1.1 Syllables

Most linguists would agree that syllables seem to be basic units of speech production and perception (Abercrombie Reference Abercrombie1967: 37; Lieberman Reference Lieberman1977: 120–121), but syllables are notoriously difficult to define either articulatorily or auditorily (Laver Reference Laver1994: 113–114; Rogers Reference Rogers2000: 267–268; Zec Reference Zec and de Lacy2007: 161). In prototypical cases, the number of syllables in an utterance matches the number of peaks of sonority (Goldsmith Reference Goldsmith and Goldsmith2011: 194), where sonority is taken to be a scale of intrinsic audibility on which individual segments can be ranked (Laver Reference Laver1994: 503–505; Parker Reference Parker2008). Syllables also seem to be intuitively natural units for ordinary native speakers of most languages, at least in the sense that there is general agreement on how many syllables a word contains, although not necessarily on where the boundary is between one syllable and the next (Clark and Yallop Reference Clark and Yallop1990: 97; Steriade Reference Steriade and Frawley2003: 193; Duanmu Reference Duanmu2009: 1–2).

There is also a consensus that phonotactic constraints apply to syllables (Pike Reference Pike1947: 180–181; Fudge Reference Fudge1969: 254; Zec Reference Zec and de Lacy2007: 162). For the most part, any string of phonotactically admissible syllables in a language is a phonotactically admissible word in that language, although not all phonotactic generalizations are syllable based (Zec Reference Zec and de Lacy2007: 192).

Most researchers today who would identify themselves as phonologists analyze Japanese (i.e. modern Tokyo “standard” Japanese) as having both moras and syllables: one-mora light syllables (short syllables), two-mora heavy syllables (long syllables), and even three-mora superheavy syllables (extra-long syllables), although this last category is marginal. There is, however, no colloquial Japanese word that denotes a Japanese syllable in this sense, and ordinary native speakers know how to count moras but not how to count syllables (Vance Reference Vance2008: 115–116; Labrune Reference Labrune2012b: 116).

1.2 Moras

A mora is understood as a unit of syllable weight (i.e. quantity) in languages that distinguish between light (i.e. short) and heavy (i.e. long) syllables (Trubetzkoy Reference Trubetzkoy and Baltaxe1969 [1939]: 173–181; Davis Reference Davis and Goldsmith2011: 103–108). Japanese moras are intuitively isochronous units of rhythm for native speakers, and scholars familiar with traditional Japanese language research often use the musical term haku ‘beat.’ Thus, in a well-known categorization of languages into rhythmic types (Pike Reference Pike1943: 35; Abercrombie Reference Abercrombie1967: 96–98), Japanese is the textbook example of a mora-timed language, in contrast to syllable-timed languages such as Spanish and stress-timed languages such as English (Ladefoged Reference Ladefoged1982: 226; Rogers Reference Rogers2000: 270–271; Kubozono and Honma Reference Kubozono and Honma2002: 20–21).

Languages do not, however, exhibit a straightforward relationship between syllable weight and duration (Davis: Reference Davis and Goldsmith2011: 132–133), and experimental work on Japanese has not been able to demonstrate even approximate isochrony of moras. Of course, no linguist would maintain that every mora has precisely the same duration, even at a fixed tempo. Phonetic research on mora timing in Japanese has focused on trying to show that compensation effects make average mora durations more nearly equal than would be expected from inherent segment durations. A thorough review of this research concludes that the evidence for such compensation is not persuasive (Warner and Arai Reference Warner and Arai2001), and it has even been suggested that moraic isochrony in Japanese is merely an illusion caused by the relative proportion of vowels and consonants in utterances (Ramus, Nespor, and Mehler Reference Ramus, Nespor and Mehler1999).

The remainder of this chapter assesses Japanese moras and syllables as psychological units. The arguments that researchers offer involve language-internal phenomena and behavior on psycholinguistic tasks. Moraic isochrony is not at issue.

1.3 Subsyllabic Constituents

The standard terminology for the parts of syllables presupposes that the boundaries between syllables can be determined, although it can accommodate ambisyllabicity, that is, a segment that is both the end of one syllable and at the beginning of the next. The nucleus of a syllable is the portion with the highest sonority (Zec Reference Zec and de Lacy2007: 163) and is prototypically a vowel, although many languages allow at least some consonants to function as nuclei (i.e. syllabic consonants). A syllable must have a nucleus, and some languages allow syllables with only a nucleus, as in the monosyllabic English word /ɔ/ awe. The non-nuclear segments in a syllable make up the margins. Consonants in the margin preceding the nucleus are the onset, and those in the margin following the nucleus are the coda. In the monosyllabic English word /plænt/ plant, for example, the nucleus is /æ/, the onset is /pl/, and the coda is /nt/. A syllable with a coda is closed, and a syllable without a coda (e.g. /pli/ plea) is open.

It is often claimed that in English and many other languages, the nucleus and the coda in a closed syllable form a unit and that syllables therefore have the hierarchical structure in Figure 7.1 (Fudge Reference Fudge1969: 268; Blevins Reference Blevins and Goldsmith1995: 212–216; Treiman and Kessler Reference Treiman and Kessler1995). (The symbol σ stands for a syllable.) The unit containing the nucleus and the coda (if there is a coda) is called the rhyme.

Figure 7.1 Hierarchical structure: onset + rhyme

Two main types of evidence are usually cited for the rhyme. First, English shows a strong preference for dividing between the onset and the rhyme in speech errors (Fromkin Reference Fromkin1971: 32; Kubozono Reference Kubozono1989: 266), as in /ren trɛk/ rain trek for intended /tr^en r^ɛk/ train wreck, in intentional blends, as in /brʌnč/ [brʌ̃ntʃ] brunch from /br^ɛkfəst/+/l^ʌnč/ breakfast+lunch, and in language games (Haraguchi Reference Haraguchi and van de Weijer2003: 49), as in Pig-Latin /ɪtsple/ it-splay for /spl^ɪt/ split. (The symbol ^ marks onset-rhyme division points in the source forms.) Not all phonologists regard such phenomena as persuasive evidence for the rhyme as a constituent (Davis Reference Davis1989; Goldsmith Reference Goldsmith and Goldsmith2011: 172).

The other type of evidence involves phonotactics, specifically, the claim that phonotactic restrictions are much more stringent between the nucleus and the coda than between the onset and the nucleus (Fudge Reference Fudge1969: 272–283; Haraguchi Reference Haraguchi and van de Weijer2003: 48; Duanmu Reference Duanmu2009: 40). For example, /a͜u/ is an admissible English nucleus, as in /ka͜unt/ count, and /mp/ is an admissible coda, as in /kæmp/ camp, but this nucleus and this coda cannot co-occur (Haraguchi Reference Haraguchi and van de Weijer2003: 48): */ka͜ump/. (A preceding asterisk indicates that a form is deviant.) Nonetheless, it is not true in every language that an admissible onset followed by an admissible rhyme is always an admissible syllable, and some phonologists therefore find the phonotactic evidence for the rhyme unconvincing (Zec Reference Zec and de Lacy2007: 177).

In languages with weight distinctions, the fact that onsets are never relevant to syllable weight (Hyman Reference Hyman1985: 6; Davis Reference Davis and Goldsmith2011: 117) has been used to motivate structures like the one in Figure 7.2, with the onset attached directly to syllable node and weight represented straightforwardly by the number of mora nodes (Zec Reference Zec and de Lacy2007: 176; Duanmu Reference Duanmu2009: 8). (The symbol µ stands for a mora.) The first mora in a heavy syllable is often called the head mora, and it typically has a higher sonority threshold than the non-head mora (Zec Reference Zec and de Lacy2007: 183), that is, the class of segments that can fill the head mora position is limited to a higher portion of the sonority scale.

Figure 7.2 Hierarchical structure with moraic nodes representing weight

4.1 Vowel Length and Sequences of Identical Vowels

As noted in Section 2, there is a phonetic distinction in Japanese, at least in careful pronunciation, between a long vowel and a sequence of two identical short vowels, as the words /ǰiR/ jī ‘G’ ([dʑiː]) and /ǰi+i/ ji-i ‘intention to resign’ ([dʑiˇi]) illustrate. Intuitively, the second mora of /ǰiR/ behaves like a special (i.e. dependent) mora, but the second mora of /ǰi+i/ behaves like an ordinary (i.e. independent) mora, and traditional accounts in Japan treat most or all V moras as ordinary moras. For example, in the commentaries included in successive editions of the pronunciation dictionary issued by Japan’s public broadcasting corporation (NHK), Kindaichi (Reference Kindaichi and Kyōkai1966: 17–18, Reference Kindaichi and Kyōkai1985: 20–21, Reference Kindaichi and Kenkūyjo1998: 105–106) consistently recognizes only three special moras: /N/, /Q/, and /R/.

For Kindaichi (Reference Kindaichi and Kyōkai1966: 10), a special mora is just a mora consisting of one of the three “special phonemes” (tokushu-onso), that is, /N/, /Q/, or /R/. As he himself pointed out many years ago (Kindaichi Reference Kindaichi1950a), the phonetic distinction between a long vowel and two identical short vowels in sequence is a problem for any phonemic analysis that treats long vowels as double short vowels, and he saw this as a decisive argument in favor of the vowel-length phoneme /R/. A syllable advocate could argue that /ǰi+i/ has two light syllables and /ǰiR/ has one heavy syllable, allowing the difference to be represented as /ǰi_σi/ versus /ǰi_μi/ (Vance Reference Vance2008: 60–61), but this analysis violates a widely accepted principle, namely, that “lexical items do not contrast minimally in their syllabic divisions” (Steriade Reference Steriade and Frawley2003: 190).Footnote ⁸ Since sequences of two identical short vowels almost always straddle a morpheme boundary (see Section 2), the difference between [dʑiː] and [dʑiˇi] could be attributed to the difference between /ǰii/ and /ǰi+i/, with a syllable boundary corresponding to the morpheme boundary in the latter, but this solution will not work for /honoo/ ‘flame’ pronounced [honoˇo] (see Section 2). In any case, not all phonologists agree that lexical items cannot differ minimally in syllabification (Duanmu Reference Duanmu2009: 59).

4.2 V₁V₂ Sequences Ending in a High Vowel

V₁V₂ sequences are problematic when V₂ is a high vowel (i.e. /i/ or /u/). Most phonologists who ascribe light and heavy syllables to Japanese recognize a fourth type of special mora in addition to /N/, /Q/, and /R/, namely, “the second half of diphthongal vowel sequences” (Kubozono Reference Kubozono and Tsujimura1999: 32). Most such sequences end in /i/, but some end in /u/. A diphthong can be described as a change in vowel quality within a syllable that is noticeable enough to make it sound as if there is a sequence of two phonetically distinct vowels. Depending on the language, a diphthong like [a͜i] can be a single phoneme (/a͜i/) or a sequence of two phonemes (/ai/ or /ay/), just as a stop+fricative sequence like [ts] can be an affricate (one phoneme) in some languages but a cluster (two phonemes) in other languages (Hyman Reference Hyman1985: 2; Duanmu Reference Duanmu2009: 6–7). Nonetheless, there is a tendency to misinterpret the term diphthong as implying a single-phoneme analysis, and Kubozono’s phrase diphthongal vowel sequence is presumably an attempt to avoid such misunderstanding. There is no question that all Japanese diphthongs should be analyzed phonemically as sequences of two vowels, and the more compact term quasi-diphthong will be used in the remainder of this chapter to denote the Japanese two-vowel sequences in question.Footnote ⁹

In a mora-plus-syllable analysis, like model B in Figure 7.4, it is often hard to tell whether a V/i/ or V/u/ sequence is a quasi-diphthong because it is hard to tell whether the two vowels are in the same syllable. In an analysis with no distinction between syllables and moras, like model A in Figure 7.3, a quasi-diphthong can be defined as a V/i/ or V/u/ sequence in which the second vowel behaves like a dependent mora, and the corresponding challenge is deciding whether there is such dependence. If all onsetless V moras were dependent, then all V₁V₂ sequences would be quasi-diphthongs. The remainder of this section is devoted to arguing that there is a distinction between dependent and ordinary V moras, although some instances of /i/ in V/i/ and /u/ in V/u/ are not readily categorizable.

The evidence for quasi-diphthong status involves accent patterns, so a brief introduction to the (modern Tokyo) Japanese pitch-accent system is necessary here (see also Chapter 8, Section 2.1, this volume). Some words are accented and others are unaccented, and the intonation pattern on a phrase is determined in part by whether or not the word(s) it contains are accented and, if so, by the location of the accent(s). An accent is manifested by a steep fall from a relatively high pitch to a relatively low pitch. The smallest intonational units in Japanese are accent phrases, and an accent phrase has at most one accentual fall. Most of the examples cited in this chapter consist of a content word in isolation or followed by a particle, and these are all pronounced as accent phrases. An unaccented accent phrase has no steep pitch fall.

In the remainder of this chapter, the location of an accent is marked in phonemic transcription by a downward-pointing arrow.Footnote ¹⁰ In /ka^↓makiri/ kamakiri ‘praying mantis,’ for example, there is a relatively high pitch on /ka/ and a relatively low pitch on /ma/, /ki/, and /ri/, and /ka/ is the accented mora (or accented syllable). Virtually all textbooks and pronunciation guides for Japanese use a phonetically imprecise system for describing the pitch patterns on accent phrases that categorizes each mora as either high-pitched (H) or low-pitched (L), and in this system, the pitch pattern on /ka^↓makiri/ is HLLL.Footnote ¹¹ In an accent phrase that begins with an ordinary mora and is not initially accented, the first mora is L and the second mora is H, as in /pori+bu^↓kuro/ pori-bukuro ‘plastic bag’ (LHHLL). This phrase-initial LH pattern is sometimes called initial lowering.

In principle, a noun consisting of n ordinary moras (or, equivalently, n light syllables) can have any of n+1 accent patterns: it can have an accent on any one of its n moras or it can be unaccented. Special moras, on the other hand, cannot ordinarily bear accent. Consequently, in a mora-plus-syllable analysis like model B in Figure 7.4, the number of possible accent patterns for a noun is one more than the number of syllables, regardless of whether the syllables are heavy or light, and this relationship is the basis for saying that syllables, not moras, are the accent-bearing units in Japanese (McCawley Reference McCawley1968: 59; Kubozono and Honma Reference Kubozono and Honma2002: 37–38), as illustrated in the examples in Table 7.2. The downward-pointing arrow appears between the two moras of an accented heavy syllable because, according to traditional descriptions, the first mora of such a syllable is H and the second mora is L. Phonetically, however, the pitch just falls smoothly from the beginning to the end of an accented heavy syllable.

It is resistance to bearing accent that has led researchers to treat some V/i/ and V/u/ sequences as quasi-diphthongs. If onsetless high-vowel moras repelled accent consistently (or at least nearly consistently), they could simply be identified as special moras, always dependent on an immediately preceding ordinary mora. In fact, however, there is no such consistency. Onsetless /i/ and /u/ often bear accent, and in a mora-plus-syllable analysis, such instances are treated as separate syllables, as in /hi_σro_σi^↓_σzu_σmu/ hiroizumu ‘heroism.’ In a mora-qua-syllable analysis like model A in Figure 7.3, the phonemic transcription would be the same, but an onsetless /i/ or /u/ that carries accent is clearly not dependent on a preceding ordinary mora.

