summarizing an article in a phonological topic.

How (not) to do phonological typology: the case of pitch-accent

Larry M. Hyman

Department of Linguistics, University of California, Berkeley, CA 94720-2650, USA

AbstractIn this paper I argue for a property-driven approach to phonological typology. Rather than seeking to classify or label

languages, the central goal of phonological typology is to determine how diﬀerent languages systematize the phonetic sub-

stance available to all languages. The paper focuses on a very murky area in phonological typology, word-prosodic sys-

tems. While there is agreement that certain properties converge to characterize two prosodic prototypes, tone and

stress, the term ‘‘pitch-accent ”is frequently adopted to refer to a defective tone system whose tone is obligatory, culmina-

tive, privative, metrical, and/or restricted in distribution. Drawing from a database of ca. 600 tone systems, I show that

none of these properties is found in all systems claimed to be accentual and that all ﬁve are amply attested in canonical

tone systems. Since all one can say is that alleged pitch-accent systems exhibit signiﬁcant constraints on the distribution

of their tonal contrasts, they do not constitute a coherent prosodic ‘‘type ”. Rather, alleged ‘‘pitch-accent ”systems freely

pick-and-choose properties from the tone and stress prototypes, producing mixed, ambiguous, and sometimes analytically

indeterminate systems which appear to be ‘‘intermediate ”. There thus is no pitch-accent prototype, nor can prosodic sys-

tems be treated as a continuum placed along a single linear dimension. The paper concludes that the goal of prosodic typol-

ogy should not be to classify languages, but rather the properties of their subsystems.

Keywords: Culminativity; Metrical structure pitch-accent; Privativity; Stress; Tonal density; Tone; Typology

1. Introduction

The goal of this paper is to argue for what I will term

PROPERTY-DRIVEN TYPOLOGY , particularly as it applies

to phonology, which I will illustrate via some of the claims made about so-called pitch-accent languages.

Within the literature, there have been two ways of talking about typology (and by extension, phonological

typology). The ﬁrst deﬁnes the goal of typology as the classiﬁcation of languages according to their properties.

Thus, for Hage`

ge (1992, p. 7) , typology strives to provide ‘‘ ...a principled way of classifying the languages of

the world by the most signiﬁcant properties which distinguish one from another. ”Vajda (2001) takes a similar

position with respect to phonological typology: ‘‘... it is possible to classify languages according to the

phonemes they contain ...typology is the study of structural features across languages. Phonological typology

involves comparing languages according to the number or type of sounds they contain. ”A second way of

doi:10.1016/j.langsci.2008.12.007

E-mail address: [email protected]

Available online at www.sciencedirect.com

Language Sciences 31 (2009) 213–238

www.elsevier.com/locate/langsci talking about typology is with direct reference to the properties themselves. AsPlank (2001, p. 1399)puts it,

‘‘Typology ...is not so much about the classiﬁcation of languages as about the distributions of individual

traits—units, categories, constructions, rules of all kinds—across the linguistic universe; these distributions,

not languages as such, are the primary objects of comparison. ”Both ways of talking about typology seem

always to be available: ‘‘ ...all synchronic typologies have this Janus-like nature in that the same data can

be utilized either for a typology of linguistic properties or a typology of individual languages ”(Greenberg,

1974, p. 14 ). The question is whether it makes a diﬀerence. I believe it does.

Consider, brieﬂy, what it has meant to be identiﬁed as a ‘‘P language ”, where P is some property. In the case

of ‘‘click language ”, this means the

PRESENCE of at least one click in the consonant inventory. In the case of

‘‘open-syllable language ”, this means the

ABSENCE of even one coda in the syllable inventory. (One could,

of course, invent the term ‘‘coda language ”instead.) A basic question is whether ‘‘click language ”and

‘‘open-syllable language ”are ‘‘types ”or just practical labels to designate presence vs. absence in inventories?

If they are types, then other things ought to follow from them, i.e. we should expect other properties to cor-

relate with them. This of course assumes, following Hage `

ge, that we have correctly chosen ‘‘the most signif-

icant properties which distinguish one [language] from another. ”Since the designation is recognized by

linguists, being a ‘‘click language ”is apparently felt to be more signiﬁcant than being, say, an ‘‘implosive lan-

guage ”or a ‘‘ﬂap language ”, even though we cannot draw inferences about cooccurrence tendencies given the

limited geographical distribution of clicks in the world’s languages. We might have more luck with ‘‘open-syl-

lable language ”, but why not ‘‘oral language ”(a language which has no nasality in its consonant or vowel

inventory) or ‘‘mellow language ”(a language which has no fricatives)? Finally, what about a language which

has one marginal click, or another which has marginal closed syllables (e.g. derived by a late deletion of ﬁnal

/i/)? Does it make sense to talk about continua such that a language with multiple clicks is a ‘‘very click lan-

guage ”and one with few closed syllables is an ‘‘almost open-syllable language ”? Of course we can always ﬁnd

a way to quantify properties such that the languages in which they occur can be placed along a linear contin-

uum. For instance, since all languages have oral stops, one can scale languages by the number of stop vs. non-

stop phonemes in their underlying inventories or by calculating the percentage of stops in the lexical entries or

in texts. On one end, the score may be 100% if the consonant system of Central Rotokas is analyzed as /p, t, k,

b, d, g/ rather than /p, t, k, b,|,g/( Firchow and Firchow, 1969; Maddieson, 1984 ). Depending on the analysis,

Rotokas could be referred to as a ‘‘stop language ”or a ‘‘high stop language ”. At the other end might be one of

the 13 three-stop languages in the UPSID database ( Maddieson and Precoda, 1990). For example, Aleut is

analyzed with only /t, k, q/ among a system of 24 consonant phonemes and might therefore be designated

a ‘‘low stop language ”.

While designations such as ‘‘high stop language ”and ‘‘low stop language ”may cause amusement, this is

exactly what the ﬁeld has been subjected to in the area of word-prosodic typology—but with a diﬀerence:

Whereas I have indicated one or another quantitative method to represent all languages along a linear stop

continuum, those opining that there is a continuum from stress to pitch-accent to tone typically fail to provide

an explicit algorithm which would place languages along a unidimensional scale. Of course, it might be pos-

sible to take a single prosodic property and scale it. For example, if languages were mapped according to their

‘‘tonal density ( Gussenhoven, 2004, p. 35 ), i.e. the percentage of syllables (or perhaps words) that require a

tone feature, we might get something like the continuum in (1).

On the one end is English, where no syllables have a lexical tone vs. Mandarin at the end where only certain

‘‘neutral ”tone syllables lack a lexical tone. In subsequent discussion I will have occasion to comment further

on the scale in (1), but note here that claims of a prosodic continuum are typically not based on a single prop-

erty, e.g. tonal density, but rather on a combination of properties, which, when all is said and done, may actu-

ally contradict each other, depending on what we decide to count, e.g. syllables vs. words. For example, a

language in which every word has to have a high tone may still have a greater percentage of toneless syllables

than another language in which a word need not have a high tone. In any case, a rigorous quantitative study of

[±tone] on either syllables or words has, to my knowledge, not been done. Instead, the use of the term ‘‘con-

tinuum ”usually refers to the fact that there are languages that somehow fall in between the clear notions we

(1) English- - -

-------W.Basque- ---------Tokyo Japanese- ---------Luganda- - - -------Mandarin

214 L.M. Hyman / Language Sciences 31 (2009) 213–238 associate with the endpoints of the scale in (1): English is a ‘‘stress language”, while Mandarin is a ‘‘tone lan-

guage ”. What about the intermediate languages? For some, one or more of these fall into an ‘‘accentual ”cat-

egory with English. Among the three intermediate languages in (1), there is a common feeling that W. Basque

is the least tonal, while Luganda is most tonal. However, in the absence of explicit criteria and argumentation,

where one draws the line between accent and tone becomes largely a matter of opinion—which, in turn, is no

way to do typology. In this study I present further evidence for the position that there is no pitch-accent prototype ( Hyman and

Wilson, 1992; Hyman, 2001a,b, Hyman, 2006 ). Rather, intermediate languages such as those in (1) ‘‘pick and

choose ”various properties which tend to cluster in prototypical stress vs. tone systems. In other words, if stress

systems typically have the four properties S1–S4 and tone systems typically have the four properties T1–T4, the

intermediate systems may have various combinations of these, for example [S1, S2, T3, T4], [S1, T2, S3, T4],

and [S1, T2, T3, S4], each of which exhibits two stress and two tone properties. It is these properties that con-

stitute the object of study in property-driven typology and which allow us to come up with an objective char-

acterization of a prosodic system. I begin in Section 2by ﬁrst providing deﬁnitions and prototypes of stress and

tone systems. In Section 3I show that a deﬁnition of a ‘‘pitch-accent system ”is too vague to be useful and that a

prototype cannot be established. Four potential properties of so-called pitch-accent systems are shown to be

available to canonical tone systems as well: obligatoriness, culminativity, privativity, metricality. I conclude

in Section 4with some ﬁnal remarks about phonological typology and prosodic systems.

2. Stress and tone systems While there is agreement that certain properties converge to characterize two prosodic prototypes, tone and

stress ( Martinet, 1960; Garde, 1967; McCawley, 1970; Hyman, 1977a, 2006; van der Hulst and Smith, 1988;

Hayata, 1999; Ding, 2006 , etc.), a common tripartite word-prosodic typology recognizes a third kind of pro-

sodic system, ‘‘pitch-accent ”, as in (2).

Widely accepted deﬁnitions are also provided in (2a,b), where stress is identiﬁed with abstract metrical struc-

ture and tone with pitch features present at the word level. Stated this way, stress and tone have virtually noth-

ing inherently in common: Stress is a structural property in which syllables are metrically hierarchized as

relatively strong vs. weak (however this contrast is realized phonetically), while tone is a featural property

referring to contrastive relative pitch. The question is: How do we deﬁne pitch-accent? To paraphrase McCaw-

ley (1970) , is it a system that comes close to one of the other two but does not quite make it, perhaps having an

incomplete metrical structure or only a slight need for pitch features? Alternatively, might pitch-accent refer to

a language which has both a stress and tone system? Either way, it would not be a third category in the sense

of having a non-overlapping deﬁnition, as stress and tone have.

Lehiste (1970, p. 1) once stated that ‘‘... a certain degree of vagueness seems to characterize most discus-

sions of prosodic features. ”Concerning word-prosodic typology, a lot of the vagueness and confusion has

stemmed from the lack of an explicit distinction between what constitutes a

DEFINITION vs. a PROTOTYPE . A def-

inition provides a minimum requirement that all instances of the phenomenon must meet: If a language does

not have word-level metrical structure, it does not have a stress system (as so deﬁned). If it does not have

word-level pitch features, it does not have a tone system. In this sense deﬁnitions are categorical: Depending

on the actual deﬁnition, something is vs. is not a stress system, and similarly with respect to tone system.

Viewed this way, the question is what must be present in a language to constitute a pitch-accent system (rather

than something else). A prototype, on the other hand, provides a cluster of properties found in the ‘‘best ”or ‘‘clearest ”case of the

category. Before proposing what a prototypical stress or tone system might look like, let us consider in (3) the

tone vs. stress properties that are often cited ( Martinet, 1960; Garde, 1967; Hyman, 1977a; van der Hulst and

Smith, 1988 , etc.).

(2) a. stress system : a language with word-level metrical structure, e.g. English

b. tone system : a language with word-level pitch features, e.g. Mandarin

c. pitch-accent system : a language with word-level

WHAT ?

L.M. Hyman / Language Sciences 31 (2009) 213–238 215 We have already mentioned the structural vs. featural distinction, which is deﬁnitional of stress vs. tone. The

remaining stress properties are also deﬁnitional. Adopting Prague School terminology, the function of stress is

constrastive in the sense that its presence on one stress-bearing unit implies its absence elsewhere. In other

words, stress contrasts are syntagmatic. The stress-bearing unit is necessarily the syllable, as it is not possible

for stress to contrast on the ﬁrst vs. second mora of a long vowel (vs. tone or ‘‘diacritic accent”). Since stress is

often predictable, there is no requirement that it be present in underlying representations. Rather, stress must

be present at the lexical level, i.e. the output of the lexical phonology. Finally, the domain within which stress

must be present is the output word: By deﬁnition, if a language has a stress system, every word has a primary

stress. Turning to the characterization of tone in (3), all but the ﬁrst property represent a prototype and are not

deﬁnitional. Like segmental features, tone typically exhibits a distinctive function in the sense that one asks

for each tone-bearing unit (TBU) what its tone is, e.g. H(igh), L(ow), or M(id). In other words, tonal opposi-

tions are prototypically paradigmatic. In (3) I have suggested that the prototypical TBU is the mora, where the

understanding is that each vowel (and possibly sonorant consonant) will carry a distinctive tone. Also in the

prototypical situation, tone will be present in underlying representations as a property of input morphemes. The comparisons in (3) reveal that all of the properties are deﬁnitional of stress, but only the ﬁrst is required

of tone. This partly accounts for the presence of intermediate systems. In some prosodic systems tone may

have a contrastive syntagmatic function, and the TBU may be the syllable. In addition, in cases where tone

has only a grammatical function, tones may be introduced by morphological rules rather than being features

on concatenated input morphemes. For example, it could be argued that the Somali noun roots in (4) are

underlyingly toneless:

As seen, in this declension class of nouns the masculine morpheme assigns a H tone (

0) to the penultimate vo-

wel, while the feminine morpheme assigns a H tone to the ﬁnal vowel ( Hyman, 1981; Saeed, 1999). Other vow-

els are toneless, produced on a relatively low or mid pitch. In the case of the long vowels in (4b), Somali

distinguishes rising [aa ´

] and falling [a ´

a] pitches. Since there is at most one H tone per word, which is limited

to the penultimate or ﬁnal vowel and is largely predictable from morphological features, Somali is far from

anyone’s ideal or prototypical tone system—in fact, it is one of the ‘‘intermediate ”languages which has some-

times been called ‘‘accentual ”.

