Professional Documents
Culture Documents
(Studies in Generative Grammar 51) Jeroen Maarten Van de Weijer, Tetsuo Nishihara - Issues in Japanese Phonology and Morphology (2001, de Gruyter) PDF
(Studies in Generative Grammar 51) Jeroen Maarten Van de Weijer, Tetsuo Nishihara - Issues in Japanese Phonology and Morphology (2001, de Gruyter) PDF
WDE
G
Studies in Generative Grammar 51
Editors
Harry van der Hulst
Jan Köster
Henk van Riemsdijk
Mouton de Gruyter
Berlin · New York
Issues in Japanese Phonology
and Morphology
Edited by
Jeroen van de Weijer
Tetsuo Nishihara
Mouton de Gruyter
Berlin · New York 2001
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter G m b H & Co. KG, Berlin.
Preface ν
Shosuke Haraguchi
The Accent of Tsuruoka Japanese Reconsidered 47
Takeru Honma
How should we Represent 'g' in loge in Japanese Underlyingly? . . 67
/
Haruo Kubozono
Hidetoshi Shiraishi
Prosodie Structure and Sandhi Phenomena in the Saru Dialect
of Ainu 141
Shin-ichi Tanaka
The Emergence of the 'Unaccented': Possible Patterns and
Variations
Shohei in Japanese Compound Accentuation
Yoshida 159
An Element-Based Analysis of Affrication in Japanese 193
Taro Kageyama
Word Plus: The Intersection of Words and Phrases 245
Takayasu Namiki
Further Evidence in Support of the Righthand Head Rule
in Japanese 277
Yosihiro Masuya
Two Different Kinds of Rhythm: Japanese and English 327
Noriko Yamane
sC Clusters as Complex Segments: Evidence from the Contrastive
Phonology of English and Japanese 357
Introduction
we argue that phonology and syntax are more or less intermingled with
each other.
1. Problem settings
Β C
a. Branching in phonology 2
Word
Foot Foot
b. Branching in morphology
Word
Stem Suffix
Stem Suffix
I I
Christ ian ity
Heaviness in Interfaces 5
c. Branching in syntax
Tense
John Tense
Tense ν
John ν
ν loves
loves ν
loves Mary
noun phrase branches: It is very unnatural for a language that the noun
phrase Dogs can passivize while the dog cannot or the other way around.
Also, we cannot imagine a language in which who can Wh-move while
who the hell cannot. This shows that branching functions as an output
condition for almost all structure building processes in all components of
grammar related to forms (we will come back to morphology later be-
cause morphology behaves like phonology in some respects and like syn-
tax in others) as opposed to meanings.
However, there is a group of movement rules that seems to be triggered
by some kind of branching/heaviness. English Heavy NP Shift and Japa-
nese Scrambling are canonical cases of this. The common feature of the
two processes is that they both are optional and they both do not conform
to the "Last Resort" view of movement rules which is assumed as stan-
dard in current syntactic theories like the minimalist program (Chomsky
1995). In the current theory of syntax, movement must be triggered by
some purely (morpho-)syntactic reasons. For example, a movement pro-
cess which is generally called Passive must be triggered by the Case-as-
signing/checking properties of verbs and nouns, and Wh-Movement is
also triggered by the necessity of attraction of the feature [+Wh] by
COMP. Because there is a well-established trigger, it is currently assumed
that all movement processes must be obligatory. However, a process like
Heavy NP Shift is optional, and, moreover, it does not have a good can-
didate of its trigger. Heavy NP Shift occurs only because the NP is
"heavy" in some sense, as illustrated in (2).
formation can affect syntax, and vice versa (Selkirk and Tateishi 1988,
1991). No, this does not show that syntax is not a separate module. It is
the nature of the component called syntax that has been wrongly un-
derstood. Tateishi (1999) points out that the semantics-oriented view of
syntax as is currently assumed in the minimalist program cannot stand
as is, because such a view cannot properly "spell out" phonological
strings out of syntactic trees and cannot properly describe various phe-
nomena related to phonology-syntax interface, such as word order and
the distinction between function and content words. Akasaka (1996),
on the other hand, argues that the semantics/LF-oriented view of syn-
tax cannot properly explain a simple syntactic fact of Binding. Based
on these studies, we both conclude that syntax must be more "phonol-
ogy/form-oriented." Contrary to the view of syntax in Chomsky (1995)
that syntax builds up a structure which can be properly interpreted in
semantics and that phonology "strips off" a phonological matrix out of
such a structure (SPELL OUT), we say that syntax builds up a struc-
ture which is pronounceable, and semantics "strips off" semantic infor-
mation out of it (Interpret a). This, we believe, is an appropriate form
of components of grammar which can ultimately explain form/meaning
interfaces.
In what follows, we will first introduce heaviness effects in phonology
and morphology (sections 2 and 3), and then point out problems related
to syntactic heaviness effects: Scrambling and Heavy NP Shift (section
4). Section 5 summarizes our view of grammatical components and their
interactions, and concludes the paper.
2. "Heaviness" in phonology:
Foot, accent, and weight-sensitivity
Except for (4a), where, with the final syllable extrametricality, only one
candidate is left to place accent, this pattern of stress assignment can be
accounted for in terms of branching into moras. The main stress feet in
(4b) and (4c) have the following structure:
η a a
Heaviness in Interfaces 9
This means that, whatever the exact content is, we have constraints of the
following kind in an OT-type constraint form.
10 Yukiko Akasaka and Koichi Tateishi
( 1 1 ) ANTEPENULT 9
Place an accent on the antepenultimate mora.
3. "Heaviness" in morphology
(15) s p a : ΜΜΟ^ΗΜΟ^Ι]
tea:
Minorai] [Moral]]
CUe: [kj[MoraU][MoraU]]
The derived words such as hypocoristics are minimally bimoraic and the
minimal size of content words is also bimoraic. Based on this and other
facts from other languages, McCarthy and Prince (1986) conclude that the
minimal size of the word universally is a bimoraic foot.
Like English, the size of hypocoristic forms (if we ignore suffixes) is all bi-
moraic. This is the second argument for the minimal complexity require-
ment (17).
a. Bimoraic Patterns
sutoraiki > suto 'strike'
demonsutoreeshon > demo 'demonstration'
negachibu > nega 'negative'
hisuterii > hisu 'hysteria'
modan booi > mobo 'modern boy'
etc.
b. 4 mora patterns
paasonaru konpyuutaa > paso-kon 'personal computer'
purinto kurabu > puri-kura 'Print Club (a kind of instant camera
booth popular among youths)'
Hirakata Paaku > Hira-Paa 'Hirakata Park (the name of an
amusement facility in Osaka)'
dezitaru kamera > dezi-kame 'digital camera'
poritikaru saiensu > pori-sai 'political science'
Kimura Takuya > Kimu-Taku (the name of a popular singer/actor)
Heaviness in Interfaces 13
Note that the bimoraic patterns like nega above all branch on the root
node, while ill-formed examples like *kon for konbineeshon do not.
That is, when the words are analyzed into syllables, we see bisyllabic
branching in nega and other bimoraic cases and no branching in kon and
other ill-formed bimoraic cases. In compounded 4 mora cases, we see
branching into words on the root node.
(22) [Word [Word [Sy|p [ Mora a] ] [Sy,s [ Mora o] ] ] [ Word [ Syl k [MoraO] [ M o r a n ] ] ] ]
Also, the well-formed trimoraic forms like konbi show branching into syl-
lables on the root node.
(25) [ N bláckbird]
[ N tówel rack]
[ N Énglish teacher] 'English teacher (not a math teacher)'
If we call the domain on which the main stress of the compound falls Ac-
centual Phrase (Acc) 16 , we see the following schematic patterns.17
(28) [ n A B ] : [ A c c A ' B ]
U N / A C C A B ] C]:[AccA'BC]
The compound stress rule assigns the main stress on the leftmost word in
the Accentual Phrase, and, if two or more Accentual Phrases exist in one
compound word, the rightmost Accentual Phrase bears the main stress,
Heaviness in Interfaces 15
following the Nuclear Stress Rule (Chomsky and Halle 1968) or some
regularities which correspond to it, which assigns the main stress on the
rightmost element. To establish this Accentual phrasing, we need the fol-
lowing constraint:
However, the pattern does not apply to more complex compound nouns.
With the left-branching structure, the above compound pattern applies.
(33) a. [ NP [ N .[ N dogs]]]
b. dogs
That is, dogs is dogs, and we do not need any unnecessary projection of it
unless dogs concatenates with something else and forms a new phrase.
Thus, we can say:
Of course, nouns like dogs concatenate with some other words, like a
transitive verb likes. Likes also does not project into unnecessary projec-
tions, so the concatenation of likes and dogs (called Merge in recent
terms) derives the following structure:
d
(35) [likeslikeS °gs]
(40) [TenseTense [ v ... V ... DP-Object]] -> [ Tense DP-Objcct, LTenseTenSe [v ...
V-tJ]]
(44) Self-Embedding
A non-terminal node of category X constructs a self-embedding
structure iff
[χΑΥΒ]
where A and Β are not empty and Y either is or contains some ele-
ment of category X.
Note that the self-embedding effect is not caused by the processing bur-
den of the stack of DPs with the same case-marker. If this is the case, the
so-called multiple-subject construction must all be unparsable. This sim-
ply is not the case.
Also note that the self-embedding effect is not caused by structural ambi-
guity. For example, the following case of multiple across-the-board 0-as-
signment causes no processing disorder.
Even though (47) can have two possible 0-marking patterns, we can inter-
pret the sentence with the two patterns with no difficulty.
As the self-embedding constructions are all formed by legitimate syn-
tactic structure-building, the only option left to us is only to explore the
possibility of phonological explanation. In this respect, note that the In-
tonational Phrase (i.e. the domain where the intonation pitch contours
are reset to its higher position) building in Japanese always places the
boundary at the clause initial position.
[utt[mtp[phpHanako-ga]][IntP[PhPJiro-ga]][IntP[PhPTaro-ga]][IntP[PhP.
Michiyo-ga][ PhP byooki-da-to it-ta-to hookoku-shi-ta-to omot-ta.]]]
(where Utt is Utterance)
In (49), we can immediately note that there are three Intonational Phras-
es with a single Phonological Phrase in a row at the utterance-initial posi-
tion. In terms of intonational rhythm, this is very unstable, as prosodie
categories as we have discussed above prefer binary rhythm patterns. On
the other hand, (43b) with Scrambling of the embedded IPs has binary in-
tonational phrases if we assume that Scrambling in this sentence is IP-ad-
junction.
Heaviness in Interfaces 23
Note that in (51b), there are more arguments whose roles are not yet de-
termined in the middle of the sentence processing than in (43a) or (46),
because, until the word byooki-da-to appears, the exact semantic role of
the five phrases preceding it cannot be determined. This shows that the
self-embedding effect cannot be a matter of sentence processing.
We would say that the following constraint takes control over the
Scrambling in the self-embedding cases at least.
Thus, we would like to say that at least some cases of Scrambling are pro-
sodically-determined. We will come back to the theoretical implications
of this in Section 5.
The syntactic tree itself does not include any ordering information.
For Kayne, the very basic word order for the world's language thus is
SVO, and the SOV order as seen in Japanese is derived by application
of Move a.
Fukui and Takano (1998), on the other hand, agree with Kayne in as-
suming orderless trees but differ in their mechanism of word-order deter-
mination. Putting aside trivial details, they claim that the word-order de-
termination is done with a universal syntax-phonology mapping rule
which says, "The projection of the head comes later." As the head of a
phrase always come at the end in their mechanism, the default word order
for them is SOV, and an English-style SVO order is "derived."
For both Kayne (1994) and Fukui and Takano (1998), the problem
which Heavy NP Shift poses is the same: Heavy NP Shift appears to be
a rightward movement. As we all know, Heavy NP Shift places a
"Heavy" DP in the final position of the sentence. If we assume that the
position of the Heavy DP is a VP-adjoined position, we have the follow-
ing structure: 27
(56)
VP
VP "Heavy" DP
rest of the VP. In either theory, the Heavy DP cannot be placed on the
right, which it must be.
The solutions to this problem for them look similar. Let us first intro-
duce Kayne's account of Heavy NP Shift. Kayne (1994) proposes that
Heavy NP Shift is a kind of Scrambling, and claims that Heavy NP Shift
involves leftward movement, not rightward movement, following his gen-
eralization about asymmetric c-command. Thus, an example like (57):
(59) John gave [[to Bill]i [Xo [[all his old linguistic books] [Y° [e]¡...
Although Fukui and Takano (1998) deal with Heavy NP Shift only in
passing, they say that Heavy NP Shift is a process in the phonological
component, as Chomsky (1995) suggests. Although we cannot say any-
thing conclusive about what they think of when they say that Heavy NP
Shift is phonological, we can at least say that their spirit is similar to
Kayne's in that they think that Heavy NP Shift is optional and is not trig-
gered by anything syntactic.
The two theories of word order determination have one claim in com-
mon. Heavy NP Shift does not have any syntactic trigger. They do not
specify what the trigger of Heavy NP Shift is, but, from our point of view,
syntactic processes without syntactic trigger are likely to have a phono-
logical trigger. With Heavy NP Shift, we may expect that the notion of
heaviness may be phonological in nature. Thus, in the next section, we will
Heaviness in Interfaces 11
review the theories of triggers of Heavy NP Shift and conclude that the
trigger of Heavy NP Shift is in fact phonological.
(61) To whom did you give the book your mother bought you for your
birthday?
- a. I gave the book my mother bought me for my birthday to my son.
b. *?I gave to my son the book my mother bought me for my birth-
day.
Then, what is "heaviness" for Heavy NP Shift? Ross (1986) has already
had an answer to this problem. Ross defines "complex" NP (the term for
Heavy NP by Ross) as follows:
(64) may well be paraphrased as follows, because the only case of a noun
phrase dominating S in English is an adnominal relative clause construc-
tion which thus necessarily is right-branching:
By (65), examples like (66) nicely fit in, as the shifted DP is right-branch-
ing.
However, the definition in (65) does not necessarily fit all Heavy NP Shift
constructions. Consider the following example with Heavy NP Shift of a
left-branching DP:
Moreover, there are two kinds of examples which show that the notion
"branching" here is indeed phonological. First, even though it is branch-
ing in a strict sense, examples like (69) generally do not undergo Heavy
NP Shift.
This is perhaps due to the fact that function words such as pronouns are
phonologically weak and they do not constitute a prosodie category of
their own. Second, even though it is not branching syntactically, exces-
sively long words can undergo Heavy NP shift.
In (71), even though the two words are branching at the word level, they
are not at the phrasal level like all the examples above. This shows that
something other than syntactic branching is at work.
As Heavy NP Shift appears to be purely optional as the studies above
say, we cannot immediately make it follow from some phonological con-
straint. However, we can at least say that Heavy NP Shift moves some-
thing "heavy" phonologically or stays in situ (if other elements indeed
move, as Takano (1998) argues). We basically follow Zee and Inkelas
(1990) and propose that "Heaviness" can be defined as branching at the
Phonological Phrase level.29 In the relative clause (complex NP) case, the
clause always makes an independent Phonological Phrase (and an inde-
pendent Intonational Phrase, as we have argued for above), so that the
complex NP always branches into two phonological phrases at least, the
head noun and the relative clause. With (66) and (67), if we assume with
Selkirk (1986) that, in English, the boundary of a Phonological Phrase is
marked at the right boundary of X max , we can have the following prosodie
structure with multiple "heavy" stacks of Phonological Phrases.
->• [phphis][phpmother's][phpfriend's][phpfather] or
[ PhP his mother's] [ PhP friend's] [ PhP father]
Examples like my fiancée do not usually undergo the shift because pro-
nouns are phonologically weak. The excessively long words can undergo
the shift because they are too long to pronounce in a single phonological
phrase. For example, pneumonoultramicroscopicsilicovolcanoconiosis
would be pronounced with several Phonological Phrases as in what fol-
lows:
All these data show that, even though the Structural Description of
Heavy NP Shift involves phonological notions, Heavy NP Shift is sensi-
tive to syntactic distance, such as subject/object asymmetry and syntactic
locality. D o we have to assume such a mixture of levels in a single move-
ment process?
We believe that these facts can be explained once we think seriously of
how the rightward shift can be established in the aforementioned order-
less tree theories like Kayne (1994) and Fukui and Takano (1998). In
Kayne's and Fukui and Takano's theories, the word order information is
mapped based on the hierarchical information in tree structures. Due to
the nature of their mapping procedures, any moved element must move
to the left. In Kayne's theory, adjunction of an element X to, say, an IP
creates the following structure:
(81) It [ vP [the woman who saved a kid's life hit by a car] ν [ VP impressed
John] very much].
The verb impress raises to the projection of v, the light verb, and adjoins
to it.33
Now that rightward movement is not available, the only choice for us to
have (76a) would be to move the PP about Jonnie's problems between to
and all of the teachers. Such an operation is not allowed in syntax, because
it is a lowering and because it is an adjunction to an intermediate projec-
tion of a preposition. About the indirect object case (76b), we already
have an explanation. Suppose we follow Takano, which we do, and as-
sume that the natural (hierarchical) order of elements in a ditransitive
construction is Goal-Theme, as we do in the DP-PP construction derived
from the PP-DP construction. If so, we have, for example, John gave [the
man in the garden] a book about roses as the base. Now, this construction
have already all its Case checking relations satisfied as is (of course, by
later NP-movement and other checking mechanisms) and is in fact li-
censed as a grammatical sentence in English. With the Goal-Theme or-
der, there is no need for the shift of either object. As there is no option of
rightward movement in grammar, this is the only ordering relation we
have for this sentence and there is no Heavy NP Shift. (77) can also be ex-
plained if we assume that Goal comes first hierarchically. If Goal comes
first among internal arguments, we only have I told [a man who had a kind
face] (Goal) that we were in trouble. Again, without rightward movement,
we cannot have Heavy NP Shift. Heavy NP Shift (or, more precisely,
Heavy NP Non-Shift) is triggered by phonological branching, and its ap-
parent purely syntactic behavior is a by-product of the facts about English
sentence structures.
About Scrambling, we have a very interesting alternative. Boskovic
and Takahashi (1998) propose that Scrambling is not a matter of move-
ment. Rather, they say that the Scrambling construction is base-generat-
ed. According to them, later operations "lower" Scrambled elements to
its 0-marked positions.
(85)[NAB]:[ACCA'B]
UN/ACCA B] C] : [ Acc A' Β C]
[ n A [ n / A c c B C]]: [ Acc A][ Acc B· C]
(86)
dog
dog
(87)
statistics
statistics
In (88b), for the verb make to come in the final position is highly predict-
able by the use of the word point, and thus the Nuclear Stress Rule choos-
es point as the target of the main stress.
If the above arguments are all correct, the organization of grammar will
be as follows:
(89)
Semantics
^ / ^ ^ I n t e r p r e t Alpha
Lexicon, Interpretable
Numeration Tree
Merge, Move and \^Speel-Out?
Prosodie Category
Formation (Stress?)
Phonetic
Representation
(90) {α,{α,β}}
This reads as "a merger of a and β with the head a," that is, a's projec-
tion.36 The righthand member of the set expresses the syntactic and se-
mantic relations of elements, and the lefthand member of the set express-
es the headedness of the node. However, we immediately find that the
lefthand member tells us a redundant information, because, once we get
to know what a and β are, we immediately know which of the two is likely
to be the head of the phrase. This set-theoretic notation is also mysterious
in that, even though Chomsky considers syntax as communicating with
Semantics and Phonology, syntactic objects in Chomsky's tree do not con-
tain any phonological information although it does contain semantic in-
formation in some form. We conjecture that it is the lefthand member of
the node set which should contain phonological information like prosodie
categorization. The Interpretable Tree for us is a tree which contains both
relational (semantic/ syntactic) and prosodie information bits.37
In the minimalist program, derivations are intended to make a struc-
ture both semantically and phonologically interpretable. In the current
syntactic discussions, we see arguments about how we can make a struc-
ture semantically interpretable, such as issues on quantifier scope, Wh-
movement and reconstruction and 0-marking, and operations for that
purpose are syntactic derivations. In this sense, syntax and semantics in
the current syntactic theory are parallel. In the discussions with these in
mind, one basic assumption has been considered something inviolable:
Inclusiveness. Inclusiveness rules all syntactic derivations and bans some-
thing other than information from lexical items inserted in a tree. This
perhaps follows the tradition of compositional formal semantics. Howev-
er, for a tree to become phonologically interpretable, it is unavoidable to
insert something not in the lexicon. For example, it is simply impossible
to have a complete stress grid structure up to the Utterance level in the
lexicon. Phonology is just not made this way. However, as we have seen
above, rules like the Nuclear Stress Rule, which adds a stress grid, must
be applicable in syntax. We would say that the Inclusiveness theme must
be loosened to allow for pure insertion of grids and prosodie categories
for the phonological purpose. Unless we do so, we cannot have syntactic
derivations which are both semantically and phonologically interpret-
38 Yukiko Akasaka and Koichi Tateishi
Note that both anaphors and pronominals, except for reciprocals which
necessarily involve some quantificational operation, are variables in se-
mantics, at least truth-functionally.
(93) a. Because Spell-Out now splits off all phonological materials out of
a tree, all that is left after Spell-Out should be phonetic implemen-
tations. This in a sense is well fit into the framework of grammar like
Chomsky's (1995), because there is only a real interface with mo-
tors and perceptions after a "fork" of organization of grammar. So,
we need to re-examine and see whether anything "phonological" is
left after Spell-Out.
Notes
* The order of the authors is alphabetical and the two authors are equally responsible for
the content of this paper. The authors thank the editors of this book for the chance to
contribute this paper to the present volume. We have benefited much from comments
from an anonymous reviewer of this volume, Francis Michinao Matsui, and Elisabeth
Selkirk. We would also like to thank our colleagues at Osaka University and Kyoto Uni-
versity of Foreign Studies for their encouragement and comments. All mistakes and er-
rors which are left in this paper are, of course, solely our own.
1. However, since SPE's (Chomsky and Halle 1968) unfortunate neglect of the notion of
syllable and the use of [± tense] for length distinctions, the weight distinction in phonol-
ogy had also been neglected until the (re)birth of the metrical theory as presented by
Liberman and Prince (1977).
2. This is an abbreviated tree. There are nodes like mora under Syl(lable), for example.
3. Of course, there are many exceptions to the regularities that we introduce here, as is
mentioned in studies like SPE and others. In this paper, we will ignore all these excep-
tional cases and only discuss the core cases.
4. We will ignore the so-called "Ambisyllabicity" (Kahn 1976). However, the existence of
ambisyllabicity itself may show that the sensitivity to branching is real in a sense.
5. Yamada (1990) argues that Yamato nouns basically have an antepenultimate accent
pattern and that Japanese accent is predictable in all vocabulary classes. Yamada bases
his argument on the fact that this pattern is dominantly attested in derived words such
as compounds and newly coined words like foreign borrowings. However, we think that
these facts do not save us from specifying the place of accents in all those words with
unpredictable accent patterns.
6. There are some identifiable groups of exceptions to this regularity. Most typical of these
are those words whose accent patterns are taken from accent patterns of original lan-
guages, e.g. Chómusukii 'Chomsky,' ákusento 'accent.'
7. Japanese syllable structure is basically (C)V(V)(C), where the coda C is either a nasal
or the first half of a geminate. All segments except onsets are counted as moras. There
are some exceptional trimoraic syllables (Tateishi 1997,1998b).
8. Some speakers shift the accent to the right, due to devoicing of the vowel [u], which is
not substantially relevant to the point here.
9. Probably, this derives from the final two moras constituting a foot and from the foot in
question being extrametrical (Tateishi 1985,1992), together with an End Rule of some
kind.
10. In (12), *BR(SEG) designates a heavy syllable that is accented. We need this kind of des-
ignation to capture the fact that accent never falls on the second mora of a heavy sylla-
ble in other vocabulary classes either. We avoid formalizing this constraint as
ALIGN(ACCENT, LEFT, SYL), because this formulation is partly vacuous and redundant
as accent necessarily falls on the left (and right) edge of a light syllable anyway. Note
that the constraint in question is not undominated in an Optimality term as an anony-
mous reviewer claims, because we occasionally find non-head accentuation even in the
Tokyo dialect of Japanese (Tateishi 1997).
11. In McCarthy and Prince's theory of prosody, onsets are outside of the moraic structure.
We follow their way of representing moras.
12. As onsets and extra codas are adjuncts to syllables in McCarthy and Prince (1986), we
can say that these are not counted as branches. We avoid using the term foot explicitly,
because the status of the notion foot in morphology is not as clear as that in the stress
Heaviness in Interfaces 41
theory, and because using the term foot may make it impossible to distinguish between
a non-branching degenerate foot and a binary foot. What we need here is a binary
branching foot.
Function words are naturally exceptions to this generalization, as they often behave
as a syntactic affix.
13. Examples like sinkuro 'synchronization (sinkuronaizeeshon), synchronized swimming
(sinkuronaizudosuimingu)' perhaps fit into the bimoraic compounding pattern.
14. Perhaps, in the case of nicknames, the word must be only minimally complex because
hypocoristics generally are used with vocative suffixes like -san, -chan, -ko and so on.
Combined with these suffixes, the word becomes truly complex. On the other hand,
without the help from such suffixes, contraction forms must be truly complex/heavy.
Cases like retii < remontii 'lemon tea' are probably formed following such complexity
requirement.
15. We only discuss the location of the main stress.
16. We call Accentual Phrase what McCawley (1968) calls minor phrase. It comes between
Phonological Word and Phonological Phrase and bears at most one accent.
17. Compounds are of course not only formed out of nouns. Non-nominal compounds are
often subject to idiosyncracies.
18. "Length" of words in Japanese for this purpose is measured by the mora count. Roughly
speaking, if a word has three or more moras, it is long (McCawley 1968).
19. Incidentally, this Accentual Phrasing works as the domain of another Japanese phono-
logical rule called rendaku (Otsu 1980, Ito and Mester 1986, Fukazawa and Kitahara,
this volume) and the rule of rendaku also has a right-branching effect.
20. As Elisabeth Selkirk (personal communication) has correctly pointed out, in Japanese,
a right-branching Adjective-Adjective-Noun sequence, e.g. [ao_i [uma_i [ame]]] 'blue
tasty candy', is prosodically parsed into [ PPh ao_i uma_i ame]. This means that adjectives
are regarded as non-maximal in syntax-phonology mapping, if we follow Selkirk and
Tateishi (1991) and assume that it is the left edge of a syntactic maximal projection that
coincides with a boundary of a Phonological Phrase. More elaborated investigations on
the syntax of attributive adjectives are in order.
21. For example, in a language with bisyllabic stress alternation, of course a word can have
an odd number of syllables. In such a case, Universal Grammar should give us an option
of having a degenerate monosyllabic foot and of not parsing a left-over syllable into a
foot.
22. Kuno (1973) actually uses relative clause constructions. The point of self-embeddedness
is kept intact, however.
23. There are cases where (48) is pronounced with one Intonational Phrase: [ IntP Taro-ga
Hanako-ga tensai-da-to it-ta], with no resetting of the pitch contour and continuous
lowering of pitch. In such cases, our intuitive interpretation is either that Taro-ga is fo-
calized or that Hanako-ga tensai-da-to it-ta is defocalized. We only discuss the default
out-of-the-blue cases in this paper.
24. Perhaps, this constraint must be interpreted as a relative one. Other factors like focus
and topic may make a unary-branching Intonational Phrase sound better.
25. The terms NP and D P are here used interchangeably.
26. A asymmetrically c-commands Β if and only if A c-commands Β but not vice versa.
27. The point still goes through even if Heavy NP Shift is IP-adjunction.
28. Note that this sentence may be fine only if we put emphatic stress on books. Stress
marks the boundary of some kind of prosodie category, and this strengthens our posi-
tion that Heavy NP Shift is triggered by something phonological.
42 Yukiko Akasaka and Koichi Tateishi
29. See also Verhijde and Van de Weijer (1992) for a similar conclusion independently
reached.
30. Why is a heavy NP "shifted"? We conjecture that this would be due to the fact that pro-
sodie heaviness (non-unarity) obligatorily creates binary rhythm. With two words we
can compose a hierarchical structure, but we cannot with only one.
A
b. Binary Phrase
*
* *
[A B]
With something binary and with a rhythmic head at the end, we can easily adjust the
rhythm of the whole utterance. Thanks go to an anonymous reviewer for this point.
We need an account of truly focal "light" NP Shift.
(ii) She bequeathed to the library her PAPERS and to the radio station her CD's.
(ill) *
*
[PAPERS]
Thus, heaviness here corresponds to some prosodie hierarchy which is either forced by
binarity or focus.
31. There is another interesting syntactic fact pertaining to Heavy NP Shift. A D P can shift
over the object-oriented secondary predicate, while it cannot shift over the subject-ori-
ented secondary predicate.
(i) a. Jews never eat t raw [fish over two days old].
b. *John left t angry [the reception for the ambassador from Ulan Bator],
33. Here we simply follow Fukui and Takano's (1998) assumption that the syntactic trees
must keep "expanding" throughout the course of derivations, which they attribute to
Chomsky (1995). Thus, there is no head movement per se and the so-called head move-
ment is also a case of phrasal adjunction.
34. The reason of ungrammaticality of (82) perhaps is attributed to an independent matter
of Case-checking of the subject. Even if we shift the object and the adverb later by some
sort of Scrambling, they should come to the left of the verb and we would have * It John,
very much, impressed the woman who saved a kid's life hit by a car.
35. Perhaps, we are leaving the minimalist spirit of strict modularity among grammatical
components. According to our view of grammar, syntactic trees can contains anything
phonological and semantic, including something like phonemes. Being invisible and be-
ing irrelevant are separate matters, we believe.
36. Adjoined nodes may have a different form, which is not substantially relevant to the dis-
cussion in this paper.
37. See Akasaka (2000) and Akasaka and Tateishi (2000) for details.
References
Akasaka, Yukiko
1996 Binding as the Realization of Overt Variables. Master's thesis, Kyoto Univer-
sity of Foreign Studies.
2000 Extended Prosodie Syntax: Case Studies in English. Master's thesis, Osaka
University.
Akasaka, Yukiko and Koichi Tateishi
2000 "Metrical Branching and Phonology-Syntax Interface," Phonological Studies
3, 61-42
Bolinger, Dwight
1972 "Accent is Predictable (If You are a Mind Reader)." Language 48, 633-644.
Boskovic, Zeljko and Daiko Takahashi
1998 "Scrambling and Last Resort." Linguistic Inquiry 29, 347-366.
Chomsky, Noam
1972 "Remarks on Nominalization." Studies on Semantics in Generative Grammar,
7-61, Mouton, The Hague.
1986 Barriers. The MIT Press, Cambridge, MA.
1995 The Minimalist Program. The MIT Press, Cambridge, MA.
Chomsky, Noam and Morris Halle
1968 The Sound Pattern of English. Harper and Row, New York.
Fiengo, Robert
1974 Semantic Conditions on Surface Structure. Doctoral dissertation, MIT.
Fukui, Naoki
1993 "Parameters and Optionality." Linguistic Inquiry 24, 399-420.
Fukui, Naoki and Yuji Takano
1998 "Symmetry in Syntax: Merge and Demerge." Journal of East Asian Linguistics
7,27-86.
Hale, Kenneth
1980 "Remarks on Japanese Phrase Structure: Comments on the Papers on Japa-
nese Syntax." MIT Working Papers in Linguistics 2,185-203.
44 Yukiko Akasaka and Koichi Tateishi
Haraguchi, Shosuke
1991 A Theory of Stress and Accent. Foris, Dordrecht.
Hayes, Bruce
1980 A Metrical Theory of Stress Rules. Doctoral dissertation, MIT.
1982 "Extrametricality and English Stress." Linguistic Inquiry 13, 227-76.
1995 Metrical Stress Theory; Principles and Case Studies. The University of Chicago
Press, Chicago.
Itô, Junko
1990 "Prosodie Minimality in Japanese." CLS 26,213-239.
Itô, Junko and R. Armin Mester
1986 "The Phonology of Voicing in Japanese: Theoretical Consequences of Mor-
phological Accessibility." Linguistic Inquiry,17, 49-73.
1995 "Japanese Phonology." The Handbook of Phonological Theory, ed. by John
A. Goldsmith, 817-838, Blackwell, Oxford.
Jackendoff, Ray
1977 X' Syntax: A Study of Phrase Structure. The MIT Press, Cambridge, MA.
Kahn, Daniel
1976 Syllable-Based Generalizations in English Phonology. Doctoral dissertation,
MIT.
Kayne, Richard S.
1994 The Antisymmetry of Syntax. The MIT Press, Cambridge, MA.
Kubozono, Haruo
1988 The Organization of Japanese Prosody. Kurosio Publishers, Tokyo.
Kuno, Susumu
1973 Nihon Bunpo Kenkyu [Studies in Japanese Grammar], Taishukan, Tokyo.
Liberman, Mark and Alan Prince
1977 "On Stress and Linguistic Rhythm." Linguistic Inquiry 8, 249-336.
McCarthy, John and Alan Prince
1986 "Prosodie Morphology." ms., University of Massachusetts, Amherst and
Brandeis University, Waltham.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. Mouton, The
Hague.
Miyagawa, Shigeru
1989 Syntax and Semantics 22: Structure and Case Marking in Japanese. Academic
Press, San Diego.
1997 "Against Optional Scrambling." Linguistic Inquiry 28,1-25.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems." MIT Work-
ing Papers in Linguistics 2, 207-227.
Poser, William J.
1990 "Evidence for Foot Structure in Japanese." Language 66, 78-105.
Prince, Alan and Paul Smolensky
1993 Optimality Theory: Constraint Interaction in Generative Grammar. Technical
Report 2, The Rutgers Center for Cognitive Science, New Brunswick.
Rochemont, Michael S. and Peter W. Culicover
1990 English Focus Constructions and the Theory of Grammar. Cambridge Univer-
sity Press, Cambridge.
Ross, John Robert
1986 Infinite Syntax! Ablex Publishing Corporation, Norwood, NJ.
Heaviness in Interfaces 45
Saito, Mamoru
1985 Some Asymmetries in Japanese and Their Theoretical Implications. Doctoral
dissertation, MIT.
1992 "Long Distance Scrambling in Japanese." Journal of East Asian Linguistics 1,
69-118.
Saito, Mamoru and Naoki Fukui
1998 "Order in Phrase Structure and Movement." Linguistic Inquiry 29, 439-474.
Saito, Mamoru and Hajime Hoji
1983 "Weak Crossover and Move-α in Japanese." Natural Language and Linguistic
Theory 1, 245-259.
Selkirk, Elisabeth
1984 Phonology and Syntax: The Relation between Sound and Structure. The MIT
Press, Cambridge, MA.
1986 "On Derived Domains in Sentence Phonology." Phonology Yearbook 3, 371-
405.
Selkirk, Elisabeth and Koichi Tateishi
1988 "Constrains on Minor Phrase Formation in Japanese." CLS 24, 316-336.
1991 "Syntax and Downstep in Japanese." Interdisciplinary Approaches to Lan-
guage: Essays in Honor ofS.-Y. Kuroda, ed. by Carol Georgopoulos and Rob-
erta Ishihara, 519-543, Kluwer Academic Publishers, Dordrecht.
Takano, Yuji
1998 "Object Shift and Scrambling." Natural Language and Linguistic Theory 16,
817-889.
Tateishi, Koichi
1985 Accent of Compound Nouns in the Standard Japanese: A Nonliner Approach.
Master's thesis, International Christian University.
1989 "Theoretical Implications of the Japanese Musician's Language." Proceedings
ofWCCFL 8, 384-398.
1992 "Maximal Non-Exhaustive Invisibility." Kansai Linguistic Society 12,134-144.
1997 "Nihon-go Tokyo-Hogen Washa-ni Okeru Akusento-no Yure-to Onsetsu-
Kozo [Variability of Judgement on Accentuation among Tokyo Japanese
Speakers and the Syllable Structure]." a paper presented at the 115th meeting
of the Linguistic Society of Japan, Kyoto University.
1998a "Foot or Syllable?: Binarity Effects in Japanese and English Phonology." ICU
English Studies Vol. 7, Society of ICU English Studies, 1-22.
1998b "Non-Repair Strategies in Japanese Syllables." Phonological Studies 1, 123-
130.
1999 "Are Interface Studies Possible?" Journal of the Phonetic Society of Japan 3,
4-19.
Verhijde, Mark and Jeroen van de Weijer
1992 "On Deriving Minimal NPs." Proceedings of the Leiden Conference for Junior
Linguists 3, ed. by L.C.J. Barbiers, M. den Dikken and C.C. Levelt, 293-306.
Leiden University Press, Leiden.
Webelhuth, Gert
1989 Syntactic Saturation Phenomena and the Modern Germanic Languages. Doc-
toral dissertation, University of Massachusetts, Amherst.
Yamada, Eiji
1990 "Stress Assignment in Tokyo Japanese." ms., Fukuoka University, Fukuoka.
46 Yukiko Akasaka and Koichi Tateishi
Introduction
As noted above, the existence of this type of dialect would be rather sur-
prising. This is because if lexical items have η moras (or syllables), we
would expect that the largest number of their accentual pattern be n+1,
taking the unaccented lexical items into account. This n+1 pattern is ob-
served in (3).
(4)
No. of Gloss In isolation ~ + s a ~+do ~+no ~-i-gara
Mora 'Acc' 'and' 'Gen' 'from'
2a. nose hana hanasa hanado hanano hanagara
L L L LL L L L L L L LLLL
2b. leg 'ásí 'ásisá 'ásidó 'àsino 'ásigára
LH LLH LLH LL L LLHL
2c. shoulder kadá kadása kadádo kadáno kadágara
LH LH L LH L LH L LH LL
2d. autumn 'àgi 'ágisa 'ágido 'ágino 'ágigara
HL H LL HLL HLL HLLL
3a. fish sagana saganasa saganado saganano saganagara
LL L L L LL L L L L L L L L L L L LL
3b. head 'ádamá 'ádamasá 'ádamadó 'ádamano 'ádamagára
L L H L L LH L L L H L L L L L L LHL
3c. heart kogoró kogorósa kogoródo kogoróno kogorógara
L LH L LH L L LH L L LH L L LH L L
3d. rabbit 'úsági 'úságisa 'úságido 'úságino 'úságigara
LH L LH LL LH L L LH L L LHLLL
3e. helmet káBudo káBudosa káBudodo káBudono káBudogara
H L L H L L L, H L L L H L L L H LLLL
50 Shosuke Haraguchi
All the L-toned nouns of the 2a and 3a types are analyzed as unaccented
by all linguists, while the other nouns with an Η-tone are analyzed as ac-
cented, with the accent on the Η-toned mora. When the accusative
marker -sa, the Genitive marker -no, -do 'and' corresponding to -to in
Tokyo Japanese, and -gara 'from' corresponding to -kara of Tokyo Japa-
nese, are attached to the nouns, the unaccented nouns remain L-toned.
However, in the accented type 2b and 3b nouns, only L-tone appear
when the Genitive marker -no is attached to a noun, which suggests that
the word-final accent is deleted when it is adjacent to the Genitive mark-
er. Also, before the other particles, the Η-tone appears one-mora to the
right, which is interpreted by most Japanese linguists as a shift of accent
to the right. In other cases, the accent seems to surface in its underlying
position.
Haraguchi (1979), paying attention to the fact that accent shifts to the
right only when the vowel of the post-accented mora is [e, o, or a], while
the shift is blocked when the vowel is [i or u], proposes that the seemingly
peculiar properties of Tsuruoka Japanese can be accounted for if we as-
sume the following accent shift rule contingent on the height of the post
accented vowel.
Assuming this accent shift rule, we can analyze the nouns of the 2c and 3c
types as having an underlying penultimate accent, which is moved one
mora to the right after the application of the accent deletion rule before
no. Haraguchi (1979) also proposes that, assuming this accent shift rule,
this accentual system is nothing but a variant of Tokyo-type pattern: the
n+1 accentual pattern.
Let us now examine four-mora nouns in (6).
The Accent of Tsuruoka Japanese Reconsidered 51
(6)
No. of Gloss In isolation ~+sa ~+do +no ~+gara
Mora 'Acc' 'and' 'Gen' 'from'
4a. chicken niwadori niwadorisa niwadorido niwadorino niwadorigara
LLLLLLLLLLLLLL LLLLL LLLLLL
4b. wild pig 'ínosisí 'ínosisisá "ínosisidó 'ínosisino 'ínosisigára
L L LH L L L L H L L L L H LLLLL LLLLHL
4c. sound of feet 'asi'odó 'ási'odósa 'ási'odódo 'ási'odóno 'ási'odógara
LLLH LLLHL LLLHL LLLHL LLLHLL
4d. onion tamanégi tamanégisa tamanégido tamanégino tamanégigara
L L H L L L H LL L L H LL LLHLL LLHLLL
4e. morning glory 'ásága'o 'ásága'osa 'ásága'odo 'ásága'ono 'ásága'ogara
LHLL LHLLL LHLLL LHLLL LHLLLL
4f. mushroom mázidage mázidagesa mázidagedo mázidageno mázidagegara
HLLL HLLLL HLLLL HLLLL H LLLLL
Parallel to what we have seen in (4), the L-toned nouns are unaccented
and the final-accented nouns of the 4b type lose accent before the Geni-
tive marker no. The nouns of the 4c type have an underlying accent on the
penultimate mora, shifting it to the right by the accent shift rule in (5).
Those of the 4d type have the same surface and underlying accent on the
penultimate mora, and those of the 4e and 4f types have an underlying ac-
cent on the initial mora, with the subsequent accent shift to the second
mora in the case of the 4e type.
Consider now the cases of one-mora nouns, which have just two types,
unaccented and accented.
(V)
No. of Gloss In isolation ~+sa ~+do ~+no ~+gara
Mora 'Acc' 'and' 'Gen' 'from'
la. mosquito ka kasa kado kano kagara
L LL LL LL LLL
lb. picture 'é 'esá 'édo 'eno 'egára
H LH HL LL LHL
terpreted as a shift of accent to the right. When the particle do 'and', cor-
responding to in Tokyo Japanese, is attached, the accent does not shift to
the right for some reason unclear to me at present. When the Genitive
marker no is attached, the word-final accent is deleted, while it is shifted
to the right in the case of the particle gara 'from', corresponding to kara
in Tokyo Japanese.
I assume that the accent shift rule in (5) applies to lexical items in the
lexicon and to what McCawley (1968) calls the 'minor phrase'. Recall that
this rule moves the accent to the right if the height of the following vowel
is non-high. Once we assume this rule, the data in (4, 6, and 7) can be an-
alyzed as follows:
Let us first discuss the tonal system of nouns in this dialect, and determine
what the best way is to analyze its melodic properties. The basic tonal
facts of nouns are illustrated in (4, 6,7) in section 2.
I will begin with a discussion of the accentual system of this dialect. A
careful examination of the above facts shows that accent seems to be
shifted to the right when the following mora contains a non-high vowel [e,
o, or a]. One case which does not conform to this accent shift is 'é 'picture'
in (lb) followed by do 'and'. For some unexplained reason the phrase
'édo does not undergo accent shift. Thus, if this is a fact, we must regard
this phrase as an exception to the rule.
The all-L tone in the no-phrases of the lb, 2b, 3b, and 4b types compels
us to set up an accent deletion rule, which deletes accent immediately in
front of the Genitive marker no:
Notice that unless we assume that this rule applies before Accent Shift in
(5), we would not account for the difference between the 2b, 3b, and 4b
types and the 2c, 3c, and 4c types. This shows that we need derivational
information to distinguish the b classes and the c classes.
54 Shosuke Haraguchi
An examination of the list immediately shows that only one H tone ap-
pears in a word or a phrase and everything else is L. Let us first discuss
how this fact is analyzed within the framework of Haraguchi (1977) and
(1979). The basic tone melody of this dialect is LHL. Assuming that the
H tone is associated with the accented mora first, as in numerous other
Japanese dialects, all other moras are associated with the L tone of the
LHL melody. Notice also that the absence of a contour tone (i.e., HL fall-
ing tone or LH rising tone) indicates that this dialect prohibits association
of two or more tones to a single tone-bearing unit, which leaves the L tone
unassociated with the peripheral mora of the initial-accented word or of
the final-accented word.
The remaining thing for us to do is to account for the all-L tone of un-
accented words. One possible solution is to assume that Tsuruoka Japa-
nese has no system to associate the basic tone melody LHL with unac-
cented words, which leaves the unaccented words toneless. The toneless
words later receive the default L tone by a default tone association con-
vention. The second possible solution is that, just like Tokyo Japanese, the
H tone of the LHL melody is associated with the final mora of the unac-
cented words, which brings about the following schematic melodic shape:
(10) . . . X X X X
\ l / I
L H L
The H tone associated with the unaccented mora is then deleted and the
floating L tone is associated with the toneless final mora. The third possi-
ble solution is to assume that the H tone of the LHL melody is associated
with the initial mora of a word. Thus, the derivation is the mirror image
of the second solution, as indicated in (11):
(11) X X X X
I I / /
L H L
Which solution is the best? It is rather difficult to argue for one over the
others. One might claim that the best guess would be to choose the last
one, taking the following observation by Nitta (1994: 85) into consider-
ation.