If accent falls on the first mora or on neither mora of a V₁V₂ sequence, the accent pattern usually provides no information about whether V₂ is syllabic (i.e. independent). For example, there is no reliable way to tell whether (in a mora-plus-syllable analysis) /ka^↓i+ro/ kai-ro ‘circuit’ should be treated as /ka^↓_μi_σro/ or as /ka^↓_σi_σro/ and whether /ni+kai+me/ ni-kai-me ‘second time’ should be treated as /ni_σka_μi_σme/ or as /ni_σka_σi_σme/. Initial lowering can be diagnostic if the second mora of a word is an onsetless V and the first mora does not bear accent, since initial lowering is optional in words that begin with a heavy syllable (see Section 5.1). For example, /dai+gaku/ dai-gaku ‘university’ would be /da_σi_σga_σku/ if initial lowering is obligatory (LHHH) and /da_μi_σga_σku/ if initial lowering is optional (LHHH~HHHH).Footnote ¹²

It has been noted many times that the default location for accent in recent borrowings and foreign names is the syllable containing the antepenultimate mora (or the first syllable if the word is shorter than three moras) (McCawley Reference McCawley1968: 133–134; Kubozono Reference Kubozono and Tsujimura1999: 43; Kubozono and Honma Reference Kubozono and Honma2002: 36–38). In many loanwords, a sequence of the form (C)V/i/ arguably behaves like a heavy syllable with respect to default accent. For example, /taipura^↓itaR/ taipuraitā ‘typewriter’ has default accent if it is syllabified as /ta_μi_σpu_σra_μ^↓i_σta_μR/, with /i/ treated as a special mora.

Comparable examples involving (C)V/u/ sequences are much harder to find, but verb forms provide some relevant data. When the citation form of a verb or adjective combines with the interrogative particle ka and the particle has a low and non-rising pitch, the combination expresses a kind of acquiescence. For example, Kowareru ka means something like ‘So (it’s) going to break.’ When the verb form is accented, like /koware^↓ru/ kowareru ‘to break,’ its accent is retained before ka in this pattern: /koware^↓ru ka/. When the verb form is unaccented, an accent appears immediately preceding ka, as in /agaru^↓ ka/ (cf. unaccented /agaru/ agaru ‘to rise’) and /susumu^↓ ka/ (cf. unaccented /susumu/ susumu ‘to advance’). The examples in Table 7.3 show what happens when ka in this meaning follows the citation form of an unaccented verb ending in (C)/au/ or (C)/ou/ or follows the citation form of an unaccented adjective ending in (C)/ai/, (C)/oi/, or (C)/ui/. In such combinations, the accent falls on the penultimate mora rather than the final mora. This is the expected result if the verb or adjective form ends in a heavy syllable (on the mora-plus-syllable analysis) or, equivalently, in a dependent (i.e. special) mora (on the mora-qua-syllable analysis).

There are, however, V/i/ and V/u/ sequences that do not behave like quasi-diphthongs. Pronunciation dictionaries (NHK Hōsō Bunka Kenkyūjo 1998; Kindaichi and Akinaga Reference Haruhiko and Kazue2001) give /fukui^↓+ši/ for Fukui-shi ‘the city of Fukui,’ with /u/ and /i/ in separate syllables (/fu_σku_σi^↓_σ+ši/), although many speakers seem to prefer /fuku^↓i+ši/ (implying (/fu_σku^↓_μi_σ+ši/). There are foreign province and state names that show the same kind of variability. When such a name combines with /šuR/ shū ‘province,’ name-final accent appears. Typical examples are /arubaRta^↓+šuR/ Arubāta-shū ‘the province of Alberta’ (where /ta/ is a short syllable and an ordinary mora) and /orego^↓N+šuR/ Oregon-shū ‘the state of Oregon’ (where /goN/ is a long syllable or a sequence of an ordinary mora followed by a dependent mora). In /hawai^↓+šuR/~/hawa^↓i+šuR/ Hawai-shū ‘the state of Hawai‘i’ and /irinoi^↓+šuR/~/irino^↓i+šuR/ Irinoi-shū ‘the state of Illinois,’ however, many speakers prefer the form in which the onsetless /i/ mora carries the accent. In general, there is considerable variability and uncertainty regarding the dependent status of the second vowel in (C)V/i/ sequences and, to a lesser extent, (C)V/u/ sequences (which are relatively uncommon).

When there is good reason to treat a high front vowel as a dependent mora, it is not uncommon to see it transcribed phonemically as /J/ (see, e.g., Kubozono Reference Kubozono and Haraguchi1993: 73; Kawahara Reference Kawahara2016: 170), as in /taJpura↓JtaR/ for taipuraitā ‘typewriter.’ This phonemic distinction between /i/ and /J/ has the same basic motivation as the distinction between ordinary vowel phonemes and /R/. As noted earlier in this section, if long vowels and V₁V₁ sequences are both analyzed phonemically as two identical vowel phonemes in a row, lexical items can differ minimally in syllabification or, equivalently, in the dependence versus independence of an onsetless V mora. In the same way, if V₁V₂ sequences ending in a high front are all analyzed phonemically as ending /i/, the door is open to violations of the widely (but not universally) accepted principle that minimal pairs differing only in syllabification are not possible (Section 4.1). For example, although /ha^↓ir-u/ hairu ‘to enter’ and /oi^↓-ru/ oiru ‘to grow old’ are not a minimal pair, if they are treated as /ha^↓_μi_σru/ and /o_σi^↓_σru/, they both have penultimate accent (the regular accent pattern for an accented citation form of a verb). The problem of contrastive differences in syllabification disappears if these two words are phonemically /ha^↓Jru/ and /oi^↓ru/, because /J/, like /R/, implies dependence.

It would be preferable to be able to transcribe all onsetless high front vowel moras as /i/ and use other information to determine which instances of /i/ are dependent. If this were possible for all onsetless vowel moras, there would be no need for the phonemes /J/ and /R/, and the claim that lexical items do not differ minimally in syllabification could still be maintained. The same logic applies to the much less common instances of a high back vowel behaving as a dependent mora. If dependence is not predictable, then in addition to /R/ and /J/ a phoneme /W/, distinct from /u/, would presumably be necessary in examples like /dona^↓W+gawa/ Donau-gawa ‘Danube River’ for speakers who have this pronunciation. River names ending in /gawa/ are generally accented on the syllable immediately preceding /gawa/ (or, in a mora-qua-syllable analysis, the last ordinary mora preceding /gawa/), as the examples in Table 7.4 illustrate. According to one authoritative accent dictionary (Kindaichi and Akinaga Reference Haruhiko and Kazue2001), Donau-gawa can be either /donau^↓+gawa/ or /dona^↓u+gawa/ (the same kind of variability exhibited by state names such as /hawai^↓+šuR/~/hawa^↓i+šuR/ Hawai-shū ‘the state of Hawai‘i’).

5.1 Initial Lowering

In an accent phrase that is not initially accented, initial lowering (i.e. the rise from a low pitch on the first mora to a high pitch on the second mora; see Section 4.2) is optional in some cases. In a mora-plus-syllable analysis, the generalization is that initial lowering is optional when the initial syllable is heavy (Hattori Reference Hattori1954: 246; McCawley Reference McCawley and Hyman1977: 262; Kubozono Reference Kubozono2006b: 14). The facts may actually a bit more complicated (Labrune Reference Labrune2012b: 123–124), but in a mora-qua-syllable analysis, the same generalization can be captured by saying that initial lowering is optional when the second mora of a word is a special mora. It is not clear in either analysis what the motivation is for this optional flattening of the LH contour.

5.2 Accent-bearing Units

In a mora-plus-syllable analysis, as explained above in Section 4.2, it is syllables, not moras, that are the accent-bearing units of Japanese. In a mora-qua-syllable analysis, needless to say, moras are the accent-bearing units, but since special moras ordinarily resist carrying accent, there is usually “a leftward shift of the accent when they occupy a prosodic position where an accent would be expected to occur” (Labrune Reference Labrune2012b: 124). Accented special moras do sometimes occur, however, and only /Q/ is indisputably incapable of carrying accent (ibid.: 125, n. 10).

One example of accent on /N/ involves a compounding pattern in which a place name combines with an emphatic /Q/ and /ko/ ‘child’ to form a noun denoting a person from that place. Most compounds that follow this pattern are accented on the penultimate syllable. When the place name is /ro^↓NdoN/ Rondon ‘London,’ the compound is /ro_μN_σdo_σN^↓_μQ_σko/ Rondon-kko ‘born and bred Londoner’ in a mora-plus-syllable analysis, with the accent carried by the first mora of the unusual heavy syllable /N_μQ/. The alternative would be /ro_μN_σdo^↓_μN_μQ_σko/, with the accent carried by the first mora of the superheavy syllable /do_μN_μQ/, but Kubozono (Reference Kubozono and Tsujimura1999: 50–55) argues that such trimoraic syllables are universally dispreferred and that Japanese usually avoids creating them or repairs them when they arise. In a mora-qua-syllable analysis, the atypical accent on the second /N/ in /roNdoN^↓Qko/ could be accounted for on the assumption that a special mora is dependent on the nearest preceding ordinary mora. Treating /N/ as independent and allowing it to bear accent can be understood as a way to avoid having the two special moras in /doNQ/ both dependent on the same ordinary mora /do/.

Labrune (Reference Labrune2012b: 125) claims that /R/ can also bear accent, but the example she cites is not persuasive. Accent dictionaries (NHK Hōsō Bunka Kenkyūjo 1998; Kindaichi and Akinaga Reference Haruhiko and Kazue2001) give two alternative patterns for a loanword meaning ‘chain store’: /če^↓RN+teN/~/čeR^↓N+teN/ chēn-ten, both represented in a way that implies accent-bearing /R/ in the latter. However, in the audio recording that accompanies this entry in electronic editions of the NHK dictionary, the latter alternative is pronounced /čee^↓NteN/ [tɕeˇẽnːtẽɴː], with clear vowel rearticulation, implying the sequence /ee/ rather than the long vowel /eR/. Furthermore, initial lowering (Section 5.1) is obligatory when the latter form appears phrase-initially (*HHLLL), as expected if the first two moras are in separate syllables (/če_σe···/). In contrast, the unequivocal long vowel /eR/ in /čeRsu+gi^↓NkoR/ Chēsu-ginkō ‘Chase Bank’ does not require (and typically lacks) initial lowering.

On a mora-plus-syllable analysis, the form of chēn-ten with a second-mora accentual peak can be treated as /če_σe^↓_µN_σ+te_µN/, with no trimoraic syllable. On a mora-qua-syllable analysis, this same form can be treated as /če_σe^↓_σN_σte_σN/, with no sequence of two special moras both dependent on the same ordinary mora. In neither case is there an accent-bearing special mora. However, if the second mora were /R/ rather than /e/, mora-plus-syllable /če_µR^↓_µN_σte_µN/ and mora-qua-syllable /če_σR^↓_σN_σte_σN/ would both have two dispreferred characteristics. The former has an accent-bearing special mora and a trimoraic syllable, and the latter has an accent-bearing special mora and two special moras (/R/ and /N/) both dependent on the ordinary mora /če/. It would be hard to explain why a form that violates two constraints would ever arise, given that the alternative (mora-plus-syllable /če_σe^↓_µN_σte_µN/ or mora-qua-syllable /če_σe^↓_σN_σte_σN/) avoids both violations. In short, it appears that /R/ cannot carry accent.Footnote ¹³

5.3 Accent Loss Before /no/

Some lexically accented nouns lose their accent immediately preceding the genitive particle no (Martin Reference Martin1975: 23–24). On a mora-plus-syllable analysis, this loss affects only nouns of two or more syllables that are lexically accented on the last syllable, which can be either light or heavy. For example, /a_σši_σta^↓/ ashita ‘tomorrow’ (with a light final syllable) and /ki_σno_μ^↓R/ kinō ‘yesterday’ (with a heavy final syllable) retain their lexical accents when followed by nominative ga (/ašita^↓ ga/, /kino^↓R ga/) but become unaccented when followed by no (/ašita no/, /kinoR no/). There are many exceptions, and accent retention before no may actually be the default (Labrune Reference Labrune2012b: 129–130), but the fact remains that every noun that does undergo this accent loss is lexically accented on its final syllable. Furthermore, the accent loss never affects monosyllabic nouns, regardless of whether the lone syllable is light or heavy. On a mora-qua-syllable analysis, the nouns in question can be characterized as (1) containing more than one ordinary mora and (2) having a lexical accent on the last ordinary mora.

If one is willing to posit underlying/input accents in locations where they never surface, it is possible to claim that a noun ending in a special mora that loses its accent before no has its lexical accent on that last mora (Labrune Reference Labrune2012b: 130). For example, /niho^↓N/ Nihon ‘Japan,’ which loses its accent (/nihoN no/), would be ^LEX/nihoN^↓/, whereas /hoRge^↓N/ hōgen ‘dialect,’ which does not lose its accent (/hoRge^↓N no/), would be ^LEX/hoRge^↓N/. Since a special mora cannot ordinarily bear accent (see Section 4.2), the accent shifts leftward onto the preceding ordinary mora to yield the actual output form /niho^↓N/ when not followed by no. No shift is necessary for /hoRge^↓N/, since the output matches the input. This approach cannot account for nouns ending in an accented ordinary mora that fail to undergo accent loss. The fact that /cugi^↓/ tsugi ‘next,’ for example, retains its accent (/cugi^↓ no/) while /yama^↓/ yama ‘mountain’ does not (/yama no/) cannot reasonably be attributed to an underlying difference in accent location.

The same sort of analysis is possible for some compounds with a second element (E2) ending in a special mora (Labrune Reference Labrune2012a: 227–231). For example, in descriptions that assume syllables distinct from moras, it is frequently claimed that in a compound with an E2 that is three moras or longer and accented on its final syllable as an independent word, the compound will be accented on the first syllable of E2 (Kubozono Reference Kubozono, Miyagawa and Saito2008: 177). A typical example with a light final syllable is /tera+o^↓toko/ tera-otoko ‘temple assistant’ (cf. /otoko^↓/ ‘man’), and one with a heavy final syllable is /aka+čo^↓RčiN/ aka-chōchin ‘red lantern’ (cf. /čoRči^↓N/ ‘lantern’). If the word for ‘lantern’ is underlyingly ^LEX/čoRčiN^↓/, the accent in /aka+čo^↓RčiN/ can be attributed to the underlying accent on the final mora of E2, just as in ^LEX/otoko^↓/, making it unnecessary to refer to syllables distinct from moras. As a word on its own, of course, ^LEX/čoRčiN^↓/ surfaces as /čoRči^↓N/ in order to avoid an accent on a special mora. On this approach, the accent in compounds like /koRčoR+seNse^↓i/ kōchō-sensei ‘school principal’ (cf. /seNse^↓i/ ‘teacher’) indicates that the word for ‘teacher’ is underlyingly ^LEX/seNse^↓i/, with the accent on the penultimate mora.

A major problem for an abstract analysis involving underlying accents on word-final special moras is that accent loss before no and E2-initial accent in compounds do not necessarily go together. For example, /niho^↓N/ Nihon ‘Japan’ loses its accent before no (as noted above), indicating that the underlying form is ^LEX/nihoN^↓/, but compounds like /niši+niho^↓N/ nishi-Nihon ‘western Japan’ indicate that the underlying form is ^LEX/niho^↓N/. On the other hand, /čoRči^↓N/ chōchin ‘lantern’ retains its accent before no (/čoRči^↓N no/), indicating that the underlying form is ^LEX/čoRči^↓N/, but compounds like /aka+čo^↓RčiN/ aka-chōchin ‘red lantern’ (cited above) indicate that the underlying form is ^LEX/čoRčiN^↓/.