While (3) establishes a baseline for prototypical tone, another approach to establishing the ‘‘best ”or

‘‘clearest ”tonal prototype might be to accumulate properties that make an accentual interpretation unappeal-

ing, if not impossible. As a concrete example consider a tone system which has the properties in (5).

(4) root masculine feminine

a. /inan/ ı ´

nan ‘boy’ ina´

n ‘girl’

/na

as/ na ´ as ‘stupid man’ na a´

s ‘stupid woman’

/goray/ go ´

ray ‘male ostrich’ gora ´

y ‘female ostrich’

b. /darmaan/ darma ´

an ‘colt’ darmaa´

n ‘ﬁlly’

eesaan/ eesa ´

an ‘young he-goat’ eesaa ´

n ‘young she-goat’

/dameer/ dame ´

er ‘he-donkey’ damee ´

r ‘she-donkey’

(5) a. three (or more) level tones: /H/, /M/, /L/ b. three (or more) ﬂoating tones/tonal morphemes:

H,M,L

c. all possible contours of these levels on one TBU: HM, HL, MH, ML, LH, LM

d. rules of H-, M- and L-tone spreading: /H-L/, /M-L/, /L-H/ ?H-HL, M-ML, L-LH

e. other arguments that every TBU must have a tonal speciﬁcation (100% tonal density)

216 L.M. Hyman / Language Sciences 31 (2009) 213–238 Since, as we shall see, the controversial or ambiguous cases typically involve languages which have only two

contrastive tone heights, the proposed prototype is one where three or more tone levels are distinguished.

Since one of the tones, e.g. M, might be analyzed as the absence of tone, asPulleyblank (1986) and Akinlabi

(1985) originally proposed for Yoruba, thereby aﬀecting the tonal density as discussed above, the prototypical

tone system is one where no tone can be ‘‘zeroed out ”. This would be quite hard to do if all three levels occur

as ﬂoating tones and combine to form all six tonal contours, as in M OnO [Ubangi] ( Kamanda-Kola, 2003; Ol-

son, 2005 ), or if all three tones trigger tone spreading, as in Gwari ( Hyman and Magaji, 1970). In short, the

clearest or least ambiguous tone system is one where there are multiple arguments that every tone has to be

featurally speciﬁed and that every TBU has to carry a tone. Deﬁned this way, languages such as Somali are

obviously quite far from the tonal prototype. As a result, scholars still feel compelled to address the question

originally raised by Klingenheben (1949) : Ist das Somali eine Tonsprache?

On the other hand, there is considerably less controversy concerning what constitutes a stress system. A

widely accepted deﬁnition of a language with stress is one in which there is an indication of word-level metrical

structure meeting the two core criteria in (6).

The two properties in (6) are often combined in a single statement to the eﬀect that every lexical word has

ONE

AND ONLY ONE

primary stress. In addition to meeting both of these criteria, another inviolable property of stress

systems is that the stress-bearing unit is the syllable: Syllables which are claimed to contrast two kinds of ‘‘ac-

cent ”(e.g. with rising vs. falling pitch) are not to be analyzed as having primary stress on their ﬁrst vs. second

mora. Instead, tone is always involved, as in the Somali examples darma´

an ‘colt’ vs. darmaa´

n ‘ﬁlly’. Treating

the Somali system as tonal is consistent also with the fact that verbs usually occur without a H tone, as do

most nouns in subject position: inan wa´

a dha

ay ‘a boy fell’. The /H/ of Somali thus satisﬁes culminativity

(6b), but not obligatoriness (6a). Before moving on to consider other such ‘‘intermediate ”systems, let us address the question of what the

prototype of a stress system might be. While earlier studies have suggested that word stress originates at

domain edges ( Hyman, 1977b; Bybee et al., 1998 ), the so-called demarcative function is not a good prototype

candidate in the sense of being unambiguous. A number of initial- or ﬁnal-stress languages might be amenable

to an analysis with boundary tones rather than metrical structure per se. Some of the systems which van Coet-

sem (1996, p. 39) calls ‘‘non-dominant prominence ”might fall into this category, i.e. languages where the

stress has little or no eﬀect on the segmental phonology, e.g. Finnish, Hungarian, and especially Turkish ( Levi,

2005 ). The prototype might therefore be a ‘‘dominant prominence ”stress system which has the properties in

(7). (6) a.

OBLIGATORINESS : every lexical word has AT LEAST one syllable marked for the highest degree of

metrical prominence (primary stress)

CULMINATIVITY : every lexical word has AT MOST one syllable marked for the highest degree of

metrical prominence

(7) a. stress location is not reducible to simple ﬁrst/last syllable b. stressed syllables show positional prominence eﬀectsi. consonant-, vowel-, and tone oppositions are greater on stressed syllables

ii. segments are strengthened in stressed syllables (e.g. Cs become aspirated or geminated, Vs become

lengthened, diphthongized)

c. unstressed syllables show positional non-prominence eﬀects i. consonant-, vowel-, and tone oppositions are fewer on unstressed syllables

ii. segments are weakened in unstressed syllables (e.g. Cs become lenited, Vs become reduced)

d. stress shows cyclic eﬀects (including non-echo secondary stresses)

e. stress shows rhythmic eﬀects lexically/postlexically (cf. the English ‘‘rhythm rule ”)

f. lexical stresses interact at the postlexical level, e.g. compounding/phrasal stress

g. lexical stress provides the designated terminal elements for the assignment of intonational tones (‘‘pitch-accents ”)

h. other arguments that every syllable is in a metrical constituent which can be globally referenced

L.M. Hyman / Language Sciences 31 (2009) 213–238 217 The above prototype sounds a lot like English, which has an accumulation of properties that make a non-

stress (e.g. tonal) interpretation unappealing, if not impossible. Thus, it would be hard to capture the facts

of English if the underlying system were analyzed with a prelinked H tone on the ﬁrst vs. second syllables

ofco´

nvert vs.conve ´

rt .

The above prototypes clearly identify unambiguous tone vs. stress systems. No reasonable linguist would

attempt to analyze. M OnO or Gwari with lexical accents in the place of tone or English with lexical tones in the

place of stresses. There are, however, systems which are so distant from both prototypes that they are grouped

together in a third category termed ‘‘pitch-accent ”. These systems are the subject of the next section.

3. ‘‘Pitch-accent’’ systems In this section I address the nature of word-prosodic systems which have been said to have ‘‘pitch-accent ”,

‘‘tonal accent ”or simply ‘‘accents ”. As stated by Salmons, 1992, p. 21 ), ‘‘the notion of ‘pitch accent’ languages

has long been vague ...”The most widely cited example has been Tokyo Japanese, which has presented a prob-

lem of interpretation. As seen in (8), Tokyo Japanese has been analyzed both accentually and tonally

( McCawley, 1968, 1978; Haraguchi, 1977; Poser, 1984; Pierrehumbert and Beckman, 1988 , etc.).

In (8a) the accent is represented by the down arrow (

;), which indicates the place where there is a pitch drop

from H to L. As seen, there can be at most one such drop per word, which in the case of atama ;ga‘head

(nom.)’ is realized on the nominative enclitic = ga. Since words such as sakana‘ﬁsh’ do not exhibit a pitch

drop, the so-called accent meets only the culminativity criterion in (6b), but not the obligatoriness criterion

in (6a). As seen in (8b), an analysis is also possible with a HL melody ( Haraguchi, 1977) or ‘‘pitch-accent ”

( Pierrehumbert and Beckman, 1988 ). The H is linked to the mora said to be accented in the (8a) analysis, while

the L aﬀects all subsequent moras. The approximate phonetic output in (8c), which incorporates the eﬀect of

an initial %L boundary tone, attempts also to show the diﬀerential realization of non-L moras: The highest

pitch indicated by (

0) occurs on the mora followed by the drop to L, which is likely related to the tendency

of a H (or M) tone to be raised before L (see Hyman, 2007and references cited therein). Intermediate non-

L pitches are indicated by (

–) and are realized as follows ( Pierrehumbert and Beckman, 1988; Warner,

1997 ): (i) A sequence of non-Ls will gradually rise up to the H pitch that precedes the drop to L, even possibly,

although with apparent speaker variation, if that H is ﬁnal and the L either ﬂoats or forms a slightly falling

contour with the H ( Gussenhoven, 2004, pp. 190–191 ). (ii) In the absence of a H to L drop, a sequence of non-

Ls will reach a peak on the second syllable which is lower than in (i), from which there is a gradually lowering

of pitch to the end of the sequence. Thus, the highest pitch of sa`

k a

¯ n

¯ ga

¯ is on [ka].

Let us assume for the moment that Tokyo Japanese warrants being placed into a third category. How, then,

might we deﬁne ‘‘pitch-accent ”. For Ding (2006, p. 1) , ‘‘a ‘pitch-accent system’ [is] one that generates tonal

patterns through diﬀerent placement of a marked tone on a syllable/mora in an underlying domain, which

is independent of both the syllable and the word. ”This covers Tokyo Japanese and Somali, but not Osaka

Japanese ( Haraguchi, 1977 ), where one also needs to know if a word begins L or H, or Fasu ( May and Loe-

weke, 1964 ) and Dadibi ( MacDonald and MacDonald, 1974 ), where one syllable per word contrasts /H/ vs.

/L/ tone, remaining syllables being /Ø / (cf. also Prinmi in Section3.3). Other deﬁnitions have assumed a

superclass including stress-accent and pitch-accent and focused on the realization of the latter in terms of a

ﬁxed pitch:

218 L.M. Hyman / Language Sciences 31 (2009) 213–238 ‘‘In a tonal or pitch-accent system (like Japanese, for instance), in addition to the abstract accent, there is

a constant physical property [i.e. a tone] associated with the accent...”(Hyman, 1977a, p. 4)

‘‘Hypothesis: Stress accent diﬀers phonetically from non-stress [pitch-] accent in that it uses to a greater

extent material other than pitch. ”(Beckman, 1986, p. 1 )

‘‘Pitch accent languages must satisfy the criterion of having

INVARIANT TONAL CONTOURS on accented syl-

lables, since tone is a lexical property. This is not so for pure stress languages, where the tonal contours

of stressed syllables can vary freely, being determined postlexically by the intonational system ...”

( Hayes, 1995, p. 50 )

While other systems analyzed as pitch-accent clearly combine stress and tone, Somali and Japanese are best

summed up by the following: ‘‘A pitch-accent system is one in which pitch is the primary correlate of prom-

inence and there are signiﬁcant constraints on the pitch patterns for words ...”(Bybee et al., 1998, p. 277 ). In

other words, ‘‘pitch-accent ”often means the same thing that Voorhoeve (1973)had in mind by the term

‘‘restricted tone system, ”a system in which tones are subject to ‘‘signiﬁcant constraints, ”e.g. culminativity,

as in Somali and Japanese.

In this section I would like to make the following claims concerning pitch-accent: (i) Some languages must

be analyzed with a stress system (e.g. English), some with a tone system (e.g. Yoruba), and others with both,

e.g. Texmelucan Zapotec ( Speck, 1978). However,

NO LANGUAGE MUST BE ANALYZED WITH PITCH-ACCENT . A tonal

analysis is always possible. (ii) Pitch-accent is not a coherent notion, rather a ‘‘pick and choose ”among the

properties that characterize prototypical tone vs. prototypical stress-accent ( Hyman, 2001a,b). (iii)

THERE IS NO

PITCH-ACCENT PROTOTYPE.

This is because the intermediate systems are not a coherent class. Consider for exam-

ple Donohue’s (2005, p. 40) attempt at identifying a pitch-accent prototype:

‘‘I take the following characteristics as being the deﬁning ones for a prototypical pitch-accent

system: 1. lack of metrical iteration;

2. relative pitch heights speciﬁed phonologically in the lexicon, not prosodically;

3. possibility of a contrast on monosyllabic (/monomoraic) words;

4. possibility of the entire system being described with automatic phonological processes and maximally

one single diacritic assigned lexically ...”