(12) Classes la, 2a, 3a, etc. are phonologically represented as having, LL,
LLL, LLLL, etc. but these all L melodies, as Kindaichi has indepen-
The Accent of Tsuruoka Japanese Reconsidered 55
To derive the melodies ML, MLL, MLLL, etc., what would be needed is
to simply assume that the initial H tone associated with the unaccented
mora is phonetically realized as the M tone. The M tone which appears
optionally on the second mora will be handled by assuming an optional
phonetic spreading of the M tone one mora to the right. Assuming that
this solution is correct, Tsuruoka Japanese would be just the opposite of
Tokyo Japanese in the association of the H tone with an unaccented word.
Admitting the initial plausibility of this analysis, I will elect to choose
the second solution here. This solution makes it necessary to introduce an
H Deletion rule, which erases the H tone on the unaccented mora, induc-
ing the subsequent spreading (or association) of the L tone. One of the
reasons that I choose this option is to make this dialect parallel to other
Tokyo-type Japanese in handling the association of the H tone with the
final mora of unaccented words. This analysis is supported by features of
verbs in this dialect, which are discussed in the next subsection.
Overall, we can say that tonal facts of Tsuruoka Japanese are more
complicated than those of Tokyo Japanese in that the former requires the
accent shift rule dependent on the height of vowels, the H Deletion rule
for nouns, and a couple of phonetic rules.
Tonal facts of Tsuruoka nouns discussed in section 1 can be summa-
rized as follows:
I will not go into the details of exceptions to the analysis presented above.
For further discussion of ramifications on marked irregularities and a
number of exceptions, see Haraguchi (1997) and (1998).
56 Shosuke Haraguchi
b. Accented Class
déru HL dérusage' dené' dené'sage' déNna dedéba
míru HL mírusage' miné' miné'sage' míNna midéba
kúru HL kúrusage' koné' koné'sage' kúNna kudéba
kú H kúsage' kuwané' kuwané'sage' kúna kéba
tóru HL tórusage' torané' torané'sage' tóNna todéba
kágu HL kágusage' kagané' kagané'sage' káguna kagéba
tadéru LHL tadérusage' tadené' tadené'sage' tadéNna tadedéba
toZíru LHL toZírusage' toZiné' toZiné'sage' toZíNna toZidéba
cugúru LHL cugúrusage' cugurané' cugurané'sage' cugúNna tuguréba
hará'u LHL hará'usage' harawané' harawané'sage' hará'una hara'éba
hé'ru LHL hé'rusage' he'rané' he'rané'sage' hé'Nna heréba
kaZo'éru LLHL kaZo'érusage' kaZo'ené' kaZo'ené'sage' kaZo'éNna kaZo'edéba
'ugogásu LLHL 'ugogásusage' 'ugogasané' 'ugogásauné'sage' 'ugogasuna 'ugogaséba
where e' stands for open [ε], Ζ for prenasalized [z], and Ν for moraic
nasal.)
Notice that the unaccented verbs have the Η tone on the final mora of the
Present. This means that the Η tone of the LHL basic tone melody is as-
sociated with the final mora of verbs in the autosegmental analysis. Recall
that for unaccented nouns we have chosen the second solution in which
The Accent of Tsuruoka Japanese Reconsidered 57
the H tone is associated with the final mora, with the subsequent erasure
of the H tone. Verbs are different from nouns in that this H deletion does
not apply to verbs. However, the association of the H tone with the final
mora of unaccented cases is common to both nouns and verbs.
Accented class is assumed to be assigned accent by the following rule,
which is identical to that of Tokyo Japanese:
This accent assignment applies before the accent shift rule in (5). The
stem-final or penultimate accent does not undergo accent shift because
the Present morpheme ru has a high vowel. The implication of this as-
sumption will be clear immediately below.
First consider the morpheme sage'. An examination of both the unac-
cented class and the accented one shows that this morpheme is accented.
Within the present framework assuming an accent shift rule, we must fur-
ther assume that this morpheme belongs to the so-called 'Pre-accenting'
class, which is represented schematically as follows:
(16) *
* *
sa ge'
This makes it possible for the accent shift rule to move accent to the right,
as illustrated in (16):
(17) * *
* * * * * *
Notice that this accent normally does not surface when the morpheme is
attached to accented verbs, with a few exceptions with which we are not
concerned here.
Turn next to the Negative ne\ which is derived from the underlying
form ai by Coalescence. This bound morpheme should be classified as un-
accented, because there is no indication of the existence of accent when
it is attached to unaccented verbs. The fact that it has accent when it is at-
tached to accented verb stems can be easily handled by the accent shift
rule (5) introduced independently for nouns.
Notice that the Negative Imperative na is also unaccented. What is in-
teresting is that when it is attached to the Present form of verbs, the ru is
58 Shosuke Haraguchi
turned into the geminate Ν immediately before η of the Negative ne'. This
segmental process seems to be common to most dialects of the Tohoku
district and the northern Kanto district (e.g., various Ibaragi dialects). Ir-
respective of the relative order of this segmental rule and the accent shift
rule, we can account for the reason why accent shift is blocked in cases
where the accent is followed by the geminate N. This will be clear from
the discussion below of the accent shift of the Past form.
Consider now the Conditional form (d)eba, which corresponds to
(r)eba of Standard or Common Japanese. Thus, when the Conditional is
attached to the consonant-ending (or consonantal) verbs, eba is selected
as illustrated in:
(20) (Cvo.ced)r V , V
IN J
Otherwise ta remains as it is. Just like other forms, the Past ta is unaccent-
ed by nature. Thus, when it is attached to unaccented verbs, the Past forms
have no accent at all. However, when it is attached to accented verbs, it
sometimes gets accent as illustrated in the examples of the column I:
lying high vowel, which exists when the rule in question applies, prevents
accent shift from applying to this case.
Notice that in the discussion above, I omitted a couple of seemingly re-
calcitrant cases, which will be briefly examined at this point:
Recall that the Past morpheme ta is invisible and accent is assigned to the
final syllable of the verbal stem. This means that accent is assigned to the
initial mora of these cases. The accent shift rule under consideration
should move the accent of (22a and b), while it predicts, contrary to fact,
that it should be blocked from applying to derive (22b).
The problem is how to account for the bocking of accent shift in (22b).
Since the relevant examples are scarce, I cannot say anything conclusive at
present. However, the best guess is that accent shift is permitted to apply
under certain conditions even if the following mora contains a high vowel.
I suspect that the voiced consonant has something to do with this. Howev-
er, I have to leave the problem of the blocking of accent shift open here.
The core cases discussed in this section clearly show the necessity of ac-
cent shift and of derivations.
The accented class receives accent on the penultimate mora (or equiva-
lently on the stem-final mora), which is common to almost all of the To-
kyo-type dialects.
The reader might wonder how to handle the fact that some adjectives
like siré in 2b, 'okki in 4a, and 'oisi in 4c appear to have final accent. This
fact will pose no problem if we assume that the underlying forms of these
adjectives are respectively as follows:
Assuming that accent is assigned to the underlying forms and then the fol-
lowing rules apply to derive the surface forms, we can easily handle these
properties of accentual behavior:
Notice that the word-final open e' is also the result of Coalescence, which
is universally found in a variety of dialects and languages, and subsequent
shortening in (25b):
Looking at the samples in (23), the astute reader might have noticed that
the adjective né' is classified as belonging to the accented class. However,
we have seen in the section above that the bound form ne' is unaccented.
How should we account for this apparent incompatibility?
A careful examination of facts concerning adjectives shows beyond
reasonable doubt that the adjective né' 'not' is accented. To see this, con-
sider the following cases:
62 Shosuke Haraguchi
As is clear from the comparison of (27) and (28), the Negative ne' be-
haves parallel to the other accented adjectives. Thus, I conclude that the
adjective should be classified as accented.
Returning to the bound form ne', I assume that this bound form is a
kind of compound adjective and suggest that the form should be inter-
preted in a parallel way that Halle and Vergnaud (1987) treated some cas-
es of compounds of Dakota and English. They propose that the noun man
in English is demoted to an affix in some cases. To see this, consider the
following examples cited from Halle and Vergnaud (1987: 91):
(29) a. [ae]: garbage man front man Kennedy man garage man
b. [a]: infantryman salesman outdoorman postman doorman
(30) The forms in [(29a)] are ordinary compounds with the falling stress
contour typical of such words in English. The forms in [(29b)], on the
other hand, are compounds whose second constituent is treated not
as a full word but as an affix. This is shown by the fact that man sur-
faces without stress.
- H a l l e and Vergnaud (1987: 92)
In line with their proposal, I interpret the bound adjective ne' as being de-
moted to an affix. As a result of this demotion, the accented adjective los-
es accent, and becomes an accentless affix. This resolves the apparent in-
compatibility under consideration.
The Accent of Tsuruoka Japanese Reconsidered 63
In the previous sections, I have assumed that Tsuruoka Japanese has the
accent shift rule in (5). We are now in a position to propose an alternative
analysis which does not incorporate accent shift. Instead of shifting ac-
cent, we can account for the tonal facts of Tsuruoka Japanese if we as-
sume a phonetic rule which moves the Η-tone one mora to the right, if the
mora contains non-high vowel [e, o, or a]:
Note
* This paper is in part supported by The Special Research Project of the Typological Inves-
tigation into Languages and Cultures of the East and West, The COE Project of Kanda
University of Foreign Affairs, Monbusho's Grant-in-Aid for Scientific Research (A), and
Grant-in-Aid for International Scientific Research (Joint Research).
I am grateful to two anonymous reviewers of this paper, Wayne P. Lawrence, Roger
Martin, John Shillaw, Kevin Varden, and Jeroen van de Weijer for many helpful com-
ments and suggestions on an earlier draft of this article. The remaining errors and insuf-
ficiencies are of course mine.
References
Halle, Morris and Jean-Roger Vergnaud
1987 An Essay on Stress. Cambridge, MA: MIT Press.
Haraguchi, Shosuke
1977 The Tone Pattern of Japanese: An Autosegmental Theory of Tonology. Tokyo:
Kaitakusha.
1979 "Nihongo Onchou no Shosou (Aspects of Japanese tonal systems)," Gengo no
Kagaku (Science of Language) 7, 21-69.
1991 A Theory of Stress and Accent. Dordrecht: Foris.
1997 "Tsuruoka Hougen no Akusento Saikou (Accent of Tsuruoka Japanese Re-
considered)," Report (1): Researching and Verifying an Advanced Theory of
The Accent of Tsuruoka Japanese Reconsidered 65
1. Introduction
1997a, b; Itô 1999; Mester 1999). It will be shown that a standard version
of Optimality Theory cannot account for the opaque interaction between
the constraints in Japanese phonology. In section 3, we will introduce
Sympathy Theory originally proposed by McCarthy (1997, 1998, 1999)
and review the analysis of the opaque interaction in Japanese phonology
in question. In section 4, we point out that the analysis provided by Itô
and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999) is not
in consonant with the assumption of Richness-of-the-Base and Lexicon
Optimization. In section 5, we will explore the possibility of extending the
notion of ^-selector from the set of Faithfulness constraints to the set of
Faithfulness and Markedness constraints. Concluding Remarks will be
given in section 6.
2. Basic facts
In this section, we will review the basic facts about Sequential Voicing and
Voiced Velar Nasalization and the analysis of their interaction given by
Itô and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999). It
will be shown that the standard version of Optimality Theory cannot ac-
count for the opaque interaction between the constraints.
Itô and Mester (Itô and Mester 1996, 1997a, b; Itô 1999; Mester 1999)
propose the constraint ranking in (2) in order to account for Sequential
Voicing.
(4) word
The tableau ( 5 ) shows that Lyman's Law outranks SEQVOI. This ranking
correctly chooses the candidate (b) as a winner. Lyman's Law blocks the
voicing if the input of the second member in a compound contains a
voiced obstruent.
b. satsu-taba *
(6) word
*R J
+voi
—son
Itô and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999) pro-
pose the constraint ranking in (9).
*PrWd[o m u s t outrank *g. This ranking accounts for the realization of the
voiced velar stop PrWd-initially in output for the corresponding voiced
velar segment in input, as is illustrated in the tableau (11):
(li) W o » * g
b. geta *! *
(12) *g » lDENT-IO(nas)
®· b. kagi *
Itô and Mester (1997b) report that we can find both transparent and
opaque interactions between the constraints that are relevant to Sequen-
tial Voicing and Voiced Velar Nasalization (VVN).
In (13), the stem-initial /k/ in the input form surfaces as [Q]. S E Q V O I is re-
sponsible for the voicing of the surface correspondent of the input /k/. Sim-
ple voicing of the input /k/ into [g] will create the violation of *g. Since *g
bans voiced velar obstruents from surfacing in the output, the input /k/
will surface as a velar nasal [rj]. This interaction between two constraints
S E Q V O I and *G is transparent, in the sense that the effects of the constraints,
namely, voicing and nasalization, can be observed in the output.
On the other hand, the examples in (14) show that a stem-internal [g]
(14a), which is the correspondent of an input /g/, blocks the effect of
SEQVOI, even if it is not a voiced obstruent phonetically. In other words,
in spite of its sonorant nature in the output, a stem-internal [g] does not
behave like a sonorant such as in (14c), but behaves as if it were a voiced
obstruent such as in (14b) in terms of Lyman's Law and then blocks Se-
quential Voicing. This interaction between the stem-internal voiced velar
nasalization effect and the stem-initial Lyman's Law effect (i.e. the lack of
Sequential Voicing) is opaque, in the sense that Lyman's Law can take an
effect on a form with a stem-internal [g], even if the form has no voiced
obstruent triggering Lyman's Law. In other words, a stem-internal [g] be-
haves as if it were a voiced obstruent and triggers Lyman's Law and
blocks the Sequential Voicing effect.
Itô and Mester (1997b) propose the constraint hierarchy (15) by connect-
ing the two ranked hierarchies (2) and (9). In this hierarchy, SEQVOI dom-
inates *g and LDENT-IO(nas).
^ ^ Lyman's Law
I
SEQVOI * [η
iDENT-IO(voi) *g
I
iDENT-IO(nas)
While the ranking in (15) can select the correct winner in the transparent
case (16), it will designate the wrong candidate as the winner in the
opaque case (17).
74 Takeru Honma
c. ori-Qami * *
b. saka-doge * *
desired winner *! *
c. saka-toge
wrong winner * *
d. saka-dor)e
Itô and Mester (1997b) conclude that it is not possible to capture this kind
of opaque interaction in standard O T and they present an alternative ac-
count based on the Sympathy Theory proposed by McCarthy (1997,1998,
1999). In the next section, the sympathetic analysis presented by Itô and
Mester (1997b) will be discussed in some detail and some crucial prob-
lems will be pointed out.
a. C* partitions the candidate set into two subsets (as do all con-
straints):
(i) those that do not violate C*, and
(ii) those that violate C* (to whatever degree — gradience of vio-
lation is irrelevant as long as there is some candidate that does not
violate C* at all).
b. The designated sympathy candidate (= ^-candidate) is that ele-
ment of subset (i) that best-satisfies Ή - C* (the rest of the con-
straint system), in the standard optimality-theoretic sense (Prince
and Smolensky 1993).
Let us return to the Japanese case. Itô and Mester (1997b) point out that
nasalization that occurs in the output obscures (or 'opacifies') the effect
of the interaction between Sequential Voicing and Lyman's Law, and
therefore Sympathy must be oriented towards a non-nasalizing candi-
date. In other words, a candidate within the set of non-nasalizing co-can-
didates must be referred to as the ^-candidate. In order to select the suit-
able ^-candidate, Itô and Mester (1997b) propose that ^-selector C*
should be lDENT-IO(nas). lDENT-IO(nas)® divides the output into two sub-
76 Takeru Honma
sets: the one is the set of nasalizing candidates which violate this con-
straint, and the other is the set of non-nasalizing candidates which ob-
serve the constraint. One of the candidates within the set of non-
nasalizing candidates will be selected as the ^-candidate.
(19)
/saka-toge/ Lyman's Law SEQVOI *g LDENT-IO(nas)* LDENT-IO(voi)
* *
Φ a. saka-toge
b. saka-doge *! * *
c. saka-toge * *
d. saka-dor)e * *
(21 )
/saka-toge/ Lyman's Law : IDENT- SEQVOI *g IDENT- IDENT-
b. saka-doge *! i * *! *
' r
c. saka-toge 1 * *
d. saka-doge 1 *; * *
I
The tableau (22) makes it sure that the constraints and the ranking that
are introduced into the Sympathetic analysis can also account for the
transparent case (16).
How should we Represent 'g' in toge in Japanese Underlyingly? 77
(22)
/iro-kami/ Lyman's Law IDENT- SEQVOI *g IDENT- IDENT-
^O(VOI) « W IO(voi)
'Ss a. iro-gami *! *
b. iro-kami *! *
"" c. iro-gami * * *
The analysis so far seems to be reasonable, but as we will see below, it em-
bodies problems related to the Richness of the Base assumption and the
principle of Lexicon Optimization.
The analysis of the opaque cases with velar nasals in the stem-medial po-
sition (i.e. [saka-toge] case) crucially depends on the assumption that ve-
lar nasals in the output correspond to voiced (but non-nasal) velar stops
in the input. This can be a problem, because this is not compatible with
the assumption of Richness of the Base. In Optimality Theory, grammat-
ical generalizations are expressed as the results of interactions of
Markedness/Structural constraints at the level of the output or con-
straints that state correspondences between the output and the other lev-
els (the level of input, ^-candidate, etc.) but no specific property can be
stated referring to the level of input alone:
Itô and Mester (1997b) take up this issue in the footnote (see Itô and
Mester 1997b: note 4). According to the footnote, Kazutaka Kurisu and
Philip Spaelti pointed out this issue to them. Under the richness-of-the-
base assumption, we cannot prevent an alternative form such as /saka-
toqe/ from being a possible input. As the result of the effect of faithful-
ness to input nasality (namely, lDENT-IO(nas)), this input, however, will
wrongly correspond to the output [saka-doge].
The tableau (24) shows that, for an alternative input /saka-torje/, the
constraint system that we have developed so far would select [saka-doqe]
78 Takeru Honma
(24)
/saka-toge/ Lyman's Law IDENT- SEQ Voi *9 IDENT- IDENT-
^O(VOI) IO (nas) * IO(VOI)
Φ a. saka-toge *! * * *
b. saka-doge *! * * *
desired winner *! *
c. saka-torje
Wrong winner *
d. saka-dorje
LO provides a desired result for the stem-initial [g]. The combined tab-
leaux in (26) show that two possible inputs, /geta/ and /η et a/, can corre-
spond to a desired output, [geta]. The input /geta/ will be chosen as the ac-
tual input for [geta], because the output corresponding to the input /geta/
incurs only one constraint violation and the violation is not fatal in the up-
per tableau (26), while the output corresponding to the input /geta/ incurs
two violations in the lower tableau in (26).
How should we Represent 'g' in toge in Japanese Underlyingly? 79
(26)
/geta/ *PrWdW *g lDENT-IO(nas)
*
a. geta
*
b. geta *!
/geta/ *PrWd[0 *g lDENT-IO(nas)
* *
c. geta
d. geta *!
(27)
LO input output *PrWd[0 *g lDENT-IO(nas)
<*· a. *
/geta/ [geta]
b. /geta/ [geta] *
*!
Lexicon Optimization can select a desired input for a word with word-ini-
tial [g]. This result is "desired," in the sense that the input provided by L O
is identical with the input required by the Sympathetic analysis presented
in Itô and Mester (1997b).
On the other hand, for the stem-medial case, the input provided by L O
is different from the one required by the Sympathetic analysis. For exam-
ple, in order to get the output [toge], we can assume both /toge/ and /toge/
as shown in (28) and (29), respectively:
Lexicon Optimization will choose /torje/ as the input for [toge]. The L O
tableau (30) illustrates the situation.
80 Takeru Honma
(30)
LO input output *PrWdfr) *g lDENT-IO(nas)
a. /toge/ [torje] *!
b. /toge/ [toge]
As we see in the tableaux (21) and (24) above, the Sympathetic analysis
requires that the second member of the compound [saka-toge] should be
/-toge/ but not /-torje/ at the level of input, /torje/ is predicted in the iso-
lated form but /-toge/ is required for the compound.
Itô and Mester (1997b) and Mester (1999) suggest that the crucial refer-
ence to the input made by the ^-selecting constraint C®, (namely, I D E N T -
IO(nas) in the case at hand) could be the source of the problem. They men-
tion a possible alternative that utilizes a markedness constraint against the
occurrence of [η] as the ^-selector 6 instead of an ΙΟ-Faithfulness con-
straint (namely, LDENT-IO(nas)), but they do not develop this alternative ful-
ly in their papers. It is worth while fleshing it out. Based on their sugges-
tion, if we designate a Markedness constraint as the ^-selecting constraint,
we can get the desired output [saka-toge] either from /saka-toge/ (31) or
from /saka-toge/ (32). In other words, the analysis presented here is com-
patible with the Richness-of-the-Base assumption.
(31)
/saka-toge/ LL IDENT- SEQVOI *g IDENT- IDENT-
^O(VOi) 'O(nas) IO(voi)
& a. saka-toge * *!
* * *
b. saka-doge *!
c. saka-toge * * *
d. saka-doge *t * * *
(32)
/saka-toge/ LL IDENT- SEQVOI *g IDENT- IDENT- *g*
IO(nas) IO(voi)
Φ a. saka-toge *
*! *
b. saka-doge *L * * * *
How should we Represent 'g' in toge in Japanese Underlyingly? 81
IO(NAS) IO(VOI)
c. saka-torje * *
d. saka-dorje *! * *
(33).
LO LL IDENT- SEQVOI *g IDENT- IDENT-
IO(nas) IO(voi)
a. /saka-toge/ *
*! *
[saka-toge]
<*" b. /saka-torje/ * *
[saka-toge]
Now we are in a position to answer the question posed in the title of this
paper: how should we represent 'g' in toge in Japanese underlyingly? The
answer is: we should represent 'g' in [toge] as /g/ underlyingly.
6. Concluding remarks
Notes
* This research was partially supported by the Ministry of Education, Science, Sports and
Culture, Grant-in-Aid for Encouragement of Young Scientists, No. 11710265,1999-2001.
I wish to express my gratitude to the editors of this book for offering me a chance to con-
tribute this paper to the present volume and for their patience with my delay in turning
in my manuscript.
1. I use single quotation marks (") to suggest that the representation is "orthographic."
2. Based on this allophony, Trubetzkoy (1939), for example, mentions that the opposition
between [g] and [rj] signals boundaries. "Im Japanischen besteht zwischen g und rj ein
How should we Represent 'g' in toge in Japanese Underlyingly? 83
References
Alderete, John
1997 Dissimilation as Local Conjunction. Proceedings of North East Linguistics So-
ciety 27,17-32.
Fukazawa, Haruka
1999 Theoretical Implications of OCP Effects on Features in Optimality Theory. Ph.
D. thesis, University of Maryland.
Honma, Takeru
2000 Sympathetic Analysis of German R-Vocalization. Ms., Downloadable at http:/
/member.nifty.ne.jp/honmat/papers.html.
Inkelas, Sharon
1995 The Consequences of Optimization for Underspecification. NELS 25, 287-
302.
Itô, Junko
1999 Optimality Theory I: The Foundations. Handout of the lecture at Tsukuba
University on 21st, July, 1999.
Itô, Junko and Armin Mester
1996 Rendaku I: Constraint Conjunction and the OCP. Ms. [ROA-144], Download-
able at http://www.ruccs.rutgers.edu/roa.html.
1997a Correspondence and Compositionality: The Gagyo Variation in Japanese
Phonology. In Iggy Roca (ed.), Derivations and Constraints in Phonology, 419-
462. Oxford: Clarendon Press.
1997b Featural Sympathy: Feeding and Counterfeeding Interactions in Japanese. In
Rachel Walker, Motoko Katayama, and Daniel Karvonen (eds.), Phonology
84 Takeru Honma
at Santa Cruz, Volume 5, 29-36. Santa Cruz: The Linguistics Reseach Center,
University of California.
1998 Markedness and Word Structure: OCP Effects in Japanese. Ms. [ROA-255],
Downloadable at http://www.ruccs.rutgers.edu/roa.html.
Kager, René
1999 Optimality Theory. Cambridge Textbooks in Linguistics. Cambridge: Cam-
bridge University Press.
McCarthy, John
1997 Sympathy and Phonological Opacity. Handout at Hopkins Optimality Theory
Workshop/Maryland Mayfest 1997, May 1997.
1998 Sympathy and Phonological Opacity. Ms. [ROA-252], Downloadable at http:/
/www.ruccs. rutgers.edu/roa.html.
1999 Sympathy and Phonological Opacity. Phonology 16, 331-399.
McCarthy, John and Alan Prince
1995 Faithfulness and Reduplicative Identity. In Jill N. Beckman, Laura Walsh
Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasion-
al Papers 18, 249-384. Graduate Linguistic Student Association.
Mester, Armin
1999 OT: Some Current Developments. Handout of the lecture at Tsukuba Univer-
sity on 21st, July, 1999.
Prince, Alan and Paul Smolensky
1993 Optimality Theory: Constraint Interaction in Generative Grammar. RuCCs
Technical Report No. 2. Piscataway, New Jersey: Rutgers University Center
for Cognitive Science.
Trubetzkoy, Nikolaj Sergeevich
1939 Grundzüge der Phonologie (1st ed.). Prague: Cercle Linguistique de Prague.
Principles of Phonology. Trans, by Christiane A. M. Baltaxe. Berkeley: Uni-
versity of California Press., 1969.
1962 Grundzüge der Phonologie (3rd ed.). Göttingen: Vandenhoeck & Ruprecht.
Domain-Relative Faithfulness and the OCP:
Rendaku Revisited*
Haruka Fukazawa and Mafuyu Kitahara
Kyushu Institute of Technology and Indiana University
1. Introduction
The Obligatory Contour Principle (henceforth the OCP) was first pro-
posed at the advent of autosegmental phonology (Leben 1973; Goldsmith
1976) for a restrictive representation of autosegments. It was modified
along with the development of phonological theories in the 1980's and has
been adapted to the analyses of a wide variety of languages and phenom-
ena (see McCarthy 1986 for a concise review and Yip 1988 for a typolog-
ical consideration with respect to the OCP). The frequently cited defini-
tion of the OCP is as follows (McCarthy 1986):
Let us take a look at the basic facts of Rendaku. 2 The examples in (2)
show typical compound words which have voicing at the juncture of the
two elements.
(2)
a./koi/ 'love' + /fumi/ 'letter' [koibumi] 'love letter'
b./ama/ 'rain' + /tare/ 'drop' —> [amadare] 'raindrop'
c./tsuri/ 'hang' + /kane/ 'bell' —> [tsurigane] 'hanging bell'
d./kaimono/ 'shopping' + /fukuro/ 'bag' —> [kaimonobukuro] 'shopping bag'
When the second element of the compound contains any voiced ob-
struent, Rendaku does not take place. This is known as Lyman's Law. The
examples in (3) show sample cases with hypothetical ungrammatical
forms where two voiced obstruents cannot cooccur within a word.
88 Haruka Fukazawa and Mafuyu Kitahara
(3)
a. /take/ 'bamboo' + /higo/ 'stick' —> [takehigo] 'bamboo stick'
*[takebigo]
b. /shira/ 'white' + /sagi/ 'heron' [shirasagi] 'white heron'
*[shirazagi]
c. /kita/ 'north' + /kaze/ 'wind' —> [kitakaze] 'north wind'
*[kitagaze]
d. /mukashi/ 'old days' + /katagi/ —> [mukashikatagi] 'old-fashioned spirit'
'character' *[mukashigatagi]
Voiced obstruents in the first element of the compound do not affect Ly-
man's Law at all. The examples in (4) show that two voiced obstruents can
cooccur within a word in such cases.
(4)
a./tabi/ 'travel' + /hito/ 'person' —> [tabibito] 'traveller'
b./sabi/ 'rust' + /tome/ 'stopper' —> [sabidome] 'anti-corrosive'
c./shibu/ 'bitter' + /kaki/ 'persimmon' —> [shibugaki] 'bitter persimmon'
d./ichigo/ 'strawberry' + /hatake/ 'field' —> [ichigobatake] 'strawberry field'
Ito and Mester (1986; 1998) propose that voicing in Rendaku can be con-
sidered as a morpheme which attaches to the second member of the com-
pound as an abstract prefix ρ. This prefix is assumed to bear the feature
[voice] underlyingly.
(5)
Stem 1 Stem 2
ama ρ tare
[voice]
In order for the prefix to be fully parsed at the surface, the [voice] feature
of the prefix must be realized as a part of the initial obstruent of the sec-
ond member. Ito and Mester propose a constraint REALIZE-MORPHEME to
force the feature to surface in the output. This constraint can only see
whether the input morpheme is realized in the output.
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 89
(6)
/ama ρ tare/
1 REALIZE-MORPHEME lDENT[voice]
[voice]
a. [ama + tare] *!
[voice] e
b. [ama + dare] *
/
[voice] e
OCP violation
Lyman's Law has been analyzed as a violation of the OCP on the [voice]
feature (Itô and Mester 1986; 1998). As a result of the OCP violation, the
[voice]Q feature from the ρ morpheme cannot be linked to any segment
and thus does not surface. Let us tentatively regard the OCP as a mono-
lithic constraint and incorporate it in the previous constraint ranking. In
order for Lyman's Law to block Rendaku, the OCP on [voice] has to be
ranked higher than REALIZE-MORPHEME. This ranking correctly selects
the desired output as shown in tableau (8).3
90 Haruka Fukazawa and Mafuyu Kitahara
(8)
/ k i t a + ρ + k a ζ e/
REALIZE- IDENT
1 1 OCPfvoice]
MORPHEME [voice]
[voice] [voice]
a. [k i t a + k a ζ e] *
I
[voice]e [voice]
b. [k i t a + g a ζ e] *! *
/ I
[voice]e [voice]
When the first member of the compound contains the [voice] feature,
Rendaku is not blocked even though there are two [voice] features seem-
ingly adjacent to each other. To account for this asymmetry, the domain of
the OCP constraint is specified as a morpheme in the previous analyses.
(9)
/ t a b i + ρ + h i t o/ OCP[voice]
REALIZE- IDENT
1 1 within a
MORPHEME [voice]
[voice] [voice] morpheme
a. [t a b i + h i t o] *!
I
[voice] [voice]Q
b. [t a b i + b i t o] Not violated *
I \
[voice] [voice]Q
from different sets accounts for the cases in which a phonological alterna-
tion is observed in some domain, and not observed in another within a
grammar of a language. We will briefly review cases of the Emergence of
the Unmarked (TETU) (McCarthy and Prince 1994), positional faithful-
ness constraints (Beckman 1995), relativized faithfulness constraints for
distinct morphemes (Urbanczyk 1995; 1996; Benua 1995; 1997; Fukazawa
1998; Fukazawa, Kitahara, and Ota 1998; Itô and Mester 1999).
1 1
[cor] [cor]
b. sdog + chi-ya *!
1
[cor]
(13)
/krab + en/ lDENTONs[lar] *LAR lDENT[lar]
*
a. kra.ben
*
b. kra.pen *!
As in tableau (13), candidate (a) with the voice feature in the onset position
becomes optimal due to the proposed ranking. On the contrary, in other po-
sitions, the markedness constraint * L A R penalizes the candidate which con-
tains voiced obstruents. As shown in tableau (14), candidate (b) wins, be-
cause the violation of *Lar is fatal due to the fact that lDENTONs[lar] is
satisfied by both candidates.
(14)
/krab/ lDENTONs[lar] *LAR iDENTflar]
a. krab *!
*
b. krap
copies only the first CjVj. For example, the distributive of [badá?]
(C1V1C2V2C3) 'child, offspring' is not *[b9-b9dá?] (C1V1-C1V1C2V2C3)
but [bad-badá?] (C1V1C2-C1V1C2V2C3) 'children'. On the other hand, the
diminutive form for [cabs] 'hand' is [ca-cabs] (C1V1-C1V1C2V2C3) 'little
hand', and a C1V1C2-C1V1C2V2C3 form, *[cal-cabs], is incorrect.
Urbanczyk analyzes this as the avoidance of a coda which results in a
CV-shape for the diminutive. In contrast, codas are possible in the distrib-
utive morpheme, creating a CVC-shape. Therefore, the markedness con-
straint prohibiting codas, NOCODA, is respected in the diminutive redupli-
cation at the expense of a violation of the faithfulness constraint against
deleting a segment, MAX (NOCODA > MAX). However NOCODA is violated
to satisfy MAX in the distributive (MAX > NOCODA). TO resolve this con-
flict between the two rankings, she claims that each of the reduplicative
morphemes has its own correspondence relation to the base; hence, there
are two different Base-Reduplicant (BR) relations in Lushootseed.
Consequently, two full sets of BR faithfulness constraints are generated
in the grammar of Lushootseed: BR-Diminutive (Dim): {MAX-BR-Dim,
DEP-BR-Dim, LDENT[F]-BR-Dim, ...} and BR-Distributive (Dis): {MAX-
BR-Dis, D E P - B R - D Í S , IDENT[F]-DÍS, ...}. These faithfulness constraints
are placed in a single ranking with the markedness constraints. Both the
diminutive CV-shape and the distributive CVC-shape result from ranking
M A X - B R - D Í S » NOCODA » M A X - B R - D i m .
Let us take a look at how the constraint ranking works with the actual
analysis. The given ranking accounts for the CV-shape of the reduplica-
tive forms of the diminutive.
b. cal-calss *!
The same ranking provides the correct analysis for the CVC-shape of dis-
tributive reduplication.
Thus, the two full sets of faithfulness constraints for both distributive and
diminutive reduplicative morphemes are instantiated in the grammar of
Lushootseed.
Benua (1995, 1997) shows that there are two patterns of affixation in
English. For example, certain coda clusters are simplified in both root
morphemes: [kAndem], *[kAndemn] ('condemn') 5 , and class 2 affixation
such as when -ing is attached to the root: [kAndemirj], *[kAndemmr)]
('condemning'), but not in class 1 affixation: [kAndemneysAn], *[kAnde-
meysAn] ('condemnation').
Benua notes that class 1 and class 2 affixal morphemes each display a
different correspondence relation to the output of the root morphemes;
hence, there are two kinds of Output-Output ( O O ) faithfulness relations
in English. Thus, two full sets of faithfulness constraints, namely, OO-class
1 affix: {MAx-OO-class 1 affix, DEP-OO-class 1 affix, IDENT[F] OO-class 1
affix,...} and OO-class 2 affix:{MAX-00-class 2 affix, DEP-OO-class 2 af-
fix, LDENT[F]00-class 2 a f f i x , . . . } are found in the grammar of English.
First, in an unaffixed word, the ranking between *mn] a » MAX-ΙΟ ex-
plains why the output is not [kAndemn] but [kAndem].
b. kAndemn *!
The clusters are simplified due to the ranking in (17). Now, from this out-
put form, two kinds of affixation are possible: class 1 and class 2.
In class 1 affixation, the clusters are not simplified, because D E P - O O -
class 1 affix is lower ranked than the phonological constraint *mn] 0 and
MAX-IO.
b. kAndemeysAn **1
On the other hand, the clusters are simplified in class 2 affixation, because
DEP-OO-class 2 affix and *mn] 0 are higher ranked than M A X - I O .
Domain-Relativ e Faithfulness and the OCP: Rendaku Revisited 97
b. kanta *!
98 Haruka Fukazawa and Mafuyu Kitahara
As shown in tableaux (20) and (21), a single ranking can account for the
occurrence of PNV in Yamato and its non-occurrence in Sino-Japanese.
Thus, the simultaneous attendance of the relativized faithfulness con-
straints account for the asymmetric phonological phenomena among stra-
ta in Japanese.
(22)
a. Faithfulness constraint: [F(i)](o) = list-of-violations
b. Structural constraint: S (o) = list-of-violations
Itô and Mester call [F(¿)] part of the faithfulness constraint an "instanti-
ated constraint" which is derived from the two-argument schema of the
function. They further argue that because this instantiated constraint is
derived for each input, there is a possibility that different instantiated
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 99
4. Analysis of Rendaku
times plays the role of a rule trigger. She also indicates that there are four
kinds of repair strategies for the OCP violations, namely degemination,
dissimilation, assimilation, and epenthesis. Therefore, it seems that there
are logically five types of languages from the perspective of the OCP ef-
fect. Fukazawa (1999) reexamined Yip's categorization of languages spe-
cifically for the OCP on features. She eliminates some of the possibilities
from Yip's classification such as epenthesis of a segment, and reorganizes
some of the categories such as dissimilation, assimilation, and deletion. 9
She concludes that there are four types of languages regarding the OCP
on features:
I
[voice]e [voice]
b. [k i t a + g a ζ e] *! *
/ I
[voice]e [voice]
In tableau (24), it seems that the first [voice] feature deletes to satisfy
OCP[voice] at the expense of violation of R E A L I Z E - M O R P H E M E as in can-
didate (a). However, in addition to this featural deletion, there are other
possible repair strategies to satisfy OCP[voice] according to Fukazawa's
claim, such as featural fusion, featural change, and segmental deletion.
Among them, featural change and segmental deletion are irrelevant since
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 101
they are not observed. Therefore, we examine the case where featural fu-
sion takes place.
(25)
/kita + Q + kaze/
REALIZE- IDENT
1 1 OCPfvoice]
MORPHEME [voice]
[voice] [voice]
a. [kita + kaze] *!
1
[voice]
<desired output>
»! *
b. [kita + gaze]
/ 1
[voice] [voice]
*
*®"c. [kita + gaze]
\ /
[voice]e
<wrong winner>
Candidate (c) in tableau (25) has a featural fusion for repairing the OCP
violation, but it is not the desired output. Another constraint which
penalizes candidate (c) must be introduced. UNIFORMITY[voice], which
prohibits fusion of two [voice] features, is the appropriate constraint to
specifically penalize the fused structure in candidate (25c). U N I F O R M I -
TY[voice] outranks REALIZE-MORPHEME, which leads candidate (a) to be
the output.
(26)
/kita + Q + kaze/
UNIFORMITY REALIZE- IDENT
1 1 OCPfvoice]
[voice] MORPHEME [voice]
[voice] [voice]
*
a. [ k i t a + k a z e ]
1
[voice]
*
b. [ k i t a + g a z e ] *!
/ 1
[voice]e [voice]
*
c. [ k i t a + g a z e ] *!
\ /
[voice]e
Now, let us examine another example in which the first member of the
compound contains the [voice] feature under the same constraint ranking.
102 Haruka Fukazawa and Mafuyu Kitahara
It a b i + ρ + h i t o/
UNIFORMITY REALIZE- IDENT
1 1 OCP[voice]
[voice] MORPHEME [voice]
[voice] [voice]
*«" a. [ t a b i + h i t o ] *
ι
I
[voice]
<wrong winner>
b. [ t a b i + b i t o ] *! *
Ι I
1 1
[voice] [voice] e
c. [ t a b i + b i t o ] *! *
\ /
[voice] e
<desired output>
The ranking given in (26) incorrectly lets candidate (a) win in tableau
(27). The correct output should be determined as either candidate (b) or
candidate (c).
Here, we propose that the faithfulness constraint UNIFORMITY[voice] is
relativized depending on the domain as we have already mentioned in
section 2. Similar to the ranking schema for positional faithfulness, the
general UNIFORMITY constraint UNIFORMITY[voice]-G must be ranked low-
er than the specific UNIFORMITY constraint for within a morpheme, UNI-
FORMITY[voice]-M. The overall ranking including relativized UNIFORMITY
is shown in (28).
Let us examine this ranking by reanalyzing tableaux (26) and (27). First,
when the second member of the compound contains a [voice] feature,
candidate (a) in tableau (29) wins as in tableau (26). Thus, relativizing
UNiFORMiTYfvoice] does not complicate the picture.
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 103
[voice]
b. [k i t a + g a ζ e] *!
/ I
[voice] [voice]
c. [k i t a + g a ζ e] *! *
\ /
[voice]e
On the other hand, when the first member of the compound contains a
[voice] feature, candidates in which featural fusion takes place win, which
is the desired result. In tableau (30), candidate (c) does not violate UNI-
FORMITY[voice]-M because the fusion takes place across two morphemes.
Hence, not only the OCP[voice] violation but also the violation of R E A L -
I Z E - M O R P H E M E becomes fatal in the candidate set.
\ /
[voice]e
5. Conclusion
Notes
* We would like to thank the editors of this volume, Jeroen van de Weijer and Tetsuo Nish-
ihara, for offering us a chance to contribute this paper. We also would like to thank Stuart
Davis and Laura W. McGarrity for their helpful comments.
1. Following Smolensky (1993, 1995, 1997), we assume that a local conjunction of con-
straints A and Β (represented as A&B) also plays as another constraint. In other words,
once two (or even more) constraints are conjoined into a local conjunction, it should be
treated as an independent constraint. Therefore, no quantitative factor in the evaluation
procedure is introduced there.
If Local Conjunction is a type of constraint, it must be in Universal Grammar (UG).
However, if it is in UG, it must be cross-linguistically valid. A question now arises: Are
all possible local conjunctions truly in U G ? If so, U G grows extremely large. Following
Fukazawa and Miglio (1998), we consider that the possibility of the conjunction is in UG,
in other words, the " & " operator for conjunction is in UG. However, the choice of which
two constraints to be conjoined is language specific. Furthermore, there must be specific
restrictions on which two (or more) constraints can be conjoined. For the detailed discus-
sion on the restrictions of Local Conjunction, see Fukazawa & Miglio (1998).
2. There are numerous exceptions to Rendaku and much literature is devoted to account
for those exceptions. First, though Rendaku has been considered to take place mostly in
the Yamato (Native) stratum in Japanese lexicon, some Sino-Japanese or Mimetic vocab-
ularies show Rendaku, while some Yamato compounds do not (Itô and Mester 1999,
Vance 1996). These observations are based on etymological analyses of lexical stratifica-
tion. Our position to these exceptions is that a purely phonological lexical stratification
can also be considered as a synchronic grammar in our cognitive system. For a more gen-
eral discussion on lexical stratification, see Fukazawa, Kitahara, and Ota (1998) and Itô
and Mester (1999). Second, compounds with more than two elements have systematic ex-
ceptions to Rendaku due to the internal structure of the compound (Otsu 1980). Third,
pitch accent assignment and Rendaku are argued to be related in some cases (Kibe 1978).
Sato (1989) reviews these and other exceptional cases to Rendaku. Finally, Itô, Mester
and Padgett (1995), Rice (1997), Itô and Mester (1998) and Honma (2000, this volume)
give a detailed discussion of the relationship between Rendaku and Post Nasal Voicing
and voicing specification of sonorants which apparently seems a problem for an OCP-
based analysis of Lyman's Law.
Although we are aware of those exceptions and theoretical complications around
Rendaku, our main focus in this paper is not to explore them in detail but to give a new the-
oretical direction within a general framework of Correspondence Theory and Local Con-
junction. Further research is of course necessary for a more thorough treatment of Rendaku.
3. In tableau (8), another candidate, where the stray voice is from the underlying /z/, can be
considered. A similar situation arises when the second member of the compound has a
voiced initial obstruent. A possible candidate is that the input [voice] is delinked from the
106 Haruka Fukazawa and Mafuyu Kitahara
second member and the [voice]Q is re-linked there. For example, /hana/ 'flower' + /gara/
'pattern' is realized as [hanagara] with [voice]e as the voicing feature for [g]. These cases
can be taken care of by additional faithfulness constraints for association lines such as
MAxfassoc] and DEpfassoc]. As shown in tableau (8)', candidate (a) is more faithful to
the association line configuration in the input than (c) though they are phonetically
equivalent. Both the [voice] e feature from the prefix ρ and the [voice] feature from input
[z] are delinked in (8c)', which causes two violations of MAxfassoc]. Moreover, the
delinked [voice] e is re-linked to [z], which causes a violation of DEp[assoc].
(8)'
/kita + Q + kaze/
1 1 MAx[assoc] DEp[assoc]
[voice] [voice]
*
τ a. [k i t a + k a ζ e]
I
[voice]p [voice]
*
c. [k i t a + k a ζ e]
[voice]e [voice]
Though MAx[assoc] incurs violation(s) in other cases with Rendaku, we do not include it
in the following tableaux since the point of discussion here is not to layout the entire
ranking but to review the gist of Itô and Mester's analysis on Rendaku.
4. Lombardi assumes that all features are privative. Therefore, [b] bears [voice], and [p]
does not. The definition of the constraint * L A R is "Do not have the [laryngeal] feature."
The feature [voice] is considered to depend on [laryngeal]. Thus, [b] violates * L A R be-
cause it has [laryngeal], while [p] does not.
5. It could be pointed out that the vowels in the words "condemn", "condemning", and
"condemnation" should be replaced by the vowels [θ] or [a]. We cite the data exactly
from what Benua (1997) mentions.
6. For the discussion of the constraint *mn] 0 see Benua (1997).
7. Tableaux (18) and (19) only show some parts of the recursive tableaux which are utilized
in Benua (1997). For the whole analyses, see Benua (1997).