To summarize, the necessary (but not sufficient) conditions for accent loss before no can be stated either in terms of a mora-plus-syllable analysis (more than one syllable and accent on the final syllable) or in terms of a mora-qua-syllable analysis (more than one ordinary mora and accent on the last ordinary mora). It does not appear that either analysis offers any advantage in dealing with the irregularity of this accent-loss phenomenon.

5.4 Names ending with /taroR/

The name /ta^↓roR/ Tarō is a traditional favorite for first-born sons, and it is also popular as a second element in longer names. Assuming a mora-plus-syllable analysis, these longer names are unaccented if the first element (E1) is monosyllabic, accented on the second syllable if E1 is two light syllables, and accented on the first syllable of /ta^↓_σro_µR/ if E1 is three moras or longer (Kubozono Reference Kubozono and Tsujimura1999: 45–46). This pattern is clearly productive and applies, for example, to /kyu_µR_σ+ta_σro_µR/ Kyū-tarō (the name of a cartoon character), in which E1 is a single heavy syllable. Additional examples are given in Table 7.5. There are also a few common nouns ending with Tarō, and in these as well, light and heavy monosyllabic E1s are treated the same: /yo_σ+ta_σro_µR/ yo-tarō ‘dunce,’ /pu_µR_σ+ta_σro_µR/ pū-tarō ‘day laborer.’

To formulate the generalization in terms of a mora-qua-syllable analysis, one can say that names ending in Tarō are unaccented when E1 is either a single mora or a sequence of an ordinary mora followed by a special mora (Labrune Reference Labrune2012b: 131). The same set of E1s could also be specified as those that contain only one ordinary mora. It is not clear why this set of E1s should be a natural class in a mora-qua-syllable analysis (Kawahara Reference Kawahara2016: 186), but the generalization is not difficult to state.

2.1 Tokyo Japanese

Based on the number of pitch patterns observed or permitted in the system, Uwano (Reference Uwano and Kaji1999, Reference Uwano2012a) proposed classifying Japanese pitch accent systems into two major groups, accented systems and accentless ones, with the former further divided into two groups, multiple-pattern and N-pattern systems. Of these, the accentless group refers to systems that have no fixed pitch pattern for words, namely, those where pitch is not specified at the lexical level. Accented systems, on the other hand, have pitch specifications at the lexical level. Among this latter group, multiple-pattern systems refer to those in which the number of pitch patterns increases as the word becomes longer, whereas the number is fixed to a certain integer in N-pattern systems, typically ranging between one and four, independent of the length of the word.

Of these three types, Tokyo Japanese belongs to the multiple-pattern group since it exhibits two pitch patterns for monosyllabic nouns, three patterns for disyllabic nouns, and four patterns for trisyllabic nouns. This is illustrated in (1–3). In (1–3) and the rest of the chapter, pitch accent in Tokyo Japanese is denoted by an apostrophe, whereas superscript /°/ is added to unaccented words to clearly distinguish them from accented words. Dots indicate syllable boundaries, wherever necessary, and /ga/ is a grammatical particle denoting the nominative case (nom).

1. (1) Monosyllabic nouns

   1. a.

      hi’-ga

      ‘fire-nom’

   2. b.

      hi°-ga

      ‘sun-nom’

1. (2) Disyllabic nouns

   1. a.

      ha.’na-ga	‘Hana (female name)-nom’
      a’me-ga	‘rain-nom’
      a.’ki-ga	‘autumn-nom’

   2. b.

      ha.na’-ga	‘flower-nom’
      a.ki’-ga	‘weariness-nom’

   3. c.

      ha.na°-ga	‘nose-nom’
      a.me°-ga	‘candy-nom’
      a.ki°-ga	‘vacancy-nom’

1. (3) Trisyllabic nouns

   1. a.

      ka’.bu.to-ga

      ‘helmet-nom’

   2. b.

      ko.ko’.ro-ga

      ‘heart-nom’

   3. c.

      o.to.ko’-ga

      ‘man-nom’

   4. d.

      sa.ka.na°-ga

      ‘fish-nom’

As shown in (1–3), each syllable can bear an accent, yielding n-patterns for n-syllable nouns.Footnote ² Tokyo Japanese permits one additional pattern for each word length, hence (n+1) patterns for n-syllable words. This additional pattern is called the unaccented pattern, which involves no sudden pitch fall even if the noun is followed by a grammatical particle. This pattern apparently violates the principle of obligatoriness, or the idea that every lexical word must have at least one prominence or prominent position (Hyman Reference Hyman2006), and may differentiate Tokyo Japanese from many pitch accent languages in the world (Kubozono Reference Kubozono2012a). In phonetic terms, it is often difficult to distinguish this accent pattern from the finally accented pattern shown in (2b) and (3c) when words are pronounced in isolation or in utterance-final position (Vance Reference Vance1995), but they can be readily distinguished when followed by a grammatical particle since pitch suddenly falls immediately before the particle in one case but not in the other. Because of the unaccented pattern, moreover, the multiple-pattern system illustrated in (1–3) as a whole displays a contrast in accentedness – presence or absence of a pitch accent – in addition to accent position.

Four points must be noted here. First, while nouns thus exhibit multiple pitch patterns in Tokyo Japanese, verbs and adjectives permit only two patterns, as shown in (4): (a) a pattern with an accent on the penultimate mora and (b) the unaccented pattern.Footnote ³ Since accent position is fixed in accented verbs and adjectives, pitch accent can only be contrastive in terms of accentedness in these types of words.

1. (4) Verbs and adjectives in Tokyo Japanese

   1. a.

      Accented
      ki’.ru	‘to cut’
      su’.ru	‘to print’
      mo.to.me’.ru	‘to request’
      a.o’i	‘blue, green’Footnote 4
      u.ma’i	‘tasty’

   2. b.

      Unaccented
      ki.ru°	‘to wear’
      su.ru°	‘to do’
      ma.to.me.ru°	‘to sum up’
      a.kai°	‘red’
      a.mai°	‘sweet’

Second, the multiple pitch accent patterns displayed by nouns are not equally distributed across the vocabulary. On the contrary, there are only two accent patterns that are popular across nouns of different lengths: the antepenultimate pattern with an accent on the third mora from the end of the word, as in (3a), and the unaccented pattern as in (2c) and (3d) (Haraguchi Reference Haraguchi1991; Kubozono Reference Kubozono2006a). For example, these two accent patterns combined account for more than 90% of trimoraic nouns (Kubozono Reference Kubozono2006b). Moreover, the (n+1) rule does not work in relatively long words; for example, there are very few six-mora words, if any, that are initially accented or finally accented. This led Kubozono (Reference Kubozono, Miyagawa and Saito2008) to propose that Tokyo Japanese basically has a two-pattern system for nouns as it does for verbs and adjectives and that nouns differ from verbs/adjectives in the extent to which lexical exceptions are permitted.

Third, since there are only two dominant patterns in nouns in Tokyo Japanese, the contrastive function of pitch accent in this system naturally involves an opposition in accentedness. That is, most pairs of words that contrast in pitch accent show a contrast between the accented and unaccented patterns. This is exemplified in (5).

1. (5) Accented and unaccented pairs in Tokyo Japanese

   1. a. Accented words

      mi.ya.gi’.san	‘Mt. Miyagi’
      ko’o.koo	‘filial piety’
      ta’n.go	‘Tango (place name), the tango (dance)’
      i’.on	‘ion (in chemistry)’

   2. b. Unaccented words

      mi.ya.gi.san°	‘produce of Miyagi’
      koo.koo°	‘high school’
      tan.go°	‘word’
      i.on°	‘allophone’

Finally, and related to the third point, the contrastive function of pitch accent in Tokyo Japanese is not as high as might be assumed: it only accounts for 14% of all pairs of words that are segmentally homophonous, while the remaining 86% are pairs of words that are also homophonous accentually (Sibata and Shibata Reference Sibata and Shibata1990). This has two implications. For one thing, it implies that accentuation in Tokyo Japanese is rule-governed to a considerable extent: accent patterns are more or less predictable if the structure of the word is given. Second, it also implies that the primary function of pitch accent in Tokyo Japanese is not a distinctive one but rather a demarcative or culminative one, that is, the function of signaling the unity of words in the sentence or in connected speech. This view is supported by the fact that the default position of pitch accent is the antepenultimate mora irrespective of the length of the word: an abrupt pitch fall signals that it is almost the end of the word.

2.2 N-pattern Systems

While the number of accent patterns increases as the word becomes longer in Tokyo Japanese, there are also many dialects – especially in the southern part of Japan – that have N-pattern systems, or systems where the number of accent patterns is independent of the length of the word. The number may vary from one to four, with the two-pattern system being by far the most common (Uwano Reference Uwano and Kaji1999). One of the most well-known two-pattern systems is that of Kagoshima Japanese, which has a high (H) tone on the penultimate syllable (Type A) or on the final syllable (Type B) for both nouns and verbs/adjectives (Hirayama Reference Hirayama1951; Kibe Reference Kibe2000; Kubozono Reference Kubozono2006b, Reference Kubozono2010). The choice between the two accent patterns is mostly lexically determined in short words, although it becomes more predictable as the word becomes longer or morphologically more complex (Kubozono Reference Kubozono and van Oostendorp2011; see also Section 8 below). In (6) and the rest of the chapter, high-pitched portions are denoted by capital letters.Footnote ⁵ Parentheses describe an alternative accentual analysis of this system whereby Type A is regarded as accented and Type B as unaccented (Haraguchi Reference Haraguchi1977; Shibatani Reference Shibatani1990; Kubozono Reference Kubozono2012a; see also Section 3 below for more details).Footnote ⁶

1. (6) Two accent patterns in Kagoshima Japanese

   1. a. Type A

      NA.tsu (na’.tsu)	‘summer’
      A.ka (a’.ka)	‘red’
      na.tsu.ya.SU.mi (na.tsu.ya.su’.mi)	‘summer holiday’
      KI.ru (ki’.ru)	‘to wear’
      ma.to.ME.ru (ma.to.me’.ru)	‘to sum up’

   2. b. Type B

      ha.RU (ha.ru°)	‘spring’
      a.O (a.o°)	‘blue, green’
      ha.ru.ya.su.MI (ha.ru.ya.su.mi°)	‘spring holiday’
      ki.RU (ki.ru°)	‘to cut’
      mo.to.me.RU (mo.to.me.ru°)	‘to request’

Note that this system resembles the two-pattern accent system of verbs and adjectives in Tokyo Japanese illustrated in (4) above, although surface pitch patterns are often different between the two systems.

2.3 Accentless Systems

In addition to the multiple-pattern and N-pattern systems, Uwano (Reference Uwano and Kaji1999) posits an independent type called an accentless system. This is a system where word prosody does not use pitch. In the Kumamoto dialect spoken near Kagoshima, words are usually produced with a more or less flat pitch in utterances (Maekawa Reference Maekawa and Sato1997a). Pitch is used at the sentence level in this system, too, to differentiate interrogative sentences from declarative ones, for example. However, it is not specified at the lexical level, so a single word may be produced with several pitch patterns depending on the syntactic and/or pragmatic context, without changing its lexical meaning.

The accentless systems should not be confused with one-pattern systems that represent a subtype of N-pattern systems. A typical one-pattern system is found in the Miyakonojo dialect, which is also adjacent to Kagoshima (Haraguchi Reference Haraguchi1977; Shibatani Reference Shibatani1990). In this system, all words are pronounced in the same way, usually with a high pitch in phrase-final position so that pitch is not used distinctively in this system just as in accentless systems. However, pitch must nevertheless be specified at the lexical level in this system since words would sound awkward if they were pronounced with different pitch patterns, for example, with an H tone in their initial position. Moreover, the H tone is usually bound by bunsetsu, the basic syntactic phrase consisting of a content word and one or more function words. A typical pitch pattern of a sentence is illustrated in (7).

1. (7)

   bo.ku-WA	ni.hon-ka.RA	ki.TA
   I-top	Japan-from	came
   ‘I came from Japan.’

8.1 Right-dominant Compound Rule

The right-dominant nature of compound accentuation in Tokyo Japanese is illustrated in (26–28). Whether the rightmost element is ‘short’ (monomoraic or bimoraic) or ‘long’ (trimoraic or longer) (McCawley Reference McCawley1968), the basic principle underlying this rule is to preserve the accent of the rightmost element if it is lexically accented, unless it violates the non-finality constraint prohibiting the preservation of any accent on the word-final syllable (Kubozono Reference Kubozono1995a, Reference Kubozono, Agbayani and Tang1997; cf. Poser Reference Poser1990). Thus, the lexical accent of the rightmost element is readily preserved if it does not violate this constraint as in (26). If this accent cannot be preserved due to the non-finality constraint or if the rightmost element has no accent to preserve, a default compound accent emerges on the rightmost, non-final bimoraic foot of the compound.Footnote ¹⁷ This is exemplified in (27a, b), respectively, where foot structure is minimally shown.

1. (26) o.na.ga° + sa’.ru → o.na.ga-(za’.ru) ‘long tail + monkey; long-tailed monkey’

   koo.so.ku° + ba’.su → koo.so.ku-(ba’.su) ‘highway + bus; highway bus’

   ku’.ro + ka’.ra.su → ku.ro-(ga’.ra).su ‘black + crow; black crow’

   ryuu.kyu’u + a.sa’.ga.o → ryuu.kyuu-a.(sa’.ga).o ‘Ryukyu + morning glory (flower); Ryukyuan morning glory’

1. 1. a. a’.ki.ta + i.nu’ → a.ki.ta’)-(i.nu) ‘Akita + dog; Akita Dog’

      me’.ron + pa’n → me.(ro’n)-(pan) ‘melon + bread; melon flavored bread’

      te’ + ka.ga.mi’ → te-(ka’.ga).mi ‘hand + mirror; hand mirror’

   2. b. ku.gu.ri’ + to° → ku.gu.ri’)-do ‘to go through + door; side door’

      o.na.ga° + to.ri° → o.na.ga’)-do.ri ‘long tail + bird; long tailed cock’

      ku’.ro + ga.ra.su° → ku.ro-(ga’ra).su ‘black + glass; black glass’

      mi.na.mi° + a.me.ri.ka° → mi.na.mi-(a’.me).(ri.ka)~mi.na.mi a.(me’.ri).ka ‘south + America; South America’

A major exception to the generalization illustrated in (26–27) is unaccented compounds, most of which occur due to the deaccenting nature of so-called deaccenting morphemes (McCawley Reference McCawley1968; see Giriko Reference Giriko2009 and Kubozono Reference Kubozono2017 for other factors triggering unaccented compounds in Tokyo Japanese). Thus, native morphemes such as /i.ro’/ ‘color’ and /ka.ta’/ ‘type’ as well as Sino-Japanese morphemes including /to’o/ ‘political party’ and /se’n/ ‘line’ have an effect of deaccenting the entire compounds of which they form the final member.Footnote ¹⁸

1. 1. a. pi’n.ku + i.ro’ → pin.ku-i.ro° ‘pink + color; pink’

      ne.zu.mi° + i.ro’ → ne.zu.mi-i.ro° ‘rat + color; gray’

   2. b. e’e + ka.ta’ → ee-ga.ta° ‘A + type; blood type A’

      ha’.mu.ret.to + ka.ta’ → ha.mu.ret.to-ga.ta° ‘Hamlet + type; the type of Hamlet’

   3. c. kyoo.san° + to’o → kyoo.san-too° ‘common wealth + political party; Communist Party’

      roo.doo° + to’o → roo.doo-too° ‘labor + political party; Labor Party’

   4. d. too.ka’i.doo + se’n → too.kai.doo-sen° ‘Tokaido + line; Tokaido Line’

      yo.ko.su.ka° + se’n → yo.ko.su.ka-sen° ‘Yokosuka + line; Yokosuka Line’

Compound accentuation in Tokyo Japanese is thus determined by the rightmost member of compounds, including the deaccenting morphemes in (28). One potential exception to this general rule is the accentuation of “dvandva” compounds, or compounds involving a coordinate structure. This type of compound tends to preserve the lexical accent of the initial member and lose the accent (if any) of the second member if they consist of two short members as in (29a) (Akinaga Reference Akinaga1985). If they consist of relatively long members, on the other hand, they tend to split into two accentual units or prosodic words instead of being fused into one unit (Kubozono Reference Kubozono1988, Reference Kubozono2017). This latter pattern is illustrated in (29b), where {} denotes prosodic word boundaries.

