the word — or, more precisely, at the ways in which lexical elements are recombined into

Czaykow-WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE 27

phonological phrases What is really at stake here — and what the CW pattern outlined above and described in detail in section 2.1 serves to illustrate — is that in Lushootseed there is not a one-to-one mapping between what is treated

by the syntax as a word (i.e., a morphosyntactically autonomous entity) and what is recognized as a word (the domain of phonological affixation and stress assignment) by the phonology As we shall see in section 2.2, phonological phrasing is independent of syntactic structure, and it is the status of a lexical item as a phonological clitic or as a phonological word that counts in Lushoot­seed prosodic phonology

2.1 The Phonological Phrase

The rules or constraints that build phonological phrases in Lushootseed are quite straightforward and, as noted above, bear a strong resemblance to the rules used to form syllables in many languages Each phrase in Lushootseed is built up around a phonological word which serves as a kind of phrasal nucleus

A Lushootseed sentence can consist of a single word or a string of words, each constituting its own phrase (delimited here by parentheses), as in (3):1

(hay) (X iqag w il-li) (d-suq w suq w a?)

well-then come-out-[imp] lpo-[rdp] cousin

'well then, come out of there, my cousins' (Coyote 56)

The Lushootseed phonological phrase (PhP) is frequently set off from contigu­ous phrases by an audible pause, usually of approximately 50 to 100 ms; in rapid speech this pause may be smaller, but it is usually perceptible even in these circumstances by the lack of phonological interaction between segments located on either side of a phrasal boundary As in (3), a PhP may contain a

1 The abbreviations used here are: 1, 2, 3 = first-, second-, third-person; add = additive; appl = applicative;  = phonological clitic; caus = causative; D = deictic; IP = intonational phrase; intj

= interjection; irr = irrealis; l.o.c = lack of control; md = middle; neg = negative; np = nominalizing prefix; p = plural; PhP = phonological phrase; pnt = punctual; po = possessive; prog = progressive; prt = particle; rdp = reduplication; s = singular; stat = stative; subj = sub­ junctive; top = topic-marker; trm = transmutative; W = phonological word Boundaries are marked by: § = intonational phrase; )( = phonological phrase; + = phonological affix; - = morphological affix; ][ = syntactic phrase

28 DAVID BECK

single word, but more commonly a phrase consists of a word and one or more clitics or affixes, and in many cases constitutes an entire sentence, as in (4): (4) ( C W )

(put- t-as-X u-il)

really-now [past] - [stat] -thin- [trm]

Within the phrase, the phonological nucleus bears the single stress (marked by

a peak in amplitude) Thus, in (4) the unique stress falls on /u/ in the root of the

verb tasy úil 'was getting thin'; the adverb receives no stress and becomes a

clitic Phonologically, cliticization is marked by the lack of a pause between elements and, in some cases, the beginnings of coarticulatory assimilation at the word-clitic boundary

When sentences get more complex, they consist of more than one phrase, each containing a single word, and optimally a single clitic as in (5):

(5) a (  W ) (  W )

(ti?ii sbiaw) (g w әl ?ux - )

D coyote [top] go-now

(x w ul p áy ay) (ti?il s-?abyid~s) (ti?rl c'y a?)

only worthless D np-give-3po D stone

'what he gave to Stone [was] only junk' (Coyote 32)

As these examples show, the preferred phrasal pattern is one of procliticization, with a preceding clitic joining to a word to form a sort of phrasal "onset" Words never cliticize to words or share clitics between them This is also ap­parent in (6), which shows that when a C appears between two Ws, it adjoins to its right rather than to its left:

(6) a (C W ) (  W ) (  W) (W)

(huy q w u?-t-9b-9x ) (ti?il ?aciitalbix ) (tuul?al bәk w ) (cád) then gather- [caus]-[md] -now D people from all where 'then the people were gathered together from everywhere (Little Mink 47)

(hay) (cad lu-yәc-әb-tú-bicid-әx ) (dәg w i) (si?ab) (d-sya?ya?) well-then 1s [irr]-tell-[md]-[caus]-2s-now 2s noble lpo-friend 'well then, I will tell [it] to you now, my noble friend' (Little Mink 4)

WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE 29

In terms of a model, the patterns observed up to now could be handled either by some sort of association rule linking clitics rightward to the nearest head before linking them leftward, or, alternatively, could be described in terms the ranking

of various constraints on phrase-formation in the style of Optimality Theory ( — Prince & Smolensky 1993) Given that my aim here is to describe the patterns found in the data rather than to develop a rigorous theoretical appara­tus, I will informally opt for the latter as a descriptive convention, without going to great lengths to justify it in detail For the same reason, I will not make use of what are currently the standard  alignment constraints or try to position this work with respect to relevant theoretical issues current in the  literature Such activities are left to the interested reader For our purposes here all we need is a set of simple constraints, beginning with one stating a prefer­ence for phrasal "onsets" over phrasal "codas", which I will dub in the  spirit "No PhP-Coda" (NPC):