8. It has been a controversial issue whether the markedness constraints can be split like
faithfulness constraints. Although we adopt Itô and Mester's reasoning in this paper, we
recognize that the claim will have to be supported by valid empirical evidence and further
investigation.
9. The OCP on features differs from the OCP on segments in various ways. For instance, the
repair strategies are different. OCP on segments can be repaired by assimilation, dissim-
ilation, epenthesis, metathesis, and so on, as Yip (1988) points out. On the contrary, some
of these phonological processes cannot repair violations of the OCP on features. For ex-
ample, epenthesis cannot fix a violation of the OCP on features. Consider a case where a
segment C is epenthesized between two adjacent segments A and Β which both have an
identical feature F. After epenthesis, segments A and Β are not adjacent anymore, but the
feature F are still adjacent unless the epenthetic segment C bears the oppositte value of
F. Fukazawa (1999) follows the recent proposal in phonology in which all the features are
privative. When all features are privative, the segment C never bears the oppositte value
of F. Thus, epenthesis of a segment never rescues the OCP on features. This is why Fuka-
zawa (1999) claims that the typology of the OCP on features has to be treated differently
from that on segments. For a more detailed discussion, see Fukazawa (1999).
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 107
References
Alderete, John
1997 "Dissimilation as Local Conjunction," Proceedings of North East Linguistics
Society 27:17-32.
Beckman, Jill
1995 "Shona Height Harmony: Markedness and Positional Identity," In Jill Beck-
man, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), University of Mas-
sachusetts Occasional Papers: Papers in Optimality Theory 18: 53-76, GLSA,
Amherst, Mass.
Benua, Laura
1995 "Identity Effects in Morphological Truncation," In Jill Beckman, Laura Walsh
Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasion-
al Papers: Papers in Optimality Theory 18: 77-136. GLSA, Amherst, Mass.
Benua, Laura
1997 Transderivational Identity: Phonological Relations between Words. Doctoral
Dissertation. University of Mass. Amherst.
Fukazawa, Haruka
1998 "Multiple Input-Output Faithfulness Relations in Japanese." ms., University
of Maryland, College Park. [ROA-260-0598]
1999 Theoretical Implications of O C P Effects on Features in Optimality Theory.
Doctoral dissertation, University of Maryland, College Park.
Fukazawa, Haruka, Mafuyu Kitahara, and Mitsuhiko Ota
1998 "Lexical Stratification and Ranking Invariance in Constraint-Based Gram-
mars," in M. Catherine Gruber, Derrick Higgins, Kenneth S. Olson, and Tam-
ra Wysocki (eds.), CLS 34: The Panels, 47-62.
Fukazawa, Haruka, and Miglio Viola
1998 "Restricting Conjunction to Constraint Families," in Proceedings of Western
Conference on Linguistics (WECOL 96) vol 9:102-117.
Goldsmith, John A.
1976 Autosegmental Phonology. Doctoral dissertation, MIT, Cambridge.
Honma, Takeru
2000 "How should we Represent 'g' in toge in Japanese Underlyingly?" in this vol-
ume.
Itô, Junko, and Armin Mester
1986 "The phonology of voicing in Japanese: theoretical consequences for morpho-
logical accessibility," Linguistic Inquiry 17: 49-73.
1995 "Japanese Phonology," in John Goldsmth (ed.), The Handbook of Phonolog-
ical Theory: 817-838. Basil Blackwell, Cambridge, Mass.
1996 "Structural Economy and O C P Interactions in Local Domains," Paper pre-
sented at Western Conference on Linguistics ( W E C O L ) 1996, University of
California, Santa Cruz.
108 Haruka Fukazawa and Mafuyu Kitahara
1998 "Markedness and Word Structure: OCP Effects in Japanese," ms., University
of California, Santa Cruz. [ROA-255-0498].
1999 "The Phonological Lexicon," In Natsuko Tsujimura (ed.), The Handbook of
Japanese Linguistics. Blackwell Publishers, Oxford.
Itô, Junko, Armin Mester, and Jaye Padgett
1995 "Licensing and Underspecification in Optimality Theory," Linguistic Inquiry
26: 571-613.
Kibe, Nobuko
1978 "Keetai akusentoronteki ichi koosatsu: fukugoogo akusento to gokoosee-
rendaku o megutte [A morphoaccentual analysis on compound accent, word
structure and sequential voicing]," Gobun Kenkyuu (Kyushu University) 46.
Kubozono, Haruo
1993 The Organization of Japanese Prosody, Tokyo: Kuroshio.
Leben, William
1973 Suprasegmental Phonology. Doctoral dissertation, MIT, Cambridge.
Lombardi, Linda
1995 "Why Place and Voice are different: Constraint Interactions and Featural
Faithfulness in Optimality Theory." ms., University of Maryland, College Park.
Lubowicz, Anna
1997 "Derived Environment Effects in OT." ms., University of Massachusetts, Am-
herst. [ROA-239-0198]
McCarthy, John
1986 "OCP effects: Gemination and Antigemination," Linguistic Inquiry 17: 207-
63.
1988 "Feature Geometry and Dependency: A Review," Phonetica 43: 84-108.
McCarthy, John, and Alan Prince
1994 "The Emergence of the Unmarked." ms., University of Massachusetts, Am-
herst, and Rutgers University.
1995 "Faithfulness and Reduplicative Identity," In Jill Beckman, Laura Walsh
Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasion-
al Papers: Papers in Optimality Theory 18: 249-384. GLSA, Amherst, Mass.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems," in Y. Otsu
and A. Farmer, eds. MIT Working Papers in Linguistics 2, Theoretical Issues
in Japanese Linguistics: 207-228, MIT Press, Cambridge, MA.
Prince, Alan, and Paul Smolensky
1993 "Optimality Theory: Constraint Interaction in Generative Grammar." ms.,
Rutgers University, New Brunswick, NJ; University of Colorado, Boulder.
Rice, Keren
1997 "Japanese NC Clusters and the Redundancy of Postnasal Voicing," Linguistic
Inquiry 28: 541-551.
Sato, Hirokazu
1989 "Fukugoogo ni okeru Akusento Kisoku to Rendaku Kisoku [Accent Rules
and Sequential Voicing Rules in Compounds]," in Miyoko Sugito ed., Kooza
Nihongo to Nihongo Kyooiku [Lectures on Japanese and Japanese Education]
vol. 2: 233-265.
Smolensky, Paul
1993 "Harmony, Markedness, and Phonological Activity." Paper presented at Rut-
gers Optimality Workshop-1, Rutgers University, October 23,1993.
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 109
1995 "On the Internal Structure of the Constraint Component Con of UG." Collo-
quium presented at U C L A , April 7,1995.
1997 "Constraint Interaction in Generative Grammar II: Local Conjunction." Pa-
per presented at the Hopkins Optimality Theory Workshop/University of
Maryland Mayfest 1997, May 8-12,1997.
Urbanczyk, Suzanne
1995 "Double Reduplications in Parallel," In Jill Beckman, Laura Walsh Dickey,
and Suzanne Urbanczyk (eds.), University of Massachusetts Occasional Pa-
pers: Papers in Optimality Theory 18: 499-531. GLSA, Amherst, Mass.
1996 Patterns of Reduplication in Lushootseed. Doctoral dissertation, University
of Massachusetts, Amherst.
Vance, Timothy J.
1996 "Sequential Voicing in Sino-Japanese." Journal of the Association of Teachers
of Japanese 30: 22-43.
Yip, Moira
1988 "The Obligatory Contour Principle and Phonological Rules: A Loss of Iden-
tity," Linguistic Inquiry 19: 65-100.
Epenthetic Vowels and Accent in Japanese:
Facts and Paradoxes
Haruo Kubozono
Kobe University
1. Introduction
anese have undergone consonant elision which has turned CVCV into
CVV: e.g. kofi koi 'carp'. In terms of syllable structure, this process has
had the effect of producing a heavy (bimoraic) syllable out of a sequence
of two light (monomoraic) syllables. Here, again, a conventional theory
would predict that elided consonants should be no longer visible to pho-
nological rules in the contemporary grammar of the language, and so pho-
nological rules apply to the output, not to the input of the segmental
change. While this orthodox account seems to hold in most cases, some
accent rules in Japanese exhibit cases where CVV syllables derived from
CVCV sequences pattern with underlying CVCV sequences, i.e. elided
consonants sometimes behave as if they were still present in the syn-
chronic grammar.
The purpose of this paper is to describe and generalize the peculiar be-
haviour of epenthetic vowels and elided consonants in modern (Tokyo)
Japanese in relation to a variety of accentual phenomena including loan-
word and compound accentuation. In order to achieve this goal, we will
first briefly look at the history of epenthetic vowels and elided consonants
in Japanese (section 2). We will then consider how these segments behave
in accent rules said to be sensitive to the syllable structure of words (sec-
tion 3). The final section (section 4) summarizes the basic facts and para-
doxes behind the peculiar behaviour of epenthetic vowels and elided con-
sonants, and demonstrates the difficulties they might imply for theoretical
analyses in the future.
The syllable-sensitive accent rules to be discussed in this paper are list-
ed below together with a summary of the metrical (in)visibility of epen-
thetic vowels and elided consonants for each rule. All of these rules are
more or less productive in the synchronic grammar of Tokyo Japanese;
they readily apply to new and nonsense words as well as existing words.
(1) a. the loanword accent rule known as the 'antepenultimate rule' (section 3.1.)
- Some epenthetic vowels seem invisible, while others are visible.
b. accent rules triggering 'unaccentedness' in loanwords (section 3.2.)
- Some epenthetic vowels are invisible, while others are visible.
c. the ordinary noun-noun compound accent rule (section 3.3.)
- Epenthetic vowels in SJ morphemes are generally invisible, while those in
foreign morphemes are visible. Elided consonants in native Japanese
morphemes are no longer visible.
d. the compound accent rule for personal names involving taroo (section 3.4.)
- All epenthetic vowels are visible, but elided consonants are somehow also
visible.
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 113
e. the compound accent rule for personal names with the SJ morpheme -iti
(section 3.5.)
- All epenthetic vowels are visible, but elided consonants are somehow also
visible.
f. the accent rule for verbs and adjectives (section 3.6.)
- Elided consonants in native morphemes are no longer visible.
g. the so-called 'pre-no deaccenting rule' (section 3.7.)
- Epenthetic vowels in SJ morphemes are fully visible.
As can be grasped from the summary above, epenthetic vowels are gen-
erally visible and elided consonants are generally no longer visible to
most of the syllable-sensitive accent rules. However, there are many ex-
ceptions. For example, some epenthetic vowels in foreign loanwords are
invisible to the loanword accent rule and the accent rules which yield un-
accented loanwords, although other epenthetic vowels in the same type of
words are fully visible to the same accent rules. Moreover, epenthetic
vowels in SJ morphemes behave as if they were invisible to the ordinary
noun-noun compound accent rule, while they are fully visible to the other
accent rules. Furthermore, elided consonants in native Japanese mor-
phemes behave as if they were still present in the X-taroo and X-iti com-
pounds, but not in other types of words. These puzzling facts not only
undermine the simple and transparent distinction between excrescent
(phonology-invisible) and epenthetic (phonology-visible) vowels (Levin
1987), but also seem to challenge theoretical analyses, derivational and
non-derivational alike.
Iii or lui, the two high vowels which are prone to vowel devoicing in (mod-
ern) Japanese (dots indicate syllable boundaries) 4 .
(5) a. i-onbin
tu.ki + ta.ti -> tui.ta.ti 'moon, beginning; first day of the month'
b. u-onbin
ha.ya + ku ->• ha.yau (->• hayoo) 'early (continuous form)'
c. hatu-onbin
yo.mi + te -»• yom.te -> yon.de 'to read (continuous form)'
d. soku-onbin
mo.ti + te -*• mot.te 'to take (continuous form)'
In terms of syllable structure, the segmental changes shown in (5) and (6)
had the effect of creating a heavy monosyllable (either CVV or CVC) out
of a sequence of two light syllables. This is schematically shown in (7).
The first accent rule we will consider is the loanword accent rule conven-
tionally known as the 'antepenultimate rule'. This rule places an accent
on the third mora from the end of the word in loanwords. If this antepen-
ultimate mora is the second mora of a heavy syllable, then the accent au-
tomatically shifts one mora leftwards, i.e. onto the nucleus of the syllable.
That is, an accent is placed on the syllable containing the antepenultimate
mora (McCawley 1968)5.
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 117
As can be seen from the examples in (8), this accent rule applies to
structures in which epenthetic vowels have already been inserted. That is,
it counts all the moras in the surface structure, irrespective of whether
they contain an epenthetic vowel or not. In the examples below, the ac-
cent mark (') is placed immediately after the accented mora.
A careful analysis reveals that the words showing the exceptional accent
pattern in (9c) almost invariably have the two features in (10) in common
(Kubozono 1996,1999b, Kubozono and Ohta 1997)7·8.
Note that the accent pattern in (9c) arises when the two conditions in (10)
are both met: it surfaces neither in words which have the same syllable
structure but not the same segmental structure, as illustrated in (9a), nor
in words which satisfy the segmental condition but not the condition on
118 Haruo Kubozono
Table 1. Percentage of the accent pattern in (9c) shown by each syllable structure
In sum, Tanaka's statistical study confirms the validity of the two condi-
tions in (10). This allows us to conclude that as far as trimoraic loanwords
are concerned, word-final heavy syllables are metrically stronger than
penultimate light syllables if the light syllables contain an epenthetic vow-
el; in contrast, they are metrically weaker than non-final light syllables
containing a non-epenthetic vowel. This reveals an interesting interaction
between syllable structure and segmental structure. Although the precise
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 119
formulation of this interaction is beyond the scope of this paper, the case
in question clearly demonstrates that the nature of epenthetic vowels
must be taken into consideration when generalizing the two accent pat-
terns shown by the trimoraic loanwords in (9), i.e. (9a, b) vs. (9c).
This result is especially interesting if compared with the analysis men-
tioned at the beginning of this section. As shown in (8), the antepenultimate
accent rule for loanwords does not rely on the distinction between epen-
thetic and non-epenthetic vowels; the accent pattern of p<u>'.ra.s<u>
'plus', for example, indicates that epenthetic vowels are fully visible to the
accent rule. Indeed, the very basic idea of 'antepenultimacy' is based on the
assumption that epenthetic vowels are fully visible to the rule just like non-
epenthetic vowels. However, the data given in (9) and Table 1 clearly show
that the same loanword accent rule does distinguish between the two kinds
of vowels under certain circumstances: word-initial epenthetic vowels be-
have as if they did not exist in a particular type of trimoraic words. To be
more precise, moras containing an epenthetic vowel are generally counted
by the mora-counting accent rule, but they somehow avoid bearing an ac-
cent in and only in a certain type of trimoraic words.
The first type that shows a marked tendency for unaccentedness involves
four-mora words which satisfy the two conditions in (11) (Kubozono 1996).
The validity of the generalization in (11) has been confirmed by three inde-
pendent statistical studies: Kubozono (1996) examined the accent patterns
of all of the 206 foreign placenames listed in the appendix to the NHK Ac-
cent Dictionary (1985); Tanaka (1996) looked at all the loanwords and for-
eign placenames in the same dictionary, of which 931 words have four mo-
ras; and Fukui (1998) examined all the foreign loanwords in the new edition
of the NHK Dictionary (1998) in which she found a total of 792 four-mora
loanwords. All these studies confirmed that the distinction between accent-
ed and unaccented words is heavily dependent on the syllable structure in-
volved. To be specific, unaccentedness is dominant in four-mora words sat-
isfying the condition in (11a), but not in four-mora words with different
syllable structures: e.g. a.ri.zo.na 'Arizona', ai.o.wa 'Iowa' vs. a'.ma.zon
'Amazon', ro'n.don 'London'. Table 2 summarizes the statistics reported by
these studies: it shows the percentage of unaccented words as opposed to
accented words exhibited by each of the syllable structures.
isfying the two conditions in (11) are unaccented. These figures are no
doubt extremely high in light of the fact that the same accent pattern ac-
counts for only ten percent of loanwords as a whole.
Stated conversely, four-mora words ending in an epenthetic vowel /u/
are much less likely to become unaccented even if they fulfill the syllable
structure condition in (11a). According to Fukui's data, unaccentedness
only accounts for 30 % of #HLL# and #LLLL# four-mora loanwords end-
ing in this particular type of vowel, which is considerably lower than the
ratio given in Table 2 for #HLL# and #LLLL# four-mora loanwords in
general. Instead of becoming unaccented, these words that involve an ep-
enthetic vowel word-finally tend to be accented according to the tradi-
tional antepenultimate rule. Some examples are given in (12).
among many kinds of vowels is largely invisible to the accent rule responsi-
ble for the unaccented pattern. This fact is puzzling in two ways. First, it is
strange to find that epenthetic /u/ is invisible to the accent rule for unaccent-
ed words, whereas epenthetic HI and loi are visible to the same accent rule.
A second paradox is that while epenthetic /u/ is invisible to this rule, the
same vowel seems visible to the ordinary antepenultimate accent rule. In-
deed, the accent pattern in (12) can only be accounted for by counting all
the moras in the surface structure, irrespective of whether they contain an
epenthetic vowel or not: e.g. s<u>.t<o>'.re.s<u> 'stress', ba.ri'.u.m<u>
'Barium'. One solution to this second paradox may be to assume that the
consonant immediately preceding the word-final epenthetic /u/ bears a
mora on its own while the vowel itself is phonologically invisible. Even this
solution, however, fails to explain why this peculiar behaviour is shown by
epenthetic lui, and not by the other types of epenthetic vowels.
The condition in (13a) alone does not trigger the accent type in question.
According to Fukui (1998), only one-fourth of all LHL trisyllables are un-
accented, which is about the average of all four-mora loanwords (see Ta-
ble 2). If the second condition in (13) is added to this syllable structure
condition, however, the ratio of unaccentedness more than doubles. In
Fukui's data, 41 out of 68 words (i.e. 60%) satisfying the two conditions
in (13) are unaccented. Typical examples are given in (14a), which con-
trast with the exceptions given in (14b) 10 .
The next accent rule to consider is the compound accent rule which is sen-
sitive to syllable structure as well as mora structure (Kubozono 1995,
1997). We will confine ourselves here to noun-noun compounds with a
'short' second member (N2), i.e. those compounds whose first member
(Nl) is more than two moras long and their N2 is only one or two moras
long. Generally speaking, the accentuation of compound nouns in Japa-
nese is determined by the phonological length and accent of N2, irrespec-
124 Haruo Kubozono
(17) Nonfin (head syllable) > Max-accent o Nonfin (head foot) > Edgemostness
(20) Nonfinality (head syllable, head foot) > MAX-accent > Edgemostness
The same result is obtained for words of foreign origin. Thus, bisyllabic,
bimoraic words like ba'.s<u> 'bus' and pi'.r<u> 'pill' pattern with mo.mo
and a.ka' in (22b), with a compound accent on the epenthetic vowel. This
is illustrated in (24).
While epenthetic vowels seem fully visible to the X-taroo accent rule,
elided consonants display somewhat peculiar behaviour. Bimoraic words
such as those in (6a) exhibit the compound accent pattern in (22b) and
not the pattern in (22a). This is illustrated in (25)15. In other words, these
words behave as if they were bisyllabic although they have been turned
into monosyllables by the historical process of consonant elision.
For many speakers of Tokyo Japanese, this accent pattern contrasts with
the unaccented pattern shown by SJ morphemes which are originally
monosyllabic and bimoraic. This latter pattern is shown in (26).
Recall that the native Japanese bimoraic words in (25), i.e. koi, tai and kai,
behave as monosyllabic words in the general compound accent rule de-
scribed in (21), where they form the second member of noun-noun com-
pounds. It is somewhat paradoxical, therefore, that the same morphemes
exhibit the accentual behaviour characteristic of bisyllabic words in the
compound name, X-taroo.
Finally, the compound accent patterns sketched in (22) provide an in-
teresting insight into the nature of the epenthetic vowels in the three-
mora loanwords in (9c). As discussed in section 3.1., trimoraic loanwords
are exceptionally accented on the penultimate mora if this mora is in a
heavy syllable and is preceded by an epenthetic vowel. This exceptional
accent pattern suggests that the epenthetic vowel is not fully visible to the
accent rule or, at least, that the epenthetic vowel is metrically lighter than
non-epenthetic short vowels. This account does not hold, however, when
the same trimoraic words form the initial member of X-taroo compounds.
As shown in (27), these words exhibit the accent pattern in (22c) when
combined with ta'roo to form a X-taroo compound. Thus, as far as the ac-
centuation of X-taroo compounds is concerned, trimoraic loanwords such
b<u>.ru'u 'blue' and d<o>.ra'i 'dry' pattern with ba'.ree 'volleyball',
which does not contain an epenthetic vowel.
the final syllable of the initial element. This second accent pattern is illus-
trated in (28b).
(29) a. SJ morphemes
ra.k<u>' + i.ti' -> ra.k<u>'.i.ti 'comfort, iti; Rakuiti'
e'.k<i> + i.ti' ->• e.k<i>'.i.ti 'station, iti; Ekiiti'
b. Foreign loanwords
ba'.s<u> + i.ti' ->· ba.s<u>'.i.ti 'bus, iti; Basuiti'
pi'.r<u> + i.ti' ->• pi.r<u>'.i.ti 'pill, iti; Piruiti'
The accent pattern illustrated in (29) suggests that the bimoraic words in
question behave like a bisyllabic word (e.g. hi.ko in (28b)), rather than a
monosyllabic word (e.g. koo in (28a)). This, in turn, suggests that the ep-
enthetic vowels in question are visible to the accent rule.
While the accent rule under consideration applies to the output of the
vowel epenthesis rule, the same rule seems to apply to the input of the
consonant elision rule to which native Japanese morphemes were subject.
Monosyllabic, bimoraic morphemes of native origin typically exhibit the
accent pattern in (28b), as illustrated in (30).
The next accent rule we will look at is that of verbs and adjectives. As far
as end forms ('syuusikei') are concerned, verbs and adjectives in Tokyo
Japanese fall into two accentual groups: those accented on their penulti-
mate mora and those which are unaccented. These are exemplified in
(31a) and (31b), respectively. The accentedness of verbs and adjectives is
largely unpredictable and, hence, must be specified in the lexicon. Note
also that accented verbs and adjectives are accented one mora closer to
the end of the word than nouns, which we saw in section 3.1. above.
(31) a. mi'.ru 'to look', ha.na'.su 'to speak', kan.ga.e.'ru 'to think'
u.ma'i 'tasty', a.tu'i 'hot', mi.zi.ka'i 'short'
b. to.bu 'to fly, to jump', wa.ta.ru 'to cross', ku.ra.be.ru 'to compare'
a.mai 'sweet', a.tui 'thick', tu.me.tai 'cold'
While most verbs and adjectives follow either of the two accent patterns
in (31), some words are accented on their antepenultimate mora as exem-
plified in (32).
Finally, let us consider the rule called 'pre-no deaccenting rule' (Akinaga
1985, Poser 1984, Kubozono 1999b). This accent rule has the effect of de-
leting an accent from the very final syllable of nouns. A s exemplified in
(33), it applies only if the noun is followed by the genitive particle no
(hence the term 'pre-no deaccenting rule'). (ga is the nominative particle).
This deaccenting rule offers a very good test for our discussion in this pa-
per since it fails to apply to monosyllabic nouns. This is illustrated in (34).
Recall that all these bisyllabic forms originated from monosyllables his-
torically, with a vowel inserted in the morpheme-final position. There-
fore, the morphemes in (36b) are historically equivalent to the SJ mor-
pheme ho'η 'book' in (34b), which remains monosyllabic in modern
Japanese, but they synchronically involve the same syllable and accent
structure as ha.na' 'a flower', which is a finally-accented bisyllabic word
of native origin given in (33a).
An analysis of the accentual behaviour of the bisyllabic morphemes in
(36b) reveals that these morphemes exhibit the accent pattern in (33),
and not the one in (34), when preceding the genitive particle no. In other
words, most speakers of Tokyo Japanese readily apply the pre-no deac-
centing rule to bisyllabic SJ morphemes, as shown in (37).
The fact that the SJ morphemes in question pattern with the bisyllabic ha-
ría' and not with the monosyllabic ho'η suggests that the epenthetic vow-
els contained in these morphemes are readily visible to the pre-no deac-
centing rule. This consequence is in full accordance with the findings
reported in (23) and (29a). However, it appears contradictory to the fact
reported in section 3.3, namely, the fact that bisyllabic SJ morphemes with
an initial accent pattern with accented monosyllabic morphemes when
they form the second member of noun-noun compounds. It is interesting
and, at the same time, quite puzzling to find that one and the same epen-
thetic vowel is visible to some accent rules but not to others.
4. Conclusion
In this paper we have seen how epenthetic vowels and elided consonants
interact with various accent rules that are more or less productive in mod-
ern Tokyo Japanese. The traditional idea about the interaction between
synchronic accent rules and diachronic segmental changes is that epen-
thetic vowels are fully visible, and elided consonants are no longer visible
to the synchronic accent rules. Namely, accent rules apply in the synchro-
nic grammar to the output, and not to the input, of historical segmental
changes. However, a careful analysis of accentual behaviour of words
with an epenthetic vowel or an elided consonant has shown that at least
some epenthetic vowels are invisible and some elided consonants are vis-
ible to some accent rules; i.e. some epenthetic vowels behave as if they did
not exist and some elided consonants behave as if they were still present.
The main facts are summarized below.
All these facts involve a paradox of one kind or another and, indeed,
pose difficulties for theoretical analyses. While a solution to one of these
paradoxes was hinted at in section 3.3., it remains an open question as to
how all the other paradoxes can be solved within an existing theoretical
framework, especially within the output-oriented framework of Optimal-
l y Theory (Prince and Smolensky 1993).
Acknowledgements
Earlier versions of this paper were read at a regular meeting of the phonology circle at the
Tokyo University of Foreign Studies (AA) and at the Spoken Language Working Group in
Nara. I am indebted to the audiences of these meetings, particularly to Zendo Uwano, Te-
ruhiro Hayata and Miyoko Sugito, for their valuable comments. I would like to thank the ed-
itors of this volume and the two anonymous reviewers for a number of useful comments and
suggestions for the improvement of this paper. I am also grateful to Frank Owens for check-
ing the original manuscript. All the errors that remain are my own. The work reported on in
this paper was partly supported by research grants from ATR Interpreting Telecommunica-
tions Research Laboratories and the Grant-in-Aid for Scientific Research from the Japanese
Ministry of Education, Science, Sports and Culture (Nos. 09610480,10044010,11111102).
Notes
1. In the rest of this paper I will refer to these two types of loanwords in Japanese as 'SJ mor-
phemes' (or 'SJ words') and 'foreign loanwords' (or simply 'loanwords'), respectively.
2. Tanaka (1999) reports that epenthetic vowels in loanwords in Japanese behave different-
ly from underlying vowels in music, too. In Japanese songs, there is usually a one-to-one
correspondence between the musical note and the mora in words, so that a musical note
is assigned to every mora, and vice versa. However, this correspondence does not neces-
sarily hold if the mora contains an epenthetic vowel; namely, epenthetic vowels often be-
have as if they did not exist in the mora-note alignment. Tanaka adds that this is a rela-
tively new tendency which is most eminently found in songs composed in the past two
decades.
3. From a theoretical viewpoint, Alderete (1995) argues for a constraint in the grammar
that dislikes epenthetic vowels in stressed positions—a kind of Head Faithfulness that
connects with a general theory of Positional Faithfulness. If this constraint is high-rank-
ing, then the prediction is that stress rules will avoid epenthetic vowels when counting syl-
lables. In descriptive terms, Michelson (1981,1988) and Potter (1994) report a similar ef-
fect of stress-epenthesis interaction in Mohawk. An interesting case is found in the Saru
dialect of Ainu, an almost extinct language of Hokkaido (northern Japan), where accent
rules apply before the rule of vowel epenthesis. Selkirk (1999) discusses a more compli-
cated case of stress-epenthesis interaction in Makassarese and Selayarese, where some
words appear to undergo stress assignment after vowel/consonant epenthesis while other
words appear to be derived in the reverse order. Similarly, Uwano (1992) gives some in-
teresting examples from Kagoshima Japanese, a syllable-based dialect spoken in south-
ern Japan (Kubozono 1999b), where accent (or tonal) rules exceptionally apply to the in-
put of a consonant deletion rule, despite the fact that they apply to the output of vowel
deletion rules.
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 137
4. See Tateishi (1989) and Itô and Mester (1996) for the linguistic factors responsible for the
choice of these two vowels. For the phonetics and phonology of vowel devoicing in Japa-
nese, see Beckman and Shoji (1984), Kubozono (1999a) and the references cited therein.
5. This antepenultimate accent rule accounts for the accentuation of most accented nouns
in Japanese, whether they are loanwords or words of other origins, i.e. native Japanese
or Sino-Japanese words (Tanaka and Kubozono 1999).
6. Note that /u/, not loi, is inserted after [t] in /tu.in/ 'twin' in (9c). This is probably because
of the semivowel /w/ that immediately follows the dental consonant in the source word.
7. Different scholars have proposed different generalizations about the accent pattern in
(9c). Akinaga (1985), for example, states that words ending in a 'tokusyuhaku' (a spe-
cial mora) such as a moraic nasal and a long vowel tend to attract an accent on the pen-
ultimate mora. While this rule is strikingly similar to the syllable-based generalization
proposed in (10a), it cannot fully capture the distribution of the accent pattern in (9c)
as opposed to the pattern in (9a). Quackenbush and Ohso (1990), in contrast, generalize
that words ending in a vowel in the source language tend to give rise to the accent pat-
tern in (9c). This analysis, too, is an overgeneralization of the fact because we recognize
a large number of words that fulfill this segmental requirement but nevertheless show
the ordinary antepenultimate accent pattern: e.g. ki'.raa 'killer' in (9a). More recently,
Tanomura (1999) suggests that the generalization in (10a) is too strong because the ac-
cent pattern in (9c) is not generally applicable to words ending in a moraic nasal: e.g.
p<u>'.ran 'plan' (vs. t<u>.i'n 'twin'). While this observation may be correct, it does not
argue directly against the syllable-based generalization in (10a). Because syllables with
a moraic nasal are metrically weaker than other types of heavy syllables in Japanese in
general (Kubozono 1999a, c), it should not be surprising to find that moraic nasals be-
have in the way Tanomura points out.
8. Some may argue against the generalization in (10b) and propose to attribute the pecu-
liar accent pattern in (9c) to the fact the vowels in the initial syllable/mora tend to be
devoiced or are of very low sonority. These alternative analyses can be rejected in a rea-
sonable way. First, the initial vowels in question are not always devoiced nor are they
always low in sonority. For example, unlike HI and /u/, loi does not generally get de-
voiced in Japanese but nevertheless triggers the penultimate accent: e.g. d<o>.ra'i
'dry'. In addition, lui is not prone to devoicing when it is preceded or followed by a
voiced consonant, but nevertheless gives rise to the peculiar accent pattern in (9c): e.g.
b<u>.ru'u 'blue'. Second, underlying HI and lui do not trigger the accent pattern in (9c)
although they are, other things being equal, just as subject to devoicing as their epen-
thetic counterparts: e.g. pu'.rin 'pudding', ki'.raa 'killer' and ri'.ree in (9a).
9. There are two groups of exceptions to this generalization. First, such words as
h<u>.ro'.a 'floor', s<u>.to'.a 'store', and si.ka'.go 'Chicago' are accented on the penul-
timate syllable/mora although they have the same (or similar) syllable structure as the
words in (9b). Second, such words as p<u>'.ra-n 'plan' exceptionally allow the the ac-
cent pattern in (9a) as a relatively new pronunciation along with the pattern in (9c),
p<u>.ra'-n.
10. A comparison of the examples in (14a) with those in (14b) suggests that the unaccented
accent pattern is more likely to occur if the initial light syllable as well as the final sylla-
ble contains an epenthetic vowel.
11. A functional account may be that this peculiar behaviour of SJ morphemes has to do
with the fact that they alternate with monosyllabic forms in the synchronic grammar.
Many, if not all, SJ bisyllabic morphemes are often realized as a monosyllabic form if
they are followed by a voiceless consonant in compounds: e.g. ga'.ku + koo -* gak.koo
'learning, school; school', se'.ki + ken -· sek.ken 'seat, roll; to overwhelm'. In compari-
138 Haruo Kubozono
son, foreign morphemes do not show such a synchronic alternation; for example,
bu.s<u> and ki.s<u> in (19) never manifest their original monosyllabic form in the
grammar of modern Japanese. In theoretical terms, this functional account is based on
the idea of output-output correspondence.
12. Some may object to this generalization, claiming that the dictinction between the two
accent patterns in (22a, b) has to do with whether the N1 of X-taroo compounds is a free
morpheme or not, i.e. whether it is used in isolation. It is true that many of the first
members in (22a) are bound morphemes and those in (22b) tend to be free morphemes.
However, this account does not hold in many other cases. There are monosyllabic free
morphemes which exhibit the accent pattern in (22a): e.g. ran.ta.roo 'riot, taroo; Ranta-
roo', kin.ta.roo 'gold, taroo; Kintaroo'. There are also many bisyllabic bound mor-
phemes which show the accent pattern in (22b): e.g. riki'.ta.roo (or ri.ki.ta'.roo) 'power,
taroo; Rikitaroo'. These instances suggest that the morphological status of N1 is not di-
rectly relevant to the dictinction between the two accent patterns in (22a, b).
13. Note that many of the words involving this mora/syllable structure admit the accent pat-
tern in (22c) alongside the pattern in (22b). This tendency is most evident when the
vowel in the second syllable of N1 is devoiced, i.e. either li/ or /u/: e.g. su.si' + ta'.roo >
su.si'.ta.roo, su.si.ta'.roo 'susi, taroo; Susitaroo'. This accentual variation, however, does
not affect the discussion to be developed below.
14. Some speakers of Tokyo Japanese seem to prefer the accent pattern illustrated in (22c) for
these compounds and the compounds in (24) below, e.g. ra.k<u>.ta'.roo, ba.s<u>.ta'.roo.
This variant accent pattern can probably be attributed to the phonetic devoicing of the ep-
enthetic vowel.
15. Note that some speakers of Tokyo Japanese, especially young people, seem to permit
an unaccented pattern alongside this accent pattern: e.g. tai.ta.roo and tai'.ta.roo. Most
of these speakers, however, do not admit the accented pattern in (25) for the compound
expressions in (26).
16. The syllable structure of foreign morphemes cannot be tested by this deaccenting rule
since bisyllabic foreign morphemes, whether containing an epenthetic vowel or not, are
generally accented on their initial syllable: e.g. ba'.s<u> 'bus', pi'.za 'pizza'.
References
Alderete, J.
1995 Faithfulness to prosodie heads, ms. University of Massachusetts, Amherst.
Akinaga, K.
1985 Kyootuugo no akusento (Accent of Standard Japanese). In NHK Pronuncia-
tion and Accent Dictionary.
Beckman, M. and A. Shoji
1984 Spectral and perceptual evidence for CV coarticulation in devoiced /si/ and
/syu/ in Japanese. Phonetica 41, 61-71.
Fukui, M.
1998 A remark on unaccented loanwords in Kansai Japanese, ms. Kobe University.
Hayes, B.
1995 Metrical Stress Theory: Principles and Case Studies. Chicago: University of
Chicago Press.
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 139
Sibata, T.
1994 Gairaigo ni okeru akusentokaku no iti (On the location of accent in loan-
words). In K. Sato (ed.), Gendaigo Hougen no Kenkyuu (Studies in Modern
Japanese and Dialects). Tokyo: Meiji Shoin. 338-418.
Tanaka, Shin'ichi
1996 Nihongo no onsetu koozoo to on'in gensyoo (Syllable structure and phonolog-
ical phenomena in Japanese). M A thesis, Osaka University of Foreign Studies.
1999 Kayoo ni okeru gairaigo rizumu no hensen. (The development of loanword
rhythm in songs). A paper presented at the regular meeting of the Phonetic
Society of Japan, held at Keio University on June 12.
Tanaka, Shin'ichi and H. Kubozono
1999 Introduction to Japanese Pronunciation: Theory and Practice. Tokyo: Kurosio.
Tanaka, Shin-ichi and N. Yamane
1999 Paradox resolution and unaccentedness in foreign and Sino-Japanese com-
pounds. Paper read at the 24th meeting of Kansai Linguistic Society (KLS),
Kwansei Gakuin University, October 23,1999.
Tanomura, T.
1999 Gairaigo akusento ni okeru gengo no hatuon no kanyo ni tuite. (On the influ-
ence of the pronunciation of the source language on loanword accentuation).
Nihongo Kagaku 5, 67-88.
Tateishi, K.
1989 Phonology of Sino-Japanese morphemes. University of Massachusetts Occa-
sional Papers in Linguistics 13, 209-35.
Trubetzkoy, N.S.
1969 Principles of Phonology (trans. C.A.M. Baltaxe). Los Angeles: University of
California Press. (Original work published 1958).
Uwano, Z.
1992 Kagosima Hukiage-tyoo hoogen no hukugoomeisi no akusento (Accentua-
tion of compound nouns in Kagoshima Fukiage dialect). In T. Kunihiro (ed.),
Nihongo Intoneesyon no Zittai to Bunseki (Facts and Analyses of Japanese In-
tonation) (Grant report). 91-208.
Yoshida, Y.
1999 Kanzi on to akusento (Sino-Japanese sounds and accent). Paper presented at
the March meeting of the Phonological Association in Kansai (PAIK).
Prosodie Structure and Sandhi Phenomena
in the Saru Dialect of Ainu1
Hidetoshi Shiraishi
Sakhalin Museum of Regional Studies
1. Introduction
In both (la) and (lb), the structural description of the rule is met. Never-
theless, Liaison applies only to the former. This contrast is rooted in the
difference of the syntactic structure between (la) and (lb): in (la), the Li-
aison context is located within a single syntactic phrase (modifier-head)
whereas in (lb), it spans two phrases, an inappropriate domain for Liaison.
On the other hand, a number of phonological processes are blind to such
syntactic structure. Flapping rule of American English belongs to this type.
It is applicable even across the largest syntactic unit, the sentence.
(2) Have a seat. I'll be right back. > ... sea[r] I'll...
(N&V 1986: 236)
(3) n-alternation
a. ror -un-so > roru[y] so
place of honor-at- seat
'the seat at a place of honor in a house'
b. pon yuk > po[y] yuk
little deer
'little deer'
c. yayan wakka > yaya[w] wakka
normal water
'normal water (in contrast with water from a hot spring)'
d. . . . a n w a . . . > a[m m]a
to be CON 4
'... exists and ...'
e. ... an yakka ... > a[y] yakka ~ an akka
to be CON
'...even though...'
(4) r-alternation
a. ar- rametok > a[n]rametok
real-bravery
'a real bravery'
b. yar nima > ya[n] nima
bark tray
'a tray made from bark'
c. asir cise > asi[t] cise
new house
'a new house'
d. kusur ta > kusu[t] ta
kusur LOC
'at Kusur (place name)'
(5) n-alternation
a. inani un hoski arpa=an kor pirka kus hawas sekor yaynu=an //
which to first go 4 CON good PURP sounds COMP think 4
sino wen iruska poka nesi a=ki
real bad anger even very 4 do (KT6 Kayano 1998a: 110)
Ί could not decide where to go first. I was completely frustrated.'
b. e=motoho a=nukar wa an= an //
2 origin 4 look CON to be 4
yaun mosir un iwor kor kamuy a=ne wa ...
land country to field have spirit 4 be CON (KT Kayano 1998c: 58)
Ί am looking at your origin. I am the spirit governing this field in this country
c. kanna ruyno ye yan // ye yan sekor
once again say FP: IMP COMP (KKo Kayano 1998a: 74)
'Say it again. Say it again'
146 Hidetoshi Shiraishi
(6) r-alternation 7
a. ne sinrici ka e=kopuspakar _ nokan uypehe ka opitta usa muni
that root also 2 dig fine chips also all various garbage
turano e=uhuyka.
together 2 burn (KKo Kayano 1974:145)
'You dig that root. Together with the fine chips you burn all of them'
b. kotankonnispa sine matnepo kor _ nea matnepo
the village head one daughter have that daughter
ramutu uk tek hine ...
life pull out briefly CON
(HF Tamura 1985: 60)
'the village head had one daughter. [My brother] pulled out
the life of that daughter ...'
c. nea niatus heyasi a=ninpa kor _ rapokke sinki= an hine ...
that pail towards a bank 4 drag CON then to be tired 4 CON
(KT Kayano 1974: 99)
Ί dragged that pail to the bank and then became tired ...'
d. isepo kuari cironnup kuari a=eykoysanpa kor _ cironnup
rabbit trap fox trap 4 imitate CON fox
ka isepo ka a=rayke
too rabbit too 4 kill
(KKo Tamura 1988a: 12)
Ί imitated [my father's] rabbit traps and fox traps and caught rabbits
and foxes.'
e. eci= kor mosir sekor eci= haweoka ka eaykap kunihi eci=
2PL have country COMP 2PL say also impossible NOM 2PL
ramu kor _ tane anakne somo uhekote itak=an kusu ne na
think CON now TOP NEG each other talk 4 will COP FP
(HS Tamura 1986: 46)
'Keep in mind that you cannot insist that it is your country
[if lots of people are gathering] and don't talk about this any longer.'
In both (5) and (6), the structural description of the rules is met, and yet
only r-alternation occurs. T h e apparent difference with the e x a m p l e s in
(3) and (4), w h e r e n o such discrepancy was observed, is their syntactic
construction. In (5) and (6), the relevant segments b e l o n g to separate sen-
tences, i.e. the target and trigger are separated by a s e n t e n c e boundary. It
should also b e n o t e d that the t w o words in question are not separated by
a pause.
Prosodie Structure and Sandhi Phenomena in the Saru Dialect of Ainu 147
It is worth noting that the length of the sentence does not matter for the
application of n-alternation. Consider the following.
b. sasun _ sir
to have descendants-NOM
(part of an idiom)
(9) a. tan _ ya ta
this land LOC
'at this land'
b. iwan _ suy
six times
'six times (a sacred number of the language denoting 'many')'
c. pon _ suma
little stone
'little stone'
3.1. Frequency
sists of two beats. With respect to the sequence of this metrical unit con-
sisting of two beats, the following regularity is observed.
(21) a) The left edge of a rhythmic unit should be aligned with the left
edge of a metrical unit, i.e. the first beat. 16
b)The right edge of a rhythmic unit should be placed somewhere be-
tween the second beat and the first beat of the next metrical unit.
(Okuda 1988: 40, slightly modified by the author)
(22) a II I
(poron no} 'many'
b * II I II
{po ron no}
(where || denotes the first and the second beat. The rhythmic unit is
contained in curled brackets)
(Okuda 1988: 42-43)
The next question, then, is which syntactic unit may initiate this rhythmic
unit (and which may not). As mentioned above, Okuda attributes to free
morphemes: noun, verb, conjunction, interjection, adverb, etc. On the
other hand, bound morphemes (postpositions, particles) can enter the
rhythmic unit only by cliticizing to free morphemes (1988: 37-39).
Of interest to us is the structural similarity of this rhythmic unit with the
unmarked domain of n-alternation. Similar to the latter, Okuda's rhyth-
mic unit does not require fully-fledged syntactic information (such as cat-
egory labels), as the division of free versus bound morpheme indicates.
This is exactly what we expect if the prosodie phrasing makes use only of
impoverished syntactic information. In addition, modifiers seem to form a
single rhythmic unit with its head, even though in that case the left edge of
the latter does not align with the first beat, in violation of generalization a).
(23) II I II I
{poro sin to ko} 'a big chest'
big chest
(Okuda 1988: 54)
Prosodie Structure and Sandhi Phenomena in the Saru Dialect of Ainu 153
Here the left edge of sintoko 'chest' is not aligned with the first beat even
though this can initiate a new rhythmic unit since sintoko itself is a free
morpheme (noun). This extraordinary alignment can be accounted for if
we regard the whole modifier-head phrase as forming a single rhythmic
unit. Now, recall that such a modifier-head phrase was a common n-alter-
nation domain. This 'coincidence' further supports our claim that the
same prosodie unit is provided for versification of Yukar as well as for seg-
mental rules such as n-alternation.
On the other hand, it is also true that Okuda's rhythmic unit and the
domain of n-alternation do not show a perfect match. A major discrepan-
cy between them is the phrasing of the evidential nominalizer siri (see 18).
According to Okuda, siri is usually phrased with the preceding free mor-
pheme (predicate verb), while this was never an n-alternation context in
our data. However, it seems also to be true that siri does provide an n-al-
ternation context for this speaker (Okuda p.c.), in which case this partic-
ular phrasing does not contradict our hypothesis.
Another disparity can be seen in the phrasing concerning relative
clauses. The investigation of relative clause constructions in versification
reveals that the verb in the relative clause (which is necessarily the final
constituent within the relative clause) is phrased with the following head
noun.