1. 1. a. a’.sa + ban° → {a’.sa-ban} ‘morning + evening; morning and evening’

      a’.me + ka.ze° → {a’.me-ka.ze} ‘rain + wind; rain and wind’

      ta’ + ha’.ta → {ta’-ha.ta} ‘rice field + farm; the fields’

   2. b. che’.ko + su.ro.ba’.ki.a → {che.’ko}{su.ro.ba.’ki.a} ‘Czecho + Slovakia; Czecho-Slovakia’

      bik.ku’.ri + gyoo.ten° → {bik.ku.’ri}{gyoo.ten°} ‘being astonished + being stunned; being astonished and stunned’

      i’p.pu + ta.sai° → {i’p.pu}{ta.sai°} ‘one husband + many wives; polygamy’

      to’o.zai + na’n.bo.ku → {to’o.zai}{na’n.bo.ku} ‘east & west + south & north; all directions, everywhere’

8.2 Left-dominant Compound Rule

While compound accentuation in Tokyo Japanese is basically right-dominant, there are many dialects whose compound accent patterns are left-dominant. The two-pattern accent systems found in Kyūshū – for example, Kagoshima, Nagasaki, and Koshikijima Japanese – are typical examples (Hirayama Reference Hirayama1951; Uwano Reference Uwano and Kunihiro1997; Hayata Reference Hayata1999; Kibe Reference Kibe2000). To take some examples from Kagoshima Japanese, compound words in this dialect take the Type A pattern (an H tone on the penultimate syllable) if their initial member is lexically Type A, while they take the Type B pattern (an H tone on the final syllable) if the initial member is Type B. This is exemplified in (30a, b), respectively.

1. 1. a. NA.tsu + ma.TSU.ri → na.tsu-ma.TSU.ri ‘summer + festival; summer festival’

      NA.tsu + ya.su.MI → na.tsu-ya.SU.mi ‘summer + holiday; summer holiday’

   2. b. ha.RU + ma.TSU.ri → ha.ru-ma.tsu.RI ‘spring + festival; spring festival’

      ha.RU + ya.su.MI → ha.ru-ya.su.MI ‘spring + holiday; spring holiday’

One naturally wonders here if the right-dominant and left-dominant nature of compound accentuation may have to do with the distinction between multiple-pattern and N-pattern systems discussed in Section 2, which, in turn, may be linked to the typological categorization of Japanese pitch accent into “word accent” and “word tone” discussed in Section 4 (see Kubozono Reference Kubozono2012b for arguments for this idea). Whether these parameters correlate with each other is a very important and interesting topic for future work.

8.3 Hybrid System

Interestingly, there also exists a pitch accent system that exhibits both the right-dominant and left-dominant features in its compound accentuation. This system is widely found in Kinki Japanese (Uwano Reference Uwano and Kunihiro1997; Hayata Reference Hayata1999). In the Kyoto dialect, for example, compound words display two major prosodic features, one concerning the pitch height of their initial position – high or low – and the other regarding their accentedness – accented or unaccented. Of these two features, the second one is shared by Tokyo Japanese as shown in (26–28). Kyoto Japanese is, in fact, quite similar to Tokyo Japanese in that compounds look at their final member to determine whether they are accented or unaccented. They also tend to preserve the lexical accent of their final member as much as possible, subject to the non-finality constraint.

In addition to this, compounds in Kyoto Japanese refer to their initial member to determine whether they begin with a high or low pitch. If the initial member begins with a high pitch when pronounced in isolation, this feature is inherited by the compound, as in (31a). Likewise, the compound begins with a low pitch, as in (31b), if the initial member is a low-beginning morpheme (Wada Reference Wada1942).

1. 1. a. NA.tsu + YA.su.mi → NA.TSU-YA.su.mi ‘summer + holiday; summer holiday’

      KYA.be.tsu + ha.TA.ke → KYA.BE.TSU-BA.ta.ke ‘cabbage + field; cabbage field’

   2. b. ha.RU + YA.su.mi → ha.ru-YA.su.mi ‘spring + holiday; spring holiday’

      ya.saI + ha.TA.ke → ya.sai-BA.ta.ke ‘vegetable + field; vegetable field’

Compound accentuation in Kyoto Japanese thus shows a hybrid system combining the right-dominant nature of compound accentuation found in Tokyo Japanese and the left-dominant one found in Kagoshima Japanese.Footnote ¹⁹ To use Hayata’s (Reference Hayata1999) terminology, Kyoto Japanese has both “word accent” and “word tone” and its compound accentuation involves the right-dominant preservation of “word accent” (pitch fall) and the left-dominant preservation of “word tone” (word-initial pitch pattern in this particular dialect).

9.1 Summary

This chapter has considered the diversity of pitch accent systems in Japanese with the system of Tokyo Japanese as a reference. It has shown that Japanese dialects differ from each other in many features pertaining to word accent. For example, Tokyo Japanese permits multiple accent patterns for nouns – two patterns for monosyllabic nouns, three patterns for disyllabic ones, four patterns for trisyllabic ones, etc. On the other hand, N-pattern systems permit a fixed number of accent patterns irrespective of the length of the word: Kagoshima Japanese and its sister dialects, for example Nagasaki and Koshikijima Japanese, exhibit only two patterns (Type A and Type B) no matter how long the word may be (Section 2).

Japanese dialects also show variability with respect to the distinctive phonetic feature of lexical pitch accent (Section 3). Tokyo Japanese and many other dialects employ pitch fall to differentiate one accent pattern from another, whereas some like the Narada dialect exceptionally use pitch rise for the same purpose.

Japanese dialects also display variability in how accent patterns are determined. First, they differ in the domain where pitch accent patterns are defined: Tokyo Japanese uses the word as the domain, whereas many dialects such as Kagoshima and Nagasaki Japanese realize pitch accent patterns in the domain of the bunsetsu, ‘the basic syntactic phrase’ (Section 4). Second, many dialects including Tokyo, Kagoshima, and Nagasaki Japanese permit only one prominence or underlying H tone per domain, whereas some dialects such as Koshikijima Japanese permit more than one prominence or H tone in the same domain (Section 5). Third, Tokyo and Nagasaki Japanese use the mora as the basic unit to measure phonological distances in pitch accent assignment, while Kagoshima Japanese measures phonological distances with the syllable (Section 6).

Fourth, Japanese dialects may vary in the directionality of pitch accent assignment (Section 7). Tokyo Japanese as well as Kagoshima and Koshikijima Japanese count the number of moras (or syllables) from the right edge of the domain. In contrast, Nagasaki Japanese determines the position of the word-level prominence from the left edge.

Finally, Japanese dialects display variability regarding compound accentuation, too (Section 8). Tokyo Japanese has a typical right-dominant compound accent rule by which the phonological structure of the rightmost element plays the key role in determining compound accent patterns. In contrast, Kagoshima Japanese and its sister dialects have a left-dominant accent rule whereby the prosodic property of the leftmost element is inherited by the entire compound.

These observations, which are summarized in Table 8.1, show how and to what extent pitch accent systems of Japanese differ from each other. They clearly demonstrate that Tokyo Japanese represents only one type of system among various Japanese pitch accent systems.

The situation summarized in Table 8.1 is complex enough. However, it is further complicated by the existence of hybrid systems involving two seemingly competing features pertaining to one and the same parameter. For example, a single dialect may use both the word and the bunsetsu as the domain of pitch accent assignment: it defines one accent pattern within the word, while defining another pattern within the bunsetsu domain (Section 4). Similarly, a single system may use both the syllable and the mora as counting units: Koshikijima-Teuchi Japanese counts the number of moras to determine the position of the primary H tone, while counting syllables to define the position/domain of the secondary H tone (Section 6).

Moreover, while most dialects determine the position of word-level prominence from either the left edge or right edge of the word or bunsetsu, some dialects use both the left-to-right and right-to-left procedures in the same system or even within the same word (Section 7). Finally, the left-dominant and right-dominant compound accent rules may coexist within a single prosodic system (Section 8). The existence of these hybrid systems makes Japanese pitch accent look more complex, but even more interesting and fascinating at the same time.

9.2 Future Agenda

This chapter has raised as many questions as it has solved. One interesting question for future work concerns the relationship between the various parameters that are used to describe the pitch accent systems. It was suggested in passing that the distinction between multiple-pattern and N-pattern systems is closely related to the domain parameter (word versus bunsetsu) (Section 4) and also to the nature of compound accentuation (left-dominant versus right-dominant) (Section 8). On the other hand, our analysis has shown that the multiple-pattern versus N-pattern distinction is independent of other parameters regarding the distinctive feature (Section 3), culminativity (one prominence versus two prominences) (Section 5), the basic prosodic unit (mora versus syllable) (Section 6), and the directionality of pitch accent assignment (left to right versus right to left) (Section 7). The foregoing discussion has clearly shown that N-pattern systems display variability in these parameters: Nagasaki, Kagoshima, and Koshikijima Japanese, for example, all have two-pattern systems, but they exhibit different features with respect to culminativity, the prosodic unit, and directionality. It will be interesting to examine in more depth how these parameters interact with each other and other parameters.

To answer this interesting question, it will be necessary to look at more data from a wider range of Japanese dialects. Given that many regional dialects are at the risk of extinction like the Koshikijima and Kikaijima dialects, it is indeed vital to expand our data in order to better understand the diversity of pitch accent systems, while we also continue to examine the pitch accent system of Tokyo Japanese.

Expanding the scope of analysis is another dimension in which our research on Japanese pitch accent can be extended. Most research on Japanese pitch accent in the past has concentrated more or less on the analysis of word accent within the word or bunsetsu. On the other hand, how word accent patterns are manifested in sentences and connected speech remains a largely understudied topic. If word accent studies are expanded in this direction, they might reveal new aspects of lexical accent such as the loss of word accent or accentual contrasts in focus and other constructions beyond the word and phrase.

Research along these lines will inevitably call for cross-linguistic comparisons between Japanese and other languages. The diversity of Japanese pitch accent systems is certainly an interesting topic by itself, but it will be more interesting to consider it from cross-linguistic and typological perspectives to see what insight and implications the diversity has for the nature of word accent and prosodic typology.

2.1 What is Intonation?

Ladd (Reference Ladd2008: 4) defines intonation as “the use of suprasegmental phonetic features to convey ‘post-lexical’ or sentence-level pragmatic meanings in a linguistically structured way.” Suprasegmental features are commonly defined as fundamental frequency (F0), intensity, and duration. F0 corresponds to the rate of complete cycles of vibration of the vocal folds per unit of time. A higher F0 value gives listeners the auditory sensation of a higher pitch. Thus, F0 is a physical property of speech, whereas pitch is its perceptual correlate (however, these two terms are used interchangeably in this chapter). In Japanese and many other languages (Cruttenden Reference Cruttenden1997; Gussenhoven Reference Gussenhoven2004), pitch is the primary feature in intonation; therefore, this chapter focuses on the F0 contours of the utterance.

Meanings conveyed by intonation apply to phrases or to an utterance as a whole. They include, according to Ladd (Reference Ladd2008), sentence type, speech act, focus, or information structure. A prototypical sentence-type distinction signaled by intonation is statement versus question, which in many languages is achieved by a rising boundary pitch movement at the end of the question sentence. Intonation contrasts with lexical tones (generally called “lexical pitch accents” in Japanese linguistics), which distinguish word-level meanings. In the distinction between háshi ‘chopstick’ and hashí ‘bridge,’ for example, a high pitch (denoted by the acute accent marker) falls on the first syllable in the former but on the second syllable in the latter. In Japanese, lexical pitch accent constrains prosodic phrasing above the word level; therefore, this chapter especially focuses on lexical pitch accent in Section 2.

Intonation conveys meanings in a structured way as a phonological organization (e.g. Ladd Reference Ladd2008: 3–42; Gussenhoven Reference Gussenhoven2004: 49–70). Pitch variation conveying non-linguistic information (e.g. sex, age, and emotional state) is excluded from intonation as defined here. This information can be interpreted even by listeners who do not know the language because forms and functions are directly correlated. For example, the more excited a person is, the higher is the pitch of his/her voice. In the currently defined meaning of intonation, in contrast, forms and functions are phonological in the sense that the phonetic dimension is mediated by the language’s phonological system. To confirm that intonation differs from non-linguistic pitch variation, Gussenhoven (Reference Gussenhoven2004: 49–70) demonstrates that intonation and other language components share the three design features that Hockett (Reference Hockett1958) identifies for human language: arbitrariness, discreteness, and duality of structure.

In fact, the form-function relation in intonation tends to be non-arbitrary. In many languages, rising or high pitch signals question while falling or low pitch signals statement (Ohala Reference Ohala1984). Moreover, a word produced with a higher pitch tends to be interpreted as being more informative than another produced with a lower pitch. Utterance-final rising, or high pitch in general, conveys continuation while falling or low pitch conveys finality (Gussenhoven Reference Gussenhoven2004: 71–96). Even though in most cases intonation lacks an arbitrary relation between form and function, this is not always the case. In Chickasaw, for example, the interrogative contour is falling, whereas the declarative contour is rising (Gussenhoven Reference Gussenhoven2004: 53–54), thus indicating the arbitrariness of the form-function relation in this language. While it is not easy to find evidence for arbitrariness in Tokyo Japanese, aspects of certain regional varieties of Japanese defy the non-arbitrary tendency. In the Kagoshima dialect, for example, a rising pitch contour does not signal interrogativity; the pitch contour in a question sentence is virtually identical to that in a statement (Kibe Reference Kibe, Kobayashi and Shinozaki2010). Furthermore, in the Imaichi dialect, the focused word bears the lowest pitch and the location of the pitch peak in the utterance is not associated with the focused word at all.Footnote ¹

Most contemporary researchers agree that intonation has a discrete form-function relation, indicating that phonetic dimensions are not directly correlated with semantic ones. This can be illustrated by the case in which two phonetically similar pitch contours are interpreted as signaling discrete meanings. Pierrehumbert and Steele (1989) demonstrate that the phonetic continuum of English rise–fall–rise contours is interpreted by native speakers as two discretely different intonation patterns. However, we frequently observe cases in which a certain putative linguistic message is signaled by continuously variable pitch. For example, Liberman and Pierrehumbert (Reference Liberman, Pierrehumbert, Aronoff and Oerhle1984) show that in English, the degree of emphasis and pitch range size are correlated, with more emphatic utterances having a wider pitch range. This message (i.e. the degree of emphasis conveyed by a pitch range) cannot be considered to be the meaning that the intonation as defined here conveys. Instead, the pitch range variation at issue is considered to be a phonetic modification of a single phonological category of intonation. In fact, establishing discreteness in intonation is one of the most challenging issues in the study of intonation. In Japanese as well, the question of the number of intonation patterns is yet to be resolved. We will return to these issues in Section 3.3, which concerns the inventory of Japanese boundary pitch movements.

