The Accent System of the Kyoto Dialect of Japanese ―A study on phrasal patterns and paradigms― Yuko Z. Yoshida & Hideki Zamma 0. Introduction This paper aims to explore the cross-categorical characteristics of pitch accent in the Kyoto dialect of Japanese (KJ). The discussion centres on the detailed study of various classes of nouns, case-marking particles, verb paradigms and phrasal interactions. Some sections will show interesting contrasts between KJ and Standard Japanese (SJ) in terms of pitch accent behaviour. To identify the contrasts between the two dialects, we will focus on pitch phenomena both in single phrases from various grammatical categories and in the merged phrases found in connected speech. First, the pitch pattern of a noun plus a case-marking particle in KJ demonstrates that there are two classes of so-called ‘accentless noun’. This discussion of accentless nouns leads directly to a phrasal analysis, in which we observe interesting accentual behaviour. We will claim that KJ utilises a pseudo accent – an accent which is phonologically assigned in a given accentless domain – which should be distinguished from a lexical accent. A further claim involving the inaccessibility of the domain-initial position lends support to the idea that there are two types of ‘accentless’ classes which are found cross-categorically. In terms of typological classification, KJ is presented here as an accent language, as distinct from a tone language. In some earlier studies, the pitch patterns of this dialect, like those of SJ, were treated as tone melodies (Haraguchi 1977), grouping the two dialects together as tone languages (which also make use of accentual information). Our proposal of the inaccessibility of the domain-initial position places KJ solely into the accent language group, having no recourse to tone melodies of any kind. 1. The Kyoto Dialect and Its Typological Grouping 1.1. Examples of Noun + Case-marking Particle As set out below, the KJ data of case-marked nouns reveal some interesting pitch properties in the dialect. The contrast shown below between KJ and SJ should highlight the properties in question. Note also a lexical variation: KJ contains lengthened words, i.e. bimoraic1 words, which in other dialects (e.g. SJ) correspond to monomoraic words (1). For example, ha ‘teeth’, ki ‘tree’ and e ‘handle’ in SJ are lengthened to haa, kii and ee, respectively. Pitch patterns in the two dialects do not necessarily correlate, i.e. the pitch patterns of some lexical items in one dialect cannot always be deduced from those in the other dialect. In the examples given below, a bar over a segment indicates that the relevant part is perceived with high pitch (other segments have default pitch) and a * denotes a lexical accent. Kyoto Dialect ‘glos.’ Standard Japanese (1) * * * * a. ha a ha a -ga ‘teeth (-nom.)’ ha ha-ga b. ki i ki i -ga ‘tree (-nom.)’ c. ee e e -ga ‘handle (-nom.)’ e e-ga (2) * * * * a. ka ki ka ki -ga ‘fence (-nom.)’ ka ki ka ki-ga * * * * b. ka ki (i)2 ka ki -ga ‘oyster (-nom.)’ ka ki ka ki –ga c. ta ne ta ne –ga ‘seed (-nom)’ d. ka ki ka ki -ga ‘persimmon (-nom.)’ ka ki ka ki -ga (3) * * * * a. ma ku ra ma ku ra -ga ‘pillow (-nom.)’ ma ku ra ma ku ra -ga * * * * b. ta ma go ta ma go -ga ‘egg (-nom.)’ ta ma go ta ma go –ga * * c. non applicable ‘head (-nom)’ a ta ma a ta ma -ga d. u sa gi u sa gi -ga ‘rabbit (-nom.)’ u sa gi u sa gi -ga e. sa ku ra sa ku ra -ga ‘cherry (-nom.)’ sa ku ra sa ku ra -ga 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 [ta], 2) the initial half of a geminate consonant [kippu], 3) the latter half of a long vowel [too] or a diphthong [dai], 4) a moraic nasal [hoN]. 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 morae, 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. The data above show that in both SJ and KJ, a noun in isolation, or a sequence 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 example, a sequence of a trimoraic word and a particle may show the pitch pattern, H00H (H stands for a high-pitched mora, and 0 denotes a mora without high pitch). That is, the particle portion dominates no independent accent domain, except for special styles of speech. Forms without any lexical accent divide into two classes in KJ, whereas SJ allows only one type of pitch pattern for accentless words. In SJ, all the morae up to the rightmost edge of the given phonological domain are high-pitched in a phrase which lacks a lexical marking. In a phrase (a phonological domain) with a lexical accent, all the morae up to the accented one appear high-pitched except for the initial mora (if the initial mora itself does not bear the lexical accent). To put this in another way, we observe a sudden drop in pitch immediately following the accented mora. In KJ, mono-moraic words are usually lengthened