(7) NO PHP-CODA (NPC): A phrase must not contain an enclitic following the phrasal nucleus

This prevents non-words from becoming enclitics Similarly, there must be a constraint governing phrasal onsets, though what this might consist of is not yet apparent

So far, all of the examples shown have been cases with alternating CW patterns However, clitics often occur adjacent to one another as well, as in (8):

[(ti?ii bibscәbә) (ti?ii su?suq a?s)]

/(ti?ri bibscәb+?i) (ti?ii su?suq a?-s)/

D [rdp]mink+and D [rdp]cousin-3po

'Little Mink and his cousin (Little Mink 5)

[(tud z әláxadbidә1) (ti?il pәdt as)]

/(tu-d әlaxadbid+?al) (ti?il pәdt as)/

(ti?ii stulәk w )/

D river (Bear & Fish-Hawk 10)

30 DAVID BECK

While these sentences appear to be violations of the constraint against phrasal codas, in each of these cases the first clitic in a WCCW string has, in fact, been parsed into the preceding phrase, not as an enclitic, but as an affix (shown by the "+") Affixation or phonological incorporation can be distinguished from chticization in that where ordinary clitics generally retain their own shape and original segmental material, an incorporated clitic re-syllabifies with a stem and, in most cases, either loses a mora or some phonemic material, or triggers some phonological alternation such as consonant or schwa-deletion in the word

to which it attaches — all of which are processes typical of Lushootseed level phonology (cf the reduction of the past-time prefix /tu-/ to [t-] in (4) above) In all of the examples in (8), the incorporated clitic loses its onset, becomes a part of the final syllable of the preceding word, and undergoes vowel-reduction, /i/ > [ә] in (a) and /a/ > [ә] in (b) and (c) Other examples of­fer even more striking evidence for affixation:

word-(9) a (  W+C )

[(ti?il dәx w ?ibәSәl)]

/(ti?il dәx w -?ibәs+cәl)/

D np-walk+lp.po ' for our journey' (utterance-final) (Coyote 10)

b (  W ) ( W+C ) (  W )

[(puut ?әsp il) (ti sqabati?ә?) (hik c'Åa?)]

/(puut ?әs-p'il) (ti sq-abac+ti?ә?) (hik c'Åa?)/

really [stat]-flat D high-body+D big stone

'it [was] really flat up on top of the big stone' (Coyote 4)

Here incorporated clitics — in (a) the possessive pronominal cәl 'our' and in (b) the deictic ti?ә? — lose onsets somewhat more substantial than a glottal

stop and are resyllabified with their phrasal head; in (b) the final consonant in

sqabac undergoes deaffrication ([c] > [t]) In (10), the possessive pronominal

/cәl/ seen in (9a) loses its syllabic nucleus and is reduced to [ci]:

tuudi?) (tu-sluX'luX'+cai)/ yonder [past]-elders+lp.po

(Little Mink 2)

WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE 31

b (  W+C ) (  W )

[(dii dәx W ut'asadci) ( t i ? ә ? c'ya?)]

/(dii dәx w -?u-tasa-d+cәl) (ti?ә? c'Xa?)/

[focus] np-[pnt]-paid-[caus]+lp.po D stone

'this [is] why we are paying Stone' (Coyote 11) The next example contains two instances of affixation:

(11) ( C W+C) ( C W+C ) ( C W )

[(?al su?әlә) (ti?il s?úlax ii?) (k i g w әsbәk dx s)]

/(?al s-u-?ә-әd+?ә) (ti?il s?uladx +x ii?) (k i g w ә-s-bәk -dx -s)/

on np-[pnt]-eat+of D salmon+[neg] D [subj]-np-all-[l.o.c.]-3po 'as he ate the salmon, [he] couldn't eat it all' (Bear & Fish-Hawk 23)

In the second case, the onset of the incorporated clitic xw ii? '[neg]' assimilates

to the final element in the coda of s?uladxw 'salmon' and triggers the deletion

of the /d/ in the word-final coda of its head, as does the preposition ?ә in su?әlә, derived from /s?ufәd+?ә/ There are a number of other boundary phe­

nomena associated with incorporation, and while there is by no means enough space to go into all of them here, a few more will be dealt with in the context of prefixation, which provides an even clearer contrast between affixation and cliticization For the moment it is enough to note the distinctive behaviour of affixes as opposed to clitics: in the former there is a high degree of phonological incorporation, whereas in the latter the clitic more closely resem­bles its citation form