Since a free morpheme casi is left-aligned with the second beat, the rhyth-
mic unit should be interpreted as initiated by the preceding kor (other-
wise there is a violation of generalization a)). Recall, however, that the
relative clause-head noun sequence context never provides an n-alterna-
tion context (see 19). However, it should also be noted that the total num-
ber of relative clause constructions is small in our primary source of re-
search.
Although the details of the domain for n-alternation and Okuda's
rhythmic unit are not isomorphic to each other, it remains a fact that they
exhibit a certain degree of similarity, which we consider a worthwhile top-
ic for future research.
154 Hidetoshi Shiraishi
4. Conclusion
Acknowledgments
A great deal of this paper constitutes a part of the author's M. A. thesis submitted to the Chi-
ba University Graduate School in September 1998.1 am grateful to the member of the thesis
committee, Professors Tohru Kaneko, Hiroshi Nakagawa and Kenji Kanno. Part of this pa-
per was presented at the Hoppogengobunka Kenkyukai (October 1997). Thanks to the par-
ticipants for their comments, especially Professor Suzuko Tamura and Hideo Kirikae. Many
thanks to Professor Osami Okuda for helpful comments. Thanks are also to my colleague at
the University of Groningen, Tjeerd de Graaf, Dirk-Bart den Ouden and especially Wouter
Jansen for hours of stimulating discussion. All the errors are my own. The research was par-
tially supported by the Netherlands Government scholarship (NUFFIC).
Notes
1. Ainu : an endangered language of Japan whose genetic affiliation with the neighboring
languages is unknown. The Saru dialect refers to the dialect once spoken in the villages
alongside the river of Saru, southwest Hokkaido.
2. _ = application, // = blocking of Liaison and of any other phonological process in this pa-
per.
Prosodie Structure and Sandhi Phenomena in the Saru Dialect of Ainu 155
3. We will use the conventional writing system of Ainu throughout this paper. The follow-
ing correspondences should be noted: y = IPA [j], r = IPA [r]. Hyphens indicate mor-
phological boundaries. Segments that have undergone alternation are bracketed.
4. Abbreviations: 2 = second person personal prefix, 4 = fourth person personal affix, in-
dicating the first person in the oral literature, hence in the examples below. These per-
sonal affixes are separated with a double hyphen (=) in the text. C O M P = complemen-
tizer, C O N = conjunctive particle, C O P = copula, E V I D = evidential, F P = final particle,
IMP = imperative, L O C = locative, N E G = negative, N O M = nominalizer, OBJ = ob-
ject, PL = plural, P U R P = purpose, T O P = topic. ~ indicates variation.
5. A feature-geometric approach following the model of Padgett (1991) is proposed by
Shiraishi (1998).
6. Hereafter we will provide the initials of the speakers.
7. The astute reader would notice that the examples (c-e) of r-alternation all involve the
conjunctive particle kor 'and, while'. This is due to the unbalanced number of r-alterna-
tion contexts, i.e., the r-n context outranks the other three (r-c, r-t, r-r), and it is difficult
to find an example without kor for the latter.
8. 'Adjectival verbs' refer to prehead intransitive verbs modifying the following head
noun. Ainu has no adjectives (morphologically speaking).
9. Of the 18 speakers, 7 show n-alternation across a phrase boundary and 5 do not. The
rest lack data of the context under discussion.
10. Note: identical expressions within a single speaker are counted as a single context. If
there is variation, it is counted as a case of application.
11. The evidential nominalizer can be regarded as a subcase of relative clause construction
as siri derives from the noun meaning 'appearance, state'.
12. Postpositional stranding as seen in this example is a common strategy for relative clause
construction in Ainu.
13. Examples of relative clause constructions containing relevant segments for n-alterna-
tion are, however, quite rare in our primary source.
14. This applies to the conjunctive particles yakne, yakka and yakun as well. We still have
no answer why these function words show unstable application of n-alternation for
some speakers, being inconsistent with the prediction of most literature on phrasal pho-
nology (e.g. Selkirk 1995).
15. The Shizunai district lies about 30km to the southeast of the Saru district and its dialect
differs slightly from that of Saru. A comprehensive description of the Shizunai dialect
has been published in 1986 by Kirsten Refsing.
16. Okuda's original principle contains an alternative: the left edge of the rhythmic unit can
also be placed somewhere within the latter half of the second beat. Since this is irrele-
vant for our discussion, it is ignored here.
References
Hayes, B.
1989 The Prosodie Hierarchy in Meter. In P. Kiparsky and Y. Gilbert (eds.), Pho-
netics and Phonology 1 Rhythm and Meter, 201-260. San Diego: Academic
Press.
Inkelas, S. and D. Zee (eds.)
1990 The Phonology-Syntax Connection. Chicago: The University of Chicago
Press.
156 Hidetoshi Shiraishi
1988b Ainugo [The Language of Ainu] In. T. Kamei et al. (eds.), The Sanseido En-
cyclopaedia of Linguistics, vol. 1, 6-94. Tokyo: Sanseido.
1989 Ainugo onsei shiryo [Phonetic materials of the Ainu] 6. Tokyo: Waseda Insti-
tute of Language Teaching.
1996 Ainugo jiten [The Ainu-Japanese Dictionary Saru dialect] Tokyo: Sofukan.
1997 Ainugo onsei shiryo [Phonetic materials of the Ainu] 10. Tokyo: Waseda In-
stitute of Language Teaching.
The Emergence of the 'Unaccented':
Possible Patterns and Variations
in Japanese Compound Accentuation
Shin-ichi Tanaka
Nagoya University
1. Introduction
From this brief review, it is evident that any formal analysis must reflect
these three properties in some way or other.
However, it is also the case that they do not always hold well as excep-
tion-free generalizations, since they often conflict with one another. For
instance, nagoyási 'Nagoya City' and rondónbasi 'London Bridge' do not
bear accent on N2, favoring the factors in (2) and (3) over the one in (1).
Conversely, such examples as bitámin + sii bitaminsíi 'vitamin C' and
biggu + bén biggubén 'Big Ben' carry over the lexical accent of their
head nomináis, preferring (1) to (2) and (3). In this sense, these three gen-
eralizations seem to hold true in some cases but to have exceptions in oth-
ers, and it matters in what case one factor is dominant over the others.
Thus, our task is to explicitly show how these characteristics can be cap-
tured by violable, hopefully general and well-motivated constraints and
how such constraints are ranked with respect to one another. Needless to
say, OT will be a promising framework for giving a principled account of
Japanese accent, which has long been considered too complicated to cap-
ture in a rule-based framework where any exceptions or violations to spe-
cific generalizations are not allowed, or constraints are inviolable.
162 Shin-ichi Tanaka
There are, of course, other factors in computing compound accent, but the
above three are particularly important and crucial, as is clear from Kubo-
zono's (1995, 1997) analysis. He proposes the constraints and associated
ranking relations in (4) and (5) to account for the accent distribution of
compounds with either long or short heads.1 The generalizations in (1), (2),
and (3) correspond roughly to (4a), (4b), and (4d), respectively:
(6) Examples
misó) + (siru) *! **
isi) + (a)(táma) * *! *! *
isi) + (á)(tama) * **
sinzyu) + (gai) * *! **
d. /minami + amerika/
minami) + (ame)(riká) **1 * *
minami) + (ame)(ríka) *! *! *
minami) + (amé)(rika) *! **
As shown in (6a), compound accent basically carries over the lexical ac-
cent of N2 unless it is final at the moraic or syllabic level. If N2 is final-
accented at that level as in (6b, c) or unaccented as in (6d), compound ac-
cent falls on the 'around-the-boundary' syllable within the non-final
rightmost foot. Here and below, I am assuming that foot construction pro-
ceeds from right to left (Poser 1990 and Kubozono 1995, 1997) within a
single morpheme and not across a morpheme boundary indicated here as
+ (Tateishi 1985 and Tanaka 1992). The morphology-sensitive foot pars-
ing is motivated by such examples as nagoyá)+(si), but not *nagó)(ya+si)
'Nagoya City'.
Note in the above tableau that *sinzyugái is final-accented at the syl-
labic level and thus has a single violation of N O N - F I N A L I T Y (μ\ Σ ' ) while
*misosirú is final-accented at both the moraic and syllabic levels, causing
two violations of the constraint. N O N - F I N A L I T Y (F') is motivated by the re-
traction of accent to the penultimate foot in (6b, c), which clearly shows
evidence for metrical feet from the viewpoint of accentual phonology, as
compared to oft-cited evidence from prosodie morphology (Tateishi
1989, Poser 1990, and Itô, Kitagawa, & Mester 1996, among others).
RIGHTMOSTNESS (4e) requires that, other things being equal, accent be lo-
cated as far to the right as possible, so that unaccented words have as
many violations to (4e) as their number of syllables.4 Finally, it does not
matter here whether Japanese has iambic or trochaic feet, as will be dem-
164 Shin-ichi Tanaka
onstrated in section 3.2., because FOOT TYPE is much less dominant (hence,
is not given here) in its ranking system. In fact, what we have called 'ac-
cent' so far is not the head of a foot but the head of a prosodie word.
In more general terms, this analysis regards accentual faithfulness as
quite crucial and amounts to saying something like (7):
Note that in ( l i b ) , nihónzin, isíkai, and kasíkin are also acceptable vari-
ants and that other words also bear accent on the non-final rightmost foot
when they undergo further compound formation as in mokugekisyóugen
'eyewitness's evidence,' denwabángou 'phone number,' sankakukánsuu
'trigonometrical function,' and so on. Thus, it can be said that (8) holds
true generally for compounds with Sino-Japanese heads.
and kurisumasutúrii, but (12a) and (12b) generally do not undergo accent
shift even when they are compounded with another word (unlike the
Sino-Japanese cases in (lib)), as in koukyuukafebáa 'high-class cafe bar,'
koukyuuroosuhámu 'high-quality roast ham,' and the like. The examples
in (12c) are nativized loans whose accent falls on the penultimate foot,
but such types are observed less frequently. It thus follows that the un-
marked pattern for foreign compounds parses lexical accent regardless of
its finality.
Next, the following three pieces of evidence for the ranking in (9a) con-
cern quadrimoraic heads whose lexical accent is originally non-final but
moves leftward when compound formation applies to them. For example,
the penultimate accent in (13a) shifts leftward obligatorily without varia-
tion (cf. (10), (lia), and (12b)). The antepenultimate accent of quadrimo-
raic heads in (13b) usually exhibits variation when they form compounds,
but in either case compound accent appears within the penultimate foot. 11
Moreover, compounds with unaccented quadrimoraic heads as in (13c)
always receive the same accent as (13a):
Penultimate-Accented Unaccented
yakatá) + (bune) 'houseboat' vs. abaré) + (usi) 'spirited cow'
onna) + (gó)(koro) 'women's heart' vs. yama) + (zá)(kura) 'mountain cherry tree'
denki) + (nóko)(giri) 'power saw' vs. sin) + (yóko)(hama) 'New Yokohama'
cent on the same non-final rightmost foot. Thus, unaccented heads seem
to reflect the unmarked pattern of compound accent.
Finally, let us reexamine Kubozono's (1995,1997) proposed ranking in
(5), repeated here as (15a), in terms of its reranking possibilities which
would be necessary to account for the data so far:
b. PARSE promotion: PARSE (accent) » NON-FINALITY (μ\a', F1), ALIGN-CA ... for (lib) and (12a)
C . PARSE d e m o t i o n : NON-FINALITY (μ\σ\ F ' ) > PARSE (accent), ALIGN-CA ... for (10), (11a), and (13a)
(15a, b) and (15c) would be needed for preserved accent and penultimate-
foot accent, respectively. In such parsed-accent compounds as meisyó
'place of scenic beauty,' kyuusyó 'vital spot,' and sinzitumi 'truth' in ( l i b ) ,
the reranking of N O N - F I N A L I T Y ( M ' ) » P A R S E (accent) » N O N - F I N A L I T Y (σ1,
F') (instead of (15b)) would be useless, because their accent is final at the
moraic level (note 9). In general terms, the reranking system in (15) seems
to be more complicated than the one proposed in (9a, b) in that (15a)
would have to be reranked in the three ways of (15b-d). Furthermore, the
reranking in (15d) would involve some 'non-local' process, as shown in
(16a, b), although the optimal accent in (16b) is actually a very productive
pattern. The local reranking of N O N - F I N A L I T Y ( F ' ) » P A R S E (accent),
A L I G N - C A in (16c), however, would have only the same result as (16a):
neri) + (hami)(gaki)
172 Shin-ichi Tanaka
neri) + (hami)(gáki) * * *
*!
neri) + (hamí)(gaki) * **
neri) + (hámi)(gaki) *
nerí) + (hami)(gaki) *
In contrast to (15), I will explicitly show in the next section that the new
ranking system proposed in (9), which involves only one minimal rerank-
ing, is both necessary and sufficient to account for the data in (6) and (10)-
( 1 3 ) . Furthermore, I will present some evidence for ALIGN-L ( o \ root),
which plays a vital role particularly in accounting for the accentual behav-
ior of such quadrimoraic words as those given in (13).
In (17), all the parsed-accent cases are compounds with foreign heads and
can thus be captured by (9b), whereas varied cases with native heads are
accounted for by the free ranking of either (9a) or (9b). The varied cases
with Sino-Japanese heads whose accent can exceptionally be parsed as in
nihón + ζ in ->• nihonzín 'Japanese people' and kityou + mén -> kityoumén
'methodical nature' (i.e. ( l i b ) ) are also captured by (9b). In addition,
some nativized foreign compounds in (12c), such as súupaa + mán -*
suupáaman 'superman' and sáin + pén > saínpen 'pen for signature,' sug-
gest that they are assimilated to the usual ranking of (9a). Thus, given (9),
the analysis involves only one minimal reranking to explain the data
above, as in (18)-(20). Note that the symbol '=' in (19) indicates that the
former two constraints concerned can be ranked freely, either as in (9a)
or (9b):
a. /káfe + báa/
kafe) + (báa) **
kafé) + (baa) *! *
b. /máikuro + básu/
maikuro) + (básu) * *
**
maikuró) + (basu) *!
c. /dorággu + sutóa/
doraggu) + (su)(tóa) * *
**
doraggu) + (sú)(toa) *!
174 Shin-ichi Tanaka
hidari) + (Ú)(tiwa) * **
c. /nihón + zín/
**
nihon) + (zín)
nihón) + (zin) * *
(17) (see also (13)). It is true that their accent uniformly falls on the pen-
ultimate foot, but it still remains unclear why it doesn't falls on the ante-
penultimate but the pre-antepenultimate mora as in *nuka) + (yoró)(ko-
bi) and *denki) + (nokó)(giri), as compared to (a) (bará) + (bone) /
*a)(bára) + (bone) 'rib bones' and oya)(yubí) + (hime) / *oya)(yúbi) +
(hime) 'thumb princess.' In other words, the following two questions re-
main to be explained; that is, 1) why only quadrimoraic heads exhibit tro-
chaic pattern, while all others are iambic due to A L I G N - R (PrWd, o{) and,
moreover, 2) why only words with the original antepenultimate accent
(i.e. (13b)) exhibit an iambic/trochaic variation as in neri) + (hámi)(gaki)
/ (neri) + (hamí)(gaki). (21) clearly demonstrates that both of these are
given a convincing account by the interactive effect of MAX (accent) and
A L I G N - L ( o \ root):
NON-FINALITY ALIGN-R
MAX (accent) = ALIGN-L (σ\ root)
(μ\σ\Γ) (PrWd, a-)
a. /nuka + yorokobi/
nuka) + (yoró)(kobi) **
***
nuka) + (yóro)(kobi) *
b. /dénki + nokogíri/
denki) + (nokó)(giri) * *1 **
denki) + (nóko)(giri) *
c. /sin + yokohama/
sin) + (yokó)(hama) *! **
sin) + (yóko)(hama)
d. /neri + hamígaki/
"" neri) + (hamí)(gaki) * **
***
"" neri) + (hámi)(gaki) *
The compounds in the left column originally bear accent before the
boundary due to NON-FINALITY (F'), but it shifts to the left within the pen-
ultimate foot, attracted to the new boundary, when they undergo further
compound formation as is shown in the right column (although the parsed
option is also possible as in (21d)). As a result, compound accent might
appear to be assigned either before or after the boundary as Kubozono's
A L I G N - C A defines, but this proves to come from the effect of NON-FINAL-
ITY (F ) and the usual single-edge alignment A L I G N - L (Σ 1 , root). On the
1
So far I have argued for the ranking system of NON-FINALITY (μ1, σ\ F') »
M A X (accent), in place of Kubozono's NON-FINALITY ( μ ' , o') » PARSE (ac-
cent) » NON-FINALITY (F'), on the basis of the varieties of possible pat-
terns and variations of accented compounds, and indeed, the data I have
presented thus far seem to favor the proposed ranking with respect to the
present state of Japanese grammar. However, I demonstrate in this sec-
tion that the most compelling and convincing evidence in support of this
ranking can be found when we attempt to account for such unaccented
compounds as those given in (23):14
b. Quadrimoraic outputs
kínu +ito kinuito 'silk thread'
dasí + siru ->• dasiziru 'soup stock'
nía + mísu -»· niamisu 'near miss'
dokú + gásu -»• dokugasu 'poisonous gas'
on + in ->· onin 'sound and rhythm, phonology'
mizu + hana -> mizubana 'running nose'
usu + azi -> usuazi 'light taste'
náka + niwa nakaniwa 'courtyard'
dán + zétu ->• danzetu 'extinction'
sóku + báku -»· sokubaku 'restraint'
sétu + zóku -»• setuzoku 'connection'
kótu + sétu ->· kossetu 'bone fracture'
on the penultimate foot, because its typical word form (more strictly, its
minimal word) is quadrimoraic with two roots as in (23c). Note that none
of the compounds in (23a-c) have a pitch fall anywhere in their domain,
which means that they seriously violate A L I G N - R (PrWd, Σ ' ) . Thus, my
task is to explain how such an unaccented pattern is possible at all in the
grammar of Japanese accentuation by taking advantage of the constraint-
ranking system for all the patterns seen in the preceding sections.
I propose that the key to answering this question lies in the constraint
of NON-FINALITY. Specifically, it is quite natural that Japanese grammar
has not only N O N - F I N A L I T Y (JU\ A\ F ' ) but also N O N - F I N A L I T Y (PrWd') as
a member of the N O N - F I N A L I T Y family, since the inventory of prosodie cat-
egories contains not only mora, syllable, and foot, but also prosodie word.
If we add this constraint to the ranking I have argued for, it should be lo-
cated below M A X (accent), as shown in ( 2 4 ) : 1 5
Here, the dotted lines indicate the rerankable relations of MAX (accent)
to the other three constraints: the reranking of (24a) is for foreign heads
and some archaic native and Sino-Japanese heads as seen in (18) and (19);
(24b) is for quadrimoraic heads with the antepenultimate accent as (21);
and I am claiming that (24c) is exactly what is needed for the unaccented
compounds shown in (23).
Following the definition of (4b, c), we define the requirement of NON-
FINALITY (PrWd') in (25a). Note, however, that this definition logically
implies that accent must not be present in PrWd, precisely because any
compound in Japanese forms a single PrWd of its own and hence it is al-
Possible Patterns and Variations in Japanese Compound Accentuation 179
sinzyu) + (gai) *
*!
sinzyu) + (gai) *!
Possible Patterns and Variations in Japanese Compound Accentuation 181
sinzyú) + (gai) *!
sinzyu) + (gai) *! *
*
sinzyu) + (gai)
When MAX (accent) is located at the (28a) position, the ranking produces
the parsed-accent patterns of foreign heads like (18) and some archaic na-
tive and Sino-Japanese heads like (19); when it is at the (28b) position, the
ranking captures the most productive patterns in (20) and (21), in which
accent falls on the penultimate foot; and when it is at the (28c) position,
the ranking accounts for the unaccented compounds in (26). Since I as-
sume that NON-FINALITY ( μ \ σ\ F ' ) is just a single constraint, M A X (ac-
cent) cannot be ranked in between the two of NON-FINALITY (μ', σ\ F ' ) ·
The three possibilities in (28), then, also mean that logically, there should
be four types of accent variation as shown below:
The 'local' variations in (30a-c) are captured as the minimal reranking (pro-
motion and/or demotion) of MAX (accent) and taken to be strong evidence
for its position between NON-FINALITY (μ1, Σ 1 , F11 ) and NON-FINALITY (PrWd ' )
in the present grammar (cf. (5)). The mimetics in (31) and the acronyms in
(32) also exhibit the same variation as (30b) and can be seen as arguments
for the position of MAX (accent) adjacent to NON-FINALITY (PrWd'): 20
(32) Acronyms
a. Penultimate foot
patoróoru + káa patókaa 'patrol car'
zéneraru + sutoráiki ->· zenésuto 'general strike'
haráguti + syóusuke harásyou 'Shosuke Haraguchi (person name)'
hasimoto + ryuutarou hasíryuu 'Hashimoto Ryutaro (person name)'
b. Unaccented
páasonaru + konpyúutaa ->• pasokon 'personal computer'
sékusyaru + harásumento • sekuhara 'sexual harassment'
kimura + tákuya kimutaku 'Takuya Kimura (person name)'
toyókawa + étuzi toyoetu 'Etsuji Toyokawa (person name)'
On the other hand, the 'non-local' variations in (30d) might at first sight
appear to be a case of non-minimal reranking, but I claim that their
reranking is nothing but minimal, because NON-FINALITY (μ', σ\ F ' ) and
NON-FINALITY (PrWd') tend to behave uniformly as members of the NON-
FINALITY family; namely, the data in (30d) show that the family members
184 Shin-ichi Tanaka
are rerankable as a whole with MAX (accent), although it is true that the
variations in (30d) are relatively rare cases.
Since compounds with parsed accent within the final foot are limited in
number (except for foreign heads, as stated in section 3.1.), we can safely
say that the patterns with accent on the penultimate foot and/or without
any accent are more dominant than their parsed-accent counterparts in
(30a, c, d). Hattori (1991, 1998, 1999) also concludes from her historical
survey of 821 nouns that the general accentual change has converged by
90% at the two patterns concerned, which are much more prominent
than other patterns in the present state of Japanese grammar. She found
that words with the penultimate-foot accent account for 31 % while unac-
cented words amount to 58 %, although her survey contains not only com-
pounds but also simplex nouns.21 In fact, there are many compounds
where unaccented patterns are more dominant than their penultimate-
foot or parsed counterparts. (33) exemplifies the dominance relation
among the three variable patterns based on the examples in (30b-d),
where the one indicated by '>' or '<' is considered now as "more appro-
priate as a pronunciation of Standard Japanese than the other(s)" accord-
ing to NHK's (1998: 9) Dictionary of Japanese Pronunciation and Accen-
tuation:
(33) Dominance
a. Penultimate foot / unaccented
taiónkei < taionkei 'thermometer'
kansyázyou < kansyazyou 'letter of thanks'
suihéisen < suiheisen 'horizontal line'
akitáinu > akitainu 'Akita dog'
sorámimi < soramimi 'mishearing'
mawarimiti < mawarimiti 'roundabout way'
kigókoro > kigokoro 'disposition'
nizákana > nizakana 'fish boiled with soy'
b. Parsed / penultimate foot / unaccented
sinzitumí < sinzitúmi < sinzitumi 'truth'
kityoumén > kityóumen < kityoumen 'methodical nature'
kasikín > kasikin < kasikin 'loan'
sekihán > sekíhan < sekihan 'rice boiled with red beans'
kubinekkó < kubinékko > kubinekko 'scruff of the neck'
katidokí > katidoki < katidoki 'shout of victory'
orizúru < orízuru > orizuru 'folded-paper crane'
hatomúgi > hatómugi < hatomugi 'adlay, oats'
Possible Patterns and Variations in Japanese Compound Accentuation 185
c. Parsed / unaccented
sankousyó < sankousyo 'reference book'
keisatusyó > keisatusyo 'police station'
saibansyó < saibansyo 'court of justice'
siodoki < siodoki 'favorable tide, good chance'
waruzié < waruzie 'serpentine wisdom'
wanpakumonó < wanpakumono 'naughty boy'
waraigóe > waraigoe 'laughing voice'
waribási < waribasi 'half-split chopsticks'
These data show that in the near future, more and more compounds are
likely to become de-accented, which means that MAX (accent) is going to
demote below N O N - F I N A L I T Y (μ', o\ F ' , PrWd') as a direction of accent
change in progress.
In addition to de-accentuation of compounds I have been concerned
with so far, see also Tanaka (2001, to appear) and Tanaka & Yamane
(2000) for de-accentuation of lexical classes and their different applicabil-
ity and variation from compounds.
In this paper, I have argued that Japanese grammar at present has a sys-
tem of assigning compound accent to the penultimate foot, and that is
why there are some compounds or their variants which may either parse
their lexical accent or undergo de-accentuation, as was illustrated in (28).
In other words, M A X (accent) must now be located between N O N - F I N A L I -
1
TY (μ , σ', F 1 ) and N O N - F I N A L I T Y (PrWd'), which, however, tend to demote
below N O N - F I N A L I T Y (PrWd 1 ) as a recent change in progress. Thus, my
proposed system allows us to give a principled account for what patterns
and variations are possible in the accentual system of Japanese compounds
as well as why they can tend toward accent loss. Needless to say, the exist-
ence of N O N - F I N A L I T Y (PrWd 1 ) is crucial in the system.
Note also that the movement of MAX (accent) through the constraint
hierarchy plays two important roles in characterizing the grammar of Jap-
anese: it characterizes not only the synchronic accentual systems based
mainly on word classes but also their diachronic change of the core sys-
tem from the past, as summarized in (34) (DEP (accent) is also ranked
here for the reason stated below):
186 Shin-ichi Tanaka
Synchronic Diachronic
Ranking Head
(DEP (accent))
( M A X J ) Alien Periphery
I
A
N o VOICED GEMINATES
Ç MAX Foreign
I
N o SINGLE [ p ]
MAX ^ Sino-Japanese
P O S T - N A S A L VOICING
(DEP)
Unlike (34), the demotion of MAX in (35) does not reflect the diachronic
change of the segmental system; rather, recent cultivated or younger-gen-
188 Shin-ichi Tanaka
Acknowledgements
This paper is a radically revised version of my talk delivered at the monthly meeting of the Pho-
nology Association in Kansai (PAIK), which was held on March 23 in 1999 at Kobe University.
I would like to thank the following people for their comments and suggestions on an earlier ver-
sion: Junko Ito, H a r u o Kubozono, John Mc Carthy, Armin Mester, Akio Nasu, Noriko Ya-
mane, and Yuko Yoshida. I am also very grateful to Edward T.W. Haig for stylistic improve-
ment. Special thanks go to Jeroen van de Weijer, Tetsuo Nishihara, and an anonymous
reviewer, who greatly helped me elaborate my idea on the present topic through this book
project. Any remaining inadequacies or misconceptions are my responsibility alone, of course.
Notes
1. What I call "a c o m p o u n d " in this p a p e r can be defined as a polymorphemic (i.e. non-sim-
plex) word without any 'pitch valley (pitch fall + pitch rise)' in its domain, consisting of
two (or more) words or roots. In this sense, Sino-Japanese words with two roots like
doubutu 'animal' ( d o u 'moving' + butu 'thing') are taken to be c o m p o u n d s by definition,
as well as mimetic or reduplicative words like betabeta (to) 'stickily' (beta 'the sound or
m a n n e r of sticking something', although dou, butu, and beta are bound m o r p h e m e s and
do not occur alone.
However, I will not be concerned in the present discussion with the accentual behavior
of the so-called 'superlong head compounds,' which contain five or m o r e moras in their
heads, as given below:
In these cases, it is surprising and extraordinary that even the unaccentedness and the fi-
nal accent of the original words in (ib) and (ic) are carried over in the whole compound,
just as the usual non-final accent remains intact in the case of (ia). The accentuation of
such superlong head c o m p o u n d s may involve the dominant ranking of IDENT (accent) (cf.
MAX (accent) in section 3.1.), but I will leave this issue to f u r t h e r research.
2. The reason that the accent of N2 is preserved rather than that of N1 may be attributed to
the fact that N2 is the morphological head of a compound.
Possible Patterns and Variations in Japanese Compound Accentuation 189
3. Kubozono, Itô & Mester (1997) argue that in Japanese, N O N - F I N A L I T Y at the syllabic lev-
el works for loanword truncation as well as compound accent.
4. The violations are computed at the syllabic level and not at the moraic level because ac-
cent-bearing units are syllables in Japanese. In this respect, it may be restated as A L I G N -
R (PrWd, σ 1 ), which I will adopt in section 3.1. (see also Tanaka 2001, to appear).
As for A L I G N - C A , unaccented compounds will always violate it, since they do not bear
accent at all on the 'around-the-boundary' syllable. This constraint appears to require a
rather curious alignment in that it has a double-edge disjunction in it (either at the left
edge of N2 or the right edge of N l ) . In section 3.1., I will adopt a more straightforward
version of a single-edge alignment constraint that has different effects from A L I G N - C A .
5. The traditional view of distinguishing in mora length between short and long head com-
pounds is supported in Sato (1989), Uwano (1997), and so on. See Yoshida (1995) for a
treatment of compound accent in the framework of Government Phonology, which also
makes it possible to dispense with the length of N2 for its computation.
6. Recently, Alderete (1999) has proposed a ranking system of Japanese compound accent
in OT, which is somewhat different from (4). For a non-derivational analysis of foreign
accent (simplex words), which exhibits similar pattern to compound accent, see Kataya-
ma (1995,1997).
7. In Kubozono (1997: 274, this volume), the accent on the non-final rightmost foot is
dubbed 'default compound accent' or 'emergence of the unmarked' and provided with
some generality in the accentual system of Japanese, but with a less dominant status
than the preserved accent stated in (7). I am claiming here that default accentuation
should be more dominant in general than full preservation of input accent, as will be
clear from N O N - F I N A L I T Y ( F ' ) » M A X (accent) in (9a).
8. See also note 20 for related discussion.
9. As Kubozono (1995: 36,1997: 286, this volume) suggests, there may be a possibility of
considering the original N2 words in (11a) as final-accented and hence shifted, because
their final vowels i and u are all epenthetic or underlyingly absent. For criticism of such
an analysis, see Tanaka & Yamane (2000).
As for the data in ( l i b ) , neither N O N - F I N A L I T Y (μ') » M A X (accent) » N O N - F I N A L I T Y (σ',
F ' ) nor N O N - F I N A L I T Y ( μ ' , σ') » M A X (accent) » N O N - F I N A L I T Y ( F 1 ) does not account for
the accent preservation of such cases as in meisyó, sinzitumí, and so on, because their
accent is final at the moraic level. Thus, there is no doubt that the necessary reranking
must be M A X (accent) > N O N - F I N A L I T Y ( « ' , Σ ' , F ' ) , as indicated in (9b).
10. The compound accent in (12a) is final at the syllabic level. As is well-known, there are
no final-accented loanwords at the moraic level in Japanese.
11. In section 3.2., I will present an analysis which allows the variation of either the pre-an-
tepenultimate or antepenultimate accent within the foot concerned.
12. I am not concerned here with unaccented compounds such as urá + yamá ' urayama
'hill at the back' and gárasu + tamá -> garasudama 'glass ball,' which will be discussed
in section 4.1.1 also ignore some Sino-Japanese compounds whose input accent on the
final mora is parsed as in ( l i b ) (meisyó 'place of scenic beauty,' kyuusyó 'vital spot,' and
sinzitumí 'truth'), since they are possible but rare cases, and typically have another ac-
centual pattern as their variation. See section 4.2. for details of their variation.
Note also in (17) that there are some cases which involve hiatus between N l and N2, as
in sinobíasi, niwakáame, and the like. The vowel sequences emerged after compound
formation never form a diphthong or a long vowel in Japanese, because the word
boundary prevents a syllable as well as a foot from crossing it.
1 3 . An anonymous reviewer points out that since we already had a free ranking of N O N - F I -
N A L I T Y (JU\ Σ\ F ' ) and M A X (accent) in ( 1 9 ) , we now have six possible subrankings (i.e.
190 Shin-ichi Tanaka
3 x 2 x 1 possibilities) and wonders whether all the subrankings are motivated. This
does not matter empirically, however, because the optimal candidate does not change
if we take A L I G N - L (σ1, root) into consideration in (19).
14. A n exception to (23) is úbu + ke ubuge 'downy hair,' a trimoraic word where ke is not
final-accented but unaccented: it always forms unaccented compounds such as wakige
'armpit hair,' munage 'chest hair,' and the like, even though they do not become quadri-
moraic. But it is generally the case that unaccented compounds emerge only from the
environments in (23a-c), where N2 is monomoraic or bimoraic (although ki + kokóro
-* kigokoro 'reliability' is a quadrimoraic compound exceptionally with a trimoraic
head). In other words, de-accentuation of compounds can apply only if their input head
is short or their output form is quadrimoraic. I will leave the question of how it is ac-
counted for to further research.
15. I am grateful to Armin Mester for pointing out the idea of N O N - F I N A L I T Y (PrWd').
16. Otherwise, since a compound consists of two (or more) words, it might be possible that
it contains two PrWd's, not two roots, and that the prosodie wordhood of N1 and N2
always percolates upward to the whole compound. That is, the prosodie wordhood that
matters here might be determined by the so-called 'A-over-A principle.' I do not adopt
such an assumption, because it does not hold for the Sino-Japanese cases in (23c) or the
mimetics in (30b) below, which are made up of two bound morphemes and not of two
PrWd's. Moreover, this principle does not apply to the case of English compounds.
17. By using bidirectional extrametricality, Tateishi (1992) offers a derivational account of
why trimoraic and quadrimoraic simplex words are prone to be unaccented; however,
it does not explain the cases in (23a) since the unaccented compounds are more than
four moras long, which he is not concerned with. Thus, it may be the case that unaccent-
edness of simplex words is triggered by different factors from that of compounds, even
though they have in common the fact that they have some length-effect on unaccentu-
ation (see note 1 4 ) and that their accent loss is due to N O N - F I N A L I T Y (PrWd 1 ).
18. See section section 3 . 1 and note 4 for the difference between A L I G N - C A and A L I G N - L
(σ1, root).
19. I assume here that only the members of the FAITHFULNESS family (i.e. M A X , D E P , and
IDENT) can promote or demote through the fixed constraint ranking. For detailed dis-
cussion on this topic, see section 5. Given this assumption, we focus on the rerankings
in (24a) and (24c) and ignore the one in (24b) that is mainly for quadrimoraic heads
with the antepenultimate accent as in (21). See also Tanaka (2001, to appear) and Tana-
ka & Yamane (2000) for a lexical-class-based account of how individual compounds se-
lect the specific reranking possibilities.
20. Note that in the case of mimetics in (31a), trochaic patterns such as bèta) + (beta) and
kira) + (kira) are produced. This fact might appear to pose a certain problem with the
proposed system, which would generate *betá) + (beta) and *kirá) + (kira) as it stands.
However, we must recall the definition of MAX (accent), which requires that the accent
of a head root has a correspondent in the compound. In usual compounds, the N2 accent
is respected because N2 is their morphological head, but this is not the case with mimet-
ics: I assume that the head of mimetic words is N l , and that is why MAX (accent) favors
bèta) + (beta) and kira) + (kira), excluding *betá) + (beta) and *kirá) + (kira). A L I G N - L
(ισ', root) also favors the trochaic patterns, because the head root is N l .
21. Strictly, Hattori refers to the former type not as "penultimate-foot accent" but "accent
on the syllable containing the third mora from the end." Since the dominance of the lat-
ter implies that of the former, I consider them virtually equivalent here.
22. As is well-known, native and mimetic words also undergo vowel epenthesis and conso-
nant gemination, respectively, as in kak + hazimer-u kakihajimeru 'begin to write,' yar
Possible Patterns and Variations in Japanese Compound Accentuation 191
+ oer-u yarioeru 'finish doing,' beta-ri bettari 'very stickily,' biku-ri bikkuri 'very
surprisingly,' and so on.
References
Alderete, John
1999 Morphologically Governed Accent in Optimality Theory. Ph.D. dissertation,
University of Massachusetts at Amherst.
Hattori, Noriko
1991 Mechanisms of Word Accent Change: Innovations in Standard Japanese. To-
kyo: Shinozaki Shorin.
1998 "Base Transparency in Suprasegmental Changes: Ongoing Changes in Japa-
nese and English," Language, Variation, and Change 10: 2, 85-96.
1999 "On the Convergence of Accentual Patterns: 'Drift' Observed in Variation,"
Phonological Studies 2, ed. by the Phonological Society of Japan, 37-44. To-
kyo: Kaitakusha.
Hayes, Bruce
1995 Metrical Stress Theory. Principles and Case Studies. Chicago: University of
Chicago Press.
Itô, Junko and Armin Mester
1995 "The Core-Periphery Structure of the Lexicon and Constraints on Rerank-
ing," University of Massachusetts Occasional Papers in Linguistics 18 (Papers
in Optimality Theory), 181-209.
Itô, Junko, Yoshihisa Kitagawa, and Armin Mester
1996 "Prosodie Faithfulness and Correspondence: Evidence from a Japanese Ar-
got," Journal of East Asian Linguistics 5, 217-94.
Katayama, Motoko
1995 "Loanword Accent and Minimal Reranking in Japanese," Phonology at Santa
Cruz 4,1-12.
1997 Optimality Theory and Japanese Loanword Phonology. Ph.D. dissertation,
University of California at Santa Cruz.
Kubozono, Haruo
1995 "Constraint Interaction in Japanese Phonology: Evidence from Compound
Accent," Phonology at Santa Cruz 4, 21-38.
1997 "Lexical Markedness and Variation: A Nonderivational Account of Japanese
Compound Accent," Proceedings of the 15th West Coast Conference on For-
mal Linguistics, 273-87.
(this volume) "Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes"
Kubozono, Haruo, Junko Itô, and Armin Mester
1997 "Nonfinality in Japanese Phonology," Proceedings of the 16th International
Congress of Linguistics (CD-ROM), ed. by B. Caron. Amsterdam: Elsevier.
- McCarthy, John and Alan Prince
"Generalized Alignment," Yearbook of Morphology 1993, ed. by Geert Booij
and Jaap van Marie, 79-153. Dordrecht: Kluwer.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. The Hague: Mouton.
1977 "Accent in Japanese," University of Southern California Occasional Papers in
Linguistics 4 (Studies in Stress and Accent), 261-302.
192 Shin-ichi Tanaka
N H K (ed.)
1998 A Dictionary of Japanese Pronunciation and Accentuation. Tokyo: N H K
Shuppan.
Poser, William
1990 "Evidence for Foot Structure in Japanese," Language 66, 78-105.
Sato, Hirokazu
1989 "The Accent Rule and Sequential Voicing in Compounds," The Japanese Lan-
guage and Japanese Education 2, ed. by Miyoko Sugito, 233-65. Tokyo: Meiji
Shoin.
Tanaka, Shin-ichi
1992 "Accentuation and Prosodie Constituenthood in Japanese," Proceedings of
the 5th Annual Meeting of the Tokyo Linguistic Forum, 195-216.
2001 "Japanese Accent in Optimality Theory: The Mechanism of De-accentuation
and its Different Applicability among Lexical Classes," The Interface between
Meaning and Form: A Festschrift for Dr. Minoru Nakau on the Occasion of his
Sixtieth Birthday, ed. by Shuichi Takeda, Atsuro Tsubomoto, Yukio Hirose,
Koichi Takezawa, and Nobuhiro Kaga. Tokyo: Kuroshio Shuppan.
(to appear) Accent and Rhythm: From the Basics of Phonology to Optimality Theory.
Tokyo: Kenkyusha.
Tanaka, Shin-ichi and Noriko Yamane
2000 "Paradox or Reranking?: Accent and Epenthetic vowels in Sino-Japanese,"
Proceedings of the 24th Annual Meeting of the Kansai Linguistic Society (KLS
20).
Tateishi, Koichi
1985 Accent of Compound Nouns in the Standard Japanese: A Nonlinear Analysis.
MA thesis, International Christian University, Tokyo.
1989 "Theoretical Implications of the Japanese Musicians' Language," Proceedings
of the 8th West Coast Conference on Formal Linguistics, 384-98.
1992 "Maximal Non-Exhaustive Invisibility," Proceedings of the 16th Annual Meet-
ing of the Kansai Linguistic Society (KLS 12), 134-44.
Uwano, Zendo
1997 "The Accent of Some Dialects in Japanese with Special Reference to Com-
pound Nouns," Accent, intonation, Rhythm, and Pause, ed. by Tetsuya Kuni-
hiro, Hajime Hirose, and Morio Kouno, 231-70. Tokyo: Sanseidou.
Walker, Rachel
1995 "Mongolian Stress: Typological Implications for Non-Finality in Unbounded
Systems," Phonology at Santa Cruz 4, 85-102.
1996 "Prominence-Driven Stress," Ms., University of California at Santa Cruz.
[Available at ROA-172-0197 from the Rutgers Optimality Archive.]
Yoshida, Yuko
1995 On Pitch Accent Phenomena in Japanese. Ph.D. dissertation, SOAS, Universi-
ty of London. [Published by Holland Academic Graphics, The Hague, in
1999.]
An Element-Based Analysis of Affrication
in Japanese*
Shohei Yoshida
Yokohama National University
1. Intoduction
2. Data
There are four types of affricates in standard Japanese: the voiceless and
voiced dental-alveolar affricates [ts, dz] and the voiceless and voiced
palato-alveolar affricates [tj, d3]. The voiceless affricates [ts, t j ] are allo-
phonic variants of It/, /t/ becomes [ts] before a high-back vowel, while it
194 Shohei Yoshida
becomes [tj] before a high-front vowel and a palatal glide. Likewise, /d/
becomes [dz] and [d3] before a high-back vowel and a high-front vocoid,
respectively. It should be noted at this point that the phoneme /z/ is usu-
ally pronounced as the affricate [d3] before a high-front vocoid and as
[dz] otherwise. It follows that the contrast between Idi and /z/ is neutral-
ized before a high-front vocoid and a high-back vowel in modern stan-
dard Japanese. Before /j/, however, there is one context where the phone-
mic status of [d3] is obvious. When the /w/ in the sequence /de+wa/ 'in that
case' (CONNECTIVE + TOPIC) is deleted, the resulting /dea/ becomes /djaa/
[d3a] or [d3aa] as a result of glide formation and optional compensatory
lengthening (Poser 1985). In this case, the source of [d3] is clearly /d/. In
addition, all the voiced affricates, regardless of their phonemic status, are
prone to be lenited to the fricatives [z, 3] intervocalically in rapid speech.
(1) shows the distributions of the allophonic variants of Iti, Id/ and /z/ fol-
lowed by a palatal glide and the five vowels of Japanese.1
(1)
Iii Iii Id /a/ loi lui
Iti ítjvi m [te] [ta] [to] ftsui]
Idi [de] [da] [do]
[d3V~3V] [d3i~3i] [dzui-zui]
/ζ/ [dze~ze] [dza~za] [dzo~zo]
Examples:
/tjoo/ [tjoo] 'butterfly' /kati/ [katji] 'value'
/mate/ [mate] 'Wait!' /kata/ [kata] 'shoulder'
/ato/ [ato] 'later' /natu/ [natsui] 'summer'
to-alveolar affricates when followed by the palatal glide /j/ across a word
boundary, e.g. last year [laistjia], did you [did3u:]; in Korean stem-final
dental stops become palato-alveolar affricates when followed by ¿-begin-
ning suffixes, e.g. kath+i [katji] 'together', path+i [patJi] 'plowed
field+suBJECT'; in Brazilian Portuguese the dental stops /t, d/ become
palato-alveolar affricates [tj, d3] before /i/, e.g. noite [noitJi] 'night', tarde
[tahd3i] 'afternoon'; and female Cairene Arabic speakers palatalize and
affricate dental-alveolar stops when the following vowel is III, e.g. dilwaiti
[d3ilwa?tj"i] 'now', balad+i [balad3i] 'my country'.
In addition to dental-alveolar stops, the velar stops /k, gl are often coro-
nalized and affricated to [tj", d3] before front vowels, if not pronounced
with a secondary palatal articulation as in [kJ, gi], in many languages. Affri-
cation that accompanies the coronalization of velar stops is found in Aca-
dian French (Flikeid 1988), Russian (Kiparsky 1982:124), Slavic (Chom-
sky and Halle 1968: 421-422), and Eastern Arabic dialects (Johnstone
1967). In Japanese, velar stops are dialectally fronted before HI and 1)1
(Fujiwara 1997:55-59). For example, in the Tohoku provinces, Ikl before
HI is pronounced either as the palato-alveolar affricate [tjï] or the velar-
palatal affricate [kçï], e.g. /kirei/ [tjïree]~[kçïree] 'beautiful'. These cases,
together with the rarity of the palatal stops [c] and [j.] in languages (Cat-
ford 1988: 94), indicate that in the palato-alveolar-palatal region the affri-
cates [t J, d3] are preferred over dental-alveolar stops followed by a glide
[ti, di], and the palatal stops [c, j], pointing to the unstable status of palatal
closure, and a correlation between palato-alveolar-palatal articulation and
affrication.