Duality of structure is the most controversial issue in intonation. According to Gussenhoven (2006), the existence of duality in intonation is suggested by the English calling contour or “vocative chant” (like Jo-ohn!) in which two phonological elements embody one intonational morpheme. However, we will not discuss duality of structure in this chapter.

2.2 Major Components of Japanese Intonation

We describe the Japanese intonation system in terms of two major components: boundary pitch movements (BPMs) and prosodic phrasing above the word level (prosodic phrasing).

BPMs are tones that occur at the end of the prosodic phrase and contribute to the pragmatic interpretation of the utterance (Venditti, Maekawa, and Beckman Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008: 471), for example, as showing questioning, continuation, or emphasis. This information is sometimes regarded as “modality,” which is defined as the mental attitude that a speaker assumes when s/he produces an utterance with a certain intention. According to Kori (Reference Kori and Kunihiro1997: 190), modalities include questions, strong insistence, incredulity, and emphasis, which can be distinguished solely by intonation. For example, when Ichirō ga hōmuran o utta (Ichiro nom homerun acc hit.past ‘Ichiro hit a homerun’) is articulated with a rising pitch at utterance-final mora (i.e. with a rising BPM), the sentence is interpreted as a question, whereas it is interpreted as a statement without this BPM. While BPM is sometimes referred to as “sentence-final intonation” (Kori Reference Kori and Kunihiro1997: 190), this terminology is avoided in this chapter as BPM occurs not only sentence-finally but also sentence-medially.

Prosodic phrasing is defined as the grouping of words in an utterance through suprasegmental features. In Japanese, prosodic phrasing can signal the focused word in the sentence, the syntactic constituency, and so forth. For example, when the NP utsukushii suishagoya no musume (beautiful miller gen daughter) is divided into two prosodic phrases, {utsukushii} {suishagoya no musume}, it is interpreted as ‘the miller’s beautiful daughter’ (curly brackets indicate boundaries of prosodic phrases). When, in contrast, the NP is produced with a single prosodic phrase, {utsukushii suishagoya no musume}, it is interpreted as ‘the daughter of the beautiful miller.’ These two interpretations are a consequence of differences in the syntactic structure. The former has the structure [utsukushii [suishagoya no musume]] in which the adjective utsukushii ‘beautiful’ modifies the extended NP suishagoya no musume ‘miller’s daughter’ (square brackets indicate syntactic boundaries). In contrast, the latter has the structure [[utsukushii suishagoya no] musume] in which the adjective modifies only the immediately following noun suishagoya.

3.1 Introduction

BPMs are tones that occur at the end of prosodic phrases. They include a slightly concave rising pitch movement, transcribed as LH% in the X-JToBI scheme (see Section 3.2), which typically occurs at the end of a question sentence. The sentence ending with a verb in a predicative form shown in Figure 9.1 is interpreted as a question when accompanied by LH% (left), whereas the same sentence is interpreted as a statement without LH% (right) (Uemura Reference Uemura1989; Japanese utterances can also have no BPM at all).

Figure 9.1 Question utterance with BPM and statement utterance without BPM. Ya’mano wa u’mi de oyo’gu. (LH%) (Yamano top sea loc swim) ‘Will Yamano swim in the sea?’ (left) and ‘Yamano will swim in the sea’ (right). The mora assigned a BPM is /gu/, which is underlined in this caption.

While LH% typically appears in question sentences, it does not always indicate a question. This is shown in Figure 9.2 where the sentence is interpreted as a statement even with LH%, indicating that the meanings of BPM are more abstract, as will be described in the following section.

Figure 9.2 Statement utterances with and without BPM (LH %). Ya’mano wa u’mi de oyo’gu yo. (Yamano top sea loc swim sfp) ‘Yamano will swim in the sea.’ The mora assigned a BPM is /yo/, which is underlined in this caption.

3.2 The BPM Inventory

LH% is not the only BPM in the inventory of Japanese BPM. As discussed in Section 3.4 below, there is disagreement over the number of BPMs. Figure 9.3 depicts the four main types of BPM indicated by the X-JToBI framework (Maekawa et al. Reference Maekawa, Kikuchi, Igarashi and Venditti2002), the extended version of Japanese Tone and Break Indices, or the J_ToBI system (Venditti Reference Venditti and Jun2005), which owes its theoretical foundation to Pierrehumbert and Beckman’s (Reference Pierrehumbert and Beckman1988) study. The four main types of BPM are H% (simple rise), HL% (rise–fall), LH% (scooped rise), and HLH% (rise–fall–rise). All the BPMs in this figure are attached to a phrase comprising a single word.

Figure 9.3 Main types of BPM. Ima ne (now sfp) ‘Now.’ The boundaries of the final mora /ne/ are marked.

H% differs from LH% mainly in its F0 shape. In the case of H%, F0 starts rising at the beginning of the phrase-final mora, whereas in the case of LH% it starts in the middle of the final mora. In addition to this alignment difference, pitch range is generally (but not necessarily) smaller in H% than in LH% (Venditti, Maekawa, and Beckman Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008). The resultant F0 shape for H% is a linear rise with a smaller excursion while that for LH% is a concave or scooped rise with a larger excursion.

H% in the utterance-final position generally does not signal a question interpretation. Instead, it provides information, for example, that the speaker is insisting (Venditti, Maeda, and van Santen Reference Venditti, Maeda and van Santen1998) or that he is firmly persuading the listener to agree with what has been said (Uemura Reference Uemura1989). As will be discussed in Section 3.5, H% is sometimes referred to as an “emphatic” rise in other frameworks (Kori Reference Kori and Kunihiro1997; Uemura Reference Uemura1989) as it emphasizes the phrase to which it is attached. H% is also used, according to Uemura (Reference Uemura1989), when the speaker is seeking approval, convincing the listener, inviting the listener’s attention, or blaming. H% can appear utterance-medially, and, in this case, it can also lend prominence to the phrase in which it appears (Kori Reference Kori and Kunihiro1997; Yoshizawa Reference Yoshizawa1960). It also signals a continuation of speech (Kori Reference Kori and Kunihiro1997).

LH% is most often observed at the ends of utterances and typically expresses a question. However, as mentioned in Section 3.1, LH% does not always convey the meaning of a question. Uemura (Reference Uemura1989) summarizes the functions of this BPM as an expression of intimacy or a friendly attitude toward the listener.

HL% is a rise–fall BPM in which the beginning of the rise is at the onset of the phrase-final mora with the peak at the end of the rise aligned with the middle of the mora (close to its onset). After the rise, F0 falls at the end of the mora, and its duration is lengthened. The function of HL% is akin to H%: it imparts a prominence to the phrase that the BPM is attached to. In their perception study, Venditti, Maeda, and van Santen (Reference Venditti, Maeda and van Santen1998) showed that HL% is perceived by the listener as explanatory and emphatic, and it is judged to signal continuation. Citing this study, Venditti, Maekawa, and Beckman (Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008) summarize the functions of HL% by stating that listeners expect speakers to use HL% when they are explaining a certain point and want to focus attention on a particular phrase in their explanation.

The choice of H% or HL% at least partly depends on speaking style and spontaneity. An analysis of the impression rating assigned to talks in the Corpus of Spontaneous Japanese (CSJ, Maekawa Reference Maekawa2003) showed that the rate of H% positively and negatively correlates with speaking style and spontaneity, respectively, while the rate of HL% negatively and positively correlates with speaking style and spontaneity, respectively (Maekawa Reference Maekawa2006). In other words, listeners judge H% to be more formal and less spontaneous than HL%.

The F0 configuration of the former part of HLH% is akin to HL%; however, F0 rises again after the fall in the case of HLH%. The final mora is considerably lengthened. Venditti, Maekawa, and Beckman (Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008) suggest that HLH% may be particularly characteristic of infant-directed speech (IDS) as it can give a wheeling or cajoling quality to the utterance to which it is attached. Indeed, an analysis of Japanese infant-directed speech using the RIKEN Japanese Mother–Infant Conversation Corpus (Mazuka, Igarashi, and Nishikawa Reference Mazuka, Igarashi and Nishikawa2006), which contains IDS and adult-directed speech (ADS) in Japanese, revealed a higher occurrence of HLH% in IDS than in ADS (Igarashi et al. Reference Igarashi, Nishikawa, Tanaka and Mazuka2013). However, this type of BPM occurs much less frequently than other types, even in IDS. It occurs only twelve times in eight hours of IDS produced by twenty-one mothers. The low frequency of HLH% is also confirmed by the CSJ analysis. It occurs only fourteen times in the forty-five hour core portion of the CSJ (Venditti, Maekawa, and Beckman Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008).

The X-JToBI framework describes types of BPM other than the four types discussed here (H%, LH%, HL%, and HLH%). These types are operationally considered to be variants of the main types of BPMs in X-JToBI (for details, see Igarashi Reference Igarashi and Kubozono2015: Section 3.4).

The meanings of the BPMs described here are merely a first approximation. Unfortunately, no analysis of the meanings conveyed by BPMs is uncontroversial, and a comprehensive description is beyond the scope of this chapter. However, it is reasonable to point out here that the relations between forms and meanings in Tokyo Japanese BPMs seem to lack arbitrariness, and they fit comfortably into Gussenhoven’s theory of biological code concerning form-function relations based on the effects of the production process’s physiological properties on the speech signal (Gussenhoven Reference Gussenhoven2004) (for a full discussion, see Igarashi Reference Igarashi and Kubozono2015: 548–550).

3.3 How Many BPMs Does Japanese Have?

No consensus has emerged regarding the number of BPMs in Japanese. Moreover, few quantitative analyses have been conducted for BPMs. In the following paragraphs, we will discuss the extent to which researchers agree or disagree on the inventory of BPMs. Figure 9.4 is a schematic representation of the categorical boundaries of the rising BPMs.

Figure 9.4 Schematic representation showing correspondences among categories of the rising BPMs identified by different researchers

Most researchers distinguish two types of rises: information-seeking question rise (InfoQ rise) and prominence-leading rise (Prom rise) (Venditti, Maeda, and van Santen Reference Venditti, Maeda and van Santen1998). The InfoQ rise typically occurs in a question ending, whereas the Prom rise typically occurs when a speaker is making an insistent statement, such as Yameru! ‘I will definitely quit!’

In addition to these two rises, Kawakami (Reference Kawakami1963) and Uemura (Reference Uemura1989) distinguish the InfoQ rise from what we may call the “incredulity question rise” (IncreQ rise) (Venditti, Maeda, and van Santen Reference Venditti, Maeda and van Santen1998). The IncreQ rise is typically observed in a question where the speaker is expressing disbelief, such as Yameru? ‘Will you quit??’ However, Venditti, Maekawa, and Beckman (Reference Venditti, Maekawa, Beckman, Miyagawa and Saito2008) suggest that the InfoQ and IncreQ rises are the extreme endpoints of a continuum that includes many intermediate degrees of emphatic lengthening. The continuum between InfoQ and IncreQ is shown in Figure 9.5. They also point out that the gradient nature of the relation between the phonetic dimensions and the continuum of contrasting degrees of incredulity suggests an analysis analogous to the one proposed by Hirschberg and Ward (1992) for the uncertainty versus incredulity interpretations of the English rise–fall–rise contour (transcribed as L*+H L- H%).

Figure 9.5 An InfoQ-IncreQ continuum. Sō na no (so cop nmlz) ‘Is it so?’

Thus, defining the inventory of BPMs is far from trivial, and we cannot always rely solely on native speakers’ intuition as their intuition in the case of intonational contrasts is less sharp than in the case of lexical tone contrasts (Gussenhoven Reference Gussenhoven2004: 60). As exemplified in the InfoQ–IncreQ continuum, it is difficult to decide whether two phonetically different contours convey two categorically distinct intonational patterns or are merely phonetic variants of a single pattern. Researchers have recently been attempting to develop experimental methods for establishing intonational categories (for a discussion of the experimental approaches, see Gussenhoven Reference Gussenhoven2004: 62–70).

4.1 Double-layered Model

Prosodic phrasing is the grouping of words in an utterance. Although controversies still exist on the number of levels of prosodic phrasing in Japanese and their organization (for review, see Ishihara Reference Ishihara and Kubozono2015: 570; Igarashi Reference Igarashi and Kubozono2015: 527–529), this chapter, based on the X-JToBI framework, adopts the double-layered model, with the hierarchically organized Accentual Phrase (at the lower level) and Intonation Phrase (at the higher level).

4.2 Accentual Phrase

Japanese has two types of words: accented and unaccented. The former exhibit a pitch contour with a steep fall from high (H) to low (L) somewhere in the word while the latter show no such fall. In this chapter, the term pitch accent refers to this lexically specified pitch fall in the accented words. For example, a’me ‘rain’ has an accent on the initial syllable and is, therefore, an accented word, whereas ame ‘candy’ has no pitch accent and is, therefore, an unaccented word (the accented vowel is post-marked by an apostrophe). In addition to the presence or absence of pitch accent, its location in the word is also lexically specified; for example, na’mida ‘tear’ has an accent on the initial syllable, nomi’ya ‘pub’ on the second, and atama’ ‘head’ on the final.

An Accentual Phrase (AP) is defined as having a delimitative rise to high around the second mora and a subsequent gradual fall to low at the end of the phrase and as having at most one lexical pitch accent. While a typical AP comprises one lexical word plus a following particle or multiple particles (e.g. yama’ ga ‘mountain nom,’ niwa ni’ wa ‘garden loc top’), a single AP can often contain two or more lexical words. For instance, a noun with a genitive particle followed by another noun, such as Hiroshima no omiyage (Hiroshima gen souvenir) ‘a souvenir from Hiroshima,’ often forms a single AP. Moreover, a particle can form its own AP. For example, in a sequence of an accented noun and a following accented particle, such as nomi’ya ma’de (pub up-to) ‘to the pub,’ the noun and the particle are often merged into a single accented AP, with the accent of the particle being deleted, as in nomi’ya made. Deaccenting of particles is, however, not obligatory (see Igarashi Reference Igarashi and Kubozono2015: 538). When the accent is maintained, the particle forms its own AP.

The intonation contours of a single AP are described as a sequence of tones: an unaccented AP (1a), and an accented AP (1b).

1. (1) AP tones

   1. a.

      Unaccented AP:

      %L

      H-

      L%

   2. b.

      Accented AP:

      %L

      H-

      H*+L

      L%

“H*+L” stands for pitch accent, where the asterisk indicates the tone associated with the mora that is governed by the accented syllable. (An accent is assigned to a syllable, but the accentual H is associated with a mora.) Henceforth, the mora with which the H tone of H*+L is associated will be referred to as the accented mora.

L tones with a “%” symbol are called boundary tones, with %L and L% being the initial and final boundary tones, respectively. The low tone found at the beginning of the AP is sometimes called the initial lowering in some frameworks (e.g. Haraguchi Reference Haraguchi1977). The H-tone is called the phrasal high. The %L and H- function as starting and ending points, respectively, of the AP’s initial rise. The L% serves as the endpoint of a gradual pitch fall from H- in the case of an unaccented AP, or from L of H*+L in the case of an accented AP. In Figure 9.6, the unaccented adjective amai ‘sweet’ is combined with the following unaccented noun ame ‘candy’ into a single unaccented AP, where the %L H- L% pattern can be clearly observed.

Figure 9.6 An unaccented AP amai ame ‘sweet candy’ (left), and an accented AP uma’i ame ‘tasty candy’ (right)

4.3 AP Sequences

When a speaker produces fluent utterances of the sentences in (2), s/he groups the words into several APs. The syntactic branching structure is [A [N₁ N₂]], where A is an adjective and N is a noun followed by a particle but not [[A N₁] N₂]. The grouping of words into APs depends on the interaction of various factors such as word accentuation, syntactic branching, focus, and/or discourse structure (Venditti Reference Venditti and Jun2005). A typical prosodic phrasing of these sentences is shown in Figures 9.7–9.10.Footnote ² The rise at the end of these utterances is a LH% BPM, which was discussed in Section 3.