The ﬁrst three characteristics are typical properties of tone, while the fourth falls within Voorhoeve’s notion

of a ‘‘restricted tone system ”. In any case, one cannot deﬁne a prototype negatively (‘‘lack of metrical itera-

tion ”) or in terms of optional ‘‘possibilities ”. For this it must be shown that they have something positivein

common. My impression from the literature is that there are in fact two kinds of systems that are identiﬁed as

‘‘pitch-accent ”: (i) systems which combine stress and tone, particularly those where a tonal contrast is found

only on stressed syllables, which is presumably the only reason why Hollenbach’s (1988)description of the

ﬁve-level tone system of Copala Trique [Mixtecan; Mexico] is included in a collection on pitch-accent; (ii) sys-

tems which simply fail to have enough of the prototype properties for certain researchers to feel comfortable

identifying them as tone, e.g. W. Basque, Tokyo Japanese, and Somali ( Hualde et al., 2002, pp. 578–579). The

tones of these latter languages, which do not have stress, fall at the low end of the tonal density scale in (1) and

may have syntagmatic properties reminiscent of stress (e.g. the culminative pitch drop in Tokyo Japanese).

There is, however, no reason to think that even these reﬂect a single class of accentual systems as opposed

to restricted tone systems varying along certain parameters. Given the diﬃculty of deﬁning pitch-accent in terms of something that they and only they have, and the

impossibility of establishing a single prototype which relates both Tokyo Japanese and Copala Trique, there

is every reason to doubt the coherence of such a category. Both in his writings and in his review of this paper,

Jose ´

Hualde argues for distinguishing tone, stress, and two intermediate ‘‘types ”: the Tokyo Japanese or T-

type ‘‘pitch-accent” system which has at most one lexical tone (‘‘accent” ) per word and the Swedish or S-type,

which restricts tone to occurring on stressed syllables. If these languages appear to represent convenient ref-

erence points for comparison it is because of their simplicity: the tone of both has been analyzed in terms of

presence vs. absence—vs. the Osaka/Kyoto-type with an additional lexically contrastive initial L or Copala

Trique which contrasts multiple level and contour tones only on the stressed syllable of the word. All of

L.M. Hyman / Language Sciences 31 (2009) 213–238 219 the above languages, of course, share the property of placing severe restrictions on the distribution of tones.

Reviewing the literature suggests that the terms ‘‘accent”and ‘‘pitch-accent ”have been used to refer to

defective or restricted tone systems whose mark (say, H tone) is characterizable in one or more of the ways

in (9).

The problem, however, is that all of the properties in (9) can be found in unambiguous tone systems. To see

this, I will consider each in turn, starting with obligatoriness.

3.1. Obligatoriness (

OBLIG )

As seen in the following tonal distributions in (10), Chuave [Trans-New Guinea] requires that all words

have a H feature in them ( Swick, 1966; Donohue, 1997, p. 355 ):

Whether words have one, two, three or four syllables, all combinations of H and L are observed except an all-

L pattern. Since there are so many tone patterns (e.g. 15 out of 16 possibilities on four-syllable words), it is

highly unlikely that this is a ‘‘pitch-accent ”system. Rather, it is a tone system that happens to have a restric-

tion that no word can be all L. In fact, as seen in (11), there are many tone systems which require at least one

[H] feature per word (other interpretations have been oﬀered in some cases):

One restriction that can therefore be placed on a tone system is that every word must have at least one [H]

feature. It is however not clear that this automatically means that the [H] is accentual. Consider in (12) the

word-tone distributions in Tanimuca [Tukanoan; Colombia] ( Keller, 1999):

(9) a. obligatory (‘‘at least one per word

”)

b. culminative (‘‘at most one per word ”)

c. privative (e.g. /H/ vs. / Ø/ rather than /H/ vs. /L/)

d. metrical (e.g. positionally restricted, subject to reduction/subordination in compounding or when out of focus)

(11) a. /H, LH/ Manding (general) (Creissels and Gre´

goire, 1993, p. 109 ), Foe (Rule, 1993 ),

Hup ( Epps, 2005 )

/H, HL/ Crow (Kaschube, 1954 ), Acoma (Miller, 1965 ), Kaure? (Dommel and

Dommel, 1991 ), Kham? (Watters, 2002 )

/H, HLH/ Choctaw ( Ulrich, 1989)

b. /H, HL, LH/ Dom ( Chida, 2001), Yuhup (Ospina Bozzi, 2002 ), Prinmi (Ding, 2001 )

c. /H, HL, LH, LHL/ Kairi ( Newman and Petterson, 1990), Dogon-Jamsay (Heath, 2008),

Barasana ( Gomez and Kenstowicz, 2000 ), Wanano (Stenzel, 2007 )

220 L.M. Hyman / Language Sciences 31 (2009) 213–238 As seen, there are no all-L words. However, there also are no words of the shape H-L-L. The correct gen-

eralization may therefore be that no word can end in two low tones ( L-L#), which would automatically rule

out all-L words. (There are no monosyllabic words in the language.) If correct, we might refer to Tanimuca as

exhibiting ‘‘accidental obligatoriness ”.

Another problem is distinguishing what would be an obligatory H ‘‘accent ”from an obligatory H bound-

ary tone. In Choctaw [Muskogean; Oklahoma], every word ends in a /H/ tone ( Ulrich, 1989, p. 163). In addi-

tion, some words also have an additional /HL/ assigned to either their antepenultimate or penultimate

syllable. As a result, words may have the shapes L

n-H, L n-H-L-H or L n-H-LH (the last having undergone

a rule whereby ﬁnal HL-H ?H-LH). This produces a system where the ﬁnal /H/ is obligatory, and the addi-

tional /HL/ culminative. If the latter is an ‘‘accent ”, it clearly violates

OBLIG (H).

Finally, it should be noted that tones appear to constitute the only contrastive feature that can be obliga-

tory at the word level. For example, no language requires all words to have a contrastively long vowel or nasal

feature. As seen in the next subsection, tone is not the only feature than can be culminative.

3.2. Culminativity (

CULM )

Turning to culminativity, (13) displays the word-tone patterns of Tinputz [Austronesian; Papua New Gui-

nea] ( Hostetler and Hostetler, 1975 ).

As seen, all-L words are possible in Tinputz, but

CULM (H) restricts words to a single /H/. In addition, words

may not begin with a H, unless monosyllabic, which I have analyzed as /LH/. Other languages with culmina-

tive, but non-obligatory /H/ include Somali (and several related Cushitic languages), Agiribi Kiwai ( Wurm,

1973; Donohue, 1997 ), Yucatec Mayan (Gussenhoven, 2007 ), and Ocaina [Witotoan; Peru] ( Agnew and Pike,

1957 ). While /HL/ but not /H/ is culminative in Obokuitai [Lakes Plain; Indonesia (Papua)] ( Jenison and Jen-

ison, 1991 ), both /H/ and /HL/ (and /LH/ if the word is monosyllabic) are culminative in Puinave [isolate;

Colombia] ( Guiro´

n Higuita and Wetzels, 2007 ), as seen in (14).

The fact that there is a contrast between H-L and HL-L suggests the need for underlying /H/ and /HL/ in

Puinave, where the L feature is contrastive, even though ‘‘plain ”[L] is analyzed as the default tone of /Ø /

TBUs. However, other than their distributional limitations, there is no reason to believe that /H/, /HL/

and /LH/ are somehow accented, even if the culminative H feature is ‘‘marked ”with respect to L. As Creissels

and Gre ´

goire (1993, p. 117) put it, ‘‘...le fait d’interpre ´

ter l’un des deux termes d’une opposition tonale bi-

naire comme une marque et l’autre comme l’absence de marque n’implique nullement d’abandonner la notion

de ton pour celle d’ ‘accent’. ”

Still, it is interesting to note how H tones come to be more restricted than Ls, ultimately being limited to one

H per domain. First, H tones are generally more subject to phonological constraints than L tones. Consider, for

example, the obligatory contour principle (OCP), which prohibits two identical tones from occurring in

L.M. Hyman / Language Sciences 31 (2009) 213–238 221 sequence. OCP(L) eﬀects are observed in several /L,Ø/ systems, e.g. Malinke (Kita) [Mande; Mali] ( Creissels

and Gre ´

goire, 1993 ), Galo [Tibeto-Burman; NE India] ( Post, 2007), Munduruku [Tupi; Brazil] ( Picanc o, 2005 ),

and the closely related Witotoan languages Bora ( Weber and Thiesen, 2000) and Miran

˜a(

Seifart, 2005 ), spoken

in Peru, Colombia, and Brazil. (To these I would add Tonga [Bantu; Zambia] which, like Malinke (Kita),

inserts Hs to keep underlying /L/s apart.) In addition, Pen ˜oles Mixtec [Mixtecan; Mexico], whose underlying

system is /H, L, Ø/, shows long-distance OCP(L) eﬀects ( Daly and Hyman, 2007). There are, however, many

fewer /L, Ø/ than /H, Ø/ privative systems, many of which show OCP(H) eﬀects. To see how OCP(H) almost

produces a culminative H system, consider the following word-tone patterns in Amahuaca [Panoan; Peru]

( Russell and Russell, 1959, pp. 136–140 ), which has an underlying /H,Ø/ contrast with default L:

As seen, /H/ is obligatory in Amahuaca. However, while we observe the expected two tone patterns on bisyl-

labic words, there are four (not three) patterns on three-syllable words. Since H-L-H is also possible, it is dif-

ﬁcult to consider /H/ to be accentual in the culminative sense. Rather, the system in (15) is due to two

restrictions: (i)

OBLIG (H); (ii) OCP(H). In other words, while all-L words are ruled out by obligatoriness,

H-H, H-H-L, L-H-H and H-H-H would all be possible in principle if it were not for the word-level OCP con-

straint against two H tones in sequence. (At the phrase level, when a word ending H-L is followed by another

H, H tone plateauing can result in the L (i.e. /Ø /) being raised to H (Russell and Russell, 1959, p. 152 ).

Another constraint which seems more often to target H is what Zoll (2003)terms

CLASH : the prohibition

against mapping a single tone to multiple successive TBUs. Again,

CLASH (H) appears to be more common than

CLASH (L). Thus, consider the possible word-tone patterns in Fore [Trans-New-Guinea] ( Scott, 1990) in (16),

where (H,L) = ﬂoating tones:

As seen, there are no sequences of H tones if a L is present in the input. Scott (1990)summarizes the system as

follows:

The simplest system which may be hypothesized for Fore is one in which only changes between high and

low tone are recognised as being contrastive. (p. 141)

There are no contrastive contours. (p. 147)

...tones appear to spread by increasing the domain of the L tones in preference to the spreading of H

tones ...From this it appears that H tones are to be considered as peaks of prominence or pitch tar-

gets ...(p. 147)

It is not clear that we can attribute the dispreference of sequences of H TBUs to the OCP, since multiple linking

would also be possible. Rather, as pointed out by Zoll (2003), some languages are like Fore and prefer sequences

of L TBUs, while others prefer sequences of H TBUs. Although Fore prefers Ls in this way, it will take further

erosion of Hs for /H/ to be culminative, since it presently allows all H words as well as single Hs interrupted by Ls. Besides such mapping preferences, a tendency toward culminativity can also be observed in /H, Ø/ systems

which have H deletion rules. Consider for example the possible tone patterns on nouns in Haya [Bantu;

Tanzania] ( Hyman and Byarushengo, 1984 ) as they are realized in utterance-initial subject position (vowels

lacking a tone mark are pronounced with default L):

222 L.M. Hyman / Language Sciences 31 (2009) 213–238 Each noun in (17) consists of the article-like ‘‘augment”morpheme o-, the noun class preﬁx mu- and a noun

stem, which, as seen, can have only one underlying /H/. The citation forms of these nouns are similar except

that underlying ﬁnal /H/ is realized instead on the penult ( o-mu´

-ti ,o-mu-hu ´

me ,o-mw-i:sı ´

ki) , and penultimate

/H/ is realized as a HL falling tone ( o-mu-kaˆ

zi, o-mu-tamı ˆ

:le ). When ﬁrst eliciting nouns in isolation, one

might, therefore, initially conclude that H is culminative in Haya. However, this is incorrect, both of under-

lying and surface forms. As seen in (18), the noun ‘tree’ is realized with all eight combinations of H and L tone

in context ( Hyman and Byarushengo, 1984, p. 56 ):

The underlying form of ‘tree’ is /o ´

-mu-tı ´

/, whose citation tones [o `

-mu ´

-tı `

] appear to be inverted. Since Haya

deletes a H tone on a post-pause vowel, the augment /o ´

-/ is realized L in (18a–d). The H of the stem /-tı ´

is deleted in (18a,e) when the noun occurs as the head of a genitive construction. In (18c,d,g,h) the phrase-ﬁnal

H of /-tı ´

/ spreads onto the penult, which is accompanied by the lowering of H before pause in (18d,h), perhaps

the eﬀect of an utterance-ﬁnal L% boundary tone. Despite the fact that nouns can have only one surface H in

isolation, Haya does not respect culminativity, which can be seen also on many verb forms, e.g. ba´

-ka-mu-bo ˆ

a ‘they saw him’.