Because a single clitic is a legitimate phrasal onset, proclitics are not normally incorporated, whereas enclitics are inevitably so If the enclitic is treated as a suffix and hence part of the word forming the phrasal head, the result is a fairly consistent pattern of CW phrases In terms of constraints, this indicates that there is a requirement that phrasal onsets contain one and only one cliticized element, thereby forcing a phrase boundary between the clitics in

a WCCW sequence I will refer to this constraint as "Single Phrasal Onset" (SPO):

(12) SINGLE PHRASAL ONSET (SPO) A well-formed phrase will contain a single clitic preceding the phrasal nucleus

In addition, there must be a constraint (or pair of ranked constraints) preferring suffixation over prefixation, to prevent the creation of (C W)(C C+W) strings

in situations like (11), and we also need a constraint forcing clitics to be

associ-32 DAVID BECK

ated with phrasal heads and preventing the cliticization of clitics to other clitics:

(13) PARSE: Clitics must be linked to a proper phrasal head (a phonological word)

This serves to prevent two clitics joining together either to form a phrasal nucleus or to form a single unit which could constitute a complex phrasal onset (i.e (C+C W)) Finally, we need a constraint governing affixation, one which would be ranked below both NPC (hence, suffixation takes place over enclitici-zation) and SPO I will refer to this constraint as "Don't Incorporate" (DI):

(14) DON'T INCORPORATE (DI): *C+W; *W+C

This is a simple prohibition against phonological incorporation, in the spirit of the faithfulness constraints proposed by McCarthy & Prince (1993), which work to preserve the underlying form of a phonological string

2.2 Clitic Sequences within Phrases

In the previous section we examined data where the input to the phonology consisted of sentences with strings of no more than two consecutive clitics potentially separable by a phrase boundary Sometimes, however, the grammar creates sequences of two or more clitics which can not be divided into separate phrases In some cases, usually when the position of the "stray" clitic corre­sponds to an IP boundary, it is deleted:

(15) ( W +C) § ( C W )

(tuyәcәbtúb+cәd) ø (ti?il tu-d-yәlyәláb)

(tu-yәcәb-tu-b+cәd) ?ә (ti?rl tu-d-yәlyslâb)

[past]-tell-[caus]-[md]+ls of D [past]-lpo-[rdp]uncle-of Tate-parent

'I was told [this] by my great-aunts and uncles' (Little Mink 3)

Grammatically this sentence, a passive, requires the preposition ? marking the

oblique agent; in the utterance on tape, this particle does not surface, possibly having been "erased" by its failure to associate with a phrasal nucleus It should

be noted, however, that this is not a very well-attested phenomenon and may be attributable to speaker error A more common strategy for dealing with this sort

of situation is seen in (16), where the addition of an element as a proclitic causes the incorporation of the next element closer to the head as a prefix:

WORDS AND PROSODIC PHRASING IN LUSHOOTSEED NARRATIVE 33

[(x i?  ik adsukâwdx ) (ti?il sc'âli?)]

/(x i?  i+g w ә-ad-s-?u-k aw-dx ) (ti?il sc'ali?)/

[neg] D+[subj]-2po-np-chew-[l.o.c] D heart

'don't chew on [my] heart' (Little Mink 19)

[(yәxi+huy) (x i?  әxstáb) (dәx há?ls)]

/(yәxi+huy) (x i?  i+g w ә-stab) (dәx -ha?l-s)/

because+well [neg] D+[subj]-what np-good-3s

[(hik +әwә) (qa tiisәdә) (ti?il sbiaw)]

/(hik W +9ws) (qa ti?il+?irsәd+?ә) (ti?il sbiaw)/

big+[surprise] many D+relatives+of D coyote

'the relatives of Coyote really [are] very many' (Coyote 64)

Incorporation takes place at sentence boundaries and m strings where there would otherwise be three-clitic sequences; thus, WCCCW is parsed as (W+C)(C C+W) Just as in suffixation, a number of boundary phenomena can

be observed at work marking the phonological incorporation of the

clitic-cum-affix into the word: in (16a) we have /kwi # gwә-ads?uk awdxw/ collapsing into [kwikwadsuk áwdxw]; in (16b), /kwi gwә-stáb/ > [kwәxwstäb]; and in (c) /ti?ii ?iisәd ?ә/ > [tiisoda] Compare this last example with the phrasing in (17):

[(hay) (g w әl wiliq w id9x ) (ti?ii ?iisәds)]

/(hay) (g w ә] wiliq w id-әx ) (ti?il ?iisәd-s)/

well-then intj ask-now D relative-3po

'well then [he] asked his relatives' (Coyote 5) Here there is no incorporation of the deictic to the following word, and the clitic retains all of its phonological material

In terms of our constraint-based analysis, this behaviour must be accounted for by the interaction of the constraint hierarchy Clearly, if both NPC (which prohibits phrasal codas) and SPO (which governs the well-formedness of phrasal onsets) dominate DI, then a string of proclitics would be a violation of

a higher-ranked constraint (SPO, which requires a single proclitic as the phrasal