1. 1. a.

      Amai	jagaimo	no	nimono	wa	do’re	desu	ka?
      sweet	potato	gen	stew	top	which	cop	int
      ‘Which are the sweet stewed potatoes?’

   2. b.

      Uma’i	jagaimo	no	nimono	wa	do’re	desu	ka?
      tasty	potato	gen	stew	top	which	cop	int
      ‘Which are the tasty stewed potatoes?’

   3. c.

      Amai	zuwa’igani	no	nimono	wa	do’re	desu	ka?
      sweet	snow.crab	gen	stew	top	which	cop	int
      ‘Which is the sweet stewed snow crab?’

   4. d.

      Uma’i	zuwa’igani	no	nimono	wa	do’re	desu	ka?
      tasty	snow.crab	gen	stew	top	which	cop	int
      ‘Which is the tasty stewed snow crab?’

Figure 9.7 Phrasing at the AP level. Amai jagaimo no nimono wa do’re desu ka? (sweet potato gen stew top which cop.pol int) ‘Which are the sweet stewed potatoes?’

Figure 9.8 Uma’i jagaimo no nimono wa do’re desu ka? (tasty potato gen stew top which cop.pol int) ‘Which are the tasty stewed potatoes?’

Figure 9.9 Amai zuwa’igani no nimono wa do’re desu ka? (sweet snow_crab gen stew top which cop.pol int) ‘Which is the sweet stewed snow crab?’

Figure 9.10 Uma’i zuwa’igani no nimono wa do’re desu ka? (tasty snow_crab gen stew top which cop.pol int) ‘Which is the tasty stewed snow crab?’

When a right-branching syntactic boundary exists, an AP boundary is frequently inserted. Thus in (2), the adjective amai or uma’i forms a single AP. When no right-branching boundary intervenes, an unaccented word and the word that follows it tend to be conjoined into a single AP. In (2a–b), therefore, two NPs jagaimo no and nimono wa are conjoined into an AP. When an accented word is followed by another word, the latter often forms its own AP (Vance Reference Vance2008: Section 7.6) even if there is no right-branching boundary. Thus, in (2c–d), two NPs zuwa’igani no and nimono wa form separate APs. In all the examples in (2), the VP do’re desu ka constitutes a single AP. Thus, prosodic phrasing in (2) can be couched in the form of (3), where parentheses denote the AP boundaries.

1. (3) Prosodic phrasing in (2)

   1. a.

      (amai)

      (jagaimo no nimono wa)

      (do’re desu ka)

   2. b.

      (uma’i)

      (jagaimo no nimono wa)

      (do’re desu ka)

   3. c.

      (amai)

      (zuwa’igani no) (nimono wa)

      (do’re desu ka)

   4. d.

      (uma’i)

      (zuwa’igani no) (nimono wa)

      (do’re desu ka)

4.4 Downstep and Intonation Phrase

The Intonation Phrase (IP) is defined as the prosodic phrase immediately above the AP in the hierarchy within which pitch range is specified. At the beginning of each new IP, the speaker chooses a new pitch range that is independent of the specification of the preceding AP (Venditti Reference Venditti and Jun2005: 175). This process is called pitch reset. The pitch range specification of IPs is closely connected with a phonological process called downstep. Through this process, the pitch range of each AP is compressed when it follows an accented AP. Downstep is displayed in Figure 9.11. The peak of the third AP is significantly lower when preceded by an accented AP (right) than when preceded by an unaccented AP (left).

Figure 9.11 Downstep. An utterance without a downstep: Yubiwa o wasureta onna’ wa dare desu ka? (ring acc forgot woman top who cop.pol int) ‘Who is the woman that left the ring behind?’ (left), and an utterance with downstep on the third AP: Yubiwa o era’nda onna’ wa da’re desu ka? (ring acc chose woman top who cop.pol int) ‘Who is the woman that chose the ring?’ (right). The relevant portions of the F0 contours are marked by broken-line boxes. Dotted vertical lines symbolize AP boundaries.

When multiply accented APs form a single IP, downstep occurs iteratively, thus resulting in a staircase-like F0 contour, as demonstrated in Figure 9.12. However, in a sequence of four APs (in a syntactic phrase with a uniformly left-branching structure), as in Figure 9.11, the pitch range of the third AP is frequently expanded so that a staircase-like F0 is not observed. This effect is known as “rhythmic boost” (Kubozono Reference Kubozono1988 [Reference Kubozono and Haraguchi1993]: 220–223) and is shown in Figure 9.13, in which the pitch range of the third AP is larger than the preceding AP.

Figure 9.12 Successive downstep: Ao’i ie’ o era’nda onna’ wa da’re desu ka? (blue house acc chose woman top who cop.pol int) ‘Who is the woman that chose the blue house?’ without the rhythmic effect. Vertical lines indicate AP boundaries.

Figure 9.13 Successive downstep: Ao’i ie’ o era’nda onna’ wa da’re desu ka? (blue house acc chose woman top who cop.pol int) ‘Who is the woman that chose the blue house?’ with the rhythmic effect. Vertical lines indicate AP boundaries.

When the IP boundary is inserted, downstep is blocked at this boundary; that is, pitch reset occurs, and a new pitch range is specified to the IP. Various linguistic factors result in pitch reset at the IP boundary, including syntactic constituency and focus (Kawakami Reference Kawakami1957; Kori Reference Kori and Kunihiro1997; Ishihara Reference Ishihara2007; Kubozono Reference Kubozono and Ishihara2007b). The pitch ranges of post-focal words in accented APs are significantly reduced, a process called post-focal compression. Pitch reset and post-focal compression are shown in Figures 9.14 and 9.15, respectively. The prosodic phrasing in the utterances in these figures is shown in (4), in which curly brackets represent the boundaries of the IPs.

1. (4) The prosodic phrasing in the utterances in Figures 9.14 and 9.15. Focused words are capitalized.

   1. a. {(na’oya no) (ane ga)} {(nomi’ya de) (no’nda)} (Figure 9.14, left)

   2. b. {(na’oya no)} {(ANE GA nomi’ya de) (no’nda)} (Figure 9.14, right)

   3. c. {(na’oya no) (a’ni ga)} {(nomi’ya de) (no’nda)} (Figure 9.15, left)

   4. d. {(na’oya no)} {(A’NI GA) (nomi’ya de) (no’nda)} (Figure 9.15, right)

Figure 9.14 Pitch reset and post-focal compression: Na’oya no ane ga nomi’ya de no’nda (Naoya gen sister nom pub loc drank) ‘Naoya’s sister drank in the pub,’ without (left) and with (right) focus on the second unaccented AP ane ga. Focused words are capitalized.

Figure 9.15 Na’oya no a’ni ga nomi’ya de no’nda (Naoya gen brother nom pub loc drank) ‘Naoya’s brother drank in the pub,’ without (left) and with (right) focus on the second accented AP a’ni ga.

5.1 Intonational Contrasts in Phrasal Edges

Thus far, we have described the Japanese intonation system in terms of BPM and prosodic phrasing. They are both boundary-related prosodies, meaning that BPM is localized at the edges of prosodic phrases, which are signaled by a pitch movement localized at the phrasal edges. In other words, the locations where intonational tone occurs are not typically found in the middle of prosodic phrases in Japanese.

A marginal case is the presence or absence of downstep as discussed in Section 4.4. Downstep is manifested as the manipulation of the pitch range of the AP, which may be considered a global prosodic characteristic specific to the AP as a whole rather than a local event at the phrasal boundary. However, downstep is simultaneously a cue for the presence or absence of the IP boundary and thus may also be considered a prosodic event localized at the phrasal boundary.

On the other hand, more obvious intonational contrasts that can be described in terms of types of tones (such as H% versus HL%) or the presence or absence of tones (such as with versus without BPM) are undoubtedly localized in the phrasal edges. This restriction may be considered to be one of the typological characteristics of Japanese (for a full discussion regarding these putative typological characteristics, see Igarashi et al. Reference Igarashi, Nishikawa, Tanaka and Mazuka2013: 1286–1288 and Igarashi Reference Igarashi and Kubozono2015: 556–563).

5.2 Possible Intonational Contrasts at the Beginning of the Phrase

In most studies of Japanese intonation, it is assumed that intonationally contrastive tones are localized at the phrasal end but not its beginning. However, Kawakami (Reference Kawakami1956) observed variability in the timing of the initial rise of the utterance-initial AP: rise aligned earlier or later with respect to the phrasal boundary. He also proposed that the alignment varies according to the speaker’s “emotions.” In their experimental studies, Maekawa and Kitagawa (Reference Maekawa and Kitagawa2002) showed that the F0 contour at the beginning of the utterance varies significantly depending on the speaker’s attitude and intentions (“paralinguistic information,” in their terms), such as admiration, disappointment, and suspicion. For example, in an utterance produced with suspicion, the beginning of the initial rise is delayed considerably, thus yielding a long stretch of low F0 before the rise, and the contour exhibits a concave shape in the rising movement (BPM at the end of the phrase is what is referred to as IncreQ rise in Section 3.3). An example of strong suspicion is shown in Figure 9.16 (left).

Figure 9.16 Utterances produced with a neutral attitude (left) and with suspicion (right). Yamada-san de’su ka? (Yamada cop.pol int) ‘Is it Mr. Yamada?’ The boundaries of the first and final moras (/ya/ and /ka/, respectively) are marked to indicate their lengthening in the utterance with suspicion.

The delay of the initial rise, or the long low-pitched stretch at the beginning, may be considered a manifestation of intonational contrast. Further research is required to investigate whether this pattern is indeed contrastive or a mere variant of a single pattern. If the two contours shown in Figure 9.16 (i.e. non-delayed and delayed rise) are two contrastive intonational patterns, then we need to posit an intonational contrast at the beginning of the phrase as well. One possibility may be to posit a contrast in phrasal tones, such as a H- versus LH- contrast. Another may be to posit a %L vs. %LL contrast for the initial boundary tone.

5.3 Possible Intonational Contrasts in the Middle of the Phrase

Even if there is an intonational contrast at the beginning of the phrase, it is nevertheless localized at the phrasal edge. It must be noted, however, that the Japanese intonational contour is known to exhibit variability in the middle of the phrase.

In Section 4.2, we have seen that an unaccented phrase shows a gradual fall. The fall is accounted for by interpolation between H- and L% in the X-JToBI framework. However, Sugahara (Reference Sugahara2003), based on the inter-speaker variation found in her experimental results, claims that unaccented phrases may have no gradual fall at all and may show a high plateau instead. Indeed, in Japanese spontaneous speech, it is not difficult to find unaccented phrases without a gradual fall. An example of a high plateau in an unaccented AP is shown in Figure 9.17 (middle). Gradual pitch fall (Figure 9.17 [right]) could merely be regarded as one possible realization of the pitch pattern. Future research should address whether the two contours (with a gradual fall and with a high plateau) are indeed due to two categorically distinct intonation patterns. If they are contrastive, then Japanese intonation should be regarded as having intonational contrast in the middle of an AP.

Figure 9.17 Variations in the contour of unaccented APs Ore no mondai da (I gen problem cop) ‘It’s my problem.’ An ordinal contour (left), a contour that would be accounted for by tone spreading (middle), and a rising contour without an apparent tone target (right).

In addition to a high plateau (Figure 9.17 [middle]), a gradual rise without any turning point in the contour is also observed in an unaccented AP. Figure 9.17 (right) exemplifies an F0 rise from the beginning of the unaccented AP to the final mora, without apparent targets for H- and L% (although F0 rises throughout the utterance, it is interpreted as a statement rather than a question). If this contour results from the deletion of H-, then we need to define the contrast between L% and H% for the final boundary tone (not for BPM), in addition to the tone deletion rule for H- (thus, the contour may be interpreted as a combination of the initial boundary tone %L and a final boundary tone H%). In contrast, if the contour results from the dislocation (or delay) of H-, we must posit a tone deletion rule for L% and a rule modifying the alignment of H- (thus, the contour may be interpreted as a combination of the initial boundary tone %L and the phrasal H- without a final boundary tone). In any case, given that the contour is contrastive, we may need to assume an intonational contrast in the middle of the phrase, although these putative contrasts can still be regarded as those at the phrasal edges since edge tones such as phrasal high and boundary tones are involved in the contrasts.

Intonational contrasts existing in the middle of an AP, if any, have rarely been investigated. The reason for the scarcity of research can, in my view, be attributed to the low flexibility in variability in pitch contours in the middle of an AP. The middle of an AP in Japanese is allocated to lexical tone contrast, which is manifested as the location of a sharp pitch fall, if any. To preserve lexical tone contrasts, it is impossible to delete a sharp fall in the case of an accented AP or to implement an additional sharp fall in the case of an unaccented AP. In general, it is not usual for intonational effects to cause the neutralization of lexical tone contrasts in Japanese (Kawakami Reference Kawakami1956).Footnote ³

However, there are regional dialects of Japanese without any lexically contrastive tones; these are generally called “accentless dialects.” They are expected to show more intonational contrasts as there are no restrictions imposed by lexical tones. Indeed, recent studies reveal that accentless dialects show more flexibility in varying the contours that may be involved in intonational contrasts (Maekawa Reference Maekawa1997b; Igarashi Reference Igarashi and Jun2014). Some of the contrasts occur in the middle of APs, as we will see in the following subsection.

5.4 Intonational Contrasts in the Accentless Dialects

Accentless dialects are widely scattered in non-contiguous areas of the Japanese archipelago. The high variability in the pitch contour of the accentless dialects is reported by Maekawa (Reference Maekawa1994, Reference Maekawa1999) for the Kumamoto dialect spoken in Kyūshū in which the AP basically has a rise–fall contour, %L H- L%. However, this rise–fall shape shows considerable variability. First, the alignment of H- varies. As shown in Figure 9.18a, H- can appear on virtually any syllable within the AP while in Standard Japanese it is generally aligned with the second mora. This unstable H- is called a “wandering high” by Maekawa (Reference Maekawa1994). Second, a high plateau can be observed between the rise and fall in Figure 9.18b. Maekawa (Reference Maekawa1994) accounts for this plateau by assuming the spreading of H-. Finally, the fall at the end of the phrase can be unrealized as shown in Figure 9.18c. Henceforth, these three phenomena will be called H wandering, H spreading, and L deletion, respectively.

Figure 9.18 Variations in pitch contours in AP in the Kumamoto dialect

Igarashi (Reference Igarashi and Jun2014) demonstrated that the Koriyama dialect spoken in Fukushima Prefecture also shows H wandering. Figure 9.19 illustrates the contours of six tokens of the same wh-question ‘What do you see?’ produced by a single speaker. Each utterance comprises a single AP, and the rise at the end of the AP is due to a BPM: %L H- L% LH%. We see that the location of H- in the AP varies from one token to another.

Figure 9.19 H wandering in the Koriyama dialect: Nani ga mien da-i? (what nom visible.nmlz cop-sfp) ‘What do you see?,’ produced by a female speaker born in 1957. Six tokens normalized along the temporal scale are overlaid.