The point of these examples is to expose an asymmetry between H and L tones: While there is little that a H

can do that a L cannot ( Hyman, 2001b), prohibitions on initial or ﬁnal lexical L tones are less common than

on lexical H tones. (Recall the absence of initial H also in Tinputz in (13).) The result is that isolation forms

will have fewer H tones than forms in context. By analogizing the citation forms to new contexts, languages

may come to have fewer Hs, as has happened in Bantu. The conclusion that can be drawn from the preceding is that Hs tend to be subject to more distributional

constraints than Ls, either by position ( initial H, ﬁnal H) or by juxtaposition (e.g. OCP(H)). The most

extreme case is positional culminativity, as found in Somali, where a /H/ is permitted only on the penultimate

or last vowel ( darma´

an ‘colt’, darmaa ´

n ‘ﬁlly’, not da ´

rmaan ). The view expressed here is that culminativity is

not a reliable criterion for accent. First, note that a headed metrical structure is not required to account for the

Somali facts. Let us assume that Somali nouns all have a ﬁnal H suﬃx or boundary tone, as Choctaw appar-

ently has for all words ( Ulrich, 1989). If the ﬁnal mora is extrametrical in the masculine nouns in the declen-

sion class represented in (4) above, creating a contrast between darma(an)‘colt’ anddarmaan‘ﬁlly’, the

boundary tone will automatically link to the penultimate vowel of masculines vs. the last vowel of feminines.

Similarly, in another declension class masculine - eand feminine - oare extrametrical: waraa´

b-e ‘hyena’, abee´

s-o

‘poisonous snak

e (sp.)’. In other words, rather than having an accentual H, the culminative and non-obliga-

tory tone of Somali can be analyzed as grammatically conditioned H boundary tones. This, of course, would

make Somali quite diﬀerent from Tokyo Japanese, where the lexical Hs are generally not predictable. In both

cases, however, there is little reason to think we are dealing with anything but restricted tone. A ﬁnal problem with culminativity as a criterion for accent is that stress and tone are not the only prop-

erties that can be culminative. Examples are listed in (19). (17) 1 rstem 2 rstem 3r stem

o-mu-zi ‘root’ o-mu-limi ‘farmer’ o-mu-zigaijo ‘ﬁrst born’

o-mu-tı ´

‘tree’ o-mu-ka ´

zi ‘woman’ o-mu-lu ´

muna ‘younger brother’

o-mu-hume ´

‘blind o-mu-tamı ´

:le ‘drunkard’

person’ o-mw-i:sikı ´

‘girl’

(18) a. L-L-L o-mu-ti gwa: ka ´

to ‘Kato’s tree’

b. L-L-H o-mu-tı ´

gwange ‘my tree’

c. L-H-H o-mu ´

-tı ´

ka ´

to ‘a tree, Kato!’ (Kato = vocative)

d. L-H-L o-mu ´

-ti ‘tree’

e. H-L-L a-ka-gul’ o ´

-mu-ti gwa: ka ´

to ‘he bought Kato’s tree’

f. H-L-H a-ka-gul’ o ´

-mu-tı ´

gwange ‘he bought my tree’

g. H-H-H a-ka-gul’ o ´

-mu ´

-tı ´

ka´

to ‘he bought a tree, Kato!’

h. H-H-L a-ka-gul’ o ´

-mu ´

-ti ‘he bought a tree’

L.M. Hyman / Language Sciences 31 (2009) 213–238 223 Should we thus speak of aspirated or glottalized consonant accents, length accent, mid-vowel-height accent,

and nasalized-vowel accent? While vowel length appears the most ‘‘accent-like”, it should be noted that length

and stress do not always coincide in Mam. A similar point is made by Kisseberth (2005)concerning the phra-

sal domain in Chimwiini [Bantu; Somalia], where culminative vowel length can be penultimate or antepenul-

timate, while culminative H tone can be penultimate or ﬁnal.

3.3. Privativity

Many of the systems discussed in preceding sections lend themselves to a privative /H/ vs. / Ø/ rather than

binary /H/ vs. /L/ interpretation. In the 1980s, a number of phonologists, myself included, ﬂirted with the idea

that privativity was suﬃcient to call a system accentual. (This seems to be the only reason why Clark’s (1988)

treatment of Zulu [Bantu; South Africa] appears in a collection on pitch-accent: ‘‘ ...Zulu is a tonal pitch-

accent language with the tonal melody ‘H’. ”(p. 56)). The intuition is that an accent is present vs. absent,

whereas a tone should be fully speciﬁed with pluses and minuses. The proposal that such accents might be

represented by a diacritic asterisk placed on appropriate TBUs, as Goldsmith (1984)proposed for Tonga,

and which I unfortunately applied to Somali ( Hyman, 1981), was shown by Pulleyblank (1986) to have unde-

sirable consequences. However, one might still view the H of a /H, Ø/ system as accentual by virtue of its priv-

ativity—vs. the H of a /H, L/ system. A major problem with such a position is that it is not always clear whether the non-H tone of a two-level system

should be analyzed as /L/ or / Ø/. In addition, diﬀerent phonologists have taken the position that there should be

full speciﬁcation in underlying representations, in which case some other mechanism would be needed to distin-

guish accent-like /H/ from purely tonal /H/. Another problem is that we have counterexamples of two types: First, there are languages which show accentual properties but which require /L/ to be present. In (14) we

saw that there are two diﬀerent culminative tones in Puinave, /H/ and /HL/. Guiro´

n Higuita and Wetzels

(2007) analyze Puinave with underlying /H/, /HL/, /Ø / and marginal /LH/, which is found only on monosyl-

labic words. Assuming that this analysis is correct, if we wanted to call Puinave accentual, it would not be the

absence of a L feature that is criterial, but rather the presence of a sizeable number of toneless TBUs. The

same is seen in Prinmi [Tibeto-Burman; China] ( Ding, 2001), which I have reanalyzed in (20) as having a sys-

tem of /H/, /HL/ and /LH/:

As seen, one syllable per word is marked for one of three underlying tones: /H/, /HL/ or/LH/. (Other than

carrying a tonal speciﬁcation, there is no reason to think that the toned syllable is stressed.) As indicated,

the underlying /H/ spreads one syllable to the right, while the H of /HL/ instead assigns its L to the next syl-

lable. (There are no falling tones in the language. The ﬂoating L indicated in parenthesis will be assigned to a (19) a. aspiration and glottalization in Cuzco Quechua ( Parker, 1997, p. 2)

b. length in Mam [Mayan; Guatemala, Mexico] ( Willard (2004, p. 7))

c. mid vowels in Punu [Bantu] ( Kwenzi Mikala, 1980, p. 8; Fontaney, 1980, p. 55 )

d. nasalized vowels in Karo [Tupi; Brazil] ( Gabas, 1999, p. 42n)

224 L.M. Hyman / Language Sciences 31 (2009) 213–238 following TBU.) This /H/ vs. /HL/ distinction replacesDing’s (2001)use of a diacritic feature [±spread] to

characterize the diﬀerence. The third tone, /LH/, assigns its H to the following TBU, a process perhaps con-

ﬂatable with the H tone spreading rule. When occurring ﬁnally, /LH/ is realized as a rising tone. As indicated

in the outputs in (20), default L tone is assigned to remaining toneless TBUs. What Prinmi reveals is that privativity has less to do with whether [H] and [L] features contrast than with the

existence of /Ø / TBUs. In other words, it again relates to the issue of tonal density. In the case of Haya /H, Ø/, we

saw in (17) and (18) that output conditions result in the non-realization of input Hs, e.g. on pre- and post-pause

vowels. As analyzed by Schadeberg (2000), Ekoti [Bantu; Mozambique] shows that tonal density is controlled by

INPUT conditions. Consider in (21) the observed patterns on nouns as they are realized in non-prepause position:

As seen, the underlying opposition in Ekoti is privative /H, Ø/. There is a H tone spreading rule that doubles

an input H onto the next mora. Remaining /Ø / moras are realized with default L. If one were look only at the

tonal distributions on nouns of 1–4 moras in (21a), one might conclude that /H/ is both obligatory and cul-

minative: in each case there is only one input pattern with a single /H/. However, there are three diﬀerent pat-

terns on ﬁve-mora nouns and six patterns on six-mora nouns. Schadeberg (2000, pp. 601–602)accounts for

these distributions as in (22).

Whereas optimality theory has focused on the relative ranking of output conditions, Schadeberg’s conception

clearly is that (22i–v) are ranked constraints on

INPUTS . To show this, I have relabeled the ﬁve constraints as

OCP(H),

OBLIG (H), INIT (H), FIN (H), and [Ø n-H-L] in the tableau in (23), which reveals why Ekoti allows

only one out of the eight logical tonal inputs on nouns of three moras:

(22) ‘‘Each mora may carry a distinctive Hi tone, subject to the following restrictions, in decreasing order of priority:

(i) Two Hi tones shall not be on adjacent moras. [‘‘holds throughout the language”]

(ii) Each noun shall have at least one Hi tone. [‘‘obeyed by all nouns”]

(iii) No Hi tone shall be on the ﬁrst mora. [violated by monomoraic nouns]

(iv) No Hi tone shall be on the last mora. [violated by mono- & bimoraic nouns]

(v) The only Hi tone shall not be on the penultimate mora. ”[violated by /Ø -H-Ø/]

L.M. Hyman / Language Sciences 31 (2009) 213–238 225 As seen, the most harmonic input candidate is /Ø-H-Ø/, which violates only the lowest-ranked constraint,

[Ø n-H-L], which also is the only constraint whose motivation is unclear. Considering the other inputs in

(22a), monomoraic /H/ necessarily violates

INIT (H) and FIN (H), and /Ø -H/ wins out over /H- Ø/, since

INIT (H) outranks FIN (H). The quadrimoraic input /Ø -H-Ø-Ø / satisﬁes all of the constraints, as do the three

ﬁve-mora inputs in (22b) and the six six-mora inputs in (22c).

When the full range of data are taken into consideration, we see that although Ekoti has the privative con-

trast /H, Ø/, H is clearly not culminative on nouns in inputs or outputs. Nor is /H/ obligatory, since verbs can

be toneless (as in Somali). Of course, it may be that H tones are more restricted by position in a privative /H,

Ø / system as opposed to a binary /H, L/ one, but this hardly means that we are dealing with a third ‘‘pitch-

accent ”category. Rather, syntagmatic constraints and processes are simply more frequent and potentially

operate within larger domains in privative vs. binary systems. In a number of works, Charles Kisseberth

has shown that an underlying H tone of a verb (underlined) can be displaced onto the following word. While

the (i) examples in (24) consist of a toneless verb + toneless noun object, the /H/ of the verb root shifts onto

the same toneless nouns in the (ii) examples—to the ﬁnal in Digo, and to the penult in Giryama and Zigula:

Such phrasal displacement is apparently possible only in languages which have an underlying /H, Ø/ contrast.

Thus, except for the docking of ﬂoating tones, tonal displacement has not been documented in any /H, L/ or

/L, Ø/ system. Since words can have more than one H tone in such languages, we are not dealing with

‘‘accent ”in the culminative sense. Rather, such long-distance mobility is a logical endpoint of the syntagmatic

potentiality of tones, particularly as they operate at the phrase level. While the draw of the penultimate or ﬁnal

syllable may be viewed as ‘‘attraction to accent ”(Goldsmith, 1987, p. 99; Casimjee and Kisseberth, 1992,

p. 30 ), this at best shows that tone can peacefully co-exist with a metrical system, i.e. stress. That tone can

have metrical properties has also suggested that it might form a natural ‘‘accentual ”class with stress, as

addressed in the following subjection.

3.4. Metricality

The ﬁnal kind of system to be considered is one where tones are placed or are otherwise aﬀected by metrical

processes. We have already cited systems where H tones are located at or near the right edge, e.g. the penul-

timate or ﬁnal H of Somali, the obligatory word-ﬁnal H of Choctaw, the right-displaced Hs of Giryama, Zigu-

la and Digo. We hypothesized that the right-oriented H is a suﬃx in Somali and a boundary tone in Choctaw.

This leaves the Bantu cases in (24), which are subject to two interpretations: The ﬁrst is that an input H is

attracted to a metrically strong position to the right of its input. The other is that that the H is aligned with

the right edge of a domain (modulo non-ﬁnality). In either case, a common assumption among Bantuists is

that word- or phrase-level tone displacement derives from a combination of spreading + delinking ( Casimjee

and Kisseberth, 1992, p. 31 ;Philippson, 1998, p. 321 ).