On the other hand, it appears to be the case that affrication that does
not involve palatalization is much rarer. Such a case is found in Quebec
French. In Quebec French, coronal stops are affricated before the high-
front vowels HI and lyl without concomitant palatalization (Kaye 1989:
29), e.g. type [tsip] 'type', dur [dzyif] 'hard'. Strictly speaking, however,
this case should be distinguished from the true case of affrication without
palatalization as in Japanese in that the environment in the Japanese case
involves a high-back vowel whereas that in the Quebec French case in-
volves a high-front vowel though palatalization does not manifest itself.
This does not mean, however, the affrication without palatalization is
non-existent in other languages. An instance of affrication genuinely par-
alleling the Japanese case comes from Korean, in which the impermissible
sequence of a stem-final dental stop and a suffix-initial liquid or a nasal is
broken up by the epenthetic vowel [i], and the dental stop becomes a den-
tal-alveolar affricate, e.g. path+ro [patsiro] 'toward the plowed field',
196 Shohei Yoshida
In element theory, the phonological primes are called elements. Since el-
ement theory originated as a sub-theory of government phonology (Kaye,
Lowenstamm and Vergnaud 1985, 1990), the possibility of representing
both vowels and consonants in terms of univalent elements, rather than
binary-valued features, has been pursued by researchers such as Renni-
son (1990), Harris (1990,1994), Cyran (1997), Harris and Lindsey (1995),
Williams (1998) and authors in Cyran (1998).
Elements are monovalent, being specified for their presence only. Vow-
els are constructed from the elements I, U and A. Elements have the
property of interpretational autonomy (Harris and Lindsey 1995: 40), and
are interpretable and pronounceable in isolation, unlike binary-valued
features, which, with value specifications, contribute to the realization of
a segment only in combination with several other features. A phonologi-
cal expression (henceforth 'PE') is either simplex, composed of a single
element, or complex, composed of two elements or more, and has one
head. When linearly represented, the head of a PE is indicated by locating
An Element-Based Analysis of Affrication in Japanese 197
reader is referred to Harris (1994), Harris and Lindsey (1995) and Will-
iams (1998) for acoustic signatures of independent elements in simplex
and complex PEs.
For the present purposes, the articulatory implementation of I, U, A
taken from Harris and Lindsey (1995: 55-56) are given below. The paren-
thesized comments are supplied by the present author.
A word must be said about the unroundness of the high-back vowel. In bi-
nary feature theory such a vowel must be assigned the property [-round]
and represented differently from a high-back rounded vowel. In element
theory both [u] and [ui] can be represented by the U element since a labial
activity is but one way of expanding the oral cavity. Specifically, I propose
that the unrounded high-back vowel of Japanese is represented as an emp-
ty-headed PE containing the U element.
4. Consonants
When the resonance elements I, U and A occur in non-nuclear positions
either on their own or in combination with other elements, they define
palatality, labiality (and velarity in this paper), and pharyngeality, respec-
tively. Thus the phonological objects I and U, which in isolation are real-
ized as i and u in nuclear positions, characterize palatal and labial (and ve-
lar in this paper) articulations in non-nuclear positions. For example, the
vowels i and u and their non-syllabic counterparts j and w differ only in
their syllabic affiliation.
An Element-Based Analysis of Affrication in Japanese 199
(4) Labial
w (U)
ρ (h.?.U) b (L.h.?.U)
Φ (h.U)
m (L.N.7.U)
Dental-alveolar
r U
t (h.?._) d (L.h.?._)
s (h. J
ts (?._); (h._) dz (L.?._); (L.h.J
η (L.N.?. J
Palato-alveolar
S (h.I.J
t j (?._); (h.I._) d 3 (L.?._); (L.h.I.J
ni (L.N.?._); (I)
Palatal
j (I)
ç (h.I)
200 Shohei Yoshida
Velar
k (h.?.U._) ki (h.?.U._); ( I )
g (L.h.?.U._) i] ( L . N . 7 . U J
Glottal
h (h)
Moraic nasal3
iîi (Ν)
(where the semi-colon denotes that the segments containing it con-
sist of two Root Nodes.)
language has Ipl then it will also have Ik/, and if a language has Ik/ then it
will also have It/. Specifically, I propose that when occupying non-nuclear
positions, the U element as head is manifested as bilabial articulation
whereas it is realized as velar articulation as operator. The first implica-
tional universal can be accounted for if we assume that the presence of
the U element in stops as head ((...?.U) = a bilabial stop) entails its pres-
ence as operator ((...?.U._) = a velar stop) but not vice versa. The second
implicational universal also stems from the present elemental representa-
tions of dental-alveolars and velars. That is, the presence of empty-head-
ed stops containing U (,..?.U_), which are velars, implies the presence of
empty-headed PEs without U (...?._), which are dental-alveolars.
5. Contour structures
(5) t ts tj
X X X
I / \ / \
ROOT «C T T T T
l \ / I l \ / I I \
RESONANCE · ? ? · · h ? · · h
\ / \ / \
I
The I-? Constraint prevents the I element from being dominated by the
first branching node (Root Node) dominating the occlusion element ?,
forbidding the co-occurrence of the two elements under the same Root
Node. This constraint may apply to oral stops, nasal stops and laterals, as
they all contain the ? element. 4 When the spreading of the I element from
a nuclear position to a preceding onset position is obligatory, i.e. when I
An Element-Based Analysis of Affrication in Japanese 203
is required to coexist with the ? element under a single Root Node, in lan-
guages where this constraint is operative, the Root Node and the Reso-
nance Node it dominates are fissioned into two, and I is linked to the sec-
ond Resonance Node to avoid being c-commanded by ?. It can be argued
that the fission of stops in the palato-alveolar-palatal region is attribut-
able to the fact that palatal articulation usually involves a considerably
large constriction area. The change from [t] to [tj~] is given in (7).
(7) t i tr
χ χ —^ X X
I I / \ I
ROOT RC ? r ? ?
I \ I / I l \ I
RESONANCE ·- ? · ? · « h ·
(8) tj
χ
ROOT
\
RESONANCE
In this structure, the ? element does c-command the I element since the
first branching node dominating ? (Root Node) also dominates I. In ef-
fect, two Root Nodes share a single Resonance Node. However, I reject
this structure because the initial Root Node as well as the second Root
Node contains the I element and therefore this structure does not capture
our basic insight concerning the general incompatibility of palatality and
occlusion.
204 Shohei Yoshida
(9) h Ç
χ χ > χ χ
ROOT
\
RESONANCE • h ·
I
(10) s e
χ χ —^
ROOT
RESONANCE
I/ \ A
(11) k i tj i
X ^^ h X -;
ROOT ? ·
RESONANCE ·-
/ -χ I
υ I
206 Shohei Yoshida
In this case, the spreading I element replaces the operator U in the internal
structure of /k/, suppressing velar articulation. The result is identical to the
output of the affrication of It/ before a high-front vocoid. To see why U
within /k/ is replaced by the spreading I, let us consider cases of palataliza-
tion in nuclear positions, i.e. the spreading of I in nuclear positions.
In nuclear positions, the I and U elements may combine, resulting in
front rounded vowels, though the combination is subject to parametric
variation. Japanese does not permit the combination of I and U in nuclear
positions, as manifested in the nonexistence of front rounded vowels, and
in some morphological processes. For example, when the adjectival suffix
consisting of the I element alone spreads to the final nuclear position of an
adjectival stem in dialects spoken in the Tohoku and Kanto districts, dif-
ferent results obtain depending upon the elemental content of the stem-
final position. If the stem-final position is occupied by A, the two elements
form a complex PE as the combination of I and A is allowed, whereas if the
stem-final nuclear position contains U, the spreading I replaces it.
X X X
\ /
taka+i [takee] 'high' (A) «κ (I) -- (A.I)
samu+i [samii] 'cold' (¥) < (I) - (I)
sugo+i [suigee] 'terrific' ( A . ¥ ) <t (I) -- (A.I)
This does not deny the presence of labials with a secondary palatal artic-
ulation such as [pi], [b»j] and [mi]. These sounds are contour segments
where labial PEs and the I element are linked to two distinct Root Nodes,
which will be discussed shortly.
To accord with this constraint, the combination of the U element, which
is already present in /k/, and the I element added through spreading into
this segment is avoided by one of the following strategies:
An Element-Based Analysis of Affrication in Japanese 207
(14)
a.(h.?.¥._) « I (?._); (h.I.J k -1J
b. (h.?.U._) « I (?.U._); (h.I) k ^ kç
c. (h.?.U._) « I (h.?.U._); (I) k ki
(15) k i
χ h r
ROOT j^— ? ·
RESONANCE ν
/ I
υ _ ' ι
(16) k i
h χ χ
• I
ROOT ?
RESONANCE · ·
/ \ I
U I
(16b) ρ i
h χ χ
I• I
ROOT ?
RESONANCE
υ I υ I
Given the representations of front vowels, which all contain the I ele-
ment, we expect to find languages where the fronting of velars is triggered
by nonhigh-front vowels such as /e/ (A.I) and /ae/ (I.A). Arabic dialects
spoken in Eastern Arabia (e.g. Kuwait, Bahrain, Qatar, Oman, etc.), pro-
vide such cases. In these dialects, velar and uvular stops are fronted in the
contiguity of mid- and low-front vowels as well as /i/, e.g. /kam/ [tjaem]
'how much', /halqa/ [fraltjae] 'ring' (Johnstone 1967). In these cases, too,
the spreading property is the I element.
[i], (U.I) [y], (_) [i] and affrication. 6 Therefore, I shall limit myself to Jap-
anese and present a working hypothesis that is possible at a current level
of understanding.
As noted in Shibatani (1990:161), the high-back unrounded vowel [ui]
is centralized after /s/, /z/, /t/ and pronounced as in [süi, dziïi, tsüi]. I as-
cribe this centralization to the spreading of the empty head of the coro-
nals to the following vowel.7 Consider (17) in which the centralized high
vowel [ώ] is represented as the empty-headed PE (U._) occurring in a nu-
clear position.
(17) s UI s UI
X X X X
ROOT
RESONANCE
l\ l\
• h • h
U U
The U element is the head of [ui]. However, the spreading empty head be-
comes head and U is demoted to operator, generating [tu].8 If this account
of the centralized vowel is correct, then it follows that the affrication of
It, d/ before HI and lui have one thing in common. That is, in both cases an
onset and a following nucleus share a property under the Resonance
Node, i.e. the type of affrication with concomitant palatalization involves
the I element shared by an onset and a following nucleus, whereas the
type of affrication without palatalization involves an empty head shared
by the two positions. To unify the two types of affrication, the I-? Con-
straint in (6) can be modified as follows
(19) t UI
ui t Iii
X X -> X X
ROOT T -h h •
· .
•v—
l \ 1
RESONANCE •· ?? ψ · ? · · h ·
·
I \ / \
U υ
(20) k ui
h χ χ
ROOT ?— * *
RESONANCE · _ ·
/ \ VI
υ _' υ
ι t
U-bridge
An Element-Based Analysis of Affrication in Japanese 211
9. Conclusion
The element-based representations of segments and the I-? Constraint al-
lows us to analyze the various types of palatalization in Japanese as a uni-
tary process, i.e. I-spreading. The spreading I element is either linked to a
preceding non-nuclear position, adding a secondary articulation to a seg-
ment that already occupies the position (e.g. kJ, gj, pi, bi), or is incorporat-
ed into the preceding segment. When I is incorporated into dental-alveo-
lar stops, the place-defining property of dental-alveolars (i.e. the empty
head) and I combine to yield a palato-alveolar articulation. Further, si-
multaneous affrication takes place to avoid the violation of the I-? Con-
straint. The resulting segments are palato-alveolar affricates. When the I
element is incorporated into velar stops, it either replaces their U element
or becomes the place-defining property of the second Resonance Node
without affecting their U element and empty head, yielding palato-alveo-
lar affricates and velar-palatal affricates, respectively.
Unlike the type of affrication with palatalization, a cross-linguistic ac-
count of the type of affrication without palatalization is yet to be discov-
ered. I have tentatively proposed a constraint to account for the two types
of affrication in Japanese.
Notes
* I thank Mohammad Badawy, Hee Gyong Kang, Kyong-Young Lee, Manjae Lee, Shin-
Mee Park and Olivier Tesson for providing and confirming the Arabic, French and Ko-
rean data. I gratefully acknowledge the extremely helpful comments and suggestions of
Wiebke Brockhaus, Duck-Young Lee and Jeroen van de Weijer on an earlier draft of this
paper. Infelicities which remain despite their help are only my responsibility.
1. The effects of the five vowel segments on Iti, Id/ and /z/ are unchanged whether the vowel
segments are short, long or the first elements of diphthongs.
2. The assumption that nasality and voicing are expressed by two different elements is of the
standard element theory. Nasukawa (1998) has claimed that voicing and nasality are ex-
pressed by a single element. However, the following analysis does not hinge on this issue.
3. The representation and IPA transcription given here are of the moraic nasal that is fol-
lowed by pause, vowels, glides and fricatives. I claim that the apparent velar closure of
the moraic nasal before pause (e.g. hon [horj] 'book') is due to an unreleased glottal stop
frequently heard in prepausai position, e.g. te [te 7 ] 'hand', hai [hai ? ] 'yes'. As no one, I
believe, would include the prepausai glottalization as part of the lexical representations
of these words, the closure of the prepausai moraic nasal should not be deemed part of
its lexical representation, either, [ΰι] immediately followed by the glottal closure is indis-
tinguishable from [g], When followed by an onset filled with a P E containing the ? ele-
ment, that P E spreads to the moraic nasal, making a homorganic nasal-stop/affricate
cluster.
212 Shohei Yoshida
4. I assume that the representation of the alveolar lateral approximant [1] is (?); (_).
5. It goes without saying that nothing prevents the I element from becoming a secondary
palatal articulation of dental-alveolars in languages.
6. The representation of the Korean vowel [i] as an empty nucleus is taken from Lee (1998).
7. In the GP literature, empty-headed PEs in nuclear positions have been viewed as repre-
sentations of lax vowels (e.g. Charette and Göksel 1994), while headed PEs have been
seen as those of tense vowels.
8. The term 'spreading' here is a metaphor for 'the alignment of two heads'. See Kaye, Lo-
wenstamm and Vergnaud (1985) and Lee and Yoshida (1998) for details.
References
Avery, P. and K. Rice
1989 Segment structure and coronal underspecification. Phonology 6.2,179-200.
Bhat, D. N. S.
1978 A general study of palatalization. In J. H. Greenberg, C. A. Ferguson and E.
Moravcsik (eds.), Universals of Language, Vol. 2: Phonology, 47-92. Stanford
University Press.
Bloch-Rozmej, A.
1998 In E. Cyran (ed.), 41-54.
Broadbent, J.
1991 Linking and intrusive r in English. UCL Working Papers in Linguistics 3,281-
302.
Brockhaus, W.
1995 Final Devoicing in the Phonology of German. Tübingen: Max Niemeyer Ver-
lag.
Catford, J. C.
1988 A Practical Introduction to Phonetics. Oxford: Oxford University Press.
Charette, M. and A. Göksel
1994 Vowel harmony and switching in Turkish languages. SOAS Working Papers in
Linguistics and Phonetics 4,31-52, University of London.
1998 Licensing constraints and vowel harmony in Turkic languages. In E. Cyran
(ed.), 65-88.
Cho, Y. Y.
1991 On the universality of the coronal articulator. In C. Paradis and J.-F. Prunet
(eds.), 159-180.
Chomsky, N. and M. Halle
1968 The Sound Pattern of English. New York: Harper and Row.
Cobb, M.
1995 Vowel harmony in Zulu and Basque: The interaction of licensing constraints,
h-licensing and constituent structure. SOAS Working Papers in Linguistics
and Phonetics 5, 23-40, University of London.
Cyran, E.
1997 Resonance Elements in Phonology: A study in Munster Irish. Lublin: Wy-
dawnictwo Folium.
Cyran, E. (ed.)
1998 Structure and Interpretation: Studies in Phonology. Lublin: Wydawnictwo Fo-
lium.
An Element-Based Analysis of Affrication in Japanese 213
Flikeid, K.
1988 Unity and diversity in Acadian phonology: An overview based on compari-
sons among the Nova Scotia varieties. Journal of the Atlantic Provinces Lin-
guistic Association 10: 64-110.
Fujiwara, Y.
1997 Nihongo-hougen no jissou 'The facts of Japanese dialects'. Zoku Shouwa ( •
Heisei) Nihongo-hougen no Sougou Kenkyuu Ά comprehensive study of Jap-
anese dialects in Showa (-• Heisei)' Vol. 4. Tokyo: Musashino Shoin.
Harris, J.
1990 Segmental complexity and phonological government. Phonology 7, 255-300.
1994 English Sound Structure. Oxford: Basil Blackwell.
Harris, J. and G. Lindsey
1995 The elements of phonological representation. In J. Durand and F. Katamba
(eds.), Frontiers of Phonology: Atoms, Structures and Derivations. Harlow:
Longman.
Ingleby, M., C. Chalfont and W. Brockhaus
1994 Robust techniques for recognition of new knowledge-based speech primi-
tives. In R. Togneri (ed.), The Proceedings of the Fifth Australian International
Conference on Speech Science and Technology. Canberra: Australian Speech
Science and Technology Association, vol. 1, 374-380.
Johnstone, Τ. M.
1967 Eastern Arabian Dialect Studies. Oxford:Oxford University Press.
Kaye, J., J. Lowenstamm and J. R. Vergnaud
1985 The internal structure of phonological elements: A theory of charm and gov-
ernment. Phonology Yearbook 2, 305-328.
1990 Constituent structure and government in phonology. Phonology 7,194-232.
Kiparsky, P.
1982 Explanation in Phonology. Dordrecht: Foris Publications.
Lahiri, A. and S. E. Blumstein
1984 A re-evaluation of the feature coronal. Journal of Phonetics 12:133-146.
Lahiri, A. and V. Evers
1991 Palatalization and coronality. In In C. Paradis and J.-F. Prunet (eds.), 79-100.
Lee, D.-Y.
1998 Korean Phonology: A Principle-based Approach. Munich: Lincom Europa.
Lee, D.-Y. and S. Yoshida
1998 Α-head alignment: the case of vowel harmony in Korean. In E. Cyran (ed.),
195-204.
Nasukawa, K.
1998 An integrated approach to nasality and voicing. In E. Cyran (ed.), 205-226.
Paradis, C. and J.-F. Prunet
1991 Introduction: Asymmetry and Visibility. In C. Paradis and J.-F. Prunet (eds.),
The special Status of Coronals: Internal and External Evidence. San Diego:
Academic Press, 1-28.
Poser, W.
1985 Japanese evidence bearing on the compensatory lengthening controversy. In
L. Wetzels and E. Sezer (eds.), Studies in Compensatory Lengthening. Dor-
drecht: Foris Publication, 167-186.
Rennison, J.
1990 On the elements of phonological representations: the evidence from vowel
systems and vowel processes. Folia Linguistica 24,175-244.
214 Shohei Yoshida
Shibatani, M.
1990 The Languages of Japan. Cambridge University Press.
Williams, G.
(1998) The Phonological basis of speech recognition. Ph.D. dissertation, School of
Oriental and African Studies, University of London. Also in Williams (forth-
coming).
(forthcoming) The Phonological Basis of Speech Recognition. The Hague: Holland Ac-
ademic Graphics.
Williams, G. and W. Brockhaus
1992 Automatic speech recognition: a principle-based approach. SOAS Working
Papers in Linguistics and Phonetics 2, 371-401.
The Accent System of the Kyoto Dialect of Japanese:
A Study on Phrasal Patterns and Paradigms
Introduction
(2)
* * * *
(3) * * * *
The data above show that in both SJ and KJ, a noun in isolation, or a se-
quence of a noun plus a particle, may form a phonological domain of pitch
accent. Otherwise, we should find pitch patterns where the noun portion
and the particle behave independently in terms of pitch assignment. For
The Accent System of the Kyoto Dialect of Japanese 217
The pitch patterns that are predictable from accentual information place
SJ and KJ properly within the accent-language group, comprising either
stress or pitch accent languages, rather than in the tone-language group.
From the location of the accent, those segments with high-pitch and those
without high-pitch can be predicted. In a tone language, each syllable
contains individual tonal information, high or low, and thus the location
of the high and low tones for a given word is lexically determined (i.e. ar-
bitrary). Nevertheless, the distribution of morae which are not high is pre-
dictable both in SJ and KJ, once we know the high pitch distribution. For
this reason, the present discussion of nouns in KJ does not consider a low
pitch or tone, but only deals with high pitch information.
Let us focus on the analysis by Haraguchi (1977) of the Osaka dialect,
which he chooses as being representative of the Kansai dialects, or the
Kyoto-type accent, 3 to show how lexical tone melodic information is irrel-
evant to the present discussion, and to support the typological claim
above. Haraguchi postulates two basic tone melodies, which are HL and
LHL, to differentiate the pitch patterns of the two groups of nouns and to
determine whether the word should carry high pitch on its initial mora or
not. Η of the tone melody is linked to the accented mora, or to the final
mora of an accentless phrase. The initial mora is high pitched, when the
word accent is on the initial mora, or if the pitch 'spreads' to the initial
mora from the accented one 4 or from the final mora in an accentless do-
main. Examples of the former type are more likely to be found in longer
words, such as compounds or quadrimoraic native nouns like uguisu (see
(4b)), where the accent is on the second mora of the word.
The analysis by Haraguchi accesses both accentual information and
tonal information for every lexical item, which means that the dialect is
questionably categorised both as an accent language and as a tone lan-
guage. Our claim in this paper, that KJ only refers to information relating
to accentuation and positions in the accent domain (see Section 2 for a de-
tailed analysis), makes tonal information redundant. That is, the pitch
patterns of KJ illustrated above can be accounted for without treating the
dialect simultaneously as an accent language and a tone language. We
would like to propose a simplified interpretation which need not assume
The Accent System of the Kyoto Dialect of Japanese 219
these tone melodies as lexical properties, and we argue that in this way KJ
may be properly defined as a member of the accent language group.
Section 2 is devoted to a detailed explanation of our analysis, and
broaches the issues of 1) how the diverse pitch patterns seen in the above
data appear, 2) lexical accent assignment on the antepenultimate mora,
and 3) the inaccessibility of the initial mora. To begin with, we explain
how various pitch patterns are possible in KJ, eschewing somewhat tem-
plate tone melodies.
In both stress accent languages and in pitch accent languages, the head
portion (the head of a syllable or mora—or, more precisely the nucleus)
of a given phonological domain bears the primary accent. If a domain
contains a lexical marking, then the lexically marked mora is automatical-
ly identified as the domain head. In a pitch accent language, the high pitch
originating from a lexical accent may be shared by neighbouring seg-
ments. SJ demonstrates a relatively simple sharing mechanism: the high
pitch of the head mora is shared by all the morae to the left, except the
domain-initial one. The domain-initial mora (nucleus) is inaccessible for
high-pitch sharing, perhaps to indicate the domain or word boundary
(Yoshida 1990). KJ chooses to share the high pitch of the accented mora
with all the morae to the left, if the sharing is observed within the domain.
We say if because another lexically determined option is available for Ky-
oto words: whether or not to share the high pitch with the neighbouring
segments. Compare, for example, two quadrimoraic words with antepen-
ultimate accent, murasaki 0H00 and uguisu ΗΗ00 (see (4a, b)).
A question arises as to which mora in an accentless domain should bear
headship. If the high pitch originates from a lexical accent as explained
above, then the source for the high pitch in a lexically accentless domain
220 Yuko Ζ. Yoshida and Hideki Zamma
KJ shows clearly the location of the head portion of a domain that lacks
an inherent head. One type of accentless word clearly indicates that the
domain-final mora is the head, both in the domain of a noun in isolation,
and in the domain comprising a noun and a particle (Yoshida 1999):
recent change in KJ, whereby some features of stress accent languages are
becoming apparent, assigns accent on to predictable locations.
(7)
a. 'nightingale' b. 'Kuwatani (a surname)' c. 'Kiyomizu (a place name)'
* * *
u gu i su ku wa ta ni ki yo mi zu
I 1 i
ψ ψ ή: % %
u gu i su / u gu i su ku wa ta ni / ku wa ta ni ki yo mi zu / ki yo mi zu
This change in accent location follows the default lexical accent assign-
ment process. A close observation of the accentual distribution of trimo-
raic words exemplified in (3) above clearly reveals that the default lexical
accent location is the antepenultimate mora. Sampling 115 morphologi-
cally simplex yamato (Japanese native) nouns from the word list in
Hirayama (1957), we report the following distribution.
(8)
Example number
*
initial accent (3a) 42
i ta ti 'weasel'
*
medial accent (3b) 20
ha ta ke 'farm'
final accent (3c) n.a 0
a. tu ku t ta 'make-past' tu ku t ta
* *
b. u ka Ν da 'float-past' u ka Ν da
* *
c. na ra Ν da 'queue-past' na ra Ν da
* *
d. si zu Ν da 'sink-past' si zu Ν da
* *
e. ka wa i ta 'dry-past' ka wa i ta
* *
f. ku da i ta 'crush-past' ku da i ta
* *
g. ta ta i ta 'beat-past' ta ta i ta
* * *
h. a ro o ta 'wash-past' a ro o ta ~ a ro ta
* * *
i. u to o ta 'sing-past' u to o ta ~ u to ta
The data in (7) and (9) indicate that at least the modern form of the dia-
lect shows behaviour similar to that seen in stress languages: the accent
falls at a predictable location, the antepenultimate mora. The type of pho-
netic interpretation of the head portion determines whether the language
is a stress-accent language or a pitch accent language. The accented
(head) portion receives phonetic interpretation either as prominence/
strength for a stress accent, or as pitch-height for a pitch accent in a given
domain (Yoshida 1995).
The Accent System of the Kyoto Dialect of Japanese 223
In the quadrimoraic nouns (7a) and verbs given above, there are two
possibilities; either, pitch is shared between the initial mora and the ac-
cented penultimate mora (9b-i), or it is not (9a). Recall the data (1-3) for
a general overview of this issue. As for shorter nouns, bimoraic 6 and tri-
moraic Yamato nouns, only accentless ones can vary in this way. Logically,
we should be able to observe pitch sharing on lexically accented bimoraic
and trimoraic words if the lexical accent is on the second mora (sharing
with the initial one). This possibility is exemplified by the quadrimoraic
noun uguisu (HHOO). In order to identify which factors determine wheth-
er or not pitch sharing occurs in a given domain, we turn to the case of ac-
centless nouns, which allow both possibilities.
(10)
a. ki i ki i -ga 'tree (-nom.)'
b. e e e e -ga 'handle (-nom.)'
c. ta ne ta ne -ga 'seed (-nom)'
d. ka ki ka ki-ga 'persimmon (-nom
e. u sa gi u sa gi -ga 'rabbit (-nom.)'
f. sa ku ra sa ku ra -ga 'cherry (-nom.)'
To investigate the possible reasons behind the two different pitch patterns
found in accentless domains, we must first verify whether these two types
both really lack 'accentuation' or not: if one has accentuation and the oth-
er does not, then a clear difference between them is immediately appar-
224 Yuko Ζ. Yoshida and Hideki Zamma
ent. These domains do not contain any information concerning the loca-
tion of lexical markings; however, it is not yet clear if the head mora, the
domain-final mora, receives accentuation as well as high pitch as a means
of indicating headship. The presence of accentuation is clearly shown, if it
appears word-medially; since the pitch drop is observed between the head
mora and the mora immediately following it. What is required is a reliable
test which reveals phrase-final accentuation in SJ. The juxtaposition of
two phonological phrases in connected speech exposes the accentuation
of the phrases involved, as demonstrated below in (11), where we observe
how the noun in isolation and with the case-marking particle react to be-
ing placed next to another phrase, e.g. a verb.
a. wa sa bi wa sa bi -o 'horseradish (acc.)'
* *
In connected speech, as (lla'-c 1 ) above show, only if the noun plus a par-
ticle phrase contains a lexical accent will the accent of the verb not be per-
ceived (known as phrasal downstep), since the leftmost accent is the one
projected to the noun-verb domain. As in (11c'), if no accent is found in
the noun portion, the accent of the verb becomes the accent of the entire
phrase. Bearing this in mind, let us consider the Kyoto phrases.
(12)Kyoto
*
a. wa sa bi wa sa bi - o 'horseradish (-acc.)'
b. ki tu ne ki tu ne - o 'fried bean curd (-acc.)'
c. sa ka na sa ka na - o 'fish (-acc.)'
*
a', wa sa bi - o ta be ta
b'. ki tu n e - o- ta
. be
,(*)ta * ki tu n e - o ta be ta * ki tu n e - o ta be ta
The Accent System of the Kyoto Dialect of Japanese 225
* *
c'. s a k a n a - o ta be ta * sa ka na - o ta be ta * sa ka n a - o ta be ta
The pitch patterns in (12) show that all the noun phrases in Kyoto should
bear accentuation. All the examples are subject to phrasal downstep. This
phenomenon indicates that both (12b, c) show accentuation in the do-
main-final position (Section 2.1.). The domains do not have lexical ac-
cents (the lexical accent of a noun should remain intact following its con-
catenation to a nominative marker), but they do show 'accentuation' in
the domain-final position (the head position). Here we propose the fol-
lowing for KJ:
(13) Proposal 1
Nouns which belong to the classes (lb, c), (2c, d) and (3d, e) are all
LEXICALLY ACCENTLESS. The rightmost segment of such a
word is assigned a domain ACCENT, which we call a PSEUDO
ACCENT.
noun. The final mora of the accentless phonological word (domain) takes
over the headship of that domain. Note that in SJ this is not interpreted
as an 'accent', but merely as high pitch. Yoshida (1995) proposes that the
noun-phrase and the verb form independent domains (14a). When con-
catenated, the leftmost accent of the noun-verb string is projected to the
domain of AB as the accent of the entire phrase.
(15) [ C [A][B] C ]
In SJ, where both A and Β contain lexical marking, i) if C is a lex-
ical category, Β is strong, ii) if C is a phrasal category, A is strong,
and the marked nucleus of the strong element is the head of the
domain. (Yoshida 1995:194)
(16)
*
Merging of two domains with lexical markings into one forces the output
domain to select one of the two markings as the head of the output do-
main. As is generalised in (15), the leftmost lexical marking is chosen as
the head of the merged domain, since the domains involved here are both
phrasal. In SJ, as (16a-c) show, only if there is a lexical accent within the
noun-phrase domain will the accent of the verb not be perceived. As in
(16c), if no lexical accent is found in the noun portion, the accent of the
verb becomes the accent of the merged domain.
The Accent System of the Kyoto Dialect of Japanese 111
(17)
*
a. [[wa sa bi - o ] ta be ta]
*
(*)ta]
c. [[sa ka n a - o ] ta be * [[sa ka n a - o ] [ta be ta]] *[[sakana-o] [ta be ta]]
When the noun lacks the overt case-marker, the bracketing remains the
same.
(18)
-1 (*)
a. [[wa sa bi] ta be ta ]
_ (*)
b. [[ki tu ne] ta be ta] * [[ki tu ne] [ta be ta]] * [[ki tu ne] [ta be ta]]
* *
(*)
c. [[sa ka na] ta be ta] * [[sa ka na] [ta be ta]] * [[sa ka na] [ta be ta]]
The properties of the inner domain are projected to the outer domain, and
in this [[A]B] structure, the head of A, the inner-domain, is projected as the
head of the domain AB. The inner domain lacks a lexical marking in kitune
and satana (18b, c)—unlike the inner domain of wasabi 'm (18a). Notwith-
standing, the domain A has to nominate the potential head for the outer
domain AB, thus the head of A, the domain-final nucleus, is projected to
the merged domain AB, to be interpreted as the pseudo accent.
In SJ too, there exist cases where a phrase such as a noun-particle se-
quence acquires an accent when followed by another case-marker 7 . The
228 Yuko Ζ. Yoshida and Hideki Zamma
location of the accent is indicated by the drop in pitch on the mora adja-
cent to the right of the accented mora, as observed in the phrase in (19c).
The fact that the dative noun phrase sakura-ni is not accented anywhere
in the phrase is demonstrated by the data in (19d), where the noun-phrase
is followed by an accented verb.
lect, phrases (where the noun phrase is represented as A and the verb as
B) are organised into [[A]B] and Accent Clash Avoidance is observed
following the juxtaposition of two accents. Note that in Κ J, as with other
Kansai dialects, overt case-marking particles tend to be omitted. Recall
that there is no noun class with a lexical accent on the word-final mora
(see (3)). So if there is actually a pseudo accent on the word-final mora of
the noun (domain A), then the accent should react to the presence of a
verb-initial accent.
There are two types of Accent Clash Avoidance, one deleting the pseudo
accent in the noun (23b) and the other merging the pseudo accent with
the accent of the verb. Indeed, the location of the accent in the verb holds
the key to this phenomenon. Compare this with the data presented in
(18), which is repeated for convenience in (24). The accent is on the sec-
ond mora of the verb, and is not adjacent to the pseudo accent in the noun
portion.
(24)
*
a. [[wa sa bi] ta be ta ]
*
So far, we have examined two points. Firstly, both the usagi type (25a) and
the sakura type (25b) of noun are lexically accentless; secondly, those ac-
centless nouns acquire a pseudo accent in the domain-final position. They
do belong to the same lexically accentless class, although the actual pitch
patterns of the two types are different. In this section, we pursue the ques-
tion of how the differences in pitch patterns arise.
(25) _ _
a. u sa gi u sa gi -ga 'rabbit (-nom.)'
b. sa ku ra sa ku ra -ga 'cherry (-nom.)'
Basically, the difference between the sakura type and the usagi type of
noun rests on whether the high pitch is shared by other morae or not. The
span of high-pitch sharing in an accented word extends from the accented
mora to the leftmost mora of the noun. For example:
(26) High-pitch sharing in the Kyoto dialect —from the accented mora to
the leftmost mora of the relevant domain —
u gu i su ~ u gu i su
The form in (26a) shows that the high pitch on the accented mora can be
shared by an adjacent mora. Alternatively, the same noun can be pro-
nounced as in (26b), depending on individual speaker choice. This form
manifests itself only when the accented mora bears the high pitch without
extending it to other morae. For most quadrimoraic nouns, the same re-
sult emerges: there are two alternative pitch patterns (see (24a, a')).
Note that in SJ, the initial nucleus (the head of the initial mora) is inac-
cessible for high-pitch sharing (see Section 2.1.). The same is true in KJ,
although this system differs from SJ by marking it at the lexical level. In
SJ, where high-pitch sharing is obligatory for all classes of words where
sharing is possible, any phonological domain is subject to this inaccessibil-
ity. However, KJ chooses to treat this as lexical information, so the shar-
ing of high-pitch comes about as a result of referring to lexical properties.
232 Yuko Ζ. Yoshida and Hideki Zamma
(27) PROPOSAL 2
In the Kyoto dialect, the domain-initial mora may be lexically inac-
cessible for high-pitch.
(28)
a. u gu i su ku wa ta ni ki yo mi zu
'nightingale' 'Kuwatani (a surname) 'Kiyomizu (a place n a m e ) '
*
a', u gu i su ku wa ta ni ki yo mi zu
b. χ χ
to ri 'bird' sa ku ra 'cherry'
c. _x _x
a na 'hole' u sa gi 'rabbit'
As shown in (29a), the lexically accented verb stem tabe- is always accent-
ed, whichever suffix is attached to it, although the placement of the accent
changes according to the choice of suffix. On the other hand, the lexically
unaccented verb stem asob- is always accentless. Moreover, the melody
for the accent is always HL, using the term employed in Haraguchi
(1977); that is, pre-accent morae all have high pitch (except for the first
mora, which is pitchless because of initial inaccessibility (cf. Yoshida
(1995)) or Initial Lowering (cf. Haraguchi (1977))), while the morae to
the right of the accent are all pitchless. When the word is accentless, on
the other hand, all the morae in the word have high pitch (again, except
for the initial one).
Both of these properties, however, differ dramatically in KJ, as shown
below:
(31)
*
a.
agar-u 'rise (indicative)' agat-ta 'rise (past)'
234 Yuko Ζ. Yoshida and Hideki Zamma
Compare the indicative forms and the past forms in (31). You will notice
that none of the verbs are accented in the former, while all are accented
in the latter (albeit with a difference in pitch pattern). This means that the
verb stem itself does not contain information about accentedness, and
that the latter is determined by the suffix attached to it; hence (30a). In
the cases above, the indicative suffix -(r)u is not accenting while the past
suffix -ta is accenting.
Moreover, the difference between (31a) and (31b) shows that there are
two pitch patterns both for accentless and accented words. The indicative
forms in (31a) and (31b) are both accentless, as we have just mentioned,
but only the final mora of the words in (31b) contain the high pitch. Sim-
ilarly, the accented past forms in (31b) have the initial pitchless mora,
which those in (31a) lack. This fact clearly suggests the appropriateness of
the generalization we gave in (30b). And, from the fact that the suffixes
do not play a crucial role in determining the pitch pattern, it is natural to
conclude that this is determined by the verb stems; thus in (31), the verb
stems agar- and ake- have, in Haraguchi's terms, the HL melody, while
aruk- and iki- have LHL.
(32) indicative
a. iw-u10 ki-ru agar-u ake-ru
'say' 'wear' 'rise' 'open (tr.)'
b. aw-u mi-ru aruk-u iki-ru
'meet' 'look' 'walk' 'live'
(33) polite
a. iw-1-masu ki-masu agar-i-masu ake-masu
b. aw-i-masu mi-masu aruk-i-masu iki-masu
(34) imperative
a. iw-i ki-i agar-i ake 11
b. aw-i mi-i aruk-i i ki
(35) hortative
a. iw-o ki-yo agar-o ake-yo
b. aw-o mi-yo aruk-o i ki-yo
It is clear from the examples above that these suffixes do not assign an ac-
cent to the verb: all of the words are unaccented, with high pitch assigned
to all morae in group (a) words and to the final mora in group (b) words.
Moreover, the pitch pattern is consistent: in other words, if the initial
mora of the stem is high-pitched/pitchless in one of the forms with an un-
accented suffix, then the mora is also high-pitched/pitchless with other
unaccented suffixes. The various forms of the stem aw- 'meet', for exam-
ple, all have an initial pitchless mora, while those of iw- 'say' all have an
initial high-pitched mora.
There is, however, one exception to this generalization: the LHL and
HL patterns are neutralised in all forms of C-final 2-mora verbs, except
the indicative. Note that the stem mi- shares the same pattern with ki- in
(33)-(35). Because a highly idiosyncratic procedure would appear to be at
work here, we will ignore this class of words for the purposes of the
present discussion.
236 Yuko Z. Yoshida and Hideki Zamma
Note that the pitch pattern is consistent across all the forms, including
those in the previous section: iw- has an initial high-pitched mora in all
the forms from (32) through to (37), and similarly, aw- has an initial pitch-
less mora in all the forms. It is clear that the suffixes determine accentua-
tion, while the pitch pattern is determined by the stem.
(38) provisional
£ * * Ί· -i·
iw -e-ba ki-re-ba agar-e-ba ake-re-ba
^ ^ Ψ ífc
aw-e-ba mi-re-ba a ruk-e-ba i ki-re-ba
In this form, the accent falls on the second mora preceding the suffix, ex-
cept in aw-e'-ba, where the accent falls on the mora immediately preced-
ing the suffix. Note that the forms in Section 3.2.2.2. all have the accent
on a designated mora—specifically, the mora immediately preceding the
suffix. The provisional forms are irregular in this respect.
The Accent System of the Kyoto Dialect of Japanese 237
(39) past
^ ^ ^ ^
i t-ta ki-ta a gat-ta a ke-ta
^ Η· Ψ
oo-ta mi-ta a ru i-ta i ki-ta
The accent falls on the second mora preceding the suffix, except in oo-ta,
where no accent appears, and iki'-ta, where the accent falls on the mora
immediately preceding the suffix, (a'gat-ta is also an exception, but we
will ignore this because Η-initial 4-mora verbs show a unique pattern of
behaviour; e.g. i'w-as-i-ta (where -(s)as- is a causative suffix), ki'-sas-i-ta
vs. agar-a's-i-ta, ake-sa's-i-ta. Note, however, that the pattern illustrated
by a'gat-ta is changing to the expected one, resulting in aga't-ta, as is dis-
cussed in Section 2.2. (see (9)).)
Finally, consider the colloquial negative forms:
in 3-mora words, then the accent moves rightwards. See the following di-
agrams:
b. 1
a-e-ba
In (41a), the usual assignment of accent by the suffix -ba is carried out, ac-
cording to which the accent falls on the second mora preceding the suffix.
It means, however, that the initial inaccessibility fails to be satisfied when
assigning the accent to the initial mora. To respect the lexical marking of
initial inaccessibility, it is necessary to move the accent rightwards by one
mora, as shown in (42b).
Note that this exceptional behaviour is observed only when both of the
following conditions are met: (i) the stem exhibits the OHO pitch pattern;
and (ii) the suffix otherwise assigns the accent on the initial mora of the
verb. Recall that other exceptional forms besides a-e'-ba all conform to
these conditions: e.g. oo-ta (unaccented), iki'-ta (cf. aru'-i-ta)\ aw-a'-hen,
iki'-hin (cf. aru'k-a-hen). This fact clearly shows the suitability of the
present analysis.13
(43) *
a. accent on the initial mora: e.g. a tama 'head'
b. accent on the second mora, *
where the initial one is pitchless: e.g. katana 'sword'
4. Conclusion
Acknowledgements
We'd like express our gratitude to Phillip Backley, Jeroen van de Weijer and Tetsuo Nishi-
hara, and the anonymous reviewer of this volume for invaluable comments and suggestions
for this paper. Our thanks also go to the audience of the Phonology Forum 1999, held at the
Tokyo Metropolitan University (2-4 September) where an earlier version of this paper was
presented; especially Shosuke Haraguchi, Takeru Honma, Itsue Kawagoe, Akio Nasu, Koi-
chi Tateishi, Shin-ichi Tanaka, Noriko Yamane, and Teruo Yokotani. This paper is dedicated
to Jion.
240 Yuko Ζ. Yoshida and Hideki Zamma
Notes
1. The unit employed in this paper is the mora, which refers to any of the following: 1) a
vowel preceded by zero or one consonant e.g. [e] or [fa], 2) the initial half of a geminate
consonant [kippu], 3) the latter half of a long vowel [too] or a diphthong [da/], 4) a mo-
rale nasal [hoiV]. The issue of the unit, mora vs. syllable, is not central to our discussion.
Briefly, however, KJ abounds with lexical items with accents even on those special mo-
rae, 3) & 4), e.g. on the latter half of a diphthong koi 'carp' and on a moraic nasal as in
toNbi 'black-eared kite'. This contrasts with the SJ case, in which lexical accents are not
found on those morae, the non-head portion of a syllable.
2. A recent change in the dialect form is found in this type of noun. In Hirayama (1957),
the accented final mora in bimoraic words is lengthened when in isolation (i.e. without
a particle), although the preference of many speakers today is to use the final accented
form without lengthening.
3. Some minor dialectal variations of the data between the Kyoto and the Osaka dialects
should not affect the basic analysis; in other words, most of the Kyoto data fits well with
that of the Osaka dialect.
4. In KJ, no such words are attested—lexically accented bimoraic or trimoraic simplex Ya-
mato words which are subject to pitch spreading.
5. In SJ too, the rightmost mora serves as the head of the accentless domain (Yoshida
1999).
6. Here the term 'bimoraic' also refers to lengthened monomoraic forms such as kii 'tree'
(ki in SJ).
7. This is also pointed out by Koichi Tateishi and Teruo Yokotani, and we have always
agreed with the idea that the case-marked nouns obtain final-accents when followed by
another particle. We still believe the genitival -no, however, the case marker which is
discussed later in Section 3.1.2., cannot be generalised with other case-markers. Geniti-
val -no at the same time functions to 'nominalise' the phrase as a pronoun (equivalent
to one in English), as in the form wasabi-no, meaning 'one with wasabi\ This function
is unique to this genitival particle, and perhaps it is reasonable to assume a lexical accent
on the particle, unlike all other cases.
8. A correlate between the two types of lexical marking, lexical accent marking and the
inaccessibility of the initial mora, can be noted here. As section 1.1. illustrates, KJ con-
tains no lexically accented words which chooses to share high-pitch. Lexical accent
marking may have accompanied this inaccessibility in that lexical domain. However,
wherever we see high-pitch sharing we also observe accent assignment i.e. quadrimoraic
nouns (see 2.2.) and the lexically accentless nouns. Those nouns dominate a lexical do-
main without any markings for the lexical accent. In such a domain, where there is no
lexical-accent information, the inaccessibility information was not available from the
outset. Perhaps some of the accentless words, by analogy, set the lexical marking of in-
accessibility, along with accent assignment for quadrimoraic words, or along with the as-
signment of pseudo accent for lexically accentless trimoraic words. This correlate, how-
ever, requires more detailed consideration, and is therefore not pursued further in this
discussion.
9. The stem-final segment of C-final verbs undergoes some changes called onbin in the
past and the gerund forms. It is also observed in (31) and (39) below. The segmental
change to the vowel in oo-ta in (39b) is also classified as onbin in Japanese linguistics.