and there are three possible pitch patterns (see (1)) – one pattern more than we find in SJ. Four possible pitch patterns in bimoraic words include two types of ‘accentless’ words: (2c) represents a word without a lexical accent, where high pitch falls on the ultimate mora of the phonological phrase, while (2b) shows a word with high-pitch extending throughout the whole phonological phrase. Two other possibilities find the words accented either on the initial (2a), or the second, coinciding with the final position (2b) of the word. KJ lacks examples of trimoraic words with the accent on the final mora, unlike SJ, where this type is illustrated by forms such as atama’ga (3a). The lexical accent may fall on the initial (3a) or the second (3b) mora. Also for trimoraic words (see (3)), there are two types of accentless words, as explained for bimoraic words. Longer words without internal morphology do exist, though not in abundance. In fact, as we discuss later (Section 2.2.), longer words are subject to accent assignment unlike words with a lexical accent. (For convenience, the symbol for that assigned accent is indicated here by *, the same as that of the lexical accent.) Kyoto Dialect ‘glos.’ Standard Japanese (4) * * * * a. mu ra sa ki mu ra sa ki -ga ‘purple (-nom)’ mu ra sa ki mu ra sa ki -ga * * * * b. u gu i su u gu i su –ga ‘bush warbler (-nom)’ u gu i su u gu i su –ga As for quadrimoraic morphologically simplex Yamato words, there are only two patterns, both of which have an accent on the second (i.e. the antepenultimate) mora. One type allows the high pitch to be shared with morae to the left, whereas the other type does not. Based on the observations above, the next section considers how KJ should be categorised in terms of prosodic typology. 1.2. Typological Grouping of the Kyoto Dialect 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. arbitrary). Nevertheless, the distribution of morae which are not high is predictable 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 irrelevant 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. H 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 one4 or from the final mora in an accentless domain. 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 language. Our claim in this paper, that KJ only refers to information relating to accentuation and positions in the accent 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 Yamato words which are subject to pitch spreading. domain (see Section 2 for a detailed 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 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 templatic tone melodies. 2. Accents and Domains In this section, we propose a definition of pitch accent languages and show how our analysis works. First, the source of high pitch is discussed in relation to the headship of a phonological domain, and then the phenomenon formerly treated as ‘spreading’ is re-analysed as part of the phonetic interpretation of the headship in the relevant phonological domain. 2.1. Headship of a Phonological Domain and High Pitch 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 automatically identified as the domain head. In a pitch accent language, the high pitch originating from a lexical accent may be shared by neighbouring segments. 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 Kyoto words: whether or not to share the high pitch with the neighbouring segments. Compare, for example, two quadrimoraic words with antepenultimate accent, murasaki 0H00 and uguisu HH00 (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 must be identified. Also, note that not only in accented quadrimoraic words but also in lexically accentless words, the two options of pitch sharing (to share or not to share in the sense explained above) are available. Neither (3d) nor (3e) above bear a lexical accent; furthermore, the former is of the type where only the word- (or phrase-) final mora is high-pitched, whereas pitch sharing is observed in the latter. An accentless domain, regardless of its morphological organization, always possesses a high-pitched portion. In other words, the accentless domain should have a head portion which behaves like a lexical accent to ‘provide’ the high pitch. In this regard, Kaye (1990) proposes the Licensing Principle, which applies to all phonological domains, including pitch accent domains. (5) The Licensing Principle (Kaye 1990: p. 306, ll. 11-13) All phonological positions save one must be licensed within a domain. The unlicensed position is the head of this domain. 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): (6) Note: [ ] indicate the phonological domain relevant for pitch a. i) [ki i] ii) [ki i –ga] ‘tree (-nom.)’ b. i) [a na] ii) [a na –ga] ‘hole (-nom.)’ c. i) [u sa gi] ii) [u sa gi –ga] ‘rabbit (-nom.)’ Only the rightmost mora is high-pitched, which assumes the headship of the domain. The other type of accentless word, which is high-pitched throughout the domain, as in sakura-(ga) HHH(-H), should follow the same principles for selecting its head. The rightmost mora is the head of the domain, where high-pitch is shared by all the other morae,5 if the noun is lexically determined as such. Our task here is to identify the lexical cue for the two classes of accentless words, which will be discussed in detail later (see 3.2.1.). The next subsection considers the default lexical accent location in KJ, which 5 In SJ too, the rightmost mora serves as the head of the accentless domain (Yoshida 1999). supports our classification of the dialect as an accent language. A recent change in KJ, whereby some features of stress accent languages are becoming apparent, assigns accent on to predictable locations. 2.2. Antepenultimate Accent The following general tendency in present-day KJ affects quadrimoraic words. The accent location in quadrimoraic words, both verbs and nouns, is changing— word-initial accents are being replaced by accents on the antepenultimate mora. Accents at fixed locations suggest that KJ belongs to the group of accent languages which perhaps conform to metrical accent assignment, as in the case of SJ, in which an accent is assigned on the antepenultimate mora (Yoshida 1995). Quadrimoraic nouns including morphologically simplex/complex nouns and many proper nouns which were once accented on the initial mora are now accented on the second mora. (7) a. ‘nightingale’ b. ‘Kuwatani (a surname)’ c. ‘Kiyomizu (a place name)’ * * * u gu i su ku wa ta ni ki yo mi zu ↓ ↓ ↓ * * * * * * 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 assignment process. A close observation of the accentual distribution of trimoraic words exemplified in (3) above clearly reveals that the default lexical accent location is the antepenultimate mora. Sampling 115 morphologically simplex yamato (Japanese native) nouns from the word list in Hirayama (1957), we report the following distribution. (8) Example number initial accent (3a) * i ta ti ‘weasel’ 42 medial accent (3b) * ha ta ke ‘farm’ 20 final accent (3c) n.a 0 rising accentless (3d) su zu me ‘sparrow’ 12 high accentless (3e) ka su mi ‘haze’ 41 If lexically accented, 42 out of 62 accented trimoraic yamato words bear word-initial (i.e. antepenultimate) accent. Trimoraic yamato words with word-final accent are not attested in the dialect; in fact, this dialect avoids placing lexical accents on the final mora of the word. Observation of bimoraic words confirms that a word-final accent is not favoured: those final accented nouns are subject to lengthening (see footnote 2). The fact that about 68% of all accented words bear word-initial (antepenultimate) accent proves that the default location of the lexical accent in KJ is the antepenultimate mora. The change in accent location noted at the beginning of this section for quadrimoraic nouns is cross-categorical. A change observed in the Kyoto verbs also involves a shift in preference towards antepenultimate accent. These old forms are taken from Hirayama (1957), while the new forms are based on our own recordings of native speakers. (9) OLD FORM NEW FORM * * a. tu ku t ta ‘make-past’ tu ku t ta * * b. u ka N da ‘float-past’ u ka N da * * c. na ra N da ‘queue-past’ na ra N da * * d. si zu N da ‘sink-past’ si zu N 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 dialect shows behaviour similar to that seen in stress languages: the accent falls at a predictable location, the antepenultimate mora. The type of phonetic 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). In the quadrimoraic nouns (7a) and verbs given above, there are two possibilities; either, pitch is shared between the initial mora and the accented 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, bimoraic6 and trimoraic 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 (HH00). In order to identify which factors determine whether or not pitch sharing occurs in a given domain, we turn to the case of accentless nouns, which allow both possibilities. 3. The Special Status of Domain Edge 3.1. Domain Final Position 3.1.1. Domain-final Mora and ‘Accentuation’ In 2.1, we explained that the domain-final mora serves as the head of a domain in which no inherent accent marking exists. As discussed above, there are two types of domain: only one type bears high pitch on the head mora (10a,c,e), while the other type shares the pitch with all the other morae in the domain (10b,d,f). (10) a. ki i ki i -ga ‘tree (-nom.)’ b. ee 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 other does not, then a clear difference between them is immediately apparent. These domains do not contain any 6 Here the term ‘bimoraic’ also refers to lengthened monomoraic forms such as kii ‘tree’ (ki in SJ). information concerning the location 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 being placed next to another phrase, e.g. a verb. (11) Standard Japanese * * a. wa sa bi wa sa bi -o ‘horseradish (acc.)’ * * b. a zu ki a zu ki -o ‘red bean (acc.)’ c. sa ka na sa ka na -o ‘fish (acc.)’ * (*) a’. wa sa bi –o ta be ta ‘(I) ate some horseradish’ * (*) b’. a zu ki –o ta be ta ‘(I) ate red beans’ * c’. sa ka na –o ta be ta ‘(I) ate some fish’ In connected speech, as (11a’-c’) above show, only if the noun plus a particle phrase contains a lexical accent will the accent of the verb not be perceived (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 ne –o ta be ta * ki tu ne –o ta be ta * ki tu ne –o ta be ta (*) * * c’. sa ka na –o ta be ta * sa ka na –o ta be ta * sa ka na –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 domain-final position (Section 2.1). The domains do not have lexical accents (the lexical accent of a noun should remain intact following its concatenation to a nominative marker), but they do show ‘accentuation’ in the domain-final position (the head position). Here we propose the following for KJ: (13) Proposal 1 Nouns which belong to the classes (1b,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. Referring to the SJ accentless word, in which the domain-final nucleus (the head portion of the final mora) acts as the domain-head, we see that the head receives the high-pitch interpretation but no accentuation. In contrast to the SJ case, KJ interprets the headship in the domain-final nucleus as accentuation, which naturally accompanies a high-pitch. The question arises as to whether this interpretation as accentuation is controlled via distinct dialectal parameters or whether it results from some other condition. In both dialects it is true that, when the noun-phrase constitutes a domain in its own right, headship is assigned. It is only when the juxtaposition of a noun phrase and a verb occurs, that the accentual difference between the two dialects comes to light. Let us therefore compare the behaviour of the two dialects in this respect. If the noun phrase in KJ acquires a pseudo accent, as proposed above, then the noun phrase should occupy its own domain [[noun phrase] verb] or [[noun phrase][verb]] rather than merging with the verb [noun phrase – verb]. Let us consider this possible concatenation in more detail. The difference in phrasal structure between SJ and KJ arises from syntactic relations. By ‘syntactic relations’ we refer here to the way in which phonological processes should apply to the phonological string – a phonological domain which can be a noun, a noun phrase, a longer phrase or a sentence. A domain must have a head (Kaye 1995), and the inherently accented mora automatically serves as the head in a lexically accented 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 concatenated, the leftmost accent of the noun-verb string is projected to the domain of AB as the accent of the entire phrase. (14) a. [[A][B]] domains: A, B, AB b. [[A] B ] domains: A, AB c. [A B ] domain: AB What follows is a general formula describing SJ accent assignment in phrases and compounds: (15) [C[A][B]C] In SJ, where both A and B contain lexical marking, i) if C is a lexical category, B 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) Of relevance here is (ii) in (15), when the category is phrasal. Below in (16) are examples repeated from (11) with bracketing to show the generalization in (15). (16) * (*) a. [[wa sa bi –o] [ta be ta]] ‘(I) ate some horseradish’ * (*) b. [[a zu ki –o] [ ta be ta]] ‘(I) ate red beans’ * c. [[sa ka na –o] [ta be ta]] ‘(I) ate some fish’ 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 domain. 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. Referring back to (12a’-c’), to compare with the situation in SJ (16a-c), we see that even the lexically accentless domains of noun phrases (11b’, c’) trigger phrasal downstep, which means that the domains of noun phrases behave as if they are accented (pseudo accent (13)). (17b) shows the instruction for assigning pitch to noun-verb sequences in KJ. The phonological parsing of the noun-verb domain supercedes that of the verb domain, where the verb does not constitute its own domain. Note, as the asterisked bracketing illustrates, if both the noun and the verb dominate one domain each, as in the SJ case, then the lexical accent of either domain should be projected in connected speech. The only domain with a lexical accent to be projected to the noun-verb domain is that of the verb. This bracketing would predict that high pitch occurs only on the accented mora of the verb, which is unattested. (17) * (*) a. [[wa sa bi –o ] ta be ta ] (*) * * b. [[ki