Tones may also be restricted to the left edge of a domain. In Mayo [Uto-Aztecan; Mexico], a /H/ is

restricted to occurring on either the ﬁrst or second syllable of a word ( Hagberg, 2006). Toneless TBUs are

pronounced L when they occur between the last H and pause, otherwise they are pronounced M. As seen

in (25), the placement of the /H/ is usually a lexical property of roots: (24) a. Digo ( Kisseberth, 1984, pp. 162, 168 )

(i) ku-henz-a mu-ganga ‘to look for a doctor’

(ii) ku-i

h-a mu-ganga ´

‘to call a doctor’ ( ?[ku-ih-a mu-ga nga ˆ

])

b. Giryama ( Philippson, 1998, p. 321 ; cf.Kisseberth and Volk, 2007 )

(i) ku-tsol-a ki-revu ‘to choose a beard’ /-tsol-/ ‘choose’

(ii) ku-o

n-a ki-re ´

vu ‘to see a beard’ /-o ´

n-/ ‘see’

c. Zigula ( Kenstowicz and Kisseberth, 1990, p. 175 ; cf.Kisseberth, 1992

(i) ku-guh-a Ma-tunguja ‘to take tomatoes’ /-guh-/ ‘take’

(i) ku-fi

s-a ma-tungu ´

ja ‘to hide tomatoes’ /-fı ´

s-/ ‘hide’

226 L.M. Hyman / Language Sciences 31 (2009) 213–238 In (25a), the root -chup- ‘harvest’ assigns ﬁrst-syllable H, while in (25b) the root - pon- ‘play’ assigns second

syllable H. In each case the H tone migrates to the left as preﬁxes are added. While other Uto-Aztecan lan-

guages assign stress from the left edge of the word, e.g. Rara ´

muri ( Caballero, 2005 ), Hagberg analyzes Mayo

with a ﬂoating H tone whose placement is determined by roots: ‘‘While most theories allow for lexically con-

trastive stress, they do not allow lexical accent to ﬂoat, and yet this is clearly what happens. Since ﬂoating tone

is incontrovertibly attested in a number of other languages, I conclude that Mayo has underlying autosegmen-

tal tone, not accent. ”(p. 7) (Hagberg discusses the possibility of initial extrametricality; another possibility

would be to distinguish /H/ vs. /LH/ roots whose tone(s) are mapped left-to-right at the word level.) Demers et al. (1999) describe a similar system in closely related Yaqui, which has two main diﬀerences: (i)

the initial or peninitial /H/ spreads to the end of the word. (ii) a long vowel is shortened in a syllable that either

precedes or follows the H tone syllable. They assume initial extrametricality and interpret vowel shortening as

follows: ‘‘We propose that the mora loss rule is part of prominence marking in Yaqui. That is, vowel duration

is a competing marker of prominence (perhaps analogous to stress clash) and in order to maximize the prom-

inence of the ﬁrst syllable bearing a high tone, a mora is dropped in adjacent syllables containing two moras. ”

(p. 46) In other words, although the mark of prominence is H tone, it may be accentual because it has an eﬀect

on neighboring syllables which we do not expect of ‘‘ordinary” H tones. Taking the two languages together,

we seem to have behavior that is unbecoming both to stress (Mayo) and to tone (Yacqui). It would not seem unreasonable to invoke metrical structure in Yaqui, if not in both languages, which

determines the placement of the one underlying /H/ per word and conditions the vowel shortening in adjacent

syllables in Yacqui. There is nothing incompatible about a language with tone and metrical structure. Pearce

(2006, 2007) shows that Kera [Chadic; Chad, Cameroon] requires iambic feet for the purpose of mapping /H/

and /L/ tones as well as determining vowel height harmony: ‘‘Kera illustrates that, even without stress at the

word level, there are languages which have an interaction between the foot and a number of other phenomena,

including tone. ”(Pearce, 2006, p. 283 ) The question is whether metrical structure = stress (as I have assumed),

or whether stress = metrical structure + something additional. In Hyman (1977a, p. 5) I distinguished between ‘‘tone which is dependent on stress ”vs. ‘‘stress which is

dependent on tone ”. The ﬁrst situation has already been alluded to and concerns languages where a tonal con-

trast is possible only on stressed syllable. (The second concerns cases where stress is allegedly assigned based

on the tonal composition of a word, something to which de Lacy (2002)has brought recent attention.) A

rather intricate example of tone depending on stress, i.e. metrical structure, is found in Seneca [Iroquoian;

US, Canada] ( Chafe, 1977, 1996; Melinger, 2002 ). Following Melinger’s recasting of Chafe’s analysis, Seneca

metrical structure and H tones are assigned as in (26).

While (26a,b) are not unusual, the H tone assignment process in (26c) is. In order to see this, consider the

schematic tone assignments in (27), where each hypothetical word has an initial extrametrical syllable followed

by two (non-ﬁnal) trochees: (25) a. Initial H b. 2nd Syll H

chu ´

pnake ‘will harvest’ (tr.) ponna´

ke ‘will play’ (tr.)

hı ´

-chupnake ‘will harvest’ (intr.) hi-po´

nnake ‘will play’ (intr.)

hı ´

-hi-chupnake ‘will always harvest (intr.) hi-hı ´

-ponnake ‘will always play’ (intr.)

(26) a. mark the ﬁrst syllable extrametrical b. build bisyllabic trochees left-to-right

c. assign a H tone to the ﬁrst syllable of a trochee iﬀ either syllable is closed

(27) a. hri (Ca´

C.Ca) (CaCa) ...

b. hri (Ca´

.CaC) (CaCa) ...

c. hri (Ca´

.CaC) (Ca ´

C.Ca) ...(

CULM(H) )

d. hri (Ca.Ca) (Ca.Ca) ...(

OBLIG(H) )

L.M. Hyman / Language Sciences 31 (2009) 213–238 227 In (27a), a H tone is assigned to the head syllable of the ﬁrst trochee, since it is closed. No H tone is assigned to

the second trochee which consists of two open syllables. In (27b), again, the ﬁrst trochee receives a H on its

initial syllable, since its second syllable is closed. In (27c), each trochee has one closed syllable, and hence two

H tones are assigned to this word in violation of

CULM (H). Finally, there are no closed syllables in (27d), so no

H is assigned, in violation of

OBLIG (H).

If we turn to the question of how Seneca should be typologized, it is clear that although its H tone is priv-

ative, it fails to meet the more accent-like properties of obligatoriness and culminativity. The trochaic metrical

structure, on the other hand, does appear to be obligatory, judging from the diﬀerent behavior of ﬁnal tro-

chees. As Wallace Chafe (personal communication) has carefully explained to me, whether a penultimate

vowel is lengthened or not can be aﬀected by whether it is in an even- vs. odd-numbered syllable, or, in met-

rical terms, depending on whether it is in the initial vs. second syllable of the trochee. In (28), I have marked

the metrical and syllable constituents on examples provided by Chafe:

(28a,b) show that the vowel of a penultimate open syllable will lengthen in either position of a trochee. A pen-

ultimate vowel will also lengthen (and block H tone assignment) in a closed syllable if the latter is trochee-ini-

tial, as in (28c), but not if the vowel is in the second syllable of the trochee, as in (28d). While lengthening is

blocked if a vowel is followed by a laryngeal consonant, and there are subsequent sound changes which par-

tially obscure the original structure, the trochaic structure is clearly implicated in both H tone assignment and

vowel lengthening. (Cf. Hayes’ 1995, pp. 222–225 treatment of closely related Cayuga, whose metrical struc-

ture aﬀects the segmental phonology.) To summarize thus far, we have seen systems which have obligatory or culminative H but little or no evi-

dence for metrical structure at the word level vs. Seneca, whose H is neither obligatory nor culminative H, but

requires word-level trochaic footing. Another metrical property is one whereby the prominent features are

‘‘subordinated ”(e.g. reduced or deleted) in phrasal contexts, typically on the head word of a compound or

head-complement construction. In the Haya forms in (29a), for example, we observe that a verb may have

one or more H tones in phrase-ﬁnal position:

These H tones are deleted, however, when the same verb forms are followed by a complement, here the proper

noun Ka´

to (Hyman and Byarushengo, 1984, p. 69 ). While it is tempting to attribute H reduction to being out

of focus, it should be noted that the head noun of a genitive construction also undergoes H deletion: o-mu-

ko ˆ

no ‘arm’ ?o-mu-kono gw’ o´

-muka ˆ

zi ‘the woman’s arm’. A more accurate characterization is that the head

word undergoes tonal reduction in the appropriate

HEAD+COMPLEMENT constructions.

It is tempting to identify Haya’s deletion of H tones as accentual, perhaps similar to cases of stress sub-

ordination which occurs in the absence of phrasal accent. However, recall from (17) and (18) that Haya

has a full tone system whose Hs are not obligatory, culminative, or metrical. The same is true of Nara

[E. Sudanic; Eritrea], where the mora ( l) is a vowel or sonorant consonant. The tonal distributions in (30)

are extracted from Hayward (2000, p. 255) , where patterns in parentheses are expected, but not exempliﬁed

in the data: (28) a. ho

i (ge:.ge

¨ ) ‘I saw it’

b. ho

i (sha.go:) ge

¨ ‘I saw her’

c. hhoi (ye

¨:s.do

¨h) ‘he’s attractive’

d. hhoi (dı´

ye ¨s) do

¨h ‘they’re attractive’

(29) a. ‘they tie up’ (etc.) b. ‘they tie up Kato’ etc.

Present habitual ba-ko ´

m-a ba-kom-a ka´

Past

1 ba ´

-a ´

-ko ˆ

m-a ba-a-kom-a ka´

Past

2 ba-kom-ı ´

l-e ba-kom-il-e ka´

Past habitual ba-a-ko ´

m-ag-a ba-a-kom-ag-a ka ´

Future

1 ba-laa-ko ˆ

m-a ba-laa-kom-a ka ´

Future

2 ba-li-ko ´

m-a ba-li-kom-a ka´

228 L.M. Hyman / Language Sciences 31 (2009) 213–238 As indicated in the left column,Hayward (2000, p. 256)isolates ﬁve tonal schemas, adding: ‘‘ ...attention is

directed towards an obvious assymmetry with regard to the distribution of the two tones, for we do not ﬁnd

a HLH contour as a counterpart to the LHL contour ...”As in Haya, the head of a construction may lose its

underlying H tone(s). However, since Nara is a head-ﬁnal language, it is the second word (W

2)in COMPLE-

MENT+HEAD

constructions which undergoes a change to all L in the cells indicated in (31).

What these cells have in common is that they convert input [ ...H-L

n-H ...] sequences to [ ...H-L n-L... ], e.g. H-

L # H-H ?H-L # L-L, H-H # L-H ?H-H # L-L. Hayward (2006) characterizes the generalization as ‘‘one

high only ”(OHO): Assuming multiply linked tones, there can only be one /H/ per domain. As a result, once

there is a pitch-drop from H to L within a word or

COMPLEMENT+HEAD construction, one cannot rise again to

H—a statement similar to the one sometimes made for Japanese ‘‘accent ”,

However, (30) has clearly established Nara as much more tonal than Tokyo Japanese. Rather than accent,

the changes indicated in (30) are better attributed to a sequential constraint on tones, HLH ( Cahill, 2002,

2008 ), which Yip (2002, p. 137) refers to as

TROUGH . Nara ‘‘repairs ”input HLH violations by lowering

the second H. A more common response is for the intervening non-H TBU(s) to be raised, such that a H tone

plateau is created. In some languages, the process is limited to a single L TBU wedged between Hs, e.g. Ki-

hunde [Bantu; Democratic Republic of the Congo] ( Goldsmith, 1990, p. 36) and Mamainde´

[Nambiquaran;

Brazil] ( Eberhard, 2007, p. 297 ). In other languages, e.g. Amahuaca [Panoan; Peru] ( Russell and Russell,

1959, p. 152 ) and Luganda [Bantu; Uganda] ( Stevick, 1969; Hyman et al., 1987) multiple TBUs may undergo

plateauing. The inputs in (32a) establish that each constituent has a H to L pitch drop in isolation:

In (32b), however, we see that the Ls of the ﬁrst two words are deleted, accompanied by H plateauing from the

ﬁrst to last H of the construction. In fact, where closely related Haya deletes H tones, Luganda instead deletes

Ls and derives a H plateau. While HLH is widespread in tone systems ( Cahill, 2002, 2008), it can be violated. Thus, as seen in (33),

Siane [Trans-New-Guinea] has a /HLH/ schema, but not /LHL/ ( James, 1994, p. 126 ):

(32) a. y-a-la´

b-a `

‘he saw’ bi-ko ´

po `

‘cups’ by-aa = Walu ´

sı`

mbi ‘of Walusimbi’

HL HL HL

b. y-a-la ´

b-a ´

bı´

-ko ´

po ´ by-a ´

a´

=Wa ´

lu´

sı`

mbi ‘he saw the cups of Walusimbi’

H ------- H---------------H L LL

;;

ØØ

L.M. Hyman / Language Sciences 31 (2009) 213–238 229 As indicated, there are relatively few lexical entries with /L/ tone, without which Siane H would be obligatory

(but not culminative). As indicated to the right, /LH/ and /HL/ are realized with maximization of L, as in Fore

in (16). However, given the /HLH/ schema, regarding H as an ‘‘accent”would be quite problematic.

From the foregoing we draw two interim conclusions, both of which raise a question in turn: First, tone can

be very dependent on metrical structure, as in Seneca. The question is whether the metrical structure should be

identiﬁed as ‘‘stress ”, or whether stress is a

KIND of metrical structure (cf. Section 4). The second conclusion is

that the tonal modiﬁcations which occur in close-knit phrasal constituents are not necessary accentual. They

may be domain-based, as in the HLH cases. The question is whether H tone reduction or plateauing tends to

occur more in privative systems than in unambiguous /H, L/ systems. Both Haya and Luganda have been ana-

lyzed with privative /H, Ø/, although in both cases a third L value is introduced (within the lexical phonology

in Luganda, at the phrase level in Haya). Except for the restricted occurrence of short HL falling tones, Nara

might also be amenable to a privative interpretation. We thus come back to the relevance of tonal density, an

issue which is taken up again in Section 4.