10. As is well-known, stem-final /w/ in Japanese only appears before /a/ and is deleted else-
where. The segment is represented in the examples only to show that the words contain
/w/ underlyingly.
The Accent System of the Kyoto Dialect of Japanese 241
11. The allomorph of the imperative suffix for V-final stems is φ (zero). The hi in 2-mora V-
final verbs in (34) is not regarded as the suffix, but as part of the preceding vowel, which
is lengthened to satisfy binarity (cf. Itô and Mester (1992)).
12. The vowel of this suffix has harmonised to the one in the stem. Moreover, the stem vow-
el in this form has lengthened, probably to satisfy the binarity (cf. Itô and Mester
(1992)) required by the suffix -hen.
13. In the case of oo-ta, it is accent deletion which is carried out in order to preserve the lex-
ical marking. This is consistent with our claim here because deletion is observed only in
those words where the initial mora would otherwise be accessible. This accent deletion
may have something to do with the fact that the oota class is consonant-ending, which
undergo onbin alternation (see note 9). Note that the consonant typically becomes the
first part of a geminate through onbin, which can hardly bear the high pitch by itself. In
order to avoid the pitch contour OHO for tat-ta 'stand (past)', for example, accent deletion
may take place. Non-geminate-ending stems may be aligned to this pattern.
References
Bloch, Bernard
1946 Studies in Colloquial Japanese I: Inflection, Journal of the American Oriental
Studies 66.
Haraguchi, Shosuke
1977 The Tone Pattern of Japanese: An Autosegmental Theory of Tonology.
Kaitakusya. Tokyo.
Haraguchi, Shosuke
1991 A Theory of Stress and Accent. Foris. Dordrecht.
Hirayama, Teruo
1957 Nihongo Oncho no Kenkyuu (A Study of Japanese Accent). Meiji Shoin. Tokyo.
Itô, Junko and Armin Mester
1992 Weak Layering and Word Binarity. ms. University of California, Santa Cruz.
Kaye, Jonathan
1990 'Coda' Licensing, Phonology 7:2, 301-330.
1995 Derivations and Interfaces, in Frontiers in Phonology: Atoms, Structures, Der-
ivations, ed. by Jacques Durand and Francis Katamba. Longman. London.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. Mouton. The
Hague.
Poser, William
1984 The Phonetics and Phonology of Tone and Intonation in Japanese, Doctoral
Dissertation. MIT.
Yoshida, Yuko
1995 On Pitch Accent Phenomena in Standard Japanese. Doctoral dissertation.
SOAS, University of London. [Published in 1999 by Holland Academic
Graphics.]
1999 Binary Inter-nuclear Licensing for Accent Assignment in Relation to Phono-
logical Domains, in Phonologica 1996: Syllables, ed. by John Rennison. Hol-
land Academic Graphics. The Hague.
II. Studies in Japanese Morphology
Word Plus: The Intersection of Words and Phrases
Taro Kageyama
Kwansei Gakuin University
Introduction
Such prefixed and compound expressions not only have the phrase-like
phonology in common but can also be freely embedded inside each other
to produce more complex expressions with iterated minor phrase bound-
aries, as exemplified by (4).
Particularly noteworthy about (7) and (8) is the fact that the ungrammatical-
l y is removed if their NP parts are replaced by compound nouns, as in (9).
The deletion rule, which can erase the head noun of the NP structure in
(10b), is forbidden to affect only the second member of the compound
noun inu-goya 'doghouse' in (10c). In the same vein, the deletion cannot
destroy the integrity of the complex expressions involving the phrase-like
prefixes. The same degree of ungrammaticality as (10c) is exhibited by
(11a) below, where only the base is deleted leaving the prefix behind. This
Word Plus: The Intersection of Words and Phrases 251
(12a) is impeccable, since the relative clause with the verb iru 'be', said of
an animate subject, modifies the following human noun as a whole. In
contrast, (12b) is completely ruled out because the relative clause with
the verb aru 'be', selecting an inanimate subject, is intended to modify
only the first member (booeki-gaisya 'trading company') of the com-
pound. This is parallel to the infelicity of, for example, *a very [dark-
room], where the adverb very cannot modify the adjective dark in the
compound noun darkroom. The word status of booeki-gaisya\syatyoo
'trading-company president' is confirmed by the fact that (12b) becomes
grammatical if booeki-gaisya 'trading company' gains independence as a
phrase with the genitive marker, as in (12c).
The preceding three observations clearly indicate that the prefixed and
compound expressions with a minor phrase boundary make up syntacti-
cally inviolable units and are therefore qualified as words. Additional ev-
252 Taro Kageyama
idence for the word status is provided by the fact that the complex expres-
sions in question are amenable to lexical or morphological conditions that
generally pertain to word formation.
As is well known, word formation processes are lexically governed, re-
stricted notably by the vocabulary strata determined by the origin of in-
dividual words. In Japanese there is a pronounced tendency for a native
morpheme to combine with another morpheme of native origin and for a
Sino-Japanese morpheme to combine with another Sino-Japanese mor-
pheme, although words composed of mixed vocabulary strata are occa-
sionally encountered. This combinative restriction is illustrated by (13).
It seems that only the native moto- exceptionally meshes with both Sino-
Japanese and native bases.
In (16), zen- 'all' and itibu- 'part' are eligible for word formation, but seman-
tically similar words, hanbun 'half and kahansuu 'majority', resist comound-
ing and instead call for a phrase structure supported by the genitive particle.
Lexical idiosyncrasies of this kind are characteristic of word formation.
This section has demonstrated the word status of the peculiar phrase-
like expressions on the basis of phenomena involving lexical integrity and
lexical idiosyncrasies. Because of the word status, we will refer to the
phrase-like words, both prefixed and compounded, as Word"1" (Word
Plus). This is intended as a new morphological category that is larger than
ordinary words but still belongs to the morphological as opposed to the
syntactic domain. For example, the morphological structures of zen \ dai-
tooryoo 'ex-President' and Reagan | zen | daitooryoo 'ex-President Re-
agan' will be represented as follows.
(17) a. N* b. N+
prefix
zen Ν Ν prefix
'president'
In (17a), the phrase-like prefix zen- 'ex-' selects N + as its base. This c-se-
lection property need be specified in the lexical entry of the prefix, along
254 Taro Kageyama
The phrase-like accent contour and the absence of sequential voicing sug-
gest that the constituent parts of W + are independent units connected
with each other very loosely.
The phonologically loose connection of W + is correlated with the inter-
nal semantics. Compare the meanings of a regular compound (19a) and a
W + compound (19b).
Here we are concerned only with inbound anaphora. Sproat (1988) and
Ward et al. (1991) resort to different morphosyntactic principles to ac-
count for the impossibility of inbound anaphora. For Sproat (1988), in-
256 Taro Kageyama
The doo- in (22a) means 'same, identical' and the complex word involving
it carries the regular lexical accent. On the other hand, the homophonous
prefix in (22b), our present concern, has entirely different properties: it is
pronounced with a phrasal accent and has the function of referring to a
previously mentioned entity. Thus, doo\syuu 'the said state' is bound by
Washington-syuu 'state of Washington' in the preceding clause.
Before delineating the referential properties of the second instance of
doo-, we should first show that this prefix is truly referential rather than
deictic. This point is significant, because if the function of this prefix were
deictic, then its apparent referentiality could be reduced to pragmatic in-
ferences just like the English outbound anaphora in (20). We prove the
point by comparing doo- with two other W + prefixes which also have ref-
erential functions: hon- 'this, the present' and too- 'the said, in question'.
These two prefixes are purely deictic, as is shown by the fact that sentences
like (23a, b) can be used merely by pointing to the entities in question,
without reference to any linguistic antecedent. If we follow Sag and
Hankamer's (1984) distinction between "deep anaphora" and "surface
anaphora", the referential function of these two prefixes is characterized
as deep anaphora, which is determined pragmatically or deictically. In
stark contrast to these two is the prefix doo-, which turns out to participate
in surface anaphora. This prefix invariably requires an overt linguistic an-
tecedent and cannot be used to point deictically to an entity placed before
the speaker's eyes or pragmatically to an entity evoked in the brain.
It is thus found that the W + prefix doo- is endowed with the function of
creating a referential expression which presumably corresponds to pro-
nominals. While the English personal pronouns are distinguished by gen-
der, person, and number, the Japanese doo- can be prefixed to an unlim-
ited number of Sino-Japanese head nouns to give rise to pronominal
expressions of various semantic types. For example, doo \ sya 'the said
company' can refer to any antecedent NP identifiable as a business firm,
doo I daigaku 'the said university' to any NP identifiable as a university,
and doo \ koku 'the said country' to any NP identifiable as a country.
Having clarified the pronominal function of the prefix doo-, we now
provide examples of inbound anaphora involving this prefix inside.
258 Taro Kageyama
inherent in the prefix doo-, which itself belongs to the W + category. Thus,
if ordinary DP pronouns are employed, the islandhood of W + expressions
is restored with respect to both inbound and outbound anaphora. Witness
the following examples where the doo-words in (27a) and (26b) are re-
placed by ordinary pronouns.
The two sets of examples in (32) and (33) demonstrate the parallelism be-
tween NP structure and W + structure. Regardless of the presence or ab-
sence of the genitive particle, doo-pronominals are accepted if they are
embedded within a larger branching structure and are not directly gov-
erned by the antecedent. This is a strong indication that the morphological
branching inside W + is visible to the syntax, at least for the purpose of in-
terpreting the reference of doo-words. It should be stressed, however, that
we are not claiming that the syntax can freely access the internal structure
of W + . Were this the case, our arguments in section 1 for the syntactic ato-
micity of W + would lose ground. Rather, our claim is that the islandhood
is removed by virtue of the inherent referential property of the prefix doo-.
For another, there are a myriad of examples which have one and the same
semantic relation and yet bear different accentual patterns. The examples
in (34d) and (34e) should be compared with those in (36), which have the
same semantic relations but display the regular lexical accent.
notion of subcategorization, though tenable for affixes, does not hold for
compound words like those we brought out above. In order to account for
those compounds, it will be necessary to set up a prosodie rule that com-
bine two minor phrases into a compound. This, however, begs the funda-
mental question of exactly what causes the difference between regular
compounds and minor-phrase compounds.
Our approach is neither semantic nor prosodie, but morphological. It is
proposed that the accent patterns are direct reflections of the morpholog-
ical architecture of compound and affixed words. To prove this point, we
leave out semantic and prosodie considerations, concentrating our atten-
tion on the morphological composition.
It is known that Japanese words are grouped into several classes, de-
pending on their internal makeup (Kageyama 1982). The basic units of
word formation are single morphemes, which may be either native or
Sino-Japanese. In the case of Sino-Japanese morphemes, the smallest
building block is represented by single Chinese characters like hoo 'visit'
and bei 'America', which in turn are combined to form two-character
words like hoo-bei 'a visit to America', which may be further expanded to
bigger words like hoo-bei yotei 'a schedule of a visit to America'. The or-
derly processes of word formation observed in such examples will be eas-
ily formalized in terms of the level-ordered morphology. For example, the
process of compounding single Sino-Japanese morphemes is located at an
earlier level than that of compounding two-character words.
Separation of these two levels accounts for the fact that the level 1 com-
pounding is strictly limited to single morphemes and cannot apply to
words written with two or more Chinese characters. Compare hoo-bei
with the ungrammatical makeups in (38).
(39) a. Root: single morphemes like hoo- 'visit' and bei 'America'
b. Stem: two-morpheme words like hoo-bei 'a visit to America'
c. Word: words that can be used independently, typically with more
than two morphemes
These building blocks are combined with each other by rules like (40) to
make larger and larger units.
We need not concern ourselves with the technical details of these repre-
sentations. What is important is that the difference in 'size' among W, S,
264 Taro Kageyama
Faced with examples like (42), one might wonder whether the coordinat-
ed elements as a whole are embedded inside the complex words or they
are derived by conjunction reduction from a fuller structure as in (45).
Although the reduction analysis cannot be dismissed off hand, it has some
unfavorable consequences. First, the suggested reduction rule deleting
part of a word results in a blatant violation of the lexical integrity princi-
ple. Second, some additional conditions will have to be introduced to ac-
count for the fact that R- and S-level words do not allow the deletion.
Third, the deletion analysis claims that the resultant structure is some-
thing like kokuritu- 0 | naisi kooritu-daigaku, where the morphophono-
logical break falls right after the first element kokuritu- 'national', but the
native intuition dictates that the coordinated part [kokuritu- naisi koori-
tu-] 'national or prefectural/municipal' is a cohesive unit.
Word Plus: The Intersection of Words and Phrases 265
Besides naisi and oyobi, which are of native origin, there is a Sino-Jap-
anese morpheme which typically concatenates two or more words: ken
'and, cum', as in soori-daizin ken gaimu-daizin 'Prime Minister and (con-
currently) Minister of Foreign Affairs' and sinsitu ken ima 'a bed-cum-sit-
ting room'. In the light of this conjunctive morpheme, it is not unreason-
able to treat naisi 'or' and oyobi 'and' as sublexical coordinators that
directly conjoin words.
As expected, these coordinators are able to appear inside the W + ex-
pressions as well. Below are some examples from Kageyama (1993).
Insofar as the examples in (42) and (46) are concerned, the W + expres-
sions are no different from ordinary compounds with lexical accent.
However, a notable difference is detected if another conjunction, to
'and', is taken into account. The canonical function of this conjunction is
to connect NPs rather than mere words. Substitution of to for naisi and
oyobi in W-level compounds (42) yields outrageous results like (47).
(51) a.
prefix prefix
deceased Ν
Somehow related to these peculiarities is the fact that moto- can take
what appears to be a genitive NP. Observe the following contrast between
the two semantically similar prefixes, moto- and zen- 'ex-'.
Only moto- allows a genitive phrase after it. More examples follow in
(55), where the genitive particle no seems relatively acceptable, with the
same meaning as the compound versions without the genitive.
It is not the case, however, that moto- can attach freely to any type of NPs.
Descriptive modifiers like adjectives and relative clauses are totally ruled
out.
(58) haha no hi 'Mother's Day', nomi no iti 'flea market', ama no zyaku
'devil's advocate', ama no gawa 'the Milky Way', mago no te 'back
scratcher'
accent, like haha no hi purezento 'Mother's Day present' and nomi no iti
biyori 'fine weather suitable for flea markets'. Such lexical compounding
is not possible with the genitive phrases in (55). This leads us to conjec-
ture that the exceptional occurrence of the genitive in (55) might be due
to the contamination of the two different usages of the prefix moto- as a
W + prefix and as the phrasal determiner taking an NP structure (53). Put
another way, the no-phrases in (55) are attached directly to moto- (by mis-
taken analogy) and are consequently subjected to the general principles
of word formation like the naming function.
This conclusion implies that there is no necessity to postulate move-
ment either on the part of the prefix moto- or on the part of its base at any
level of derivation, like the LF movement of affixes proposed by Pesetsky
(1985) or the syntactic incorporation of Ν to D suggested by Miller (1993:
104). The absence of movement will be ascertained by the scope relation
of moto-. It is a general rule that the prefix takes the base (i.e. the material
it attaches to) in its scope. In (59a), moto- takes scope over syusyoo
'Prime Minister', whereby the proper name (Thatcher) that appears to
the left of moto- is outside the scope. This is reasonable, since 'ex-' desig-
nates only the official position.
The scope relation makes (59b) meaningless. This example could make
sense only in a hypothetical situation where Prime Minister Thatcher
changed her family name after leaving her office.
To sum up, we have observed that probably due to its native status, mo-
to- has the exceptional property of expanding its base from Word + to a
genitive N'. The N', however, must be restricted to conventionalized
nameworthy concepts and cannot include descriptive modifiers, because
the whole moto- complex counts as a "word".
on the rear (Fudge 1984, Ladd 1984). Obviously, the former compounds
correspond to the Japanese compounds with a lexical accent, and the latter
to the Japanese W + compounds with a phrasal accent (Kubozono 1995).
The phrase-like stress pattern is also observed with such prefixes as non-,
ex-, and anti- which have their own stress (Bates 1988).
The similarity with the Japanese W + becomes even more striking when
we notice that those complex expressions allow internal coordination.
Expressions like these, which have hitherto been analyzed simply as com-
pound words on a par with regular Word-level compounds, are now iden-
tified as candidates for the W + category.
It is interesting to note, as an anonymous review pointed out to me, that
the prefixes in (61) and (62) are borrowings from Latin or Greek, as all
the Japanese phrase-like prefixes except moto- 'former' are borrowed
from Chinese. Presumably, this will not be a mere coincidence but follow
from the nature of loan morphemes which is as yet to be explicated. Here
I can only speculate that native morphemes naturally fuse together into
tight units, forming words of relatively low levels (Stem, Root, Word),
whereas borrowed morphemes tend to resist total assimilation into cohe-
sive word units and some are merely added to the outermost shell of word
structure as an extra layer, Word + . An in-depth study is obviously neces-
sary on the nature of loan morphemes.
The Word + analysis can also provide an effective explanation for the
occurrence of inflectional elements inside complex words. Consider the
position of the plural -s in two types of compound words:
(64) onlookers
If we assume that the plural inflection creates the Word"1" category, the in-
ternal plural marker in the compounds in (63b) can be explained by the
following derivation.
(65) look + -er -> [looker] N > [[looker] N -s]N+ - [ [[looker] N -s]N+ [on] P+ ] N+
The adjective-forming suffix -fui in (66a) is stressless with its vowel re-
duced to a schwa, whereas the -fui in (66b), denoting the amount held by
a container, carries a stress on its own and its vowel is fully pronounced
as [-ful]. Allen (1978) accounts for the distinction of the two types of -ful
by assigning the adjectival -ful to Level 2 and the fully pronounced -ful to
Level 3 in her extended level-ordering hypothesis. Our alternative analy-
sis is to treat Allen's Level 3 as constituting a distinct category of W + . This
analysis is supported by the fact that the expressions in (66b) can have the
plural inflection inside them, although a recent tendency is to put the plu-
ral on the right edge.
The examples in (63b), (66b), and (68b) have their internal structures ac-
cessible to the syntax, so that the plural and comparative/superlative in-
flections, which are generally held to take place in syntactic structure,
sneak into the branching W + structure.
Once W + is acknowledged as a legitimate, albeit marginal, morpholog-
ical category, a systematic analysis can be offered to examples like (70)
which have been slighted as simply idiosyncratic.
5. Conclusion
Recently, Bisetto and Scalise (1999) have taken up Italian complex ex-
pressions which look like compound words but are analyzed as syntactic
phrases because of the insertability of lexical material. From this they
conclude that morphology and syntax can be distinguished as different
components. The phrase-like expressions discussed in this paper pose a
Word Plus: The Intersection of Words and Phrases 273
(70)
syntactic XP
phrasal accent, reference by prefix doo-,
structure X'
plural and comparative/superlative inflections
Word +
morphology Word
structure Stem lexical accent
Root
Acknowledgements
I am grateful to the editors of this book for inviting me to contribute, and to two anonymous
reviewers for suggestive comments on the draft.
Word Plus: The Intersection of Words and Phrases 275
References
Ackema, P.
1999 Issues in morphosyntax. Amsterdam: John Benjamins.
Allen, M.
1978 Morphological investigations. Ph.D. dissertation, University of Connecticut.
Baker, M.
1988 Incorporation. Chicago: University of Chicago Press.
Bates, D.
1988 Prominence relations and structure in English compound morphology. Ph.D.
dissertation, University of Washington.
Bisetto, A. and S. Scalise
1999 Compounding: morphology and/or syntax? In L. Mereu (ed.), Boundaries of
morphology and syntax. Amsterdam: John Benjamins. 31-48.
Borer, H.
1998 Morphology and syntax. In A. Spencer and A.M. Zwicky (eds.), The hand-
book of morphology. Oxford: Blackwell. 151-90.
Bresnan, J. and S. Mchombo
1995 The lexical integrity principle. Natural Language & Linguistic Theory 13,181-
254.
Di Sciullo, A.-M. and E. Williams
1987 On the definition of word. Cambridge, Mass.: MIT Press.
Fudge, E.
1984 English word-stress. London: George Allen and Unwin.
Halle, M. and A. Marantz
1993 Distributed morphology and the pieces of inflection. In K. Hale and S.J. Key-
ser (eds.), The view from building 20. Cambridge, Mass.: MIT Press. 111-76.
Inkelas, S.
1989 Prosodie constituency in the lexicon. Ph.D. dissertation, Stanford University.
Kageyama, T.
1982 Word formation in Japanese. Lingua 57, 215-58.
1989 The place of morphology in the grammar. In G. Booij and J. van Marie (eds.)
Yearbook of Morphology 2. Dordrecht: Foris. 73-94.
1993 Bunpoo to gokeisei [Grammar and word formation], Tokyo: Hituzi Press.
1999 Word formation. In N. Tsujimura (ed.), The handbook of Japanese linguistics.
Oxford: Blackwell. 297-325.
Kiparsky, P.
1982 Lexical morphology and phonology. In The Linguistic Society of Korea (ed.),
Linguistics in the morning calm. Seoul: Hanshin Pub. Co. 3-92.
Kubozono, H.
1995 Gokeisei to on'in-koozoo [Word formation and phonological structure]. To-
kyo: Kurosio Publishers.
Ladd, D. R.
1984 English compound stress. In D. Gibbon and H. Richter (eds.), Intonation, ac-
cent and rhythm. Berlin: Mouton de Gruyter. 253-66.
Lieber, R.
1980 On the organization of the lexicon. Ph.D. dissertation, MIT.
1992 Deconstructing morphology. Chicago: University of Chicago Press.
Miller, D. G.
1993 Complex verb formation. Amsterdam: John Benjamins.
276 Taro Kageyama
Namiki, T.
1985 Gokeisei [Word formation], Tokyo: Taishukan.
Pesetsky, D.
1985 Morphology and logical form. Linguistic Inquiry 16,193-246.
Poser, W.
1990 Word-internal phrase boundary in Japanese. In S. Inkelas and D. Zee (eds.),
The phonology-syntax connection. Chicago: University of Chicago Press. 279-
88.
Postal, P.
1969 Anaphoric islands. CLS 5, 205-39.
Rappaport Hovav, M. and B. Levin
1992 -er Nomináis: implications for a theory of argument structure. In T. Stowell
and E. Wehrli, (eds.), Syntax and semantics 26: Syntax and the lexicon. New
York: Academic Press. 127-53.
Roeper, T. and M. Siegel
1978 A lexical transformation for verbal compounds. Linguistic Inquiry 9,199-260.
Sag, I. and J. Hankamer
1984 Towards a theory of anaphoric processing. Linguistics and Philosophy 7, 325-
45.
Selkirk, E.
1982 The syntax of words. Cambridge, Mass.: MIT Press.
Shibatani, M. and T. Kageyama
1988 Word formation in a modular theory of grammar. Language 64, 451-84.
Sproat, R.
1988 On anaphoric islandhood. In M. Hammond and M. Noonan (eds.), Theoreti-
cal morphology. San Diego: Academic Press. 291-301.
Taylor, J.
1996 Possessives in English. Oxford: Clarendon Press.
Ward, G., R. Sproat and G. McKoon
1991 A pragmatic analysis of so-called anaphoric islands. Language 67, 439-72.
Williams, E.
1997 Lexical and syntactic complex predicates. In A. Alsina et al. (eds.), Complex
predicates. Stanford: CSLI Publications. 13-28.
Further Evidence in Support of the Righthand Head
Rule in Japanese*
Takayasu Namiki
Ibaraki University
Introduction
The aim of this paper is to discuss further evidence in favor of the Right-
hand Head Rule (henceforth RHR) in Japanese and similar evidence in
English. After reviewing the arguments for the RHR in English by Wil-
liams (1981), I adduce three additional types of evidence for the RHR in
Japanese on the basis of Japanese compounds: (1) recursive compounds,
(2) reversible compounds, and (3) the abundance of complex elements in
the lefthand position in contrast to the scarcity of complex elements in the
righthand position in Japanese compounds. Furthermore, I show similar
evidence in English. I touch on questions such as whether there is a dif-
ference in expandability between the lefthand constituent and the right-
hand constituent of a compound in lefthand head compound languages as
there is in righthand head compound languages. Finally, I suggest replac-
ing the term "Righthand Head Rule" with the term "Final Head Rule", a
more general term than the former.
In his well-known paper, Williams defines head as in (1) and proposes the
Righthand Head Rule (2) on the basis of two arguments from English,
which are given in (3) and (4).1
(3) "It is generally the case that a suffix determines the category of a
word of which it is a part. Thus, for example, in English we have
the following:
X-ism Ν X-ize V X-ish A
V-ist Ν X-fy -»> V
V-ion ->· Ν
V-er Ν
A-ness Ν
(4) "A similar situation holds for most compounds - the righthand
member determines the category of the whole:
[[offjp [white]A ]A [[dry]A [dock]N ]N [[bar]N [tend] v ]V"
(Williams 1981: 249)
Although there have been many books and articles dealing with the R H R
or the notion "head of a word" (Lieber 1980, 1992, Kageyama 1982,
Namiki 1982, 1985, 1994, Selkirk 1982, Zwicky 1985, 1993, Trommelen
and Zonneveld 1986, Di Sciullo and Williams 1987, Scalise 1988, 1992,
and Nishihara et al. this volume), I will not summarize them here. I will
simply assume that we can conclude that the RHR generally holds in
English on the basis of Williams' arguments alone. I will also assume that
we can conclude that the RHR generally applies to Japanese on the basis
of arguments in Kageyama (1982) and Namiki (1982), which are similar
to Williams'. In the following sections, I present three more types of evi-
dence from Japanese compounds that the RHR holds in Japanese.
of the whole compound (cf. (1) above), and [2] has a "kind of" or "IS A"
relation between the whole compound and itself (cf. Namiki 1994). How-
ever, since every constituent of the recursive compounds in (5) is a noun,
it should be noted that the first criterion for defining the head of a com-
pound (i.e. the determination of the lexical category of the whole com-
pound) is vacuously satisfied, and that the second criterion is necessary in
deciding which noun is the head in the compounds in (5).
i. basu-TSUUGAKU tsuugaku-BASU
'bus' 'GOING TO SCHOOL' 'going to school' 'BUS'
= 'going to school by bus' = 'bus for going to school'
j. kyouiku-KANKYOU kankyou-KYOUIKU
'education' 'ENVIRONMENT' 'environment' 'EDUCATION'
= 'educational environments' = 'environmental education'
(lOf), and (10g); a verb in (10h). Other lexical categories have not been
found in my data. On the other hand, the lefthand constituents in the
compounds under consideration vary: in (10a), (10b), (lOf), and (10g),
two nouns are coordinated; in (10c), (10e), and (10h), a noun and a com-
pound noun are coordinated; in (lOd) a noun and a noun phrase are co-
ordinated. Notice that hitosuji, whose characteristic was emphasized in
and around (9a), is used again in (10c), where coordinated nouns are used
as its lefthand constituent.
In all the examples in (10), it seems impossible to assume that some
omission or deletion process is involved in forming the compounds at is-
sue (the bracketed parts), since such an assumption would lead to the
claim that, for example, *yuumoa afureru to sesou-bunseki afureru kora-
mu-kiji (cf. (10h)), and %shufu aite to hanayome-shugyou no ojousan aite
no ryouri-gakkou (cf. (lOd)) are synonymous well-formed expressions.
But the former expression, in fact, is ill-formed. The latter is non-synon-
ymous with (lOd), because the latter expression, though well-formed,
means that there are two cooking schools, one for housekeepers and the
other for young women, but (lOd) means that there is one school for both
housekeepers and young women.
= 'the coldness particular to women who are especially conscious of their own
beauty' (Namiki 1993:16)
d. [huruhonya no nyoubou] NP goroshi (Kageyama 1993: 327)
'secondhand book seller' GEN 'wife' 'killing'
= 'the killing of the wife of a secondhand book seller'
e. [[anokoro no chuuryuu -katei] NP [sodachi]] no hitotsu no pataan
'those days' GEN 'middle-class' 'family' 'being raised' GEN 'one' 'pattern'
= 'one type which was typical of the people who were raised in middle-class
families those days' (Teruhiko Kuze, 1996, Furemo Sede, Koudansha, p. 20)
f. [[rajio no buhin] NP zukuri] ya...
'radio' GEN 'parts' 'making' 'and'
= 'the making of radio parts and ...' (Souji Shimada, 1986, Kakei Toshi, Koudansha,
p. 257)
Kageyama (1993) states that phrases are generally excluded in words and
he furthermore maintains that
Judging from his examples in (12b) and (13), it seems that the Japanese
coordinating conjunction to ('and') cannot coordinate two words in a
righthand constituent of a compound. But he points out that other coor-
dinating conjunctions oyobi ('and' (formal)) and naishi ('or' (formal))
can do so, on the basis of the following examples (Kageyama 1993: 339):
Oyobi and naishi are formal counterparts of to ('and') and matawa ('or'),
respectively.
Two more examples show that coordinating conjunctions other than to
can coordinate two words in the righthand constituent of a compound.
Consider the following:
In this section I briefly deal with evidence for the RHR in English similar
to that for the RHR in Japanese given so far.
In every case the head of the compound occurs in the rightmost position.
These recursive compounds constitute one type of evidence for the RHR
in English, as in Japanese.
It is semantically clear that in each pair the righthand constituent is the head
of compounds. The difference in referents in each pair can also be seen, for
example, in the fact that sugar maple is a countable noun, as maple is, and
maple sugar is an uncountable noun, as sugar is. This is similar to the differ-
Further Evidence in Support of the Righthand Head Rule in Japanese 291
There are many examples where complex elements appear in the lefthand
constituent of English compounds. They are divided into a few subtypes,
as in section 4.
Lieber (1992: 11) calls examples like those in (21e, f) "phrasal com-
pounds."
There are relatively many instances where phrases occur in the lefthand
constituent of English compounds. Consider the following:
There are not so many examples where clauses occur in the lefthand con-
stituent. Some of them are given below:
I do not know of any explicit statement about the scarcity of complex el-
ements in the righthand constituents of compounds in English. But, to my
knowledge, it seems harder to find the compounds whose righthand con-
Further Evidence in Support of the Righthand Head Rule in Japanese 293
stituents are expanded. I have found only one such example. 5 Consider
the following:
7. Residual problems
There are many languages in the world which have been claimed to have
righthand head compounds. They include not only Japanese and English
but also Korean, Dutch, German, etc. Of course, there are languages
which have been claimed to have lefthand head compounds, such as mod-
ern Italian (Scalise 1988, 1992), Vietnamese (Lieber 1980), and Indone-
sian (Subandi 1998). But I suspect that there are differences between the
former type of language and the latter type of language, that is, differenc-
es in the kinds of evidence in support of the righthand head compounds
and evidence in support of the lefthand head compounds. Specifically, I
would like to raise questions such as whether there are as many recursive
compounds in the latter type of language as in the former type of lan-
guage,6 whether there are as many reversible compounds in the latter as
in the former, and whether there is a difference in expandability between
294 Takayasu Namiki
Notes
* I wish to express my gratitude to the editors of this book, Jeroen van de Weijer and Tet-
suo Nishihara, for offering me a chance to contribute this paper to the present volume. I
would like to thank Edward Quackenbush, Brent de Chene, Jeroen van de Weijer, Nori-
aki Yusa, and two anonymous reviewers for giving me valuable comments and sugges-
tions on my earlier drafts, and Edward Quackenbush and Brent de Chene for improving
my drafts stylistically. I am also grateful to my students at Ibaraki University and Ibaraki
Christian College for collecting part of my data on compounds in Japanese and English.
Any remaining errors and inadequacies are my own.
1. This definition in (1) means that if both X and the head of X are eligible members of cat-
egory C, then X's membership of C is equivalent to the head of X's membership of C. Wil-
liams himself notes that there are exceptions to the R H R : a prefix en- and exocentric
compounds such as push up and run down (Williams 1981: 249-250). Namiki states that
there are at most seven prefixes in English which are category-determining and so con-
stitute exceptions to the R H R : a- (as in afire and aloud), be- (as in behead and belittle),
de- (as in defrost and demilitarize), dis- (as in disbar and disable), en- (as in encage and
enrich), out- (as in outdistance and outsmart), and un- (as in unleash) (Namiki 1982: 27-
28, and Namiki 1985: 21-22).
2. Kubozono (1995: 104-105) states that right-branching compounds are prosodically
marked and left-branching compounds are prosodically unmarked in Japanese and Eng-
lish. H e also maintains that the number of left-branching compounds (as in a. and b. be-
low) is much larger than that of right-branching compounds (as in a', and b'. below) in
Japanese and English.
Further Evidence in Support of the Righthand Head Rule in Japanese 295
In all the examples above, the head of compounds appears as the rightmost constituent.
Kubozono (1999: 140), furthermore, claims that the phonological markedness of right-
branching structure is observed in languages such as Chinese and Italian. See also Akasa-
ka and Tateishi (this volume).
3. Italicized initial consonants in the righthand constituents of these reversible compounds
indicate that so-called "rendaku" (sequential voicing) has occurred in the initial conso-
nants at issue. For details of rendaku, see Otsu (1980) and Akasaka and Tateishi (this vol-
ume), among others. Saka- in (5d) is the variant form of sake which is specific to the left-
hand constituent of compounds in Japanese (cf. saka-gura and saka-mushi).
4. An anonymous reviewer has suggested to me a similar structure ([XP] Y), attributing the
difference in expandability to the difference in categorial status between the lefthand
constituent and the righthand constituent, and bringing Cinque (1993) to my attention.
5. Ed Quackenbush has given me some more examples like (27): automobile design and
manufacture, lawmakers and breakers, and missile launching and recovery. The deletion
of the repeated words has applied to these examples, too.
6. Scalise (1992) denies the existence of recursive compounds in modern Italian, stating "In
general Italian compounds are not recursive ... Compounds such as ... capostazione can-
not serve as constituents of other compounds." (Scalise 1992:196)
7. In older Japanese, letters of words and sentences were written from top to bottom, and
even in contemporary Japanese, letters are still printed from top to bottom in newspapers
and novels.
8. An anonymous reviewer has pointed out to me that the Righthand Head Rule is not a
rule but a parametric value (are compound heads on the right or the left?).
References
Akasaka, Yukiko and Koichi Tateishi
this volume "Heaviness in Interfaces," this volume.
Allen, Margaret
1978 Morphological Investigations, Doctoral Dissertation, University of Connecti-
cut.
Bauer, Laurie
1983 English Word-Formation, Cambridge University Press, Cambridge.
Cinque, Guglielmo
1993 "A Null Theory of Phrase and Compound Stress," Linguistic Inquiry, 24,239-
297.
Di Sciullo, Anna-Maria and Edwin Williams
1987 On the Definition of Word, The MIT Press, Cambridge, Massachusetts.
Fabb, Nigel
1984 Syntactic Affixation, Doctoral Dissertation, MIT.
296 Takayasu Namiki
Fujihara, Chiaki
1995 "Eigo ni okeru 'Hyphenated Phrasal Word' ni tsuite [On 'Hyphenated Phras-
al Word' in English]," Echuudo, 25,169-202.
Fukazawa, Haruka and Mafuyu Kitahara
this volume "Domain-relative Faithfulness and the OCP: Rendaku Revisited," this vol-
ume.
Kageyama, Taro
1982 "Word Formation in Japanese," Lingua, 57, 215-258.
1993 Bunpou to Gokeisei [Grammar and Word Formation], Hituji Shobou, Tokyo.
this volume "Word Plus: The Intersection of Words and Phrases," this volume.
Kubozono, Haruo
1995 Gokeisei to On'inkouzou [Word Formation and Phonological Structures],
Kurosio Shuppan, Tokyo.
1999 Nihongo no Onsei [Speech Sounds of Japanese], Iwanami Shoten, Tokyo.
Lieber, Rochelle
1980 On the Organization of the Lexicon, Doctoral Dissertation, MIT.
1992 Deconstructing Morphology, The University of Chicago Press, Chicago.
Morita, Junya
1985 "X-bar Morphology and Synthetic Compounds in English," English Linguis-
tics, 2,42-59.
Namiki, Takayasu
1982 "The Notion 'Head of a Word' and Core and Periphery Word Formation: In-
teractions between Affixation and Subcategorization," Studies in English Lin-
guistics, 10, 21-41.
1985 Gokeisei [Word Formation (in English)], Taishukan Publishing Company, To-
kyo.
1991 "Fukugou-Keiyoushi to Kakudai-Junjozuke no Kasetsu [Compound Adjec-
tives and the Extended Ordering Hypothesis]," Gendai Eigogaku no Shosou
[Aspects of Contemporary English Linguistics], ed. by Shuji Chiba et al.,
Kaitakusha Publishing Co. Ltd., Tokyo.
1992 "Eigo no Zougoryoku [The Productivity of English Word Formation ]," Ni-
hongogaku [Japanese Linguistics], 11, 5, 66-74.
1993 Fukugou-Keiyoushi no Keitai-Tougoteki Kenkyuu - Fukugougo no Fuhenteki
Tokuchou o Motomete - [Morpho-Syntactic Studies of Compound Adjectives
in English: Toward Universalities of Compounding], a Report of a Grant-in-
aid for Scientific Research (C) (Grant No. 03610235), Ibaraki University.
1994 "Heads and Subheads of Compounds," Synchronic and Diachronic Ap-
proaches to Language: A Festschrift for Toshio Nakao on the Occasion of His
Sixtieth Birthday, ed. by Shuji Chiba et al., Liber Press, Tokyo.
Nishihara, Tetsuo, Jeroen van de Weijer, and Kensuke Nanjo
this volume "Headedness in Morphology: Tendencies in the Truncation of Loanword
Compounds in Japanese," this volume.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems," MIT Work-
ing Papers in Linguistics, 2, 207-227.
Scalise, Sergio
1988 "The Notion of 'Head' in Morphology," Yearbook of Morphology, ed. by
Geert Booij and Jaap van Marie, 229-245, Foris, Dordrecht.
1992 "Compounding in Italian," Rivista di Linguistica, 4,1,175-199.
Further Evidence in Support of the Righthand Head Rule in Japanese 297
Selkirk, Elisabeth
1982 The Syntax of Words, MIT Press, Cambridge, Massachusetts.
Shimamura, Reiko
1986 "Lexicalizatio of Syntactic Phrases," English Linguistics, 3,20-37.
Spencer, Andrew
1991 Morphological Theory, Blackwell, Oxford.
Subandi
1998 "Nichi-I-go no Fukugou-meishi no Kouzou ni tsuite [On the Structures of Com-
pound Nouns in Japanese and Indonesian]," Kokusai Shinpojium Hikaku Goi
Kenkyuu [Proceedings of an International Symposium on Comparative Studies
of Vocabulary], ed. by Ikudou Tajima, 66-74, Nagoya University.
Trommelen, Mieke and Wim Zonneveld
1986 "Dutch Morphology: Evidence for the Righthand Head Rule," Linguistic In-
quiry, 17,147-169.
Williams, Edwin
1981 "On the Notions 'Lexically Related' and 'Head of a Word'," Linguistic Inqui-
ry, 12, 245-274.
Zwicky, Arnold
1985 "Heads," Journal of Linguistics, 21,1 -29.
1993 "Heads, Bases and Functors," Heads in Grammatical Theory, ed. by Greville
G. Corbett et al., 292-315, Cambridge University Press, Cambridge.
Against Headedness in Compound Truncation:
English Compounds in Japanese
Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Miyagi University of Education, Leiden University, and
St. Andrew's University
1. Introduction
tained, unlike what might be expected on the basis of the RHR. The ques-
tion arises how truncation is governed, if not (or only partly) by the RHR.
Another pattern of truncation of loanword compounds in Japanese,
which raises the same question, is also observed: in this pattern, both
halves of loanword compounds in Japanese are truncated, as in waado
purosessaa waa puro ('word processor'). In fact, this pattern of trunca-
tion is the most common in Japanese (see Itô 1990, Kubozono 1993), al-
though it is not common in other languages, such as English or Dutch. 1
Facts like these again raise the question of what constraints there are on
truncation, and what the role is that headedness plays.
In our analysis of the different truncation patterns, we suggest that, in
the framework of Optimality Theory, the pattern of truncation of loan-
word compounds in Japanese is governed by a language-particular hier-
archy of universal constraints on truncation. Constraints that play a cru-
cial role rally for the retention of the leftmost part of a compound,
retention of material from all parts of the input, and the formation of a
proper prosodie word.
2. Preliminaries
The example above represents a very productive process, and other ex-
amples of this type are presented in (3).
It should be noted that these patterns are observed in Japanese, but not
in Chinese. In other words, these Sino-Japanese compounds were truncat-
302 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
ed after they had been borrowed into Japanese. Umegaki (1963) suggests
that the primary unit of word formation in Japanese is based on a two-syl-
lable prosodie unit; secondary units are four-syllable units (2 + 2) and
three-syllable units (2 + 1 or 1 + 2). The term 'syllable' that Umegaki
(1963) uses corresponds to the concept of 'mora', which is nowadays
more commonly used in Japanese phonology. From a phonological view-
point, it is obvious that the output of truncation of both loanword com-
pounds and Sino-Japanese compounds in Japanese is basically composed
of four moras (2 + 2): all examples so far conform to this pattern (where
of course the mora nasal counts as a single mora). Moreover, Itô (1990)
and Kubozono (1995) suggest that the basic unit of prosodie structure in
Japanese is a two-mora structure, and truncations of loanword com-
pounds are based on a four-mora pattern (2 + 2), which we might regard
as a foot. We will return to this insight below, and try to give a more pre-
cise account of these patterns then. First, let us turn to a brief discussion
of headedness.
3. Head of a word
Based on the cases which we have seen so far, languages generally seem
to choose either the left-hand member or the right-hand member as the
304 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
For this reason, Vogel (1990) argues that neither the morphology nor the
phonology of the English compounds plays a decisive role. She therefore
concludes that the second hypothesis must also be rejected.
Finally, we examine the third hypothesis, H3. According to Vogel, Ital-
ian mainly has left-headed compounds: this is the productive pattern (cf.
(8) above). Vogel concludes that the left element of English compounds
in Italian is retained because of the structure of Italian and thus that H3
is adequate. This predicts that it makes no difference from which lan-
guage compounds are borrowed: regardless of the source language, the
left-hand element will be retained. In fact, Vogel shows that compounds
of German origin have the same pattern of truncation as the English com-
pounds, as illustrated in (14):
If Vogel's Hypothesis 3 in (12) above is correct, and also holds for lan-
guages other than Italian, the right-hand element of English compounds
in Japanese is expected to be retained, leading to front truncation, be-
cause Japanese has right-headed compounds (see (6b) above; Kageyama
308 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Some more examples are given in (16) (cf. also (1) above).
Kubozono (1993) suggests that back truncation, which is the second most
common pattern, can be explained as a result of HI suggested by Vogel
(1990) (see (12) above). Kubozono (1993) gives the following examples:
Against Headedness in Compound Truncation: English Compounds in Japanese 309
However, this approach does not deal with double truncation, and hence
does not achieve an integrated analysis of truncation. We will attempt this
in the next section.
6. Analysis
As McCarthy and Prince (1994, 1995a: 322) claim, the concept of mini-
mal word no longer has an actual status as a primitive template in any
language in the OT era; instead its effect is derived from the interaction
of such metrical constraints as F T B I N , PARSE-SYLL, A L I G N - F T - L (see Be-
nua 1995: 118-19). In our paper, however, we will for the sake of conve-
nience continue to refer to the minimal word, which is equivalent to a bi-
moraic foot in such quantity-sensitive languages as Japanese. The two
constraints together demand that outputs have a structure conforming to
that in (18).
In what we think is a relatively recent development, truncations of
three moras are becoming increasingly popular, e.g. Misu-Do from Mis-
utaa Doonatsu 'Mister Doughnut [a doughnut franchise]', tere-ka from
terefoN kaado 'phone card'. Thus, the long vowels in the second member
of the original compound are shortened, which shows that forms which
have a monomoraic syllable structure with a short vowel are better
formed than bimoraic structures with a long vowel in the second PrWd.
Below, we will propose a constraint ranking accounting for this.13 Howev-
er, three-mora outputs are also found even when there is no long vowel
in the input, such as in rabu-ho from rabu hoteru 'lit. love hotel', Roi-Ho
from Roiyaru Hosuto 'Royal Host [the name of a restaurant]', and Bura-
Bi from Burakku Bisuketto 'Black Biscuit [the name of a singer]', where
there is no phonological reason to avoid *rabu-hote, *Roi-Hosu, and *Bu-
ra-Bisu. Another template, consisting of three moras, is probably more
appropriate for these forms, which we have to ignore here (cf. Itô (1990:
237, note 29) for a similar view).