tu ne –o ] ta be ta ] * [[ki tu ne –o] [ta be ta]] *[[ki tu ne –o] [ta be ta]] (*) * * c. [[sa ka na –o ] ta be ta ] * [[sa ka na –o] [ta be ta]] *[[sa ka na –o] [ta be ta]] When the noun lacks the overt case-marker, the bracketing remains the same. (18) * (*) 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] atructure, 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 sakana (18b,c) – unlike the inner domain of wasabi in (18a). Notwithstanding, 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 sequence acquires an accent when followed by another case-marker7. The location of the accent is indicated by the drop in pitch on the mora adjacent 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 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. Genitival –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. noun-phrase is followed by an accented verb. (19) SJ (Noun –Particle 1 –Particle 2) a. b. c. sa ku ra sa ku ra –ni sa ku ra –ni –mo ‘cherry tree’ ‘cherry tree –DAT.’ ‘cherry tree –DAT. –also d. * e. (*) sa ku ra –ni tu ke ru sa ku ra –ni –mo tu ke ru ‘to attach (something) to a cherry tree’ ‘to attach (something) also to a cherry tree’ Similarly in Kyoto, in the same sequence of particles an accentuation appears immediately in front of the second particle: (20) Kyoto (Noun-Particle 1-Particle 2) i) a. b. c. ki tu ne ki tu ne –ni ki tu ne –ni –mo ‘fox’ ‘fox –DAT.’ ‘fox –DAT. –also ii) a. b. c. sa ku ra sa ku ra –ni sa ku ra –ni –mo ‘cherry tree’ ‘cherry tree –DAT.’ ‘cherry tree –DAT. –also However, the accentual behaviour of KJ in the sequence consisting of a case-marked noun and a verb shows a departure from the SJ pattern. In SJ, the phrase-final mora does not acquire an accent (11c’), whereas an accent appears on the final mora of the noun-particle sequence in Kyoto (12b’,c’). 3.1.2. Evidence for Pseudo Accent and Domain Organisation In support of the acquired accent that we will label, ‘pseudo accent’, the following evidence may be cited. There exists a phenomenon in SJ which is sensitive to the domain-final accent. If, in KJ, we observe a similar kind of phenomenon that reacts to a phrase-final pseudo accent, then we may confidently claim the existence of such an accent. The Genitival marker –no has the effect of deleting a noun’s accent when in juxtaposition. Precisely, when the particle follows a noun with word-final accent, the accent of the noun is not perceived. Yoshida (1995) analyses the Genitival marker as a lexically accented particle (see also footnote 7) that induces a reaction – Accent Clash Avoidance – to eliminate adjacent accents. For example, the accent of azuki in SJ falls on the final mora; when –no is suffixed to the noun, however, the accent of the noun is not perceived (21b). If the accent falls elsewhere, word-initially or medially, the accent of the noun remains (21a). Note that the behaviour of –no is unique in comparison to other case markers such as –ga (nom.), which always respect the lexical accent of the nouns. (21) Replies to the question: ‘Which flavour did you choose?’ * * (*) a. wa sa bi [[wa sa bi] –no] ‘horseradish (-Gen.)’ * (*) * b. a zu ki [[a zu ki] –no] ‘red bean (-Gen.)’ * c. sa ka na [[sa ka na] –no] ‘fish (-Gen.)’ Accent Clash Avoidance is only relevant in this [[A]B] domain structure. In SJ, the syntactic relation between a noun and a particle is as illustrated in (21), that is, [[noun] particle], where by default the lexical property of the noun is projected to that of the noun-particle sequence. Compare this to those cases where a verb with initial accent follows a word-finally accented noun without any overt case-marker: the two accents are adjacent. (22) Accent Clash Avoidance – only in noun-particle sequences in SJ [[A] B] – not in phrasal sequences [[A][B]] * (*) a. [[wa sa bi] [ta be ta]] ‘(I) ate some horseradish.’ * (*) b. [[a zu ki] [ta be ta]] ‘(I) ate red beans.’ * c. [[sa ka na] [ ta be ta]] ‘(I) ate some fish.’ As in (22b), this avoidance phenomenon does not apply to the structure [[A][B]]. Also note that in SJ, the domain-final mora only bears the headship of the lexical-accentless domain (Yoshida 1995,1999), and does not result in an actual accent. In SJ, headship in the domain-final position only serves to support high pitch, so the segments to the right of that mora (in phrasal sequences, etc.) are not subject to any loss of high pitch; the sudden drop in pitch is only observed in the post-accented position. In fact, Accent Clash Avoidance in KJ offers clear evidence for the suggested domain structure and for the proposed pseudo accent. In this dialect, 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 KJ, 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. (23) Pseudo Accent: x * (*) a. [[wa sa bi] mo ro ta] (x) * b. [[ki tu ne] mo ro ta] Accent Clash Avoidance --> delete x (x) * * c. [[sa ka na] mo ro ta] * sa ka na mo ro ta Accent Clash Avoidance --> merge x & * 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 second mora of the verb, and is not adjacent to the pseudo accent in the noun portion. (24) * (*) a. [[wa sa bi] ta be ta ] (*) * b. [[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]] As seen above, the domain-final pseudo accent in KJ reacts when placed next to another accent. This reaction supports the claim that, in KJ, there is an acquired accent in the domain final position, which we call here a pseudo accent. 3.2. Domain-initial Position 3.2.1. Inaccessibility of Domain-initial Position 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 accentless 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 question 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 ― a. * * 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 pronounced 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 result emerges: there are two alternative pitch patterns (see (24a,a’)). Note that in SJ, the initial nucleus (the head of the initial mora) is inaccessible 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 inaccessibility. However, KJ chooses to treat this as lexical information, so the sharing of high-pitch comes about as a result of referring to lexical properties. (27) PROPOSAL 2 In the Kyoto dialect, the domain-initial mora may be lexically inaccessible for high-pitch. In KJ, we do not observe high-pitch sharing if the initial mora is lexically marked as being inaccessible8. 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 contains 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 domain 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 inaccessibility, along with accent assignment for quadrimoraic words, or along with the assignment of pseudo accent for lexically accentless trimoraic words. This correlate, however, requires more detailed consideration, and is therefore not pursued further in this discussion. (28) * * * a. u gu i su ku wa ta ni ki yo mi zu ‘nightingale’ ‘Kuwatani (a surname) ‘Kiyomizu (a place mname)’ * * * a’. u gu i su ku wa ta ni ki yo mi zu ‘nightingale’ ‘Kuwatani (a surname) ‘Kiyomizu (a place mname)’ b. x x to ri ‘bird’ sa ku ra ‘cherry’ c. x x a na ‘hole’ u sa gi ‘rabbit’ The variations found in (28a, a’) may be attributed to idiolectal differences – specifically, whether the domain-initial mora is deemed inaccessible (28a) or not (28a’). The words without a lexical accent may contain this lexical marking to indicate the inaccessibility, viz. whether to allow pitch sharing or not. The unmarked case is observed in tori or sakura (28b), where high-pitch sharing occurs. If marked (28c), no pitch sharing is observed, leaving only the final position of the word (domain) high-pitched. If the distribution of the two wprd types in (28b) and (28c) are compared, referring to the table in (8), the sakura type of occurs three times more frequently than the usagi type. If this analysis -- that is, an analysis which utilizes initial inaccessibility to account for the L-initial pitch pattern -- is on the right track, then we can adequately account for the system of accentuation in KJ, in that we can attribute the difference between the two unaccented pitch patterns to the presence or the absence of initial inaccessibility, rather than to the basic tone melodies, HL or LHL, as has been done in previous studies such as Haraguchi (1977). This is a desirable result for the analysis of the Japanese accentual system, because it conforms to the widely-held view of Japanese as an accent language, as we argue in Section 1.2. Moreover, this analysis can account for the data relating to verbal inflection and to noun accentuation in KJ. In what follows, we will discuss these two issues, to show the suitability of the analysis. 3.2.2. Evidence for Initial Inaccessibility: Accentuation in Verbal Inflection To outline the accentuation system of verbal inflection in KJ, let us first compare it with the one in SJ, which is well-studied in the literature (cf. Bloch (1946), McCawley (1968), Haraguchi (1977, 1991), Poser (1984), etc.): (29) Standard Japanese a. Accentedness (i.e. accented or unaccented) is determined by the stem: * * tabe-ru ‘eat (indicative)’ tabe-ta ‘eat (past)’ asob-u ‘play (indicative)’ ason-da ‘play (past)’9 b. There is only one pitch pattern for the accented verb; in Haraguchi’s (1977) terms, it is the HL melody. As shown in (29a), the lexically accented verb stem tabe- is always accented, 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: (30) Kyoto Japanese a. Accentedness is determined by the suffix. b. There are two pitch patterns; i.e. HL and LHL melodies. Recall that the accentedness of verbs in SJ is determined by the stem, rather than by the suffix. In addition, while SJ has only one pitch pattern, KJ has two. Let us exemplify these properties using concrete examples: (31) a. * agar-u ‘rise (indicative)’ agat-ta ‘rise (past)’ * ake-ru ‘open (tr.) (indicative)’ ake-ta ‘open (tr.) (past)’ b. * 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. aruk-u ‘walk (indicative)’ a ru i-ta ‘walk (past)’ * iki-ru ‘live (indicative)’ i ki-ta ‘live (past)’ 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. Similarly, 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. 