4. Summary and conclusions In the preceding sections we have seen that it is possible to deﬁne ‘‘tone ”and ‘‘stress ”, as well as propose

unambiguous ‘‘prototypes ”for each. In contrast, we have not been able to provide an independent deﬁnition

of ‘‘pitch-accent ”which was said to characterize systems which either combine stress and tone or place ‘‘sig-

niﬁcant constraints ”on the distribution of their tones. We also suggested that there is no pitch-accent proto-

type, and that no language must be analyzed as pitch-accent: A tonal analysis is always possible. In Section 3

we considered four properties which may create the impression that a system is accentual: obligatoriness, cul-

minativity, privativity, and metricality. This permitted us to look at a number of ‘‘intermediate ”tone systems

which have one vs. another of these properties, but which we were not able to assign to a third prosodic type.

Take the oft-made suggestion that a pitch-accent language is one in which the tonal contrasts can be indicated

with by a presence vs. absence of a single mark per word, as in W. Basque, Somali, and Tokyo Japanese. While

these languages seem to group together in this way, do they form a class with Osaka Japanese and Luganda

which require two marks to indicate where the H begins and where it ends, or Fasu and Prinmi, which require

one mark for location and another to choose from two or three contrasting tones? Finally, how does Seneca ﬁt

into this? The basic system does not require any marks, rather ﬁrst builds trochaic feet, some of which then are

assigned a H to their ﬁrst syllable. Should some of these (or comparable) languages be folded in with stress

systems (Seneca) and others with tone (Luganda, Prinmi)? To see what is wrong with this kind of thinking, let us for the moment accept that there are three types of

prosodic systems (stress, pitch-accent, tone) and ask the following question: What is the relation between

them? Let us contrast three possibilities. The ﬁrst, shown in (34), is that their relation can be represented in

terms of trees:

In (34a), stress (S), pitch-accent (PA) and tone (T) are assumed to be three independent types of prosodic sys-

tems. In (34b), stress and pitch-accent are considered to be two types of accentual system as opposed to tone,

230 L.M. Hyman / Language Sciences 31 (2009) 213–238 while (34c) shows just the reverse: pitch-accent and tone are two types of pitch systems vs. stress. Of the three,

(33b) seems to mirror the traditional terminology ‘‘stress-accent”vs. ‘‘pitch-accent ”(Hyman, 1977a,b )or

‘‘stress-accent ”vs. ‘‘non-stress accent ”(Beckman, 1986 ). (34a–c) fail, however, to capture the intuition that

pitch-accent systems are somehow intermediate between stress and tone systems. This is shown in (34d), where

pitch-accent belongs both to the class of accentual languages and the class of pitch languages. While questionable when applied to restricted tone systems, (34d) might seem most appropriate for systems

which have both metrical structure (stress) and tone. This possibility is also captured in what Voorhoeve

(1973) calls the ‘‘box model ”, where languages are typologized by speciﬁc criteria. Returning to the deﬁnitions

of word level stress and tone in Section 2, we can ‘‘box ”languages according to whether they meet one, both,

or neither deﬁnition (cf. Hyman, 2006, p. 237):

Unlike the conception in (34), this approach makes no claim as to whether the (tone) languages in the ﬁrst row

are more similar to each other than the (stress) languages in the ﬁrst column. Nor is there any suggestion that

the prosodic systems which ﬁt each value of {Stress, Tone} will seem correctly placed once the full typology is

ﬁne-tuned by the addition of other properties. In other words, some languages will be closer to the prototypes,

providing unambiguous multiple evidence for stress vs. tone, while others will diverge from these prototypes in

diﬀerent ways. Of course, this leaves unresolved the question of how to deal with ambiguous or indeterminate cases.

Some scholars will not be happy with the claim that the languages in the right column of (35) do not have

word stress. Goedemans and van der Hulst (in press) , for instance, would like to include stress under a

broader, universal notion of accent: ‘‘A comprehensive typology of accent manifestation remains to be devel-

oped, but given the broad area of cues and functions it is likely that many more languages may have word

accent than just those in which accent is manifested as ‘pitch’ or ‘stress’. As a working hypothesis, we might

assume that all languages have accent. ”The issue has already come up as to whether all metrical struc-

ture = stress. If so, how might we view a language which invokes metrical structure only in its prosodic

morphology? The second problem concerns controversial systems such as W. Basque, Tokyo Japanese, and Somali,

which could easily be analyzed with tone, metrical structure, or both. In Hyman (2001a, 2006)I emphasized

a deﬁnition of a tone system as ‘‘one in which an indication of pitch enters into the lexical realization of at least

some morphemes. ”The intention is to abstract away from analyses which use diacritics (e.g. asterisks) to sim-

ply indicate pitch and recognize that this is just another way to say that these are

VERY reduced tone systems.

These systems may also be metrical, but that needs to be demonstrated other than by the intuitition that cul-

minativity = accentual. The goal has to be to set up an explicit and comprehensive set of criteria which can be

applied to diﬀerent systems in a linguistically signiﬁcant way. For example, word stress is both obligatory and

culminative and targets syllables, and word tone is obligatorily a property of morphemes. It is in this last connection that the third approach to word-prosodic typology falls ﬂat. A number of schol-

ars have used the term ‘‘continuum ”to refer either to the relation between stress and pitch-accent, between

tone and pitch-accent, or between all three. The claim has even been made that criteria may themselves be

continua, not binary as presented in Hyman (2006)and above. In the abstract to his paper, Hualde (2006)

takes the position that along the continuum there may even be multiple intermediate prototypes:

Typological classiﬁcations of prosodic systems based on binary features (such as presence vs. absence of

lexical tone, presence vs. absence of contrastive stress, etc.) can be very useful as a ﬁrst approximation to

linguistic diversity in this domain ...In this talk, I will argue that a more useful approach may be in

terms of a number of recognized prosodic prototypes . A statement such as ‘‘language L is like Swedish

except that x ”may provide us with a more clear picture of the prosody of language L than its classiﬁ-

cation in terms of binary features. ”[my emphasis—LMH]

L.M. Hyman / Language Sciences 31 (2009) 213–238 231 Of course linguistic properties may converge and produce similarities or identity in prosodic systems.Hual-

de et al. (2002, p. 548) thus assign Northern Biskaian Basque to what they call a ‘‘T(okyo)-type pitch-accent

language ”. But what does it mean to say that such ‘‘prototypes ”fall along a a stress—pitch-accent—tone

CONTINUUM ? In other words, how do we obtain a continuum such as the one in (36), repeated from (1):

While the above languages show various degrees of tone from zero (English) to a lot (Mandarin), where does a

metrical tone system such as Seneca ﬁt in? Stacey (2004, p. 41)makes the same point with respect to Creek:

‘‘ ... metaphorical linearity of the continuum poses great diﬃculties when trying to classify a language such as

Creek, which implements tone under a metrical system ...Furthermore, the continuum is unable to account

for diﬀerent manifestations of the same system ...”

I can think of three ways to derive a single-dimensional continuum such as in (36). The ﬁrst would be to

appeal to a single continuous or quantitative property. In Section 1I suggested tonal density about which

Gussenhoven (2001, p. 15296) states the following:

A phonological typology of tone might be based on tonal density: how many locations are speciﬁed for

tone, that is, have tonal associations? In other words, languages diﬀer in the selection of the tone-bearing

unit: in the ‘densest’ case they specify every mora for tone, and in the sparsest case they just mark the

phrasing. When more than one tone is associated with a syllable, like HL or LH, the combination is

known as a contour tone, while a single tone is a level tone.

There is no doubt that a tonal density coeﬃcient could be devised for each language which would result in our

ability to list all the languages of the world from least to most tonally dense. Let us assume that we all agree on

analytical choices concerning the levels and domains to be examined and the methodological problem of how

to count. While the endpoints would be clear, the question, however, is whether there would be any internal

organization to the intermediate languages. In other words, would tonal density simply give us a continuum,

or would we gain insight into the typologically diverse intermediate systems? Suggestion: Someone should do a

pilot study and see what it yields. Since Hualde (2006) did not advocate a single dimension such as tonal density, but rather examined the

kinds of criteria considered in this paper (e.g. obligatoriness and culminativity), a second way that a discrete

continuum could conceivably be obtained is if the criteria fell into a subset relation, as represented in (37a). (36) English- - - -------W.Basque- ---------Tokyo Japanese- ---------Luganda- - - -------Mandarin

(37)

232 L.M. Hyman / Language Sciences 31 (2009) 213–238 In (37) I have posited four stress-like properties (P 1–P 4) which are in a subset relation in the sense that P 1im-

plies P

2,P 3,P 4;P 2implies P 3,P 4;andP 3implies P 4. Since the systems so deﬁned fall into a scalar hierarchy, the

ﬁve possible languages can be ranked from a value of 4 (L

1= the most stress-like) down to a value of zero

5= the least stress-like). There is exactly one combination of properties that deﬁnes each step on the

continuum.

The question is whether there are in reality four properties, P

1–P 4, which occur in such a subset relation or

whether the properties are in fact independent from each other. If the latter, we obtain the ‘‘pick-and-choose ”

distributions in (36b). The values are the same (4–0), but this time four combinations give a value of 3 and 1,

while six combinations give a value of 2. If P

1–P 4= obligatoriness, culminativity, privativity, metricality, we

can easily see in (38) that the top two are in fact independent ( Hyman, 2006, p. 245):

Since all four combinations of

OBLIG (H) and CULM (H) are attested, the two properties are not in a subset rela-

tion and hence do not deﬁne a continuum. This leaves one remaining possibility: weighting of properties. Assuming a possible score of 100 (=very stress-like) vs. 0 (=not stress-like at all), we can assign either the

same (i.e. 25) points to P

1–P 4, or a weighted amount to each, for example P 1(40) + P 2(30) + P 3(20) + P 4(10).

Systems which satisfy P

1completely would receive 40 points, while those in the mid range of satisfying P 1

might receive 20 points. This could be a way, then, of quantifying what Hualde (2006)has in mind in stating

that obligatoriness is itself a continuum (i.e. multivalued). His example is that languages diﬀer in the degree to

which they tolerate unaccented/toneless grammatical elements, e.g. the deﬁnite (but not indeﬁnite) article in

Spanish. Languages clearly do diﬀer in ‘‘stress density ”: Some languages treat grammatical elements as stress-

less clitics and particles, while others require them to be treated the same as lexical words and hence receive

stress. The idea is that languages could be placed on a continuum according to their degree of non-prosodiﬁed

(stressless, toneless) elements, which might result in W. Basque being placed near certain stress languages (e.g.

Spanish), rather than in a box with the tone systems in (35). The notions of tonal density, stress density, and weighted criteria appear to be attempts to come to terms

with the multidimensional nature of the problem: There is not just one structural property involved in char-

acterizing prosodic systems, but most of these properties aﬀect prosodic density. We thus see the diﬃculty of

placing languages along a continuous scale, let alone naming them. In practice those interested in word-pro-

sodic typology are all engaged in determining how phonologies function

AS SYSTEMS . Typology should be

PROPERTY-DRIVEN , whether one chooses to ask ‘‘what’s possible? ”or ‘what’s where why? ”’( Bickel, 2007, p.

239 ). To see how this might be done in word-prosodic typology, let us consider what has been said about

nasality. The typological question is how the phonetic feature of nasality is structured in phonological systems.

As seen in (38) there are at least ﬁve nasal ‘‘types ”, which are listed in (39) in the order of relative frequency in

the world’s languages ( Cohn, 1993; Clements and Osu, 2003 ):

As is well known, nasality is a consonant feature in most languages. Contrastive nasality on vowels is not rare,

but does tend to cluster in certain areas, e.g. West Africa. Nasal prosodies on whole morphemes are less com-

mon, and are especially well-known from Amazonian languages. Finally, approximately a dozen languages

have been reported to lack nasality in any form, including Lushootseed, Quileute, Piraha ˜, Central Rotokas,

and several Lakes Plains languages of Indonesia (Papua): Doutai, Iau, Obokuitai, Sikaritai and Kirikiri. In

(39) Nasality may be underlyingly constrastive:

a. on consonants only: /m, n, N/ e.g. Lahu

b. on vowels and consonants: /

,u ˜,a

˜/, /m, n,

N/ e.g. Bambara

c. on vowels only /

˜,a

˜/

e.g. Klao

d. on whole morphemes /CVC/ N e.g. Barasana

e. absent entirely e.g. Lushootseed

L.M. Hyman / Language Sciences 31 (2009) 213–238 233 addition to examining how or why these common/rare or geographically skewed distributions are found, the

above distinctions allow other questions to be raised, such as: Why is it that there are underlying systems with

nasality contrasting only on vowels, as in (39c), but no surface system where nasality is

REALIZED only on vow-

els? Note, ﬁnally, that although ﬁve diﬀerent ‘‘types ”are attested, there does not seem to be a need to give each

one a name, or to ask which ones are the more closely related to each other? (If asked, I’d suggest (39c,d).)