An additional constraint that plays a role demands that the leftmost
two moras are preserved in the output, and not the rightmost two moras,
or just any two moras, in double truncation. Furthermore, even when
doubly truncated forms are not preferred, the retained element usually
turns out to be the leftmost (minor) PrWd of the major (i.e. compound-
level) PrWd (e.g. suupaa from suupaa maaketto). This "leftmostness" that
double truncation and back truncation share should be generalized by a
uniform constraint. To capture this uniformity, we will propose (20),
312 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
μ μμ μ μμμμ μ
pawaa suteariNgu
The optimal output reflects the first two moras of the first input word and
the first two moras of the second input word. On the other hand, a form
in which the last two moras of the first word are taken, and—for in-
stance—the first two moras of the second word (resulting in *waa-sute)
would also result in a well-formed moraic structure (i.e. pass both M I N W D
and P A R S E - P R W D ) but would not pass higher-ranked LEFTMOST. The form
*pawa-su is ill-formed, because the second output word su fails to satisfy
the M I N W D requirement. Finally, the form *pawa satisfy the constraints
Against Headedness in Compound Truncation: English Compounds in Japanese 313
pawa-sute
waa-sute *!
pawa-su »!
pawa * !•
μμ μ
bukku
the last part, and in the final two examples parts of both words in the
original compound are used. We assume these different choices are lex-
ically determined, though all are driven by the fact that no regular, pho-
nological truncation is possible. Thus, importantly, the fact that the trun-
cation of tekisuto bukku is tekisuto is, in our opinion, not a result of the
fact that this compound is somehow marked as exceptional, but rather
of the fact that the normal pattern of truncation fails for phonological
reasons.
As was already mentioned above, long vowels are normally shortened
at the end of the second prosodie word in truncation outputs. We can put
this down to a phonological constraint against final long vowels, which
operates in structures like (18) above. The constraint could be defined as
follows:
Examples include furi-ma from furii maaketto 'flea market', where Sfuri-
mela would be incorrect, depa-ga from depaato gaaru 'lit. department
store) girl' (*depa-gaa), hai-ka from haiwei kaado 'lit. highway card'
(*hai-kaa), as well as Misu-Do from Misutaa Doonatsu (*Misu-Doo) and
tere-ka from terefoN kaado (*tere-kaa), both of which were mentioned
above. Unlike the CODA-COND, this constraint NFLV holds of loanword
truncated structures only, not of underived loanword compounds (e.g.
rabu retaa 'love letter', dansu paatii 'dance party'), Sino-Japanese com-
pounds (both underived and truncated) (e.g. dantai koosyoo 'collective
bargaining' and its truncated form dan-koo) nor hypocoristics (e.g. Maa-
chan, Taa-kun, Koo-chan). To account for the restricted nature of NFLV,
we first assume that this constraint belongs to the 'foreign' stratum of Jap-
anese lexicon (Itô and Mester 1995). This assumption appropriately ex-
cludes Sino-Japanese vocabulary and hypocoristics from the NFLV effect.
Next, we will ascribe the fact that NFLV applies only to truncated forms
to the classical concept of "derivedness". The application of this lexical-
phonological idea to the loanword phonology was, to our knowledge, first
proposed by Itô (1990: 218-19). According to her proposal, such a con-
straint is enforced on derived forms only, not on underived forms and,
since abbreviated loanwords are derived from their base forms, they sat-
isfy this criterion.
Of course, the concept of "derivedness" cannot be directly incorporat-
ed into our non-derivational analysis. One of the solutions, which we will
follow here, is proposed by Benua (1995) in the framework of Correspon-
316 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Following her proposal, we will assume the following ranking among IO-
Faith (LDENT-IO[V-length]), BT-Identity (LDENT-BT[V-length]) and the
phonological constraint NFLV.
TO illustrate how these constraints and their ranking explain our data, we
will take waa-puro (from waado prosessaa), suupaa (from suupaa maa-
ketto), and pote-chi (from poteto chippusu) as examples.
•®· waa-puro
waa-pu *!
waado *!
suu-maa *!
•S· suu-ma *
suupaa *
pote-chip *!
ra pote-chi *
ra· poteto 1 *
date suupaa does not violate NFLV, because this constraint holds of only
Major PrWd; suupaa itself is a minor PrWd). Note that suupaa does not vi-
olate LEFTMOST, because it is the leftmost element (a minor PrWd) of the
major (compound-level) PrWd. Since the candidates suupaa and *suu-ma
violate the lower-ranked constraints P A R S E - P R W D and M I N W D , respective-
ly, they are on a par with each other with respect to their potential of being
realized as the actual output. The rest is outside of the OT grammar. 16
As expected, cases like poteto chippusu can be handled in the same way
as suupaa maaketto. The four-mora doubly truncated candidate *pote-chip
triggers the violation of the highly-ranked constraint CODA-COND, but the
other candidates *poteto and pote-chi pass this constraint, though they vi-
olate the lower-ranked constraints P A R S E - P R W D and M I N W D , respectively.
Semantic reasons undoubtedly also play a role in truncation. It is often
the case that the leftmost element in compounds indicates a specific prop-
erty, while the rightmost element indicates a classlike, general character-
istic. We believe this constraint is related to the tendency that Vogel
(1990) observed when she formalized Hypothesis 1 above (12). This is ob-
vious in the following examples from English, in which the rightmost ele-
ment is often lost nowadays or is optional:
7. Conclusion
Acknowledgements
We thank Laura Benua, Bert Botma, Ellen Broselow, Jan Kooij, Nancy Kula, Sang Jik Rhee,
Grazyna Rowicka, Norval Smith and Erik Jan van der Torre for valuable discussion of ear-
lier versions. Remaining errors are our own.
Notes
1. Szymanek (1989) suggests that back truncation is the most common in English. This is
also the case in French, German and Dutch. See Wiese (1996:62-63) for German, Ni-iku-
ra et al. (1996: 384) for French, and Hamans (1997:1734) for Dutch.
2. In some languages, double truncations are common. For example, see Shin (1997) for Ko-
rean, and Crowley (1997) for English in Papua New Guinea. In standard English, exam-
ples are FedEx (< Federal Express), sci-fi (< science fiction), Amtrak (< American Track)
and AMEX (< American Express).
3. Hock (1991: 196) suggests that shortening is a common tendency in linguistic change.
Lass (1987:211) also points out examples as early as from the 18th century. Scheler (1977)
suggests that the oldest examples are from the middle of 16th-century, although these are
but few.
320 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
4. Other problematic examples for the R H R concern Russian diminutives (Lieber 1992),
and French diminutives (Jaeggli 1980).
5. Van Beurden (1988) claims that in language acquisition children assume that the right-
hand element decides the whole category and thus consider the R H R as the unmarked
option, while later they have to decide the parameter about the position of head (left or
right) according to the data of their native language.
6. Huang (1998) argues that Chinese is a headless language in the morphology, for neither
the right-hand nor the left-hand element in a compound determines the lexical category
of the whole.
7. There are exceptions, of course, such as lexicalised compounds like pickpocket, which is
not a kind of pocket.
8. Kubozono (1990,1996) points out that the notion of phonological head also plays a role
in patterns of blending and nominal compounds.
9. In Italian native compound forms, back truncations such as (11) are more frequent than
front truncations. Double truncation is unknown in Italian (Piermarco Bertinetto, p.c.)
10. Haspelmath (1993) also argues for a constraint that tolerates exceptions in morphology,
antedating mainstream OT.
11. Still another reason for excluding front truncation is that, as Yonekawa (1996) suggests
in his book on the vocabulary of the young, front truncation is often adopted to create
secret languages. This suggests that they know this is the least common and thus least
understandable way of word-formation.
12. The constraint P A R S E - P R W D also accounts for such English acronyms as AIDS, PET,
WASP, etc. (see also Itô 1990) and shows that Japanese double truncation and English ac-
ronym formation are very similar phenomena. Since they are both common patterns in
both languages, this analysis shows that seemingly different languages adopt similar pat-
terns, regardless of the location of the syntactic head (right in Japanese, left in English).
13. Exceptionally, we suggest, the long vowels in suke-boo f r o m sukeeto boodo 'skate-
board', RoN-Buu from RoNdoN Buutsu 'London Boots [a pair of comedians], bata-
pii from bataa piinattsu 'buttered peanut', suke-paa from sukai paafekuto (TV) 'Sky
Perfect T V ' are retained, and sometimes there is variation in this respect. Interesting-
ly, all these exceptions happen to have bilabial stops at the beginning of the second
PrWd.
14. There are a few examples that we know of that involve skipping: one of them is paso-
koN, which somehow has replaced regular paa-koN, which was indeed attested at an
earlier stage (Kubozono and Ota 1998:190-91), and others include Ame-futo (which has
replaced the regularly back-truncated AmerikaN) from AmerikaN futtobooru 'Ameri-
can football', supa-koN from suupaa koNpyuutaa 'super computer', which might have
been formed on the analogy of paso-koN, and bifu-teki from biifu suteeki 'beefsteak',
which also violates LEFTMOST, as we noted above.
15. Itô (1990:236, note 19) gives pii-goro from pitchaa goro 'pitcher grounder' as a counter-
example, which does not sound like a familiar word to two of the three present authors.
16. The candidate suupaa does not violate MINWD, because this constraint is satisfied as
long as the output is minimally two moras long; it does not prevent the retention of an
original longer than two moras. See Itô (1990: 234).
17. Aigeo (1991: 9) treats double clipping (truncation) in English as Innovative clipping (e.g.
junior college * juco, aviation gasoline > avgas, capsule communicator • capcom).
Against Headedness in Compound Truncation: English Compounds in Japanese 321
References
Aigeo, J.T.
1991 Fifty years among the new words. Cambridge: Cambridge University Press.
Becker, T.
1990 D o words have heads? Acta Linguistica Hungarica 40, 5-17.
Benua, L.
1995 Identity effects in morphological truncation. In J.N. Beckman, L.W. Dickey
and S. Urbanczyk (eds.), University of Massachusetts Occasional Papers 18,
77-136.
Booij, G.E.
1995 The phonology of Dutch. Oxford: Oxford University Press.
Crowley, T.
1997 An introduction to historical linguistics. Auckland: Oxford University Press.
Di Sciullo, A. -M. and E. S. Williams
1987 On the definition of word. Cambridge, Mass.: MIT Press.
Greenberg, J.H.
1963 Some universals of grammar with particular reference to the order of mean-
ingful elements. In J. H. Greenberg (ed.), Universals of language. Cambridge,
Mass.: MIT Press. 58-90.
Hamans, C
1997 Clipping in modern French, English, German and Dutch. In R. Hickey and S.
Puppel (eds.), Language history and linguistic modelling: A festschrift for
Jacek Fisiak on his 60th birthday. Berlin: Mouton de Gruyter. 1733-41.
Haspelmath, M.
1993 The diachronic externalization of inflection. Linguistics 31, 279-309.
Hawkins, J. A. and A. Cutler
1988 Psycholinguistic factors in morphological asymmetry. In J.A. Hawkins (ed.),
Explaining language universals. Oxford: Blackwell. 280-317.
Hock, H.H.
1991 Principles of historical linguistics. Berlin: Mouton de Gruyter.
Huang, S.
1997 Chinese as a headless language in compounding morphology. In J. Packard
(ed.), New approaches to Chinese word formation. Berlin: Mouton de Gruyter.
261-83.
Ishiwata, T.
1993 Waseieigo to kokusaitsuuyoogo [Japanese-English and international common
languages]. In Gairaigo [Foreign words], Tokyo: Kawade-syoboo Shin-sha.
89-104.
Itô, J.
1986 Syllable theory in prosodie phonology. PhD dissertation, University of Mas-
sachusetts, Amherst [Published by Garland, New York, 1988].
1989 A prosodie theory of epenthesis. Natural Language and Linguistic Theory 7,217-
60.
1990 Prosodie minimality in Japanese. In M. Ziolkowski et al. (eds.), CLS 26-2. Chi-
cago: Chicago Linguistic Society. 213-39.
Itô, J. and R.A. Mester
1994 Reflections on C O D A - C O N D and alignment. Ms., University of California, San-
ta Cruz.
322 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Motwani, P.
1993 A dictionary of abbreviations and truncations. Tokyo: Maruzen.
Namiki, T.
1994 Heads and subheads of compounds. In S. Chiba et al. (eds.), Synchronic and
diachronic approaches to language: A festschrift for Toshio Nakao on the oc-
casion of his sixtieth birthday. Tokyo: Liber Press. 269-85.
Niikura, S. et al.
1996 Furansugo handobukku [A handbook of French], Tokyo: Hakusuisha.
Prince, A.S. and P. Smolensky
1993 Optimality theory: Constraint interaction in generative grammar. Ms., Rut-
gers University and University of Colorado.
Ralli, A.
1992 Compounding in modern Greek. Rivista di Linguistica 4,143-74.
Scalise, S.
1988 The notion of 'head' in morphology. In G.E. Booij and J. van Marie (eds.),
Yearbook of morphology. Dordrecht: Foris. 229-45.
1992 Compounding in Italian. Rivista di Linguistica 4,175-99.
Scheler, M.
1977 Der englische Wortschatz. Berlin: Erich Schmidt Verlag.
Schultink, H.
1988 Morphological heads: Evidence from Swahili. In G.E. Booij and J. van Marie
(eds.), Yearbook of morphology. Dordrecht: Foris. 247-58.
Selkirk, E.O.
1982 The syntax of words. Cambridge, Mass.: MIT Press.
1996 The prosodie structure of function words. In J.L. Morgan and K. Demuth
(eds.) Signal to Syntax: Bootstrapping from speech to grammar in early aquisi-
tion. Mahwak, NJ: Lawrence Erlbaum Associates. 187-213.
Shibatani, M.
1990 The languages of Japan. Cambridge: Cambridge University Press.
Shin, S.-H.
1997 Correspondence in Kyungsang Korean truncation. In S. Davis (ed.), Optimal
viewpoints. Bloomington, Indiana: Indiana University Press. 115-38.
Slobin, D.
1973 Cognitive prerequisites for the acquisition of grammar. In C. Ferguson and D.
Slobin (eds.), Studies of child language development. New York: Holt, Rine-
hart and Winston. 175-208.
Szymanek, B.
1989 Introduction to morphological analysis. Warszawa: Wydawnictwo Naukowe.
Trommelen, M. and W. Zonneveld
1984 The right side won, the right side won. In H.J. Bennis and W.U.S. van Lessen
Kloeke (eds.), Linguistics in the Netherlands 1984. Dordrecht: Foris. 179-87.
1986 Dutch morphology: Evidence for the right-hand head rule. Linguistic Inquiry
17,147-69.
Umegaki, M.
1963 Nihongairaigo no kenkyuu [Studies of foreign words in Japanese], Tokyo: Ken-
kyuusha.
van Beurden, L.
1988 Bantu heads on the lefthand side. In G. de Haan and W. Zonneveld (eds.),
Formal parameters of generative grammar IV (Yearbook 1988). Dordrecht:
ICG Printing. 1-13.
324 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Vogel, I.B.
1990 English compounds in Italian: The question of the head. In W.U. Dressler et
al. (eds.), Contemporary morphology. Berlin: Mouton de Gruyter. 99-110.
Wiese, R.
1996 The phonology of German. Oxford: Oxford University Press.
Williams, E.S.
1981 On the notions 'lexically related' and 'head of a word'. Linguistic Inquiry 12,
245-74.
Yonekawa, A.
1989 Shingo to ryukoogo \New words and vogue words], Tokyo: Nan-undoo.
1996 Gendai wakamono kotoba koo [A study of the vocabulary of the young]. To-
kyo: Maruzen.
Zwicky, A.M.
1985 Heads. Journal of Linguistics 21,1-29.
1992 Some choices in the theory of morphology. In R. Levine (ed.), Formal gram-
mar: Theory and implementation. Oxford: Oxford University Press. 327-71.
III. Studies in Contrastive Japanese-
English Phonetics and Phonology
Two Different Kinds of Rhythm:
Japanese and English*
Yosihiro Masuya
Konan University
Introduction
on one and the same criterion without measuring the duration of the in-
ter-stress period as well as answering after my method the question Dauer
has raised.
Before I go on to the analysis of my data, I shall briefly review some of
the speech rhythm studies published in the 1980's and 1990's. Incidentally,
since I should also like those readers who are uninitiated into statistics to
follow my arguments, I shall insert brief explanations of statistics either in
footnotes or in the text hereafter.
she said something to the effect that if the relationship between the inter-
stress interval duration (in ms) (y below) and the number of syllables in
that interval (jt below) were "perfectly linear", the "regression 5 curve"
[sic] should have zero intercept and the correlation between them should
approach or equal 1.
When she said "perfectly linear", she seems to have had in mind such a
_y-to-x relation as y = α,*. 6 But simple linear regression is usually y = a0 +
axx. When we make a test to know which is the right one, we formulate
what is called "null hypothesis" 7 —H 0 : a 0 = 0 in this context. If the data
force us to accept the null hypothesis, y - a x x is our regression. A problem
is encountered at this point, however. Whether the data force us to accept
the null hypothesis or allow us to reject it, there is some probability for a
wrong judgement. Whenever a null hypothesis is formulated, an alterna-
tive hypothesis 8 is also formulated—//! : α 0 * 0 in this context. Let me call
"p," the probability of the wrong judgement on which the null hypothesis
that must not be rejected is rejected and "p 2 " the probability of the wrong
judgement according to which the null hypothesis which must not be ac-
cepted is accepted. 1) If the data allow us to reject the null hypothesis, the
alternative hypothesis is accepted, which means that the alternative hy-
pothesis is judged right. As p x exists, we accept the alternative hypothesis
at the risk of a given error level. This error level is what is called "signifi-
cance level".9
2) If the data do not allow us to reject the null hypothesis, the null hy-
pothesis is accepted, only because the grounds for rejecting the specifica-
tion the null hypothesis holds, are not found in the data. In this case, p2
exists, but it is not easy to calculate. Although the test of the hypothesis is
done on the principle that the rejection region 10 should be determined so
as to make p2 the least after px is fixed, there is no statistical procedure for
calculating p2, if the specification the alternative hypothesis holds is
something like a * 0.11
Fletcher's theory holds, only if H 0 : a 0 = 012 cannot be rejected and we
have no other choice than to accept it. But, as I mentioned just now, the
conditions of the significance test are completely different from those es-
tablished for the regression equation with the intercept: in the case of the
regression with the intercept, it is known how great (or how little) the risk
of the judgement error in adopting a given intercept is, whereas in the
case of the regression without the intercept, the greatness of the judge-
ment error in adopting zero intercept is not known—it may be consider-
able. The case with the intercept and that without the intercept are mat-
ters on different levels. Therefore, Fletcher's comment on French rhythm
Two Different Kinds of Rhythm: Japanese and English 331
and said, "This set of figures does not appear to support claim (i)." (p. 74).
Unfortunately, very important values were ignored in his research:
showing a standard deviation by itself without any reference to the mean
value is meaningless. According to the Camp-Meidell theorem, over
95.1 % of the data lie within the range of χ ± 3 χ SD (χ: mean value of the
data; SD: standard deviation). Supposing mE= 4, sE = 0.8 for one language
and mF= 22, sF = 0.8 for another language (m for mean, s for standard de-
viation, E for one language and F for another), over 95.1 % of the data of
Language E lie within 4 ± 2.4, i.e. in the range from 6.4 to 1.6 and over
95.1 % of the data of Language F within 22 ± 2.4, i.e. in the range from 22.4
to 19.6. The ratio of the maximum value to the minimum value is 4.0 for
Language E and 1.1 for Language F: the data values are more or less sim-
ilar in Language F, whereas they are not in Language E, despite that sE =
s F . In order to examine the difference of variability between languages in
terms of the standard deviation, the comparison of the coefficients of
variation is required. The coefficient of variation is
CV = - or CV = - 100%
χ χ
(CV: coefficient of variation, s: standard deviation, x: mean value of JC).
With this hypothetical example, the coefficient of variation of Language
E is 20.0 % and that of Language F 3.6 %. I regret that Roach's efforts thus
332 Yosihiro Masuya
came to nothing and that his conclusion was nevertheless accepted and
quoted in an academic paper as if it had been meaningful.
Williams and Hiller (1994) constructed six types of unit and examined
whether English isochronicity was caused by chance and whether it was
linguistically-based. Their types of unit were:
1) "Full" unit—the one which starts "at every full (i.e. unreduced) vowel";
2) "Lexical(ly stressed)" unit—the one which begins "at every syllable
that bears the primary lexical stress of an orthographic word";
3) "Accented" unit—what begins "at every syllable that bears a pitch ac-
cent (i.e. where there is an easily-perceived peak in the pitch con-
tour)";
4) "Tense" unit—what begins with every syllable containing a tense vow-
el (i.e. diphthong or long monophthong);
5) "Random" unit—the one whose number of syllables is determined by
a random integer between 1 and 5, generated by a random number
generator;
6) "Arbitrary" unit—the one whose "number of syllables per unit [is] de-
termined by a fixed and arbitrary algorithm (e.g. every third syllable
starts a new foot)".
bles, e.g. a unit of 8 syllables; such a group should be excluded from the
calculation. Therefore, we have a number of groups of data, defined by
the number of syllables in a unit, such as the 1-syllable group, 2-syllable
group, 3-syllable group and now we have the variation of the data within
each group and the variation of the data between the groups. In such a
case as this, not in a case where the values of the two variables (x and y )
have a one/one correspondence, what is called "one-way analysis of vari-
ance" needs to be run to know how large the deduction of the variation
due to the regression14 from the variation between groups is, the latter
variation being the variation of the mean of each group. This deduction
shows the greatness of the "lack of fit" 15 of the regression line to the data.
If a lack of fit is judged to exist, the simple linear regression y = a0 + axx
is inapplicable.
After my attention was drawn to this problem in averaged syllable re-
gressions by a remark of Prof. Öta, I tried calculating three averaged syl-
lable regression lines from two of my dataseis, one from this paper and
the other from outside of it. In these calculations, following the authors of
the paper I am commenting on, I have excluded the units preceding an ut-
terance-final pause and in calculating the "lexical" regression, those units
preceded by a pause are also excluded because the authors' treatment of
them is unknown. Two of the regression lines are of genuine feet and one
is of a "lexical" unit. (The authors used the term " f o o t " in place of "unit".
But I should restrict the use of the term to what Abercrombie (1964b,
217) defines; a broadened use of a term in general makes things confused.
"Genuine feet" are " f e e t " in Abercrombie's sense.) The result is that the
existence of lack of fit is highly significant—sometimes extremely highly
significant16—with all of the three simple linear regression lines; these re-
gression lines are perfectly inadequate. It is most probable that the aver-
aged syllable regression lines Williams and Hiller calculated are also far
from adequate.
Here and there in the paper, they made a comparison of the results of
the regression analysis between the "linguistically-based" units and the
"random" unit even though the "random" regression line was not signif-
icant, and concluded that, if the results were favourable to the "linguisti-
cally-based" units, the tendency towards isochronicity in English was
larger than chance.
In arriving at such a conclusion, they were doing something meaning-
less. A regression line not being significant means that the inferred popu-
lation coefficient of the x-term17 is zero and that the x-term does not exist.
When they drew a comparison between the coefficients of χ of a signifi-
334 Yosihiro Masuya
cant regression line and a non-significant regression line, they were com-
paring a value which exists with a value which does not exist.
Earlier in the same paper, they showed the results of their calculation
of the simple linear regression of the unit duration against the number of
syllables per unit for each of their unit types and said, "The random and
arbitrary feet show generally higher r-squared values, indicating that this
factor (number of syllables) [sic] accounts for a higher proportion of the
variance than in the other foot types in the 'linguistic' foot types there
may be some other factor also at work in the determination of foot dura-
tion" (p. 431). Apart from a little obscureness of some phrases in this quo-
tation, they may have been right. So long as one judges from the calcula-
tions they made, the portion of the variation unexplained by their
regression seems to be larger in the "linguistic" units than in the "ran-
dom" and "arbitrary" units. But lack of fit had to be tested before any-
thing was said as a conclusion, and recalculations of the r-squared values
are advisable, for Prof. Öta (personal communication) has become aware
that the regular calculation of R 2 (i.e. dividing the variation due to the re-
gression by the total variation) is not happy for the regression for repeat-
ed measures, when a lack of fit does not exist. According to him, it is rea-
sonable that we should adopt for the coefficient of determination (i.e. R 2)
something which indicates how well the calculated regression equation
grasps the variation of the mean of every group; therefore, in the calcula-
tion of R 2, the variation due to the regression must be divided by the vari-
ation between groups, not by the total variation.
Van Zanten and van Heuven (1997) said in their paper, "Assuming that
the penultimate vowel of our Indonesian target words is stressed, we may
expect an anticipatory shortening effect for the stressed vowel in bisyllab-
ic as compared to monosyllabic words" and went on to say, "For longer
Indonesian words, the stress would still be on the penultimate vowel ..."
[italics mine]. It is surprising that whether a syllable is being stressed or
not is part of an assumption. To repeat Abercrombie's (1960) words,
stress is "something which is either present or absent" [italics in the origi-
nal]. If they listen to their informants' utterances, they can more or less
affirm that stress is there or that it is not there. I say "more or less" be-
cause I do not think the question Mrs Uldall (talk before the class) came
across has not been solved yet; she let several phoneticians, who were na-
tive speakers of English, listen to Abercrombie's reading of "The North
Wind and the Sun" and plot stresses on the text in order to prepare for
her (1971) paper and saw that they sometimes disagreed about the plot-
ting on some stressable unstressed syllables. Hence a few mis-plottings of
Two Different Kinds of Rhythm: Japanese and English 335
"It is ... necessary to learn to listen differently [from the way one lis-
tens to everyday speech] in order to be able to analyse speech
rhythm, whether of one's mother tongue or another language, and to
describe it in general phonetic terms. Few succeed in doing this with-
out training" [italics in the original]
and has become familiar with the sounds of the target language. (The du-
ration of those wrongly constructed feet is expected to be buried in the
variation or discarded as extraordinary values or outliers.)
Now I shall return to the "linguistically-based" unit types which Will-
iams and Hiller casually constructed. If the speech rhythm is formed by
stress-beat in English, any other units than the stress-headed like "foot"
in Abercrombie's (1964b, 217) sense are irrelevant. Williams and Hiller
said, "A lexically stressed syllable is merely a candidate for sentence-level
stress, and may not receive stress in any given utterance" [italics mine].
Then how can a "lexical" unit when there is no stress in it, work as a com-
ponent of rhythmic construction? It can do nothing more than disturb it.
In the sentence following this, they said "As the word 'stress' has been
used to mean many different things, it is necessary to test more than one
definition of it". This is not, however, a good reason for including a factor
irrelevant to rhythm; what is important is to judge by ear and determine
the factor which forms rhythm. Otherwise, phonetics is only paper pho-
netics and any studies in paper phonetics are fruitless.
Incidentally, did they test the normality of the distribution with the data
from the unit types they constructed? The data of the "lexical" units I
have constructed to examine part of their theory, have been proved by the
Shapiro-Wilk test of normality not to be normally distributed when χ = 2
and χ - 4. From such data as these, viz. those which are not (approximate-
ly) normally distributed, statistical inference cannot be derived, because
statistical inference presupposes the (approximately) normal distribution
of the data. What is capable of being achieved in such a case is descriptive
statistics and nothing else. We can only show the results of measurement
in some informative ways—by calculating the mean, the standard devia-
tion, etc. and presenting a histogram, a scattergram, etc.; we could calcu-
late regression lines but cannot make any statistical tests. Therefore, most
of their discussions are likely to be statistically worthless. When one deals
336 Yosihiro Masuya
with those cases where one is not sure of the (approximate) normal dis-
tribution of the data, the test of normality is indispensable.
When they said that "accented" units began with every syllable that
bore a "pitch accent", it showed their ignorance of the fact that a stressed
syllable might be uttered in low pitch in English (see, for example, Halli-
day (1970)) and that stress may be silent (see above). Their "linguistical-
ly-based" types of unit are all irrelevant to the rhythm of English speech.
If one is unable "to listen differently from the way one listens to every-
day speech", how can one analyse the syllable boundary (or chestable
boundary) of the Standard Scottish English enter, for example—MacMa-
hon (1998, p. 21) says that the boundary lies in the middle of /t/ for some
people?
Nakatani et al. (1981) replaced an adjective-noun phrase in some En-
glish sentences with a reiteration of "ma"-syllable, e.g. "maMA MA" for
"absurd day" and trained the "talkers" [sic] who would help their exper-
iment, so that they could speak the sentence fluently with the proper
rhythm and the proper stress pattern on the "ma"-syllable phrase and
with the normal English manner of pronunciation on the rest of the sen-
tence, in order to examine an aspect of American English rhythm. The
sentences spoken by the talkers were recorded and the "ma"-syllable
phrase was excised from each target sentence. After they most carefully
reviewed the excised phrase, the experimenters selected what they judged
acceptable and measured the duration of each syllable of the excised por-
tion. The experiment was prepared with closest attention.
Who can be sure, however, that the talkers spoke the nonsense word
phrase of each sentence, being conscious (or subconsciously aware) that
they were reading it as a stretch of normal English sound sequences and
that the experimenters listened to the recorded sentences and excised
phrases, being conscious (or subconsciously aware) that they were listen-
ing to a stretch of normal English sound sequences? If the talkers and/or
the experimenters were conscious (or subconsciously aware) that they
were speaking and hearing non-speech phrases, the results obtained will
certainly be different from what is expected from the analysis of speech.
No one would be convinced that the switch in their brain fixed to the
speech mode even at the sequence of "ma"-syllables.
Two Different Kinds of Rhythm: Japanese and English 337
The datasets for this paper are obtained from "The North Wind and the
Sun" read by four English speakers and its translation read by as many
Japanese speakers; the texts are what is produced in phonetic transcrip-
tion in "The Principles of the International Phonetic Association", 1949
edition. The texts given to the readers were reproduced in the ordinary
writing system of both languages.
The speakers are two males and two females for each language. All the
Japanese speakers are speakers of the Tokyo accent. English speakers are
all alleged RP speakers and they have more or less common accentual fea-
tures—one of them (Em2) is a Scotsman born of Scottish parents in Scot-
land but no features of the Scottish accent are detected so far as this read-
ing is concerned: according to him, his mother was educated in English
schools; he may have been, too. E m i pronounced "traveller", which ap-
pears four times in the text, consistently in two chestables, Ef2 consistently
in three chestables and Em2 and E f l sometimes in two, sometimes in
three. The words "other" and "immediately" each occur once in the text
and E m i and E f l pronounced the former in one chestable and the latter in
three chestables, uttering "-diately" in one chestable, while Em2 and Ef2
pronounced the former in two chestables and the latter in five chestables.
All of these English speakers usually exploded the syllable-final stops and
when they exploded them, that exploded stop was pronounced in an inde-
pendent chestable, e.g. "winjd 3 ]", from time to time. Such a chestable was
regarded as an appendage to the word to prevent the situation too favour-
able to the isochronic tendency from being introduced; hence, "winjd 3 }"
was treated as a one-chestable equivalent. I should call these English
speakers "speakers of educated Anglo-English". The speakers are:
In Masuya (1994), Jack Aitken, who, at my request, did for me what Ab-
ercrombie did for Mrs Uldall, also made the scansion in the same way and
answering my question, confirmed that he felt two beats there. In the
scansion for this paper, I followed them; if I felt two beats, I inserted two
carets there and if three beats, three carets.
I made the scansion of the English readings myself—and the plotting of
chestable boundaries as well for the Japanese readings—partly because
Aitken showed that he did not make the scansion as a speaker but as a hear-
er by modifying the first scansion on his second listening and partly because
a phonetician, who was a native speaker of English, was unavailable.
Pauses are part of the rhythmic stretch of words and they do not break
the rhythm; therefore, I have dealt with unfilled feet, i.e. feet beginning
with a silent stress or those consisting of a silent stress alone, in the same
way as filled feet, which are composed of a stressed chestable followed by
either (an) unstressed chestable(s) or none of such. A silent stress has
been counted as one chestable-equivalent; hence, the pause in (2) is a se-
quence of two one-chestable feet and the second foot in (3) is a three-
chestable foot. If a pause is composed of the sequence of two feet consist-
ing of a silent stress alone, I divided the whole duration of the pause by
two and if the sequence is of three silent stress feet, by three, rounding off
Two Different Kinds of Rhythm: Japanese and English 339
the quotient to the first digit. But if the pause is of a silent stress followed
by another silent stress preceding (an) unstressed chestable(s), such as
I have left them out as unmeasurable, for dividing the duration of the
words "then the" in (3) by two is meaningless.
Now I shall turn to the Japanese readings. In Japanese, chestable
boundaries may occur 1) after a vowel, e.g. arujtoki), 2) after the syllabic
"n", e.g. danjdanj, 3) in the midway through a long vowel, e.g. majati} or
4) in between a double consonant, e.g. sikjkari to}. In such a case as 4),
the duration of each chestable cannot be measured, though the whole du-
ration of the two chestables taken together can be. Two or three syllables
are usually spoken in one chestable, though not always.
I do not think pauses consist of chestables but they still form part of the
rhythm of Japanese utterances and their durations are also measured, be-
ing treated as equivalents to chestables. A short pause, between 28 and
90ms with the mean value of 64.6ms, occurs following a glottal stop; this
is regarded as part of the preceding chestable, not a genuine pause. This
is supposed to have something to do with the release of the glottal stop.
The duration of a longer pause has been divided either by the duration
value of the preceding chestable or by the mean value of the durations of
the preceding two chestables whenever there are two measurable ones
and the quotient, rounded off to the first digit, has been regarded as the
number of chestable-equivalents included in the pause.
My dataseis are indeed limited, but for time consuming work like this,
anyone's datasets might be. It is because they are limited that I have con-
ducted the statistical analysis. Statistics is the theoretical device to draw,
from some limited data, inferences about what the elements of a big
dataset, or technically a population, might be like—see note 17 about
"population". Without this theoretical inference, statistics would never
be worth using.
Statistical inference requires as a precondition that the elements of a
sample dataset should be normally distributed. When it is assumed quite
a few factors are involved in a phenomenon, such as in a speaker fixing
the so-called isochronical foot durations in an utterance, a normal distri-
bution of the measured values is generally expected, but, to make sure
and partly to find extraordinary values if any, I have run at the 5 % signif-
icance level (p < 0.05) both the Shapiro-Wilk test and the Geary test for
the English data and the Geary test for the Japanese data to discover
340 Yosihiro Masuya
whether the distribution of each set is normal—the sample size of the Jap-
anese data is too large for the former test. Every set of samples has been
found normally distributed.
The next thing to be tested is whether the datasets are independent of
one another; in other words, each set comprises samples from different
populations. If the inferred population mean and/or variance 20 of two
sample sets are (is) not regarded as the same in consequence of a statisti-
cal test, these sample sets are treated as derived from different popula-
tions; if these population parameters are statistically the same, the
datasets come from one population. Among my datasets, Jf2 and E m i are
independent as shown in Tables 1 and 2 below; Jml, Jm2 and Jfl are de-
rived from a third population and Em2, E f l and Ef2 from a fourth popu-
lation (the speakers' symbols being given to the datasets as their labels,
too). The sample data of Jml, Jm2 and Jfl, therefore, are gathered to-
gether into one dataset and those of Em2, E f l and Ef2 into another set
hereafter, which will be called "3 J's", "3 E's", respectively. Incidentally,
it is nothing but a coincidence that one is independent and that the other
three come from one population in the data of the two languages.
In Table 1, the test at the significance level ρ < 0.05 has not indicated that
the population variance is significantly different between the four speak-
ers but suggested that Jf2's population mean is significantly different
from the others'—her lowest value of the confidence interval is almost
the same as the highest values of the other speakers. In Table 2, the pop-
ulation variance of E m i does not differ significantly from that of Em2
Two Different Kinds of Rhythm: Japanese and English 341
and E f l , but it does from that of Ef2, while the population mean of Emi
is significantly different from that of the other three speakers—Emi 's
highest value of the confidence interval is almost the same as the lowest
values of the other speakers. (When χ had a value other than 3, significant
differences were not detected.) The significance test has been carried out
at the level ρ < 0.05.
Now I shall turn to the comparison between Japanese and English and
at the same time I shall show in what way statistics makes a good contri-
bution to the description of speech rhythm construction, which is a sub-
sidiary objective of this paper.
Let us look at the scattergrams of four speakers, two each selected from
the Japanese and the English speakers (Figs. 1-4).
Jml
600 χ
% ^ 500
500 -
α Ë 400 * ~ •
duration (ms)
A
υ 100
0
0 5 10 15 20
order of occurrence
600
500
1α 'sε 400 Λ •
Figure 2. Jf2's scattergram showing the durations (in ms) of chestables with pauses
included in the order of occurrence. The order of occurrence is translated into a
sequence of numbers, 1.0,1.1,1.2,... 1.9, 2.0, . . . . (Jf2 is an independent sample
data.)
Emi
1400
1200
1000
a 800
o
• duration
600 -f
3
T3
400
O
.O
200
0 Ψ
0 1 2 3 4 5 6 7
Figure 3. E m l ' s scattergram which shows the durations of feet against the foot size,
defined by the number of chestables per foot. (Emi is an independent sample
data.)
Two Different Kinds of Rhythm: Japanese and English 343
Ef2
1200 -r
•
*
•
___ •
•
c 800 •1 1
o • 1
600 • X A duration
3
•• 1 $
•
-α $
400 I 1• •
o •
o •
200
i 1 1 1 1
0 1 2 3 4 5 6
foot size (no. of ches/foot)
Figure 4. Scattergram of Ef2 as a sample of 3E's, which shows the durations of feet
against the foot size, defined by the number of chestables per foot.
3.2. Regression
I have just said in the sub-section 3.1. that a regression does not work for the
Japanese dataseis. But in order to make a comparison between Japanese and
English on one and the same criterion and look at where the difference lies
and how different the rhythmic structures of the two languages are, I want a
regression for Japanese as well. So I have played a mathematical trick and or-
ganized a stretch of Japanese speech into a foot structure like that of English.
I have let a random number generating programme generate a sequence
of random integers between 1 and 6 inclusive chosen with a similar proba-
bility with which these integers, the representation of a foot size, occur in
344 Yosihiro Masuya
English. In the English readings used for this paper, six classes of foot size
exist with a silent stress regarded as a chestable equivalent: 1-, 2-, 3-, 4-, 5-,
and 6- chestable foot. The probability with which every foot size has oc-
curred in the speakers' readings is in Table 3. In the calculation of the prob-
ability, those feet of unmeasurable duration are also counted. Every inte-
ger of the random sequence is chosen with the mean probability, which is
shown in the p-row in Table 3.
Table 3. Probability of the occurrence of each foot size in the English datasets derived
from each population. The integers represent the foot size.
ρ: mean value of probability.
1 2 3 4 5 6
3E's 0.334821 0.276786 0.285714 0.066964 0.017857 0.017857
Emi 0.287879 0.212121 0.348485 0.136364 0 0.015152
Ρ 0.311350 0.244453 0.317100 0.101664 0.008929 0.016504
If the integer 1, for example, occurs with the probability of 0.311350, it oc-
curs 31.135 % of the time.
The sequence generated is
(4) 241111222343131214333223333421311231233
3 3 2 1 2 5 2 3 3 4 3 3 3 3 2 3 2 2 4 1 3 3 1 3 3 2 3 3 3 ....
speech—far from it. But if the rhythmic structure of the Japanese speech
had such a formation, what would happen?
Below are the scattergrams of the Japanese pseudo-foot sequences de-
rived from each population compared with that of 3E's.
pseudo-feet of 3J's
^ 1600
¿ 1400
0 1200
S 1000
î 800 • duration
1 600
S 400
o 200
o
0
0 1 2 3 4 5 6
foot size (no. of ches/foot)
Figure 5. Scattergram of Japanese foot structure derived from 3J's.
Foot size is defined by the number of chestables per foot.
pseudo-feet of Jf2
^ 1600
j | 1400
"g 1200
te
JH
1000
•a 800 • duration
§ 600
400
« 200
£ 0
0 1 2 3 4 5 6
foot size (no. of ches/foot)
3E's
1200
^ 1000
t/i
Β
~ 800
c
o
5 600 • duration
3
^ 400
o
6 200
0
0 1 2 3 4 5 6 7
The upward going slope higher at the right is steeper in the Japanese data
(Figs. 5 and 6) than in the English data (Fig. 7; see also Fig. 3). What does
this visual impression tell?
It is not difficult to calculate regressions for the Japanese data. Lack of
fit does not exist in either the data of 3J's and those of Jf2. In passing, as
I said in Section 1, one-way analysis of variance is required and it has been
carried out with both Japanese and English dataseis. Although all
dataseis do not satisfy the precondition of the analysis, i.e. the homogen-
ity of variance, the test which is to be made is what is called "F-test" and
it is, fortunately, so robust that the consequences of the test are not dam-
aged for want of this precondition. The regressions are:
But the intercept has been found insignificant at the 5 % significance level
for both (a) and (b) and no other choice can be exercised than to accept
zero intercept, though the greatness of the risk of the wrong judgement is
unknown, as I said in Section 1. Thus the regressions which can be adopt-
ed for the Japanese pseudo-feet data are:
Two Different Kinds of Rhythm: Japanese and English 347
1800
duration
(ms) / Jf2
1600
/
/
1400 / /3J's
1200
κ
// /
1000 / /
800
/
/
Λ
/
600
//
/ /
/ /
400
200
-1 0 1 2 3 4 5 6 7
but if the scattergram is divided into the two parts, how the foot duration
is lengthened is seen more clearly and somewhat differently.
Emi
1400
1200
j | 1000
o 800
C
)HS • duration
3
T3
o 400
a
200
0
0 2 4 6
foot size (no. ches/foot)
Figure 9. Scattergram and the quadratic regression for Emi.
For the x-range, 1 < χ < 3, the regression is (8), giving the predicted
values 23 (1,424), (2, 481) and (3, 539).
When 4 < x, the data I have are too few and I cannot declare anything
positively but there is no room for doubt that the increase of the foot du-
ration is greater. Fig. 10 shows the relation between the regressions (7)
and (8).
Two Different Kinds of Rhythm: Japanese and English 349
1800
duration
(ms)
1600
1400
1200
1000
800
600
400
200
-1 0 6 7
Figure 10. Relation between the simple linear regression (1 < χ < 3)
and the quadratic regression in the whole x-range for E m i .
and the curve this equation draws well indicates in what way the foot du-
ration increases. The increase is greater in the range from χ - 3 to χ = 4
and at both sides of this 3-to-4 range the increase is smaller. Here again
the dividing of the scattergram into two groups gives a clearer picture.
(The long-pause feet at the end of a sentence, (1,864) and (1,766) are also
discarded, the former being three times repeated and the latter twice re-
peated.)
350 Yosihiro Masuya
3E's
1200
800
600 • duration
200
0
0 1 2 3 4 5
foot size (no. of ches/foot)
The regression for the range 1 <jc<3 is (10) which gives the predicted val-
ues, (1,470), (2,550) and (3,629) and the regression for the range 4 <x < 6
is (11) supplying the predicted values (4,813), (5,871) and (6,929).
How complicated the English rhythmic structure is, has been described in
Section 3.2 and the contrast it makes with the Japanese simple rhythmic
structure has also been stated there. This complexity of English speech
rhythm may have its source, phonologically, in the double-layeredness of
the rhythmic structure—one is the stratum of chestable and the other is
that of foot formed by the organization of chestables. Japanese has a sin-
gle layer—the stratum of chestable. Besides the stratum structure, there
seems to be various individual differences in English rhythmic construc-
tion, apart from the difference of talking speed among speakers. In the
Two Different Kinds of Rhythm: Japanese and English 351
1800
duration
(ms)
1600
1400
1200
1000
800
600
400
200
-1 0 6 7
-200 foot size
Table 4. Values predicted by the regressions. The integers 1-6 represent the foot or pseudo-
foot size. L: linear regression. Q: quadratic or quintic regression.
1 2 3 4 5 6
3J's L 244 487 731 974 1218 1461
3E's L 470 550 629 813 871 929
Q 477 537 637 818 886 963
Jf2 L 280 560 840 1120 1400 1680
Emi L 424 481 539
Q 423 465 567 730 952 1236
352 Yosihiro Masuya
Table 5. Ratio of the increase of the predicted values. The integers 1-6 represent the foot or
pseudo-foot size. L: linear regression. Q: quadratic or quintic regression.
1 2 3 4 5 6
3J's L 1.0 1.97 3.0 3.99 4.99 5.99
3E's L 1.0 1.17 1.34 1.73 1.85 1.98
Q 1.0 1.13 1.34 1.71 1.86 2.02
Jf2 L 1.0 2.0 3.0 4.0 5.0 6.0
Emi L 1.0 1.13 1.27
Q 1.0 1.10 1.34 1.73 2.25 2.92
For the Japanese pseudo-feet, the ratio multiplies as many times as the
number of chestables in foot, though in 3J's it is not exactly as many as the
number of chestables because of an effect given by the rounding off of the
predicted values at the first decimal place.25 On the other hand, with the
English feet the duration of the 6-chestable foot is nearly 3 times the du-
ration of the 1-chestable foot for E m i and approximately two times for
3E's. The ratio of E m i is not very reliable because the data are too few
for bigger feet, as I said earlier. The duration of the 3-chestable foot is
about 1.3 times for both 3E's and Emi. Interestingly, L-ratios tell us that
in 3E's the increase of the duration is smaller with bigger feet and in E m i
smaller with smaller feet; this matches well with the scattergrams. If a
scattergram is divided in parts and the regression is calculated for each
part, more precise inference can be made. This ratio difference between
the two languages clearly indicates the isochronous tendency of the En-
glish speech rhythm and at the same time, the chestable-timed nature (or
syllable-timed nature, to use the more popular term) of Japanese is equal-
ly beyond reasonable doubt.
This difference of CV's between (12) and (13) suggests that variation in
chestable length in English is considerable whereas that in Japanese is fair-
ly small and also that Japanese shows a tendency to iso-chestabicity (or iso-
syllabicity in more popular terms) whereas English displays a tendency to
isochronicity. Incidentally, in the calculation of CV's only chestables prop-
er have been dealt with, partly because Abercrombie (1967) does not have
pauses in mind when he makes the remark just cited and partly because
what we need to know here is the relation between chestables.
The CV's and the increase ratios of the predicted values of foot dura-
tion calculated by regression explicitly display the rhythmical feature dif-
ference between Japanese and English.
Notes
* Acknowledgements are extended to Prof. Minoru Öta, former Professor of Statistics
(now Emeritus Professor after retirement), Konan University—quite a few considerably
long faxes were exchanged—to Prof. Tomoyasu Taguti, Konan University, a mathemati-
cian, for telling me about pseudo random numbers, to Prof. David Rycroft, Professor of
English Literature, Konan University, for checking for grammar and last but not least, to
Dr Keiichi Masuya for designing the computer programme " G R L I N E " for me to draw
some of the regression lines for this paper.
1. About silent stress, see Jones (19507, p. 227, note 1) and Abercrombie (1964a, 20; 1968).
2. Some may criticize this term as etymologically unjustified; but I preferred the sound and
adopted it when it was suggested by Mrs Uldall. A portmanteau word is not always ety-
mologically justified. About chestable, see Masuya (1979).
3. The phonetic fonts used in this paper are those distributed on the internet by IPA. [§]: as
for the use of [s] I am following J. C. Catford; he calls this sound "apico-infradentalized
lamino-postalveolar fricative" (personal communication). A sibilant "which occurs in
most N. W. Caucasian languages" and in the articulation of which "there is a narrow ar-
ticulatory channel between the blade of the tongue and the extreme back of the alveolar
ridge ... at the same time, the apex of the tongue is in contact with the back of the lower
front teeth . . . " In my articulation of the Japanese /s/, a narrow articulatory channel is
made between the blade of the tongue and the alveolar ridge and the tip of the tongue is
touching the lower alveolar ridge or the back of the lower front teeth; I should call this
sound "apico-infraälveolarized (or infradentalized) lamino-alveolar fricative", also fol-
lowing Catford. If this sibilant precedes H I , it is palatalized. About this Caucasian sibilant,
see Catford (1992).
4. For statistical calculations in this paper, I use Microsoft Excel 2000™ and HB for Win.
5. A regression is an equation statisticians calculate to look at the relation of one variable
to other variables. If the relation is between one variable and another, the regression is
said to be simple.
354 Yosihiro Masuya
6. y represents an inference of y. Such equation represents the variation of the data and is
called "regression". If the equation has one χ term and represents a straight line, it is
said to be "simple linear".
7. The null hypothesis is a hypothesis set up to make a statistical test of the matter in ques-
tion. It is symbolised as H0 and Greek letters, which are reserved for theoretical values,
are used for the specification of the hypothesis: a 0 is an inferred theoretical value for a 0
in the regression, which is to be calculated from the data.
8. The hypothesis which is to be accepted if the null hypothesis is rejected. Symbolised as
Hv Again Greek letters are used for the specification of the alternative hypothesis.
9. If the probability of the error being made is 5 %, the matter in question is said to be sta-
tistically significant at the 5 % significance level and the significance level is sometimes
expressed by ρ < 0.05.
10. Rejection region is the range of calculated values within which the null hypothesis is re-
jected.
11. If the alternative hypothesis is something like Hx\a = α, [α, Φ 0], p2 is calculable.
12. / / , : a 0 * 0.
13. About regression, see note 5. Some details are in Masuya (1997, chap. 13).
14. The variation due to the regression is the differences between the mean of the measured
values and the values of y, i.e. the values calculated by the regression equation. In cal-
culating this variation, each difference is squared and the effect of positive/negative is
avoided.
15. About "lack of fit", see Masuya (1997, chap. 13).
16. The significance levels from the dataset outside this paper, that from this paper and that
of "lexical" units are: ρ < 0.0001, ρ < 0.00000001, ρ < 0.0013, respectively.
17. A population is a large set of elements—too large to take a measurement of each ele-
ment on the variables of interest. In our case, for example, the data obtained from the
life time's reading of a text having done and to be done by a speaker forms a population
and (an) actually recorded reading(s) of the speaker is (are) a sample of the population.
Incidentally, "population coefficient" is the coefficient to be calculated by all the con-
cerned elements in that population.
18. The recordings of the Japanese readings were made in 1995 by Miss K. Mori, Mr Y. Fuji-
no and Mr K. Nanjö, the then graduate students of Konan University, thanks to the in-
termediary act of Prof. K. Imai, Gakushuin University and the English readings were
recorded at the Phonetics Laboratory, Oxford University in 1996 by Mr Neil MacPhee,
an instructor of English at Gakushuin University.
19. In this paper the more popular binary system of feature representation is employed.
20. Population mean, population variance: the mean, the variance, respectively, as a param-
eter in a population.
21. The 95% confidence interval is the interval somewhere of which the value of the in-
ferred parameter is believed to fall with the probability of 0.95.
22. Data values will be shown in the form of (*, y), as is done in mathematics.
23. The terms of a regression consist of the estimators of the parameters in the population
and the y-values are inferred predictions.
24. (11) is significant at the 8 % level, while the other regressions are significant at the 5 %
level. If ρ < 0.05, (11) is not significant, which means that the x-term does not exist and
that foot durations are expected to be somewhat constant in the range 4 < χ < 6.
25. If the acceptance of the null hypothesis about the intercept of the Japanese regressions
had been an error of judgement, the correct predicted values and the correct ratios—
the latter being in b r a c k e t s - w o u l d have been (1, 283) [1.0], (2, 527) [1.86], (3, 770)
[2.72], (4,1014) [3.58], (5,1257) [4.44], (6,1501) [5.30] for 3J's and (1,286) [1.0], (2,566)
Two Different Kinds of Rhythm: Japanese and English 355
[1.98], (3,846) [2.95], (4,1126) [3.94], (5,1406) [4.92], (6,1686) [5.90] for Jf2. The result
is not very different.
References
Abercrombie, David
1960 "'Stress' and some other terms". Paper given to the National Phonetic Collo-
quium, Leeds, March 1960. Produced in Work in Progress, 9 (1976,51-53) and
reproduced in Abercrombie (1991, chap. 10).
1964a "A phonetician's view of verse structure", Linguistics, 6. Reproduced in Ab-
ercrombie (1965, chap. 3).
1964b "Syllable quantity and enclitics in English", in Abercrombie et al. (1964,216-
222). Reproduced in Abercrombie (1965, chap. 4).
1965 Studies in Phonetics and Linguistics. Ronald Mackin and Peter Strevens (Gen-
eral eds.), Language and Language Learning, 10. London: Oxford Univ. Pr.
1967 Elements of General Phonetics. Edinburgh: Edinburgh Univ. Pr.
1968 "Some functions of silent stress", Work in Progress, 2. Reproduced in Aitken
et al. (1971,147-56) and Abercrombie (1991, chap. 9).
1991 Fifty Years in Phonetics. Edinburgh: Edinburgh Univ. Pr.
Abercrombie, David, D. B. Fry, P.A.D. MacCarthy, N. C. Scott and J. L. M. Trim (eds.)
1964 In Honour of Daniel Jones: Papers Contributed on the Occasion of his Eighti-
eth Birthday 12 September, 1961. London: Longmans.
Aitken, A. J., Angus Mcintosh and Hermann Pálsson (eds.)
1971 Edinburgh Studies in English and Scots. London: Longman.
Catford, J. C.
1992 "Caucasian phonetics and general phonetics", in Paris (1992,193-216).
Crystal, David (ed.)
1982 Linguistic Controversies: Essays in Linguistic Theory and Practice in Honour
of F. R. Palmer. London: Edward Arnold.
Dauer, R. M.
1983 "Stress-timing and syllable-timing reanalyzed", Journal of Phonetics, 11, 51-
62.
Fletcher, Janet
1991 "Rhythm and final lengthening in French", Journal of Phonetics, 19,193-212.
Halliday, M. A. K.
1970 A Course in Spoken English: Intonation. London: Oxford Univ. Pr.
Hammerich, L. L., Roman Jakobson & Eberhard Zwirner (eds.)
1971 Form and Substance: Phonetic and Linguistic Papers Presented to Eli Fischer-
J0rgensen 11th February, 1971. Copenhagen: Akademisk Forlag.
Jones, Daniel
19507 An Outline of English Phonetics. 7th ed. Cambridge: W. Heffer & Sons, Ltd.
Lehiste, Ilse
1977 "Isochrony reconsidered", Journal of Phonetics, 5, 253-63.
MacMahon, M. K. C.
1998 Basic Phonetics. Glasgow: Univ. of Glasgow.
356 Yosihiro Masuya
Masuya, Yosihiro
1979 "Syllable and 'chestable'", The Journal ofKonan University, Faculty of Letters,
34,1-6. Reproduced in Masuya (1997, chap. 6)
1994 "Isochronicity of Lowland Scottish English", The Journal ofKonan Universi-
ty, Faculty of Letters, 33 <English Studies>, 108-27. Reproduced in Masuya
(1997, chap. 13).
1997 Phonetics and Phonology: Selected Papers. Tokyo: Kobian-Shobo.
Nakatani, Lloyd H., Kathleen D. O'Connor & Carletta H. Aston
1981 "Prosodie aspects of American English speech rhythm", Phonetica, 38, 84-
106.
Odé, Cecilia & Wim Stokhof (eds.)
1997 Proceedings of the Seventh International Conference on Austronesian Linguis-
tics, Leiden, 22-27 August, 1994. Amsterdam: Rodopi.
Paris, Catherine (ed.)
1992 Caucasologie et mythologie comparée, Actes du Colloque international du
C.N.R.S.-/V'- de Caucasologie (Sèvres, 27-29 juin 1988). Paris: Peeters.
Roach, Peter
1982 "On the distinction between 'stress-timed' and 'syllable-timed' languages", in
Crystal (1982, 73-79).
Shen, Yao & Giles G. Peterson
1962 "Isochronism in English". Studies in Linguistics, Occasional Papers, 9. Buffa-
lo, New York: Department of Anthropology and Linguistics, University of
Buffalo.
Uldall, Elizabeth T.
1971 "Isochronous stresses in R. P.", in Hammerich et al. (1971, 205-210).
van Zanten, Ellen & Vincent J. van Heuven
1997 "Effects of word length and substrate language on the temporal organisation
of words in Indonesian", in Cecilia Odé & Wim Stokhof (eds.) (1997, 201-
213).
Wenk, Brian J. and François Wioland
1982 "Is French really syllable-timed?", Journal of Phonetics, 10,193-216.
Williams, Briony and Steven M. Hiller
1994 "The question of randomness in English foot timing: a control experiment",
Journal of Phonetics, 22,423-439.
sC Clusters as Complex Segments: Evidence from
the Contrastive Phonology of English and Japanese
Noriko Yamane
Nagoya Bunri University
1. Introduction
The purpose of this paper is to present new findings about the uniqueness of
sC clusters (i.e., Is/ and voiceless stop consonant) from the aspect of English
historical phonology and Japanese loanword phonology. It is well-known
that sC clusters behave differently from other 'normal' consonant clusters,
and instead show similar characteristics with those of complex segments.
The work on complex segments has been extensively developed by Sagey
(1986), Lamontagne (1993), van de Weijer (1996), among others, who share
a certain consent that a complex segment designates itself as a single seg-
ment which consists of a combination of manner and/or place of articulation,
whose sequence may sometimes violate the sonority hierarchy. Since sC
clusters consist of both [+ continuant] and [-continuant], and violate sonor-
ity hierarchy in the onset position,1 the latter two researchers assume that sC
clusters are categorized as complex segments. A lot more evidence for the
assumption has been presented both empirically and theoretically. Howev-
er, little is known about how sC clusters behave in historical phonology and
loanword phonology. I would like to address two independent claims, (i) In
English historical phonology, not all sC clusters behave uniformly; the di-
achronic survey on sC clusters reveals asymmetrical behavior according to
the point of articulation of the voiceless stop part, (ii) In Japanese loanword
phonology, the complex status of sC is not particular to Present-day English
(henceforth, PE) or the source language, but is also reflected in the target
language of Japanese; this will be clear from how sequences of consonants
are realized in gemination. Both findings will support the idea that sC clus-
ters in PE should be analyzed as complex segments.
kirk 1982: 347-349, among others), whether they are in the onset or the
coda position. This treatment seems to be reasonable for PE. What I
would like to argue in this section is that this is not always correct, from a
diachronic viewpoint. Specifically, the status of sC clusters has shown an
Cor/Lab asymmetry in the coda position.
As the above figure shows, sC clusters in the onset, whether they are coro-
nal sC clusters (i.e. st) or non-coronal sC clusters (i.e. sp and sk), have al-
ways behaved like complex segments from OE through PE, but they have
not in the coda position. In OE, no coda sC clusters show complex status,
but focusing on the diachronic analysis of sC in the coda position, coronal
sC clusters begin to show monopositional status, historically earlier than
non-coronal sC clusters. The chronological point when this Cor/Lab
asymmetry is observed is ME. It is not until ModE that this asymmetry
starts to disappear, or that sp and sk as well as st start to behave as one
unit. The observation outlined here will be clarified in the discussion from
section 2.1. through section 2.4.
Thus the words in (2 i a) and (2 ii a) do not undergo HVD, while the words
in (2 i b) and (2 ii b) do.
As for high vowels after the sC clusters, they are also deleted.
Every sC cluster can cooccur with long vowel /a:/, but only st allows a
great variety of long high vowels and diphthongs. Examples are shown
below. 3
g a s p
As the above illustration shows, words like gaspe did not take part in
MEOSL. The reason is that the stray mora μ is compensated by the pre-
ceding vowel, whose syllable is not open unlike tale.5
Following this argument, words with medial st are also expected to be
prevented from lengthening. However, unlike words with sp or sk, words
with st do take part in lengthening. 6 In order to derive lengthening of such
words, we need to posit the following process, which we call Complex
Segment Formation (henceforth, CSF) of /st/ as below.
(7) CSF of st
a. /s/ Delinking b. Stray Code Adjunction
μ μ μ μ
\s t! /I
s t
Based on the assumption that all segments in the rhyme bear a mora in
English, both /s/ and Iii in the coda position own a mora. But Is/ delinks
from the mora as in (7a), and it is reassociated with the coda-final mora
by Stray Coda Adjunction, as in (7b). Thus st changes its status from an
ordinary consonant cluster into a single complex segment. Note that
this procedure targets st only, which shares unmarked place feature
[cor] as well as [voice].7
Given that CSF applies after the original MEOSL triggered by schwa
loss, words with word-medial st will automatically undergo lengthening.
Let us call the original MEOSL 'the first MEOSL', whose application is
limited to the structure CVCV. Let us call the lengthening triggered by
CSF as 'the second MEOSL', which applies to the structure CVst.
362 Noriko Yamane
(9) PVFL
(i) sC Clusters
a. _ st: blast, cast, caste, fast, last, mast, past
b. _ sp: asp, clasp, gasp, grasp, rasp
c. _ sk: ask, bask, cask, casket, flask, mask, task
(ii) Voiceless Fricatives
a. _ s: brass, class, glass, grass, mass, pass
sC Clusters as Complex Segments 363
(10) PVFL
(i) Before sC
a. Delinking b. Spreading c. Stray Coda Adjunction
o σ σ
f!\\
μ μ μ
r
μ / μ\ \ μ 7i\
μ μ μ
I ni
g a s p g a s p g a s p
(11) True Cluster Analysis ->• No OCP violation due to linear ordering
a. C s C b. s C C
Fj [cor] Fj [cor] F Fj
PL PL PL PL
I / \ / \ I
[ - cont] [ - cont] [+ cont] [ - cont] [+ cont] [+ cont]
Fj Fj [cor] F [cir] Fj
In complex segments, two place features behave as if they are linearly un-
ordered. As for the sC clusters, the two place features with their head
[cont] features are assumed to belong to separate tiers, which are domi-
nated by a single place node. The adjacency condition is satisfied here
thanks to this non-linear ordering of the relevant features. Thus the aster-
isked examples below the representations militate against the OCP, and
the infrequency of the examples shown is appropriately explained. 10 We
can see here that this analysis supports the status of complex segments in
both onset and coda.
Based on this argument, it seems reasonable to conclude that sC clus-
ters in PE are monosegmental units in both onset and coda.
If sC clusters are true consonant clusters, the syllable codas will contain
two consonants, which violates *COMPLEX (CODA). On the other hand, if
the sC clusters are complex segments, the root nodes will contain two
place features, which violates *COMPLEX (SEG). The change of the status
of sC can be captured with the conflict of these constraints.
(14) a. O E b. P E
* COMPLEX ( S E G ) *COMPLEX ( C O D A )
I I
*COMPLEX (CODA) *COMPLEX ( S E G )
4 I
st, sp, sk [st], [sp], [sk]
(16) a. * P L / L A B , LAB: TWO Labial Place features are not associated to one
place node.
b. * P L / L A B , COR: Both a Labial Place feature and a Coronal Place
feature are not associated to one place node.
c. * P L / C O R , COR: TWO Coronal Place features are not associated to
one place node.
(i) Is ΰ
*
a."®· [st]
b. s t *!
(ii) Is pi
a. [sp] *!
b."®· s ρ *
(i) Is t/
a.®" [st] *
b. st *!
(ii) Is ρ/
a.·2" [sp] *
b. s ρ *!
(21) * C <V>
/X
[cont] ! [stop]
Place I
Place
(22) DEP-V: Every vowel of the output has a correspondent in the input.
(23) DEP-C: Every consonant in the output has a correspondent in the input.
This paper will focus on words with sequence C1C2# (C¡C2C3# is not taken
up here). Note that the 'potential geminate consonants' notated above in-
dicate stops, affricates and fricatives, which are all [ - sonorant] conso-
nants. The generalization above appropriately captures the fact that the
sequence whose C t is a sonorant consonant (either nasal, liquid or glide)
is not realized in gemination. However, this condition overgenerates the
gemination patterns; even if Q is a [- sonorant] consonant, it does not al-
ways turn up with geminated forms. So the sequence in question actually
falls into two types; Gemination CC and Anti-Gemination CC. In order
to predict this asymmetrical distribution, attention must be paid to the
number of contrasting features of place, [continuant] and [sonorant], be-
tween Cj and C2. The figure below presents the grouping of Gemination
CC and Anti-Gemination CC.
sC Clusters as Complex Segments 371
(i) Gemination CC
(ii) Anti-Gemination CC
(26)
Quebec French Deletion Japanese Gemination
Input ctc2#
Phonological Process C 2 Deletion Cj Insertion
Target non-salient C 2 salient Cj
Featural Contrast 5Ξ 2
Output Q< >
(#: word-boundary)
French deletes non-salient C2, when the featural contrast is two or less,
producing Cj< >. But Japanese inserts salient C t when the contrast is two
or more, producing C 1 <C 1 >C 2 .
The apparent distinction is unified in the theory of featural contrast.
Both phenomena serve to enhance the acoustic salience of Q . The anal-
ysis for Japanese will be presented in section 3.3.
(27)
Contrasting features Ν Type of CC
st [cont] 1 => Anti-Gemination CC
sk pi, [cont] 2 => * Gemination CC
sp pi, [cont] 2 => * Gemination CC
St would have one featural contrast, but sk and sp would have two. Using
the featural contrast approach, the number of contrast, two, is supposed
to trigger gemination.
However, sk and sp systematically avoids gemination as in the exam-
ples below.
(28)
Contrasting features Ν Examples (English > Japanese)
st [cont] 1 <= cast > kyasufo, test > te.suίο
sk [cont] 1 <= desk > deiufcu, task > tasu&u
sp [cont] 1 <= crisp > kumupu, grasp > gurasupu
sC Clusters as Complex Segments 373
This fact indicates that all sC clusters should be uniformly grouped into
Anti-Gemination CC whose featural contrast is less than two. If we adopt
the idea that sC clusters are complex segments, whose featural contrast is
only one (i.e., [cont]) as the apparent distinction of place features are nul-
lified by the single place node that they share, based on the representation
in (21) for instance, the gemination pattern is correctly predicted.
However, in contradiction to the representation in (21), sC clusters in
Japanese are finally broken up with vowel epenthesis, unlike Moderrn
Standard Hindi and Sinhalese. This point will be discussed in 3.4.
The next section will analyze the Gemination CC and the Anti-Gemi-
nation CC respectively in the framework of OT, where the status of sC
clusters will be further confirmed.
MAXDUR(CJ) DEP-C
i i
*C1<C1>C2 c , < c > c -2
MAXDUR(Q) MAX
V
DEP-C
Even if the ranking problem is answered in such a way, this ranking over-
generates the gemination; contrary to the fact, every input cluster CjC 2 in
the input emerges as C j < C j > C 2 in the output. Furthermore, it does not
tell us what kind of information and how much information in the input
is preserved in the output of the loanword phonology.
A solution to these problems can be found with the reconsideration of
MAX. MAX plays a role in specifying contrasts, in that it favors preserving
contrasts of featural specification of inputs (see Kirchner 1998). Accord-
ing to Flemming (1995: 44), this kind of function of faithfulness is taken
over by the "Maintain Contrast" constraints.
(36)
c,c2
1/ CONTRAST = (n)
(n)
a Q Q
- 1/
(n)
b. Ci C, C,
*
1/
(0)
(37) CONTRAST = 1
[ ' ] MAXDUR(CÌ)
I (cut-off point indicator)
CONTRAST = 2
CONTRAST = 3
Gemination does not occur when the contrast is one, and occurs when it
is two or more. So the cut-off point for the susceptibility of gemination
can be drawn between CONTRAST = 1 and CONTRAST = 2 . Overriding C O N -
TRAST = 2 and CONTRAST = 3 , M A X D U R ( C J ) serves as an indicator of the cut-
off point.
Based on the discussion above, the dominance relations among MAX-
D U R ( C ] ) , CONTRAST = Ν and D E P - C will emerge as below. (The ranking in
(31b) has been added to that in (37).)
(38) CONTRAST = 1
MAXDUR(CJ)
I \
CONTRAST = 2 DEP-C
CONTRAST = 3
Let us see if the dominance relation given here correctly produces the
output. (For graphic simplicity, the candidates are expressed with the rel-
evant part only, i.e. the realization of C,C 2 .)
Observe Gemination CC first.
sC Clusters as Complex Segments 377
(39) Gemination CC
The candidates here are all immune to CONTRAST = 1, because they have
no correspondence of Ν = 1 in the inputs: Candidates in (39 i) with Ν = 3
in the input are submitted only to CONTRAST = 3 for evaluation, and those
in (39 ii) with Ν = 2 in the input are submitted only to CONTRAST = 2 . As
for the next constraint M A X D U R ( C ] ) , it is violated in ungeminated candi-
dates (39 i a) and (39 ii a), while it is not by the geminated ones (39 i b)
and (39 ii b). Thus, the geminated candidates emerge as winners.
Next let us examine the cases of Anti-Gemination CC.
(40) Anti-Gemination CC
CONTRAST = 1 MAXDUR(C,) CONTRAST = 2 CONTRAST = 3
the minimum featural contrast (N = 1) in the input. Thus the featural con-
trast approach in OT not only captures the gemination patterns, but also
answers the problems.
Furthermore, the analysis here provides us with a significant advan-
tage: it elucidates the mechanism of gemination. Notice that the gemina-
tion is expressed as a process which turns two or more featural contrasts
into zero. The reason is clear: The presence of many featural contrasts re-
quires speakers to make much articulatory effort, so speakers attempt to
minimize their number. Gemination can then be considered as an articu-
latory-based phonological process which favors speakers.14
(42) CODACOND
*C]a
(43) CODACOND » D E P - V
Combining the constraint hierarchies (38) and (43), based on the assump-
tion in (41), the overall ranking emerges as below.
(44) CODACOND
DEP-V
CONTRAST = 1
MAXDUR^)
I \
CONTRAST = 2 DEP-C
CONTRAST = 3
Let us make sure if the constraint hierarchy above allows vowel epenthe-
sis to occur. (For the lack of space, the constraints below M A X D U R ( C , ) are
omitted.)
(45)
* *
c. <*~ te.su.to
d. tes.sto *
*!
e. tes.su.to * *!
(47)
b. te.su.t *! * *
c. te.su.to * *
d. tes.sto *
*!
e. tes.su.to * *!
f. <*• tes.to * *
b. ta.ku.t *! * *
c. ta.ku.to * *
d. tak.kto *
*!
e. tak.ku.to *
*!
f. tak.to *! * *
Multiple winners are allowed for sC as in (47 i), but are not for normal con-
sonant clusters as in (47 ii).18 This is thanks to the dominance of the sylla-
ble structure constraint or C O D A C O N D , which makes candidate (47 i f) op-
sC Clusters as Complex Segments 381
timal, in that /s/ in coda has a doubly-linked place associated with the
following I I I . Thus hierarchy (44) accords with the vowel deletion in casu-
al speech as well.
Vowel epenthesis therefore follows from the dominance of syllable
structure constraints. The constraint hierarchy subsumes the observation
that the integrity of sC is maintained against consonant insertion (thus
s<s>C is ill-formed), but not against vowel insertion (thus s<V>C is well-
formed). It is also advantageous that the analysis here captures the vowel
deletion in casual speech, which may suggest that the complex status of
sC survives at some level in the output.
4. Conclusion
We have discussed the unique behavior of sC clusters. In the former part,
dealing with English, the diachronic survey revealed the C O R / L A B asym-
metry. It was found that st started to behave like one unit, chronologically
earlier than sp and sk. Based on the observation that the status of sC var-
ies according to each synchronic system, we shifted our attention in the
latter part of this paper to another synchronic system, that of Japanese
loanword phonology. The investigation of gemination patterns forced us
to categorize sC as Anti-Gemination CC, whose featural contrast is only
one. This number is ascribed to the unique representation of the complex
segment, whose place features are nullified by a single place node, unlike
normal consonant clusters. The treatment of sC in such a way enabled the
correct prediction of gemination patterns, using the theory of featural
contrast in tandem with OT. The analysis developed here has additional
advantages. First, the mechanism of gemination is explained in terms of
minimization of articulatory effort and maximization of acoustic salience.
Second, the prima facie paradox, i.e. a consonant does not break up the
integrity of sC but a vowel does, is resolved in OT. Lastly, it makes it vis-
ible that the status of complex segment is mirrored in the output of Japa-
nese loanword phonology.
Acknowledgements
This work partially contains the basic insights of Yamane & Tanaka (2000), but is a revised and
extended version giving an OT-based account for gemination of word-final consonant clusters
in Japanese loanword phonology. I wish to express my gratitude to the following people for
their helpful discussions and suggestions on several points in the paper: John Kingston, John
382 Noriko Yamane
Me Carthey, Lisa Selkirk, Shin-ichi Tanaka, Jeroen van de Weijer, and two anonymous review-
ers. Thanks also go to James Landkamer, who kindly gave me helpful suggestions on grammat-
ical errors. Last but not least, I thank the editors, Jeroen van de Weijer and Tetsuo Nishihara,
for having let me join this valuable project. Of course, I alone am to blame for any remaining
inadequacies or misconceptions therein.
Notes
1. This point has made researchers assume that sC in the onset is 'extrasyllabic'. But a dis-
cussion on the extrasyllabic approach is beyond the scope of this paper, which focuses on
sC in coda position, not onset position. As for the 'extrametrical' approach, we need to
add some comments. It would have difficulty in treating the second MEOSL and PVFL
in section 2.2. through 2.3.; e.g. the bimoraic tas<t> would turn out to be the trimoraic
ta:s<t>, which would militate against lengthening based on mora preservation.
2. Wells (1990) describes [a:, t œ II se]· The RP form is [a:]. Pronunciations marked with the
symbol t are widespread in English among educated speakers, but nevertheless judged
to fall outside RP. In England, [ae] is a localized northern form (though it is standard in
AmE) (see Wells 1990: xii). [a:] in sp and sk also has the same variety.
3. Examples are collected from Fergusson (1985) and Gunji (1991), and I cross-referenced
their pronunciation with Wells (1990). Rare examples are omitted.
4. [oi] is an exception in the GVS. It traces back to ME /i:/, which was supposed to result in
[ai] through GVS.
5. The first MEOSL applies to tale in the following way.
t a : 1
What compensates the stray mora depends on the weight of the stressed syllable. Com-
pensation by a vowel occurs when the preceding syllable is light as above. But when it is
already heavy as in (6c), the stray mora is compensated by a final consonant as in (6d).
Notice that the word-medial sC in (6a) is heterosyllabic. But if the word-medial sp is
also taken as a complex segment, two ways of syllabification are logically possible. One is
ga. sps and the other is gasp.3. The former would make the initial syllable light, then the
stray mora delinked from schwa should be compensated by a vowel, but it is contrary to the
fact. The latter syllabification would create a heavy syllable, so the compensation by a vow-
el is not expected. But the uncompensated stray mora here needs to disappear without rea-
son, which would weaken the principle of structure preservation. Thus both syllabifications
are problematic, and we do not regard word-medial sC clusters as complex segments.
6. Among normal consonant clusters, obstruent plus liquid clusters such as able, bible, fable
underwent lengthening (see Jordan 1974: sec. 25, Remark 2 and Nakao 1985:132-3). This
is because the initial syllable is formed to be open by Sonority Contour Principle, (cf. Ya-
mane 1998a). Thanks to Jeroen van de Weijer for addressing this question.
sC Clusters as Complex Segments 383
7. It may be ascribed to the OCP effect, as an anonymous reviewer indicates, but for now
the statement, "the more similar the segments, the stronger the interaction" (Itô &
Mester 1995:205)) best describes my intuition. An empirical evidence is also found in OE
Homorganic Lengthening of Id (81.1%) (Minkova & Stockwell 1992). We assume that
the strong interaction oí Id, sharing [cor] and [voice], triggers lengthening (Yamane 1997).
8. St words which underwent PVFL are limited to the exeptions to the second MEOSL.
So st in these words did not acquire monomoraicity until PVFL began. Thanks to an
anonymous reviewer for addressing this point.
9. PVFL can be considered to be a postlexical change. Yamane (1996a, b) demonstrates
that a constraint-based system can treat this kind of change as a constraint-reranking.
10. Takeru Honma (p.c.) points out that if we assume that the OCP targets only specified
features, with the underspecification of [cor], we will gain an account for Cor/Lab asym-
metry in the sense that tVst type is found more often than *pVsp and *kVsk (cf. Davis
1991, Cairns 1988). But the OCP effect on the continuant part of coronals as in (12b) (i.e.
the sCVs type is less frequent than the sCVf type of words) cannot be captured, unless
[cor] is specified. A fuller study of the underspecification problem lies outside the scope
of this paper, but is an important topic. For the details, see McCarthy & Taub (1992).
11. * COMPLEX is also sustained by the implicational universale described by Prince & Smo-
lensky (1993:186): "If the segment inventory of a language includes a complex segment
with primary place π and secondary place ψ, it has a simple segment with place π and a
simple segment with place ψ."
12. Each system of synchronic phonology outlined agrees with the generalization "banning
the worst of the worst" (Prince & Smolensky 1993:180). From a diachronic viewpoint,
the way of constraint reranking here conforms to the principle of family invalidation;
"it is always the lowest constraint in the family of ranked constraints that is overridden
first when that family is subject to change" (Yamane 1996b). The earlier realization of
st complex follows from this.
13. sC in onset is an unmarked sequence, from a typological viewpoint (Morelli 1998). But
whether this holds true of sC in coda remains to be answered.
14. Then does the gemination bring good results only to the speakers? The analysis devel-
oped here states that gemination is desirable for both speakers and hearers. As for the
speakers, the articulatory effort is minimized, since the number of contrasts is reduced.
As for the hearers, the acoustic salience is enhanced, since Q is prolonged. Featural lev-
el or C level, gemination turns out to be in favor of both speakers and hearers. It is no
wonder that phonological phenomena is caused by the balance between speakers and
hearers (cf. Yamada 1994, 1997). Both Quebec French and Japanese seem to achieve
such a balance, although each function is realized at different levels as below.
References
Alderete, John
1995 Faithfulness to Prosodie Heads. Ms., University of Massachusetts, Amherst.
Benua, Laura
1997 Transderivational Identity: Phonological Relations Between Words. Doctoral
dissertation, University of Massachusetts, Amherst.
Broselow, Ellen
1987 Non-obvious Transfer: On predicting Epenthesis Errors, Interlanguage pho-
nology, ed. by G. Ioup and S. Weinberger, 261-278, Newbury House, Cam-
bridge, MA.
1991 The Structure of Fricative-Stop Onsets, Handout presented at the workshop
Properties of Phonological Feature Organization, University of California,
Santa Cruz.
Cairns, Charles E.
1988 Phonotactics, Markedness and Lexical Representation. Phonology 5, 209-36.
Côté, Marie-Hélène
1997 Phonetic Salience and the O C P in Coda Cluster Reduction. CLS 33: The Main
Session, 57-71, Chicago Linguistic Society.
Davis, Stuart
1991 Coronals and the Phonotactics of Nonadjacent Consonants in English, Pho-
netics and Phonology 2: The Special Status of Coronals, ed. by C. Paradis and
J-F. Prunet, 49-60, Academic Press, San Diego and London.
Dobson, Eric J.
1968 English Pronunciation 1500—1700. vol. II. Clarendon Press, Oxford.
Fergusson, Rosalind
1985 Rhyming Dictionary.Market House Books, England.
Flemming, Edward S.
1995 Auditory Representations in Phonology. Doctoral dissertation, University of
California, Los Angeles.
Fudge, Erik C.
1969 Syllables. Journal of Linguistics 5, 253-86.
Fujimura, Osamu
1997 Syllable Features and the Temporal Structure of Speech. Proceedings of LP'
96, ed. by B. Palek, 91-111, Charles University Press.
Fujiwara, Yasuaki
1990 Koeishi Inritsu Kenkyuu (A Metrical Study of Old English Poems). Keisuisha,
Hiroshima.
sC Clusters as Complex Segments 385
Yamane, Noriko
1995 An Integrated Account of Middle English Open Syllable Lengthening. Κ 15,
36-46, Kansai Linguistic Society.
1996a Rekishiteki On-in Henka to Saitekisei Riron: Schwa Shoushitsu no Baai [Di-
achronie Phonological Change and Optimality Theory: The Case of Schwa
Loss]. On-in Kenkyuu: Riron to Jissen [Phonological Studies: Theory and
Practice], ed. by On-in Ron Kenkyuu Kai, 167-72, Kaitakusha, Tokyo.
1996b Diachrony in Optimality Theory: The Directionality of Change & The
Change of Optionality. Talk given at the symposium 'Phonology and its Inter-
faces' on the 14th National Conference of the English Linguistic Society of Ja-
pan.
1997 More on Middle English Open Syllable Lengthening: Is/ + Stop Clusters. Talk
given on the 15th National Conference of the English Linguistic Society of Ja-
pan.
1998a Chuu-Eigo ni okeru Kai-onsetsu Chou-onka Saikou: Isl to Heisaon no Ren-
ketsu [More on Middle English Open Syllable Lengthening: Is/ + Stop Clus-
ters]. JELS 15, 261-270, English Linguistic Society of Japan.
1998b Great Vowel Shift, Chain Shift, and Local Conjunction. Talk given at the
workshop 'Great Vowel Shift and the Phonological Theory' on the 16th Na-
tional Conference of the English Linguistic Society of Japan.
1999a On-in Riron ni okeru Tsuuji-Henka no Yosoku-Kanousei : sC Rensa no Baai
[The Predictability of Diachronic Sound Change in Phonological Theory: The
Case of sC Clusters]. Journal of the Phonetic Society of Japan 3:1,29-40, Toyo
Shuppan, Tokyo.
1999b Local Conjunction in the Great Vowel Shift: An Answer to the Structural Co-
herence Problem from Optimality Theory. Ms., Nagoya Bunri University.
Yamane, Noriko and Shin-ichi Tanaka
2000 English Word-Final sC Clusters in Japanese Loanword Phonology. Phonolog-
ical Studies 3, ed. by the Phonological Society of Japan, 37-44, Kaitakusha, To-
kyo.
Author Index
Fukazawa, H. and M. Kitahara 41fnl9, Itô, J. and R.A. Mester 9,41fnl9, 67-69, 71-
83fn3, 83fn4, 85-109 77,79-81, 83fn3, 83fn4, 85-86, 88f, 91, 94,
Fukazawa, H. and V. Miglio 105fnl 97-98,105fn2,106fn3,106fn8,137fn4,
Fukazawa, H., M. Kitahara and M. Ota 91, 186f, 241nll, 241fnl2, 310, 314
94, 97,105fn2 Itô, J., R.A. Mester and J.E. Padgett 97,
Fukui, M. 120,122f 105fn2
Fukui, N. 19 Itô, J., Y. Kitagawa and R.A. Mester 163
Fukui, N. and Y. Takano 19, 25ff, 31, 38, Itô, J.: see also Kubozono, Itô and Mester
43fn33
Fukui, N.: see also Saito and Fukui Jaeggli, O.A. 320fn4
Johnstone, T.M. 195, 208
Göksel, Α.: see Charette and Göksel Jones, D. 353fnl
Goldsmith, J.A. 85, 369 Jordan, R. 382fn6
Greenberg, J.H. 304
Gunji, T. 382fn3 Kager, R. 71, 77
Kageyama, T. 245-76, 278,283-84,286-89,
Hale, K.L. 18 303, 307f
Halle, M. and A.P. Marantz 245 Kageyama, T.: see also Shibatani and
Halle, M. and J.-R. Vergnaud 62 Kageyama
Halle, M.: see also Chomsky and Halle
Kahn, D. 40fn4
Halliday, M.A.K. 336
Katamba, F.X. 303
Hamans, C. 319fnl
Katayama, M. 189fn6, 378-79
Hammond, M.T. 384fnl8
Kawagoe, I. 370
Hankamer, J.: see Sag and Hankamer
Kayano, S. 145ff, 148ff
Haraguchi, S. 3, 47-65, 215, 218, 232, 233f
Kaye, J.D. 195, 220, 225
Harris, J.M. 193,196f, 198ff, 200
Kaye, J.D., J. Lowenstamm and J.-R. Verg-
Harris, J.M. and G.A. Lindsey 193,196f,
naud 193,196-97, 200, 212fn8
198ff, 200
Kayne, R.S. 24ff, 31,38, 42fn32
Haspelmath, M. 320fnl0
Kibe, N. 105fn2
Hattori, N. 184,190fn21
Kim, S. 358
Hawkins, J.A. and A. Cutler 306
Hayes, B.P. 3, 5, 8,118,142,180, 360 Kindaichi, H. 49
Heuven, V.J. van: see van Zanten and van Kindaichi, K. 142,144
Heuven Kiparsky, R.P.V. 195, 273
Hirayama, T. 221, 240fn2 Kirchner, R.M. 375
Hock, H.H. 319fn3 Kitagawa, Y.: see Itô, Kitagawa and Mester
Hoji, H.: see Saito and Hoji Kitahara, M. 378
Honma, T. 67-84,105fn2, 383fnl0 Kitahara, M.: see Fukazawa and Kitahara;
Huang, S. 320fn6 Fukazawa, Kitahara and Ota
Kohler, K.J. 357
Ingleby, M., C. Chalfont and W. Brockhaus Kubozono, H. 15-16, 86,111-42,136fn3,
197 137fn4,137fn7,159ff, 164ff, 168,171,176f,
Inkelas, S. 83fn5, 261 180,188fnl, 189fn7,189fn9,246,261,270,
Inkelas, S. and D. Zee 141 294fn2, 295fn2, 299-302, 308-309, 320fn8
Ishiwata, T. 301 Kubozono, H. and S. Ohta 117,314,320fnl4
Itô, J. 13, 67-69,71, 300, 301, 302, 313-14, Kubozono, H., J. Itô and R.A. Mester
320fnl2, 320fnl5, 320fnl6 189fn3
Autor Index 391
Quackenbush, H.C. and M. Ohso 137fn7 Takahashi, D.: see Boskovic and Takahashi
Takano, Y. 19, 27, 29, 32, 42fn32
Ralli, A. 303 Takano, Y.: see also Fukui and Takano
Rappaport Hovav, M. and B.C. Levin 255 Tamura, S. 144,145ff, 148ff
Refsing, K. 155fnl5 Tanaka, S. 118,120,136fn2,159-92,189fn4,
Rennison, J.R. 196 190fnl9
Rice, K.D. 105fn2 Tanaka, S. and H. Kubozono 129,137fn5
Rice, K.D.: see also Avery and Rice Tanaka, S. and N. Yamane 126,185,189fn9,
Roach, P.J. 331 190fnl9
Rochemont, M.S. and P.W. Culicover 27 Tanomura, T. 137fn7
Roeper, T. and M.E.A. Siegel 249 Tateishi, K. 3,7,11,38,40fn7,40fn9,40fnl0,
Ross, J.R. 18, 24, 28 137fn4,163,190fnl7, 240fn7
Tateishi, K.: see also Akasaka and Tateishi;
Sag, I.A. and J. Hankamer 257 Selkirk and Tateishi
Sagey, E.C. 357, 364 Taub, A.B.: see McCarthy and Taub
Saito, M. 18-19 Taylor, J. 268
Saito, M. and H. Hoji 18 Trommelen, M. and W. Zonneveld 278, 303
Saito, M. and Ν. Fukui 19
Trubetzkoy, N.S. 83fn2,116
Sato, H. 105fn2,189fn5
Scalise, S. 278, 280,293, 295fn6, 299, 304
Uldall, E.T. 327, 334, 338,353fn2
Scalise, S.: see also Bisetto and Scalise
Umegaki, M. 302
Scheler, M. 319fn3
Urbanczyk, S.C. 87, 91, 94, 97
Schultink, H. 303
Uwano, Z. 136fn3,189fn5
Selkirk, E.O. 29, 36, 39, 41fn20, 42fn30,
136fn3,141-42,147,155fnl4, 246, 263,
Vance, T.J. 105fn2
278,290-91,304,310, 357f
Selkirk, E.O. and K. Tateishi 7, 22, 41fn20 Vergnaud, J.-R.: see Halle and Vergnaud;
Shibatani, M. 205, 209, 300f Kaye, Lowenstamm and Vergnaud
Shibatani, M. and T. Kageyama 245,248-49 Verhijde, M.M. and J.M. van de Weijer
Shimamura, R. 292 42fn29
Shin, S.-H. 319fn2 Viola, M.: see Fukazawa and Viola
Shiraishi, H. 155fn5 Vogel, I.B. 299, 302, 305-308, 318
Shoji, Α.: see Beckman and Shoji Vogel, I.B.: see also Nespor and Vogel
Sibata, T. 49,119
Siegel, M.E.A.: see Roeper and Siegel Walker, R. 166
Slobin, D.I. 306 Ward, G„ R.W. Sproat and G. McKoon 255f,
Smolensky, P. 85,105fnl, 365, 366, 383fnl5 259
Smolensky, P.: see also Prince and Smolen- Weijer, J.M. van de 357, 369
sky Weijer, J.M. van de: see also Nishihara, van
Spaelti, P. 77 de Weijer and Nanjo; Verhijde and van de
Spencer, A.J. 290 Weijer
Sproat, R.W. 255 Wells, J.C. 382fn2, 382fn3
Sproat, R.W.: see also Ward, Sproat and Wenk, B.J. and F. Wioland 329
McKoon Wiese, R. 319fnl
Stockwell, R: see Minkova and Stockwell Williams, B. and S.M. Hiller 332-33, 335
Subandi 293 Williams, E.S. 270,277-78,289, 294fnl, 299,
Szymanek, B. 319fnl 302-303
Autor Index 393
Williams, E.S.: see also Di Sciullo and Will- Yoshida, S.: see also Lee and Yoshida
iams Yoshida, Y. 133,189fn5,219-22,226,228-29,
Williams, G. 196ff, 200 233,240fn5
Williams, G. and W. Brockhaus 197 Yoshida, Y. and H. Zamma 215-41
Yamada, E. 40fn5
Zamma, H.: see Yoshida and Zamma
Yamada, N. 383fnl4
Zanten, E. van and V.J. van Heuven 334
Yamane, N. 357-87, 382fn6, 383fn9, 383fnl2
Zee, D. and S. Inkelas 29
Yamane, N. and S. Tanaka 371, 381
Zee, D.: see also Inkelas and Zec
Yamane, N.: see also Tanaka and Yamane
Yip, M.J.W. 85, 99,106fn9 Zonneveld, W.: see Trommelen and Zonne-
Yokotani, T. 240fn7 veld
Yonekawa, A. 301,320fnll Zubizarreta, M.L. 36, 39
Yoshida, S. 193-214 Zwicky, A.M. 278, 304-305
Language Index