H spreading can be observed in other accentless dialects. The two panels in Figure 9.20 contrast utterances without (left) and with (right) H spreading in the same wh-question produced by a single speaker in the Imaichi dialect spoken in the Northern Kanto district (Igarashi Reference Igarashi and Jun2014: 482). Both utterances comprise a single AP with a BPM (LH%) at its end. The AP in this dialect also has a rise–fall contour (%L H- L%), although the rise tends to be more concave than that in the Kumamoto dialect. In the utterance in the top panel, F0 continuously rises from the beginning of the AP to around the middle of the third word and then falls toward the end of the AP. In the utterance in the bottom panel, a sharp initial rise ends around the beginning of the second word, followed by a high (slightly rising) plateau that persists until approximately the middle of the final word.

Figure 9.20 Utterances without (top) and with (bottom) H spreading in the Imaichi dialect: Doko no narazumono ni nagur-are-ta? (where gen gang dat punch-pass-pst) ‘What gang were you punched by?,’ produced by a female speaker born in 1984.

L deletion is also observed in the Imaichi dialect (Igarashi Reference Igarashi and Jun2014: 482). Figure 9.21 shows two tokens of the same wh-question produced by the same speaker. Both utterances comprise two APs with a BPM at the end of the second AP. They differ, however, in that the fall seen in the second AP in the left panel is not found in that in the right panel.

Figure 9.21 Utterances without (top) and with (bottom) L deletion in the Imaichi dialect: Odawara no dare ni nagur-are-ta? (Odawara gen who dat punch-pass-pst) ‘Who in Odawara were you punched by?,’ produced by a female speaker born in 1984.

As the observed variability in the contours is much more obvious here than in Tokyo Japanese, especially in the case of L deletion, it seems reasonable to assume that they are not merely phonetic variants of a single intonational category but are involved in intonational contrasts, conveying distinct pragmatic meanings. Issues concerning meaning differences in these contours are addressed in Maekawa’s (Reference Maekawa1999) perception study of the Kumamoto dialect in which differing H- alignments affect the perceived politeness of the utterance, with a later alignment yielding more “polite” judgments. The utterance with the L deletion in combination with the latest H- alignment led to most “polite” judgments. Although further research is necessary to establish categorically distinct intonational patterns in the accentless dialects, it is suggested that intonational contrasts are involved in the different contours under investigation. This in turn suggests that more intonational contrasts exist in the accentless dialects than in Tokyo Japanese.

2.1 Morphophonological Characteristics

In the literature of traditional Japanese grammar, mimetics have been grouped together with native Japanese words (cf. Kageyama and Saito Reference Kageyama, Saito, Kageyama and Kishimoto2016). In recent literature, however, it is more common to treat mimetics separately from Japanese native words, following McCawley (Reference McCawley1968), who treats Japanese vocabulary as constituting four types of lexical strata: native Japanese words, mimetics, Sino-Japanese words, and foreign words (loanwords from Western languages). This stratification is primarily based on phonological characteristics.

However, there are other reasons to suggest mimetics should be distinguished from native words. For example, mimetics have distinctive morphophonological forms. Akita (Reference Akita2009: 107–109) shows that mimetics fit into one of the fifteen “morphophonological templates,” arguing that the templates can be used to differentiate mimetics from non-mimetic words. The templates cover information on accent, root types (mono- versus bi-moraic) and morpheme types (reduplicated, suffixed). Table 10.1 shows the templates in square brackets, preceded by examples of mimetics.

In the table, the representations are adjusted from Akita’s original. First, the templates are simplified by removing the pitch fall symbols (^). An acute accent is placed on some reduplicated mimetics (e.g. gúigui ‘jerking’), thus contrasting accented and the unaccented forms. Second, the templates are rearranged from Akita’s root-type-based organization to morpheme-type-based, grouping mimetics into three major morphological types: (I) reduplicated, (II) non-reduplicated, and (III) the CVCCVri forms (C=consonant, V=vowel).

Reduplicated mimetics have reduplicated roots, indicated by the presence of the morpheme boundary, which are represented by the hyphenation in the template (e.g. CVV-CVV). While many reduplicated mimetics are accented (on the initial vowel) as in (a–d), some are unaccented as in (e–g). The same sequence may be accented or unaccented, in which case, there is usually a contrast in meaning (e.g. gátagata ‘rattling sound’ versus gatagata ‘rough (surface)’).

Non-reduplicated mimetics, alternatively called “suffixed mimetics,” consist of a root with a suffix: N (=mora nasal), Q (=mora obstruent), V (=Vowel), or ri.Footnote ³ As indicated in the parentheses following the examples in the table, mimetics in this group must be marked by the quotative particle to when they appear in a sentence (cf. (3)).

The CVCCVri form, sometimes labeled “emphatic” or “ri-suffixed,” is similar to the reduplicated form in that it consists of four moras, but the internal structure is distinct: in the CVCCVri form, the accent falls on the last vowel before ri, and the consonant following the initial vowel is moraic (N or Q) as in jiNwári ‘warmly moved.’

One of the most interesting characteristics of mimetics is that many denote an event or state (Kita Reference Kita1997), whose function is normally carried out by verbs. Like verbs, mimetics express aspect (Hamano Reference Hamano1998) but unlike verbs, mimetics indicate aspect by their overall morphological form rather than by means of a particular aspectual morpheme.

While there is no coherent form–aspect correspondence in the CVCCVri form, there is a regular correspondence between the form and the aspectual meaning in both the reduplicated and non-reduplicated forms. Reduplicated mimetics, accented or non-accented, express unboundedness, which is characterized by various terms depending on the situation, such as: iterativity (pókapoka ‘hitting repeatedly’), continuity or durativity (kírakira ‘glittering continuously’), and atelicity (zarazara ‘being coarse’). By contrast, non-reduplicated mimetics express boundedness, which may be characterized by terms such as: brevity (kaa ‘a short cry of a crow’), completion of one cycle (pokaQ ‘hitting once’), punctuality (baN ‘a bang’), and telicity (koroQ ‘(someone) dies’). Moreover, the repetition of the form with a short phonological break iconically depicts the number of occurrences of the event. For example, toN expresses the sound of ‘a knock,’ but toN toN expresses ‘two knocks,’ or toN toN toN ‘three knocks,’ and so forth. Furthermore, if the form contains a prolonged vowel, it represents a prolonged duration. For instance, while gyuQ expresses a brief contracting motion of a hand, gyuuuuuuQ expresses that this contraction is maintained for a longer period. Most CVCCVri forms are aspectually non-specific. Although some may express boundedness if they have non-reduplicated counterparts, as in paQkuri ‘split open’ (cf. pakuQ ‘pop open’), most express meanings other than aspect proper, such as degree (meQkiri ‘considerably’) and quantity (e.g. taQpuri ‘a lot’).

2.2 Semantic Characteristics

According to Hinton, Nichols, and Ohala (Reference Hinton, Nichols and Ohala1994: 1), sound-symbolism refers to “the direct linkage between sound and meaning.” As convincingly shown in Hamano (Reference Hamano1998), Japanese mimetics are a quintessential example. For instance, in a CV-root-based mimetic, palatalization means “childishness” or “excessive energy,” or the word initial p expresses the meaning of “taut surface; explosive movement” (p. 99), or the final -ri of a CVCV-root-based mimetic expresses “quiet ending” (p. 174). Japanese mimetics also display what Hinton, Nichols, and Ohala call “imitative sound-symbolism” wherein the word itself mimes the meaning, as with “onomatopoeic words and phrases representing environmental sounds” (p. 3), but the entire system goes far beyond the category of onomatopoeia.

In traditional Japanese grammar, mimetics are frequently classed into semantic groups (cf. Kindaichi Reference Kindaichi and Asano1978). The most common grouping is a tripartite classification of giongo ‘phonomimes,’ gitaigo ‘phenomimes,’ and gijōgo ‘psychomimes’ (translations due to Martin Reference Martin1975: 1025). This classification is adopted in many recent linguistic studies.

1. 1. a. Phonomimes (giongo):

      Animal cry:	kórokoro	‘croaking of a frog’
      Human voice:	géragera	‘guffawing’
      Sound:	bashaQ	‘sound of a splash’
      	páchipachi	‘popping sound’

   2. b. Phenomimes (gitaigo)

      Manner:	hírahira	‘fluttering’
      	gúruguru	‘going round and round’
      Condition:	kushakusha	‘being all crumpled’
      	betobeto	‘sticky’

   3. c. Psychomimes (gijōgo)

      Bodily sensation:	múzumuzu	‘itchy’
      	chíkuchiku	‘prickling’
      Psychological state:	íraira	‘irritatedly’
      	gaQkari	‘feeling disappointed’

Phonomimes cover mimetics expressing an animal cry, a human voice, and sounds more generally. Japanese abounds in mimetics representing environmental sounds; for example, bashaQ ‘sound of a splash,’ kátakata ‘rattling sound,’ páchipachi ‘popping sound.’ Phenomimes cover expressions of manners (dynamic events) and conditions (states). Japanese has a variety of mimetics expressing manner of motion; for example, pyókopyoko ‘hopping,’ gúruguru ‘spinning,’ and zúruzuru ‘sliding.’ Japanese is also rich in mimetics expressing conditions that rely on tactility; for example, betobeto ‘sticky,’ tsurutsuru ‘smooth’ (Kindaichi Reference Kindaichi and Asano1978). Psychomimes cover two groups of mimetics: those expressing bodily sensations (e.g. hírihiri ‘smarting pain,’ múkamuka ‘feel nauseous’) and those expressing a psychological state (e.g. íraira ‘irritatedly’). As implied by the translations, mimetics usually have detailed meanings (e.g. yóchiyochi ‘toddling’). Because of this, they are sometimes characterized as “hyponyms” and contrasted with verbs which tend to have a more neutral meaning (aruku ‘walk’). The three types with detailed meanings in (1) constitute the semantic prototypes of mimetics.

It is essential to recognize three points about these semantic categories. First, the relationship between a mimetic and its categorical affiliation is not necessarily one-to-one, as mimetics may simultaneously express what is gathered through different senses. For instance, kórokoro can be categorized as a phonomime or a phenomime, as it can simultaneously express what is sensed from both audition (a light sound an object emits as it rolls) and vision (the manner of a small object rolling). This sharply contrasts with a non-mimetic word like mawaru ‘roll’; this vision-based descriptive word leaves the sound the object emits unexpressed, even if it has emitted a sound. Second, many mimetics extend their meanings across (sub-)categories. For instance, mimetics expressing bodily sensation are often extended to denote a psychological state, as in dókidoki, which can mean ‘heart pounding’ (sensation) or ‘feeling nervous’ (psych.), or a phonomimic sense is extended to a phenomimic sense, as in káchikachi, which can mean ‘click click’ (sound) or ‘being very hard’ (condition) (see Akita Reference Akita2013 for more on this topic). Third, a small group of “demimeticized” (Akita and Usuki Reference Akita and Usuki2016: 255) mimetics resides outside the tripartite classification in (1). Many take the CVCCVri form, expressing concepts such as quantity taQpuri ‘a lot,’ degree meQkiri ‘considerably,’ or pace yuQkuri ‘slowly.’ It is important to make note of their presence when making a generalization about mimetics in Japanese; they do not fit into the semantic prototypes in (1), even though they are mimetics in terms of their morphophonological forms.

2.3 Lexical Category

Among the forms introduced in Table 10.1, there is a major division in the pattern of how the mimetics behave in a morphosyntactic context (there are exceptions, notably CVCCVri),Footnote ⁴ with each group associated with a different lexical category.

The unaccented reduplicated mimetics (e.g. tsurutsuru ‘being slippery’) occur in some of the same environments as nouns [N] and adjectival nouns [AN], although semantically, they are adjectival in that they express a state.

1. 1. a.

      Kore	wa	tsurutsuru	(/ki/gōka)	da.		[N/AN]
      this	top	mim	wood/gorgeous	cop.npst
      ‘This is slippery (/wood/gorgeous).’

   2. b.

      Tsurutsuru (/gōka/	mizu)	ni	natta.			[N/AN]
      	water	cop.adv	became
      ‘It became smooth (gorgeous/water)’

   3. c.

      tsurutsuru (/ki)	no	yuka	(cf. ?tsurutsuru na/*ki na yuka)			[N]
      	cop.att	floor
      ‘slippery (/wooden) floor’

   4. d.

      gōka	na	yuka	(cf. gōka *no yuka)			[AN]
      gorgeous	cop.att	floor
      ‘gorgeous floor’

   5. e.

      Yoru	made	sono	tsurutsuru	ga	tsuzuita.	[N]
      night	till	that	mim	nom	continued
      ‘That smoothness continued till night (after brushing my teeth with the toothpaste).’

In the predicate position, the unaccented reduplicated mimetic is accompanied by the copula da as in (2a). This follows the pattern of N such as ki ‘tree’ and AN such as gōka ‘gorgeous,’ both of which require da. Similarly, as a complement of naru ‘become,’ the mimetic requires ni, following the pattern of both N and AN (see (2b)). In contrast, the mimetic in the prenominal position follows the pattern of nouns, requiring no to modify the head noun (see (2c)). The AN’s pattern, which usually requires na (see (2d)), seems less acceptable, though an Internet search indicates the form with na is also used, albeit much less frequently (see Uehara Reference Uehara1998 for the similar characteristic displayed by non-mimetic ANs). Further, the mimetic can be preceded by the demonstrative sono, following the pattern of nouns (e.g. sono hon ‘that book’) (see (2e)), but many ANs cannot be preceded by a demonstrative (*sono gōka ‘that gorgeous’). This distribution suggests that, overall, unaccented reduplicated mimetics more closely follow the morphosyntactic patterns of nouns.

The remaining large majority (accented reduplicated, non-reduplicated and CVCCVri forms) are usually classed as adverbs, as they syntactically modify the verb or adjective that co-occurs with them in the sentence. As adverbs, mimetics appear marked by the quotative particle to, which is obligatory for non-reduplicated forms (3a) but syntactically optional for the accented reduplicated (3b) and the CVCCVri forms (3c), as indicated by the parentheses around to (Tamori and Schourup Reference Tamori and Schourup1999: 65–68).

1. 1. a.

      Kodomo	ga	nikoQ	to	waratta.	[Non-reduplicated]	[Adv.]
      child	nom	mim	quot	laughed
      ‘The child laughed with a smile.’

   2. b.

      Kodomo	ga	níkoniko	(to)	waratta.	[Reduplicated, accented] [Adv.]
      		mim	quot
      ‘The child laughed smilingly.’

   3. c.

      Ocha	o	yuQkuri	(to)	nonda.	[CVCCVri]	[Adv.]
      tea	acc	mim	quot	drank
      ‘I drank tea slowly.’

Furthermore, these three forms can be used as verbs, as exemplified below; in such cases, the mimetics are followed by a light verb suru ‘Lit. do.’ These forms are usually called “mimetic verbs” with suru carrying the tense.Footnote ⁵

1. 1. a. Non-reduplicated: mukaQ to suru ‘get angry,’ shiN to suru ‘become quiet,’ pitaQ to suru ‘fit,’ karari to suru ‘dry’

   2. b. Reduplicated, accented: múkamuka suru ‘feel angry,’ píNpiN suru ‘be healthy,’ húsahusa suru ‘be affluent,’ pásapasa suru ‘dry’

   3. c. CVCCVri: gaQkari suru ‘get disappointed,’ hiQsori suru ‘be quiet,’ huNwari suru ‘be fluffy,’ gaQshiri suru ‘firm’

Beyond this major division, it is not easy to assign a category to mimetics. For one thing, it is unclear whether the mimetics in (4) (without suru) should be assigned a category at all, when the entire sequence with suru is already classed as verb (see Sells Reference Sells and Iwasaki2017 for more on this topic). For another thing, mimetics can appear in a sentence without any verbs as in (5a–b) or in isolation as in (5c).

1. 1. a.