126.96.36.199 Unaccented Forms Having clarified above that the accentedness of the verb is regulated by the suffix attached to it, we now consider what kind of behaviour each suffix shows. First, let us take the suffixes which make the form unaccented; that is, indicative forms, polite forms, imperative forms, and hortative forms. Examples of each are given below. (Henceforth, the examples are represented in the following specific order: 2-mora verbs and 3-mora verbs are those verbs whose indicative forms consist of 2 morae and 3 morae respectively, and C-final verbs and V-final verbs are those whose stems end with a consonant and a vowel respectively. The examples of group (a) begin with a high-pitched mora, and those of group (b) with a pitchless mora. It is important to note that the examples given in this section are merely representative forms, and thus all the verbs which fall into the given category show similar tonal behaviour.) 2-mora verbs 3-mora verbs C-final V-final C-final V-final (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-i-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 ake11 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 accent 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 unaccented suffix, then the mora is also high-pitched/pitchless with other unaccented suffixes. The various forms of the stem aw- ‘meet’, for example, 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 neutralized in all forms of C-final 2-mora verbs, except the indicative. Note 10 As is well-known, stem-final /w/ in Japanese only appears before /a/ and is deleted elsewhere. The segment is represented in the examples only to show that the words contain /w/ underlyingly. 11 The allomorph of the imperative suffix for V-final stems is ø (zero). The /i/ 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)). 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. 188.8.131.52. Accented Forms Next, let us turn to the other group of suffixes, i.e. those which place the accent on the stem. The following forms show the accent on the mora immediately preceding the suffix. (Note that there are several negative forms in KJ.) (36) negative (standardized) a. * * * * iw-a-nai ki-nai agar-a-nai ake-nai b. * * * * aw-a-nai mi-nai aruk-a-nai i ki-nai (37) negative (archaic) a. * * * * iw-a-zu(-ni) ki-zu(-ni) agar-a-zu(-ni) ake-zu(-ni) b. * * * * aw-a-zu(-ni) mi-zu(-ni) aruk-a-zu(-ni) i ki-zu(-ni) 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 pitchless mora in all the forms. It is clear that the suffixes determine accentuation, while the pitch pattern is determined by the stem. 184.108.40.206. Problematic Forms and an Analysis So far we have considered regular forms -- regular, in the sense that both accentuation and pitch patterning are consistent throughout the forms. There are, however, several forms in which the placement of the accent is irregular. First, consider the provisional forms: (38) provisional a. * * * * iw -e-ba ki-re-ba agar-e-ba ake-re-ba b. * * * * 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, except in aw-e’-ba, where the accent falls on the mora immediately preceding the suffix. Note that the forms in Section 220.127.116.11. all have the accent on a designated mora – specifically, the mora immediately preceding the suffix. The provisional forms are irregular in this respect. The past forms show similar behaviour: (39) past a. * * * * i t-ta ki-ta a gat-ta a ke-ta b. * * * 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 H-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 discussed in Section 2.2. (see (9)).) Finally, consider the colloquial negative forms: (40) negative (colloquial) a. * * * * iw-a-hen ki i-hin12 aga r-a-hen a ke-hen b. * * * * aw-a-hen mi i-hin a ruk-a-hen i ki-hin In this form, the accent falls on the second mora preceding the suffix, except in aw-a’-hen and iki’-hin, in which the accent falls, once again, on the mora immediately preceding the suffix. In all of the forms seen above, there are several word classes in which accent placement behaves in an exceptional manner. The problem to be addressed, then, is why there are such exceptions in these forms. Recall that, as we have seen in the preceding section, accent location in a verb is the property of the suffix. This being the case, why 12 The vowel of this suffix has harmonized to the one in the stem. Moreover, the stem vowel in this form has lengthened, probably to satisfy the binarity (cf. Itô and Mester (1992)) required by the suffix -hen. do the suffixes above not consistently place the accent on a designated mora? If we assume the analysis presented here, the problem above can be explained in a simple way. Specifically, it becomes possible under such an assumption to argue that the exceptions emerge because of initial inaccessibility marked on the stems. The explanation proceeds as follows. As we proposed above, the first mora of the group (b) words in the examples is marked as inaccessible. If it is the case that this marking must be respected, the accent which is required by a suffix on the first mora must move elsewhere. If no mora is available on the left side, as is the case in 3-mora words, then the accent moves rightwards. See the following diagrams: (41) a. lexical accent assignment * pitch interpretation * a-e-ba a-e-ba a-e-ba by -ba b. * a-e-ba In (41a), the usual assignment of accent by the suffix -ba is carried out, according 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 (41b). Note that this exceptional behaviour is observed only when both of the following conditions are met: (i) the stem exhibits the 0H0 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 18.104.22.168. Accentuation of Nouns Another piece of evidence comes from the accentuation of nouns containing 13 In the case of oo-ta, it is accent deletion which is carried out in order to preserve the lexical 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 Fn.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 0H0 for tat-ta ‘stand (past)’, for example, accent deletion may take place. Non-geminate-ending stems may be aligned to this pattern. three morae. Note that Hirayama (1957) lists only the following two types of accented 3-mora noun: (42) * 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’ It will be recalled from Section 2.2 that there is a clear tendency in KJ to place the accent on the antepenultimate mora in nouns. The pattern in (42a) conforms to this generalization. The one in (42b), however, does not: the accent falls on the penultimate mora. Do we have to mark all the forms with this pattern as exceptions? If we assume the analysis presented above, this pattern is not at all problematic. The forms that follow the (42b) pattern are regarded as being lexically marked for initial inaccessibility. Although the accent should be put on the antepenultimate mora, it cannot be because of this marking: note that the antepenultimate mora is the initial one in this case. Thus, the accent is carried down to the second mora, which is penultimate. Moreover, the fact that there is no high-pitch-initial noun with the accent on the penultimate mora (i.e. HH0) supports this analysis. If the penultimate accent in (42b) is assigned as it is, it is also predicted that the penultimate accent occurs in nouns with an initial high-pitch-initial mora, which is not the case. Thus, the natural conclusion is that the penultimate accenting of nouns of the (42b) type is not specifically intended, but that the accent is located there to preserve the lexical marking on the stem, thereby violating the general pattern. 4. Conclusion In this paper, we have developed a way of analysing the pitch assignment system in KJ so that it conforms to the typology of accent languages. This is achieved by means of the pseudo accent and the notion of inaccessibility, both of which are attributable to the status of domain-edges. This analysis is appealing for several reasons. First, the pseudo accent can account for the phrasal accent phenomenon, which is unique to KJ. Second, the inaccessibility of the domain-initial position can account for the exceptional behaviour of the verbal paradigm. Finally, these devices can account for many phenomena in the dialect without resorting to any underlying low pitch, which finds little empirical support in pitch accent languages. Acknowledgements We would like to express our gratitude to Phillip Backley, Jeroen van de Weijer and Tetsuo Nishihara, 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, Koichi Tateishi, Shin-ichi Tanaka, Noriko Yamane, and Teruo Yokotani. This paper is dedicated to Jion. 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. Kaye, Jonathan (1995) Derivations and Interfaces, in Frontiers in Phonology: Atoms, Structures, Derivations, 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 from Holland Academic Graphics.] Yoshida, Yuko (1999) Binary Inter-nuclear Licensing for Accent Assignment in Relation to Phonological Domains, in Phonologica 1996: Syllables, ed. by John Rennison. Holland Academic Graphics. The Hague.