This is the property-driven approach that I believe should be followed in word-prosodic typology as well, as

we attempt to transcend the temptation to see typology as ‘‘classifying languages ”. I thus agree with McCaw-

ley (1978, p. 127) : ‘‘For each of the languages discussed in this section, one can ask the question, ‘Is it a pitch-

accent language or a tone language?’ However, I think that that is a stupid question to ask ...”Phonological

typology should characterize languages by P1, P2, ...,Pn, not by labels. The questions we should ask therefore

are:

The ﬁrst question would lead to a typology of metrical functions as well a metrical realizations. It would in-

clude the well-known range of stress systems ( Hayes, 1995), but also Seneca, where the main realization of left-

to-right iterative trochee-formation is the assignment of H tones. The second question would lead to a typol-

ogy of tones sytems and their realizations. It would include the well-known range of systems from such over-

views as Hyman (2001a) and Yip (2002) , but could be extended to include systems which exploit pitch features

at the phrase level ( Ladd, 1996; Gussenhoven, 2004 ) or which combine lexical and phrasal tones ( Pierrehum-

bert and Beckman, 1988 ; see also the papers in volumes such as van der Hulst, 1999; Jun, 2005; Gussenhoven

and Riad, 2007; Riad and Gussenhoven, 2007 ).

A prerequisite to answering the above questions is to ﬁrst establish what is meant by stress and tone. If

stress is equivalent to underlying or abstract metrical structure, when is it appropriate to posit metrical struc-

ture, and the same for lexical pitch features? Even if we agreed on this, there could still be ambiguous or inde-

terminate systems. Kinga ( Schadeberg, 1973) and Nubi (Gussenhoven, 2006 ) both have an obligatory and

culminative H tone on each word. We can conclude that Kinga’s is not a stress system, since the Hs are

assigned directly to moras and produce rising and falling tones (as in Somali). However, the TBU is the syl-

lable in Nubi. As Gussenhoven suggests, it may be that Nubi is ‘‘pivotal ”: its H may be seen as metrical prom-

inence (=stress) or it may simply be a severely restricted (obligatory, culminative) tone. Whatever Nubi is, we

must ask how it is the same vs. diﬀerent from stress systems whose cues are primarily pitch, e.g. Turkish ( Levi,

2005 ). In conclusion, the need for a third category of pitch-accent has not been established, and in any case,

would seem more interpretive than demonstrable. As Gussenhoven (2004, p. 42)puts it: ‘‘Accent’...is an ana-

lytical notion and cannot be measured. [It is] thus diﬀerent from stress, which is typically an observable phe-

nomenon, and diﬀerent also from tone, whose existence is equally measurable. ”Whether dealing with a

language with obligatory or culminative tone, a language with metrical tone, or a language with both stress

and tone, we would do well to avoid using the term ‘‘pitch-accent ”as a catch-all in favor of direct reference

to the properties of what we all recognize to be a diverse collection of intermediate word-prosodic systems.

Acknowledgements

Earlier versions of parts of this paper were presented at the 2007 Annual Meeting of the Linguistic Society

of America Meeting, Anaheim, the 7th Biennial Conference of the Association for Linguistic Typology

(Paris), Universite ´

Lyon2, Universita ¨t Konstanz, Institut de Linguistique et Phone ´

tique Ge ´

ne ´

rales et Appli-

que ´

es (Paris III), University of British Columbia, MIT, New York University, and the University of Califor-

nia, Santa Barbara. I am especially appreciative of the many responses I received on these occasions, to

Wallace Chafe for his repeated help with Seneca, and to Jose ´

Hualde, Mike Kentowicz, and one anonymous

reviewer who provided very thoughtful comments and challenges to the ideas presented in this paper. (40) Which languages have word-level metrical structure (‘‘stress ”)? What are its properties?

(i.e. what do languages

DO with metrical structures?)

Which languages have word-level pitch features (‘‘tone ”)? What are its properties?

(i.e. what do languages

DO with pitch features?)

234 L.M. Hyman / Language Sciences 31 (2009) 213–238 References

Agnew, A., Pike, E.G., 1957. Phonemes of Ocaina (Huitoto). IJAL 23, 24–27.

Akinlabi, Akinbiyi, 1985. Tonal underspeciﬁcation and Yoru`

ba ´

tone. Doctoral dissertation, University of Ibadan.

Beckman, M.E., 1986. Stress and Non-stress Accent. Foris Publications, Dordrecht.

Bickel, B., 2007. Typology in the 21st century: major current developments. Linguistic Typology 11, 239–251.

Bybee, J.L., Chakraborti, P., Jung, D., Scheibman, J., 1998. Prosody and segmental eﬀect: some paths of evolution for word stress. Studies in Language 22, 267–314.

Caballero, G., 2005. The stress system of Central Rara ´

muri: root privilege, prosodic faithfulness and markedness reversals. Ms. University

of California, Berkeley. ROA-706.

Cahill, M., 2002. Some universals of tone. Handout from colloquium, Oregon SIL, July 24, 2002.

Cahill, M., 2008. More universals of tone. SIL Electronic Working Papers. Available from: < http://www.sil.org/silewp/2007/silew2007-

007.pdf >.

Casimjee, F., Kisseberth, C.W., 1992. On the tonology of depressor consonants: evidence from Mijikenda and Nguni. In: Special Session on Tone System Typology, vol. 18. Berkeley Linguistic Society, 26–40.

Chafe, W.L., 1977. Accent and related phenomena in the ﬁve nations Iroquois languages. In: Hyman, L.M. (Ed.), Studies in Stress and Accent. Southern California Occasional Papers in Linguistics 4. University of Southern California, Los Angeles, pp. 169–181.

Chafe, W.L., 1996. Sketch of Seneca: an Iroquian language. In: Handbook of American Indian Linguistics, vol. 17. Smithsonian Institution, Washington, DC, pp. 551–579.

Chida, S., 2001. On the tone of the Dom language. Ms. English translation by Yoshihiro Kawabata of Syuntaro ˆ

Tida (2000), Domugo no

to ˆ

n ni kansuru ko ˆ

satu. Masters Thesis, Kyoto University.

Clark, M., 1988. An accentual analysis of Zulu. In: Hulst, H., van der Smith, N. (Eds.), Autosegmental Studies on Pitch Accent. Foris,

Dordrecht, pp. 51–79.

Clements, G.N., Osu, Sylvester, 2003. Ikwere nasal harmony in typological perspective. In: Sauzet, P., Zribi-Hertz, A. (Eds.), Typologie des Langues d’Afrique et Universaux de la Grammaire, vol. II. L’Harmattan, Paris, pp. 70–95.

Cohn, A., 1993. A survey of the phonology of the feature [+nasal]. Working Papers of the Cornell Phonetics Laboratory, vol. 8. Cornell

University, Ithaca, pp. 141–203.

Creissels, D., Gre ´

goire, C., 1993. La notion de ton marque ´

dans l’analyse d’une opposition tonale binaire: le cas du mandingue. Journal of

African Languages and Linguistics 14, 107–154.

Daly, J.P., Hyman, L.M., 2007. On the representation of tone in Pen ˜oles Mixtec. IJAL 73, 165–208.

de Lacy, P., 2002. The interaction of tone and stress in optimality theory. Phonology 19, 1–32.

Demers, R., Escalante, F., Jelinek, E., 1999. Prominence in Yaqui words. IJAL 65, 40–55.

Ding, P.S., 2001. The pitch accent system of Niuwozi Prinmi. Linguistics of the Tibeto-Burman Area 24 (2), 57–83.

Ding, P.S., 2006. A typological study of tonal systems of Japanese and Prinmi: towards a deﬁnition of pitch-accent languages. Journal of Universal Language 7, 1–35.

Dommel, P.R., Dommel, E.G., 1991. Kaure phonology. Workpapers in Indonesian Languages and Culture, pp. 51–68.

Donohue, M., 1997. Tone in New Guinea languages. Linguistic Typology 1, 347–386.

Donohue, M., 2005. Tone and the trans New Guinea languages. In: Kaji, S. (Ed.), Cross-Linguistic Studies of Tonal Phenomena. ILCAA, Tokyo, pp. 33–53.

Eberhard, D., 2007. Mamainde ´

tone. In: Wetzels, L. (Ed.), Language Endangerment and Endangered Languages. CNWS Publications,

Leiden, pp. 287–308.

Epps, P., 2005. A grammar of Hup. Doctoral dissertation, University of Virginia.

Firchow, I.B., Firchow, J., 1969. An abbreviated phoneme inventory. Anthropological Linguistics 11, 271–276.

Fontaney, L., 1980. Le verbe. In: Nsuka-Nkutsi, F. (Ed.), Ele ´

ments de Description du Punu. CRLS, Universite ´

Lyon II, pp. 51–114.

Gabas Jr., N., 1999. A grammar of Karo, Tupı ´

(Brazil). Doctoral dissertation, University of California, Santa Barbara.

Garde, P., 1967. L’Accent. Presses Universitaires de France, Paris.

Goedemans, R., van der Hulst, H., in press. StressTyp: a database for word accentual patterns in the world’s languages. In: Everaert, M., Musgrave, S. (Eds.), The Use of Databases in Cross-Linguistics Research. Mouton de Gruyter, New York/Berlin. Available from:

Goldsmith, J., 1984. Tone and accent in Tonga. In: Clements, G.N., Goldsmith, John (Eds.), Autosegmental Studies in Bantu tone. Foris, Dordrecht, pp. 19–52.

Goldsmith, J., 1987. In: Tone and Accent, and Getting the Two Together, vol. 13. Berkeley Linguistic Society, pp. 88–104.

Goldsmith, J., 1990. Autosegmental and Metrical Phonology. Basil Blackwell Ltd., Cambridge, MA.

Gomez, E., Kenstowicz, M., 2000. Barasana tone and accent. IJAL 66, 419–463.

Greenberg, J.H., 1974. Language Typology: A Historical and Analytic Overview. Mouton, The Hague.

Gussenhoven, C., 2006. Between stress and tone in Nubi word prosody. Phonology 23, 193–223.

Guiro ´

n Higuita, J.M., Wetzels, W.L., 2007. Tone in Wa

˜nso

¨ho

¨t (Puinave). In: Wetzels, Leo (Ed.), Language Endangerment and

Endangered Languages. CNWS Publications, Leiden, pp. 129–156.

Gussenhoven, C., 2001. Suprasegmentals. In: Smelser, N.J., Paul, Baltes B. (Eds.), International Encyclopedia of the Social and the

Behavioural Sciences. Pergamon, Oxford, pp. 15294–15298.

Gussenhoven, C., 2004. The Phonology of Tone and Intonation. Cambridge University Press.

Gussenhoven, C., 2007. A Moraic and a Syllabic H Tone in Yucatec Maya. Ms. University of Nijmegen. L.M. Hyman / Language Sciences 31 (2009) 213–238

235 Gussenhoven, C., Riad, T. (Eds.), 2007. Tones and Tunes. Experimental Studies in Word and Sentence Prosody, vol. 2. Mouton deGruyter, Berlin.

Hagberg, L., 2006. The place of pitch-accent in a typology of phonological prominence. Paper Presented at BeST Conference. University of Leiden, Ms. SIL, Mexico.

Hage `

ge, C., 1992. Morphological typology. Oxford International Encyclopedia of Linguistics, vol. 3, pp. 7–8 (OUP).

Haraguchi, S., 1977. The Tone Pattern of Japanese: an Autosegmental Theory of Tonology. Kaitakusha, Tokyo.

Hayata, T., 1999. Accent and tone: towards a general theory of prosody. In: Kaji, S. (Ed.), Cross-Linguistic Studies of Tonal Phenomena: Tonogenesis, Typology, and Related Topics. ILCAA, Tokyo, pp. 221–234.

Hayes, B., 1995. Metrical Stress Theory: Principles and Case Studies. University of Chicago Press.

Hayward, R.J., 2000. Observations on tone in the Higir dialect of Nara. In: Vossen, Rainer et al. (Eds.), Mehr als nur Worte. Ru ¨diger

Ko ¨ppe Verlag, Cologne, pp. 247–267.

Hayward, R.J., 2006. The OHO constraint. In: Voeltz, E. (Ed.), Studies in African Linguistic Typology. John Benjamins, Amsterdam, pp. 155–169.

Heath, J., 2008. Jamsay Grammar. Mouton de Gruyter, Berlin.

Hollenbach, B.E., 1988. The asymetrical distribution of tone in Copala Trique. In: van der Hulst, H., Smith, N. (Eds.), Autosegmental Studies on Pitch Accent. Foris, Dordrecht, pp. 167–182.

Hostetler, R., Hostetler, C., 1975. A tentative description of Tinputz phonology. Phonologies of Five Austronesian Languages. Workpapers in Papua New Guinea Languages, vol. 13, pp. 5l–44.

Hualde, J.I., 2006. Basque, Palenquero, and the typology of word-prosodic systems. Paper Presented at the 34th Annual Berkeley

Linguistics Society Meeting, February 10–12, 2006. Abstract: .

Hualde, J.I., Elordieta, G., Gaminde, I., Smiljanic ´

, R., 2002. From pitch accent to stress accent in Basque. In: Gussenhoven, C., Warner,

N. (Eds.), Laboratory Phonology, vol. 7. Mouton de Gruyter, Berlin, pp. 547–584.

Hyman, L.M., 1977a. Tone and/or accent. In: Napoli, D.J. (Ed.), . In: Elements of Tone, Stress and Intonation. Georgetown University Press, Washington, pp. 1–20.

Hyman, L.M., 1977b. The nature of linguistic stress. In: Hyman, Larry M. (Ed.), Studies in Stress and Accent. Southern California Occasional Papers in Linguistics, vol. 4. University of Southern California, Los Angeles, pp. 37–82.

Hyman, L.M., 1981. Tonal accent in Somali. Studies in African Linguistics 12, 169–203.

Hyman, L.M., 2001a. Tone systems. In: Haspelmath, M. et al. (Eds.), Language Typology and Language Universals: An International Handbook, vol. 2. Walter de Gruyter, Berlin & New York, pp. 1367–1380.

Hyman, L.M., 2001b. Privative tone in Bantu. In: Kaji, S. (Ed.), Cross-Linguistic Studies of Tonal Phenomena. Institute for the Study of

Languages and Cultures, Tokyo, pp. 237–257.

Hyman, L.M., 2006. Word-prosodic typology. Phonology 23, 225–257.

Hyman, L.M., 2007. Universals of tone rules: 30 years later. In: Riad, Tomas, Gussenhoven, Carlos (Eds.), Tones and Tunes: Studies in Word and Sentence Prosody. Mouton de Gruyter, Berlin, pp. 1–34.

Hyman, L.M., Byarushengo, E.R., 1984. A model of Haya tonology. In: Clements, G.N., Goldsmith, J. (Eds.), Autosegmental Studies in Bantu Tone. Foris Publications, Dordrecht, pp. 53–103.

Hyman, L.M., Magaji, D.J., 1970. Essentials of Gwari Grammar. Occasional Publication No. 27, Institute of African Studies. University

of Ibadan Press, Ibadan, Nigeria.

Hyman, L.M., Wilson, S., 1992. Review of Hulst, H. van der, Smith, N. (Eds.), Autosegmental Studies on Pitch Accent (Foris, Dordrecht). Language 67, 356–363.

Hyman, L.M., Katamba, F.X., Walusimbi, L., 1987. Luganda and the strict layer hypothesis. Phonology [Yearbook] 4, 87–108.

James, D.J., 1994. Word tone in a Papuan language: an autosegmental solution. Language and Linguistics in Melanesia 25, 125–148.

Jenison, S.D., Jenison, P.B., 1991. Obokuitai phonology. Workpapers in Indonesian Languages and Culture, vol. 9, pp. 69–90.

Jun, S.-A. (Ed.), 2005. Prosodic Typology: The Phonology of Intonation and Phrasing. Oxford University Press, Oxford, New York.

Kamanda-Kola, R., 2003. Phonologie et Morphosyntaxe du M [n [: Langue Oubanguienne du Congo R.D. Munich: Lincom Europa.

Kaschube, D., 1954. Examples of tone in Crow. IJAL 20, 34–36.

Keller, N.E., 1999. Cambios tonales en la palabra verbal de la lengua tanimuca. Congresso de Lingu ¨ı

stica Amerindia y Criolla. Lenguas

Aborı ´

genes de Colombia, Memorias, pp. 72–95

Kenstowicz, M., Kisseberth, C.W., 1990. Chizigula tonology: the word and beyond. In: Inkelas, S., Zec, D. (Eds.), The Phonology-Syntax Connection. University of Chicago Press, Chicago, pp. 163–194.

Kisseberth, C.W., 1984. Digo tonology. In: Clements, G.N., Goldsmith, J. (Eds.), Autosegmental Studies in Bantu Tone. Foris Publications, Dordrecht, pp. 105–182.

Kisseberth, C.W., 1992. Metrical structure in Zigula tonology. In: Gowlett, D. (Ed.), African Linguistic Contributions (Presented in Honour of Ernst Westphal). Pretoria, Via Afrika, pp. 136–166.

Kisseberth, C.W., 2005. Accent and phrasing in Chimwiini. In: Kaji, S. (Ed.), Cross-Linguistic Studies of Tonal Phenomena: Historical Development, Tone-Syntax Interface, and Descriptive Studies. ILCAA, Tokyo, pp. 129–145.

Kisseberth, C.W., Volk, E., 2007. Theoretical consequences of depressor consonants. Paper Presented at the Workshop on Segments and Tone, Meertens Instituut, Amsterdam, June 6–7, 2007.

Klingenheben, A., 1949. Ist das Somali eine Tonsprache? Zeitschrift fu ¨r Phonetik. Sprachwissenschaft und Kommunikationsforschung 3,

289–302.

Kwenzi Mikala, J., 1980. Esquisse phonologique du punu. In: Nsuka-Nkutsi, F. (Ed.), Ele ´

ments de Description du Punu. Universite ´

Lyon

II, CRLS, pp. 7–18.

236

L.M. Hyman / Language Sciences 31 (2009) 213–238 Ladd, D.R., 1996. Intonational Phonology. Cambridge University Press.

Lehiste, I., 1970. Suprasegmentals. MIT Press, Cambridge, MA.

Levi, S.V., 2005. Acoustic correlates of lexical accent in Turkish. Journal of the International Phonetic Association 35, 73–97.

MacDonald, G., MacDonald, G., 1974. Dadibi phonology. In: Phonologies of Four Papua NewGuinea Languages. Workpapers in PapuaNew Guinea Languages, vol. 4, pp. 129–158.

Maddieson, I., 1984. Patterns of Sounds. Cambridge University Press.

Maddieson, I., Precoda, K., 1990. Updating UPSID. UCLA Working Papers in Phonetics, vol. 74, pp. 104–111.

Martinet, A., 1960. Ele ´

ments de Linguistique Ge ´

ne ´

rale. Librairie Armand Colin, Paris.

May, J., Loeweke, E., 1964. The phonological hierarchy in Fasu. Anthropological Linguistics 7.5, 89–97.

McCawley, J.D., 1968. The Phonological Component of a Grammar of Japanese. Mouton, The Hague.

McCawley, J.D., 1970. In: Some Tonal Systems that Come Close to Being Pitch Accent Systems but do not Quite Make it, vol. 6. Chicago Linguistic Society, pp. 526–531.

McCawley, J.D., 1978. What is a tone language? In: Fromkin, Victoria A. (Ed.), Tone: A Linguistic Survey. Academic Press, New York, pp. 113–131.

Melinger, A., 2002. Foot structure and accent in Seneca. IJAL 68, 287–315.

Miller, W.R., 1965. Acoma Grammar and Texts. University of California Press, Berkeley & Los Angeles.

Newman, J., Petterson, R.G., 1990. The tones of Kairi. Oceanic Linguistics 291, 49–76.

Olson, K.S., 2005. The Phonology of Mono. SIL International & The University of Texas at Arlington, Dallas.

Ospina Bozzi, A.M., 2002. Les structures e ´

le´

mentaires du yuhup maku ´

, langue de l’Amazonie colombienne: morphologie et syntaxe. The `

de Docteur de l’Universite ´

Paris 7.

Parker, S., 1997. An OT account of laryngealization in Cuzco Quechua. Work Papers of the Summer Institute of Linguistics, vol. 41. University of North Dakota, pp. 1–11.

Pearce, M., 2006. The interaction between metrical structure and tone in Kera. Phonology 23, 259–286.

Pearce, M.D., 2007. The interaction of tone with voicing and foot structure: Evidence from Kera phonetics and phonology. Doctoral dissertation, University of London.

Philippson, G., 1998. Tone reduction vs. metrical attraction in the evolution of Eastern Bantu tone systems. In: Hyman, L.M., Kisseberth,

C.W. (Eds.), Theoretical Aspects of Bantu Tone. CSLI, Stanford, pp. 315–329.

Picanc o, G.L., 2005. Munduruku ´

: phonetics, phonology, synchrony. Doctoral dissertation, University of British Columbia.

Pierrehumbert, J.B., Beckman, M.E., 1988. Japanese Tone Structure. MIT Press, Cambridge, MA.

Plank, F., 2001. Typology by the end of the 18th century. In: Auroux, S. et al. (Eds.), History of the Language Sciences: An International Handbook on the Evolution of the Study of Language from the Beginnings to the Present, vol. 2. Mouton de Gruyter, Berlin, pp.

1399–1414.

Poser, W., 1984. The phonetics and phonology of tone and intonation in Japanese. Doctoral dissertation, MIT.

Post, M.W., 2007. A grammar of Galo. Doctoral dissertation, La Trobe University.

Pulleyblank, D., 1986. Tone in Lexical Phonology. D. Reidel, Dordrecht.

Riad, T., Gussenhoven, C., 2007. Tones and Tunes. In: Typological Studies in Word and Sentence Prosody, vol. 1. Mouton de Gruyter, Berlin.

Rule, M., 1993. The Culture and Language of the Foe—Lake Kutubu People of the Southern Highlands Province, Papua New Guinea. The author, Australia.

Russell, R., Russell, D., 1959. Syntactotonemics in Amahuaca (Pano). Se ´

rie Lingu ¨istica Especial. Publicac

oes do Museu Nacional, Rio de

Janeiro, Brasil, pp. 128–167.

Saeed, J., 1999. Somali. John Benjamins, Amsterdam.

Salmons, J., 1992. Accentual Change and Language Contact: Comparative Survey and a Case Study of Early Northern Europe. Stanford University Press, Stanford.

Schadeberg, T.C., 1973. Kinga: a restricted tone language. Studies in African Linguistics 4, 23–48.

Schadeberg, T.C., 2000. The tonal system of Ekoti nouns. In: Vossen, R., Mietzner, A., Meissner, A. (Eds.), Mehr als nur Worte. Ru ¨diger

Ko ¨ppe Verlag, Ko

¨ln, pp. 597–612.

Scott, G., 1990. A reanalysis of Fore accent. La Trobe University Working Papers in Linguistics 3, 139–150.

Seifart, F., 2005. The Structure and Use of Shape-Based Noun Classes in Miran ˜a (North West Amazon). Max Planck Institut Series in

Psycholinguistics, Nijmegen.

Speck, C.H., 1978. Texmelucan Zapotec suprasegmental phonology. Work Papers of the Summer Institute of Linguistics, vol. 22. University of North Dakota, pp. 1–28.

Stacey, E., 2004. Phonological aspects of Blackfoot prominence. M.A. Thesis, U. Alberta.

Stenzel, K., 2007. Glottalization and other suprasegmental features in Wanano. IJAL 73, 331–336.

Stevick, E.W., 1969. Pitch and duration in Ganda. Journal of African Languages 8, 1–28.

Swick, J., 1966. Chuave phonological hierarchy. Paciﬁc Linguistics Series A 7, 33–l48.

Ulrich, C.H., 1989. Tone association in Choctaw. Southwest Journal of Linguistics 9, 163–174.

Vajda, E., 2001. Test materials dated August 17, 2001. Posted at:

Phonology3.htm> .

van Coetsem, F., 1996. Towards a Typology of Lexical Accent: Stress Accent and Pitch Accent in a Renewed Perspective. Winter, Heidelberg.

van der Hulst, H. (Ed.), 1999. Word Prosodic Systems in the Languages of Europe. Mouton de Gruyter, Berlin. L.M. Hyman / Language Sciences 31 (2009) 213–238

237 van der Hulst, H., Smith, N., 1988. Introduction. In: van der Hulst, Harry, Smith, Norval (Eds.), Autosegmental Studies on Pitch Accent,Foris, Dordrecht, pp. ix–xxiv.

Voorhoeve, J., 1973. Safwa as restricted tone system. Studies in African Linguistics 4, 1–22.

Warner, N., 1997. Japanese ﬁnal-accent and unaccented phrases. Journal of Phonetics 25, 43–60.

Watters, D., 2002. A Grammar of Kham. Cambridge University Press.

Weber, D., Thiesen, W., 2000. A synopsis of Bora tone. Ms. SIL. Work Papers of the Summer Institute of Linguistics, North Dakota Session 45 (Online 2001).

Willard, E.R., 2004. Dominance Eﬀects in a Dialect of Mam Maya. Ms. University of California, Berkeley.

Wurm, S.A., 1973. The Kiwaian language family. In: Franklin, K.J. (Ed.), The Linguistic Situation in the Gulf district and Adjacent Areas Papua New Guinea, Paciﬁc Linguistics C-26. ANU, Canberra, pp. 219–260.

Yip, M., 2002. Tone. Cambridge University Press.

Zoll, C., 2003. Optimal tone mapping. Linguistic Inquiry 34, 225–268. 238 L.M. Hyman / Language Sciences 31 (2009) 213–238