      Shachō	wa	uriage-zō	ni	niNmari.
      president	top	sales-increase	at	mim
      ‘The president (is) grinning (looking) at the increase in sales.’

   2. b.

      …	totsuzen	no	kōtsūjiko	no	shirase	ni	urouro.
      	sudden	gen	car.accident	gen	news	to	mim
      ‘(The mother got into) panic to hear the sudden news of (her son’s) car accident.’
      (Tamori and Schourup Reference Tamori and Schourup1999: 88)

   3. c. SutoN.

      mim

      ‘(Something) fell.’

The puzzle is whether the mimetic is an adverb in (5a–b), as Tamori and Schourup (Reference Tamori and Schourup1999: 88) assume (in their view, the clause-mate verb has simply been elided), or all the mimetics in (5) are verbal, since they are, arguably, the constituting element of the predicate, even though the mimetic has no morphosyntactic indications of a verbal element (i.e. there is neither an inflectional morpheme nor a verbal element such as suru or iu).

Given the heterogeneous patterns, it does not seem appropriate to accept the existence of a lexical category called “mimetics,” despite claims to the contrary in references to sound-symbolic words in other languages (cf. Newman Reference Newman1968). The inquiry must remain open as to whether mimetics have a lexical category at all, as questioned in Tsujimura (Reference Tsujimura2001). Alternatively, can mimetics be grouped into a traditional lexical category, such as noun and verb (or even a new hybrid category)? A final unresolved point of inquiry is how forms are related in the word-formation process.

3.1 Two-dimensional Semantic Representation

Kita’s work (Reference Kita1997) is an important cornerstone in the research history of the semantics of mimetics. It paved the way for a reoriented focus, moving from a morphophonological descriptive account of the word group – mimetics – to an explanatory investigation of why mimetics are “unique,” with a special focus on their semantics. His discussion illuminated various characteristics of mimetics hitherto taken for granted, trivialized, or simply neglected. These include: (i) the underlying principle of the form–meaning relationship of mimetics is iconicity; (ii) the semantics of mimetics have a direct appeal to sensory, motor, and affective aspects. Both characteristics clearly set the semantics of mimetics apart from those of prosaic words.

To capture the unique semantics of mimetics, Kita claims that it is necessary to posit what he calls a “two-dimensional semantic representation,” whereby the semantics of mimetics are distinguished from those of non-mimetic words. Namely, non-mimetic words belong to “the analytic dimension” in which “[a] thought or experience is represented as a proposition,” whereas mimetics belong to “the affecto-imagistic dimension” where “different facets of an experience are represented [which] include the affective, emotive, and perceptual activation in an experience” (p. 387). To defend the two-dimensional model, Kita argues that “the semantics of a mimetic and that of other parts of a sentence are not fully integrated with each other despite the fact that they are syntactically integrated” (p. 388).

While Kita’s contribution cannot be emphasized enough, his arguments are not problem-free. First, Kita’s proposal is a direct challenge to Jackendoff’s “The Conceptual Structure Hypothesis,” which states: “There is a single level of mental representation, conceptual structure, at which linguistic, sensory and motor information are compatible” (Reference Jackendoff1983: 17, emphasis in original). Kita simply says this “single level of mental representation” is only for non-mimetic words (Reference Kita1997: 380, 409), naturally assuming the conceptual structure is unable to cover the semantics of mimetics. He takes no notice of Jackendoff’s point that the conceptual structure is expected to “be rich enough in expressive power to deal with all things expressible by language … [and] to deal with the nature of all the other modalities of experience as well” (Reference Jackendoff1983: 17). This seems to imply that Jackendoff’s system is intended to be able to cover the semantics of mimetics. Kita neither explains why Jackendoff’s model is inadequate nor elucidates why having some unique semantic characteristics alone guarantees an independent dimension for mimetics. Second, Kita (Reference Kita1997) claims that, while the semantic representation of mimetic adverbs belongs to the affecto-imagistic dimension, that of “nominal mimetics” belongs to both the analytic and the affecto-imagistic dimension.Footnote ⁶ As Tsujimura (Reference Tsujimura2001: 416) aptly notes, the use of the categorial status (the notion belonging to the analytic dimension) as a key criterion for assigning mimetics to the affecto-imagistic dimension is irrational. Third, Kita makes a generalization about mimetics without encompassing all types of mimetics. As noted in Section 2.2, some mimetics express prosaic word-like meanings, deviating from the semantic prototypes. For these mimetics, the dimensional assignment leads to a clash. For instance, taQpuri ‘a lot’ must belong to the affecto-imagistic dimension since it is a mimetic, but at the same time, it must belong to the analytic dimension since it is a quantifier. Fourth and finally, when there is a certain linguistic behavioral difference between mimetic and non-mimetic words, Kita automatically attributes it to the dimensional difference, without exploring other possible causes. Examples include a phenomenon of negation, a semantic redundancy, and certain selectional restrictions, as elaborated below.

3.2 Linguistic Evidence of the Need to Separate the Dimension for Mimetics

5.1 Syntactic Condition

The relevance of the marking distinction to syntactic realization of mimetics is first discussed in Tamori (Reference Tamori1980). He notes to-marking is usually optional for reduplicated mimetics in normal clause-internal position, such as pakupaku ‘munch-munch’ (cf. (14a)), but it becomes obligatory at the postposed position (cf. (14b)) and a preferred choice at the preposed position (cf. (14c)).Footnote ¹¹

1. 1. a.

      Jon	ga	pakupaku	to/Ø	pan	o	tabeta.
      	nom	mim	quot	bread	acc	ate
      ‘John ate the bread, with a munch-munch.’

   2. b.

      Jon	ga	pan	o	tabeta,	pakupaku	to (*Ø).
      	nom	bread	acc	ate	mim	quot
      ‘John ate the bread, with a munch-munch.’

   3. c.

      Pakupaku	to (??Ø)	Jon	ga	pan	o	tabeta.
      mim	quot		nom	bread	acc	ate
      ‘With a munch-munch John ate the bread.’

      (adapted from Tamori Reference Tamori1980: 165)

While Tamori (Reference Tamori1980) pioneers in relating the mimetic’s marking distinction to the syntactic position, his discussion focuses on the clause-external positions, leaving the clause-internal positions untreated.

Following up on the topic, elsewhere I examine literary texts (309 tokens) to investigate how reduplicated mimetics distribute within a clause. I find a more dominant pattern of Ø-marked forms occupying the immediately preverbal position (151/187=81%) than to-marked forms (58/122 = 48%) (Toratani Reference Toratani2006).Footnote ¹² Although the results reveal a tendency, a critical question remains unconsidered. Is to- or Ø-marking obligatory for some mimetics at the immediately preverbal position, and if so, what types of mimetics are they? Section 5.2 partially addresses the issue, but for more on the topic, see Akita and Usuki (Reference Akita and Usuki2016) and Toratani (Reference Toratani and Iwasaki2017).

5.2 Semantic Condition

The literature suggests that the distinction of to/Ø-marking is related to different semantic factors. For instance, mimetics expressing abstract notions tend to be Ø-marked (e.g. frequency: chokuchoku-Ø iku ‘frequently go,’ degree: meQkiri-Ø heru ‘drastically reduces’ (Tamori and Schourup Reference Tamori and Schourup1999: 68–69). In addition, the marking can help identify which verb/adjective the mimetic is modifying (Toratani Reference Toratani2006: 419). For instance, (15) has a mimetic gorigori ‘sound of grating’ with two potential hosts: futoi ‘thick’ and mawasu ‘turn.’

1. 1. a.

      Mata	gorigori	to	futoi	kubi	o	mawasu.Footnote 13
      again	mim	quot	thick	neck	acc	rotate
      ‘Again, (he) rotated his thick neck gorigori (= with audible noises).’

   2. b.

      #Mata	gorigori	futoi	kubi	o	mawasu.
      again	mim	thick	neck	acc	rotate
      ‘Again, (he) rotated his noisily thick neck.’

If the mimetic is bare (i.e. Ø-marked) as in (15b), it favors a reading in which the element next to it is taken to be its host, yielding an infelicitous reading (as gorigori and futoi are semantically incompatible), but if it is to-marked as in (15a), it allows a reading whereby the verb in the distance is interpreted as its host, yielding the felicitous reading (as gorigori and mawaru are semantically compatible).

Beyond these, another much-discussed semantic factor is concerned with the mimetic-host predicate relation. Elsewhere, I argue that the distinction of to/Ø-marking indicates whether the mimetic co-occurs with a particular type of host predicate.

1. (16)

   To-marking indicates a “semantic mismatch between the predicted host and the host that actually co-occurs, whereas Ø-marking suggests a match”

   (Toratani Reference Toratani2006: 419).

This applies to the contrast observed in (17).

1. 1. a.

      Kare	wa	nikoniko	warat-te	iru.
      he	top	mim	laugh-conj	exist
      Lit. ‘He is laughing smilingly (He is smiling).’

   2. b.

      Kare	wa	nikoniko	to	akarui.
      he	top	mim	quot	cheerful
      Lit. ‘He is cheerful, smiling.’

Example (17b) has a bare (i.e. Ø-marked) mimetic. This is consistent with (16) because the mimetic–host relation is considered a match. Mimetics typically co-occur with a verb belonging to the mimetic’s superordinate category. Warau ‘laugh’ is considered to belong to the superordinate category of the mimetic as it subsumes the meaning of the mimetic nikoniko ‘smilingly,’ thereby qualifying as the predicted host of the mimetic. Example (17b) has a to-marked mimetic. This is also consistent with (16). Here, the mimetic–host relation is considered a mismatch, as the host akarui ‘cheerful’ does not belong to the superordinate category of the mimetic; naturally it is an “atypical host,” or an unpredicted host.

Although the marking distinction in (16) seems intuitively reasonable, the distinction turns out to be less straightforward. Example (16) suffers from a vague usage of the term “predicted host,” which, in turn, builds on an unexplained term “typical host.” Without clear definitions, the task of prediction concerning to/Ø marking is practically impossible.

That said, if the criteria can be set, the empirical validity of (16) can readily be tested. For the sake of argument and for present purposes, I searched literary texts and gathered 317 instances of reduplicated mimetics that occur at the immediately preverbal position; I then examined whether the marking of the mimetics conforms to the prediction made by (16), assuming “typical hosts” refer to the verbs/adjectives listed as part of the entry in a mimetic dictionary; for example, mawaru ‘turn’ is the typical host for the mimetic kurukuru ‘roll-roll.’Footnote ¹⁴ If (16) is tenable, we should expect two types of reduplicated mimetics: Ø-marked mimetics co-occurring with their typical hosts, and to-marked mimetics co-occurring with their typical hosts. This turns out not to be the case, however. When the two variables (the marking and the host) have two values each (to/Ø, typical/atypical, respectively), we get four logical combinatory possibilities. As Figure 10.1 shows, all combinations are found in my examples (the number represents token frequency or the ratio to the total).

Figure 10.1 Distribution of to/Ø-marked mimetics

Examples for each combination are shown as follows:Footnote ¹⁵ (a) Ø-marked mimetics with typical hosts (níkoniko warau ‘laugh smilingly,’ óioi naku ‘cry boohoo,’ púkapuka uku ‘float bobbingly’); (b) Ø-marked mimetics with atypical hosts (níkoniko tanoshii ‘fun smilingly’ [instead of warau ‘laugh’], nóronoro kurasu ‘live sluggishly’ [instead of susumu ‘move forward’], hírahira oyogu ‘(goldfish) swim wavingly’ [instead of chiru ‘scatter’]); (c) to-marked mimetics with atypical hosts (e.g. níkoniko to unazuku ‘nod smilingly’ [instead of warau ‘laugh’], kúnekune to aruku ‘walk meanderingly’ [instead of ugoku ‘move’], súrusuru to shimaru ‘close smoothly’ [instead of suberu ‘slide’]); and (d) to-marked mimetics with typical hosts (e.g. níyaniya to warau ‘laugh grinningly,’ súyasuya to neru ‘sleep peacefully,’ pyónpyon to tobihaneru ‘jump frisking around’). Of these, the combinations in (a) and (c) are expected from (16), but those in (b) and (d) are not expected.Footnote ¹⁶

The puzzle is why there are (b)- and (d)-type combinations at all.Footnote ¹⁷ In the case of the latter, some suggest the mimetic is to-marked when it is the focus-bearing element of the sentence (e.g. Toratani Reference Toratani2006: 419–420). In other words, the mimetic is the sole carrier of new information, with the verb carrying old information, introduced in the immediately preceding discourse (Lambrecht Reference Lambrecht1994). Even if all the instances in (d) can be assumed to be motivated by the notion of focus, the (b)-type combination, such as níkoniko tanoshii ‘fun smilingly,’ remains unexplained. At least, the dictionary-based determination of typicality of host does not work well for an account of the distribution of to/Ø-marked reduplicated mimetics. That is, their distribution is not as simple as (16) assumes. A closer examination of the contexts may shed more light on the issue.

3.1 Exclusion of Agents from Compounds of Event Predication

In the literature on lexical semantics it is generally agreed that how the arguments of a verb are realized in syntactic structure is generally predictable by the lexical semantic representation of the verb and/or the semantic roles borne by its arguments (Levin and Rappaport Hovav Reference Levin and Rappaport Hovav2005). The core linking rule holds that the agent argument of a transitive verb is realized as subject in syntactic structure, and the theme argument as direct object. The linking pattern of the canonical transitive verb kill, for example, is schematically shown in A of Figure 11.1. Only when implemented by a special operation like passive can a distorted argument realization as in B be accepted.

Figure 11.1 Normal and distorted argument realization

Generalizations on argument linking are formally stated in terms of argument structure configurations where the agent argument is situated in an external argument position (x) and the theme argument in an internal argument position (y).

1. (7) Argument structure

   1. a. transitive verbs: (x <y>)

   2. b. unaccusative verbs: (<y>)

   3. c. intransitive (unergative) verbs: (x < >)

Besides the argument structure of transitive verbs (7a), those of unaccusative verbs (non-volitional intransitive verbs that have only an internal argument) and unergative verbs (volitional intransitive verbs that are characterized as having only an external argument) are represented as in (7b) and (7c), respectively. For some researchers, an event argument is added on top of the thematic argument structures in (7).

Word-formation processes that are based on verbs as the head are generally sensitive to the distinction of external versus internal argument of the base verbs. As formulated by Lieber (Reference Lieber1983: 272), for example, English deverbal (synthetic) compounding applies either to the internal argument of a verb if it has one, as in letterwriting, or to an adjunct if it does not, as in sleepwalking. This principle correctly rules out compounds that involve a transitive verb and its external argument, as shown by the ungrammaticality of *student-writing of a letter (meaning ‘Students write a letter’). The exclusion of external arguments is in fact deemed a universal constraint that holds for noun incorporation in polysynthetic and other languages (e.g. Mithun Reference Mithun1984: 875).

Japanese N-V compounding complies with this universal restriction faithfully. Limiting attention to the argument realization in three major types of N-V compounds that function as predicates as summarized in Table 11.2, we find that there is no unequivocal instance of compounds consisting of a transitive verb and its subject.

Examples like kaeru-oyogi [frog-swim] ‘breaststroke’ and inu-kaki [dog-paddle] ‘dog-paddle’ that apparently involve an agent as the first member of a compound do not denote an action in which the named animals actually participate but only a particular manner of conventionalized action that is typically associated with these animals. On this interpretation, since the animal nouns do not represent an agent argument, they are irrelevant to the constraint on agent compounding.

In contrast to English, where compounds headed by verbs are hardly productive, Japanese abounds in productive compounding rules that apply to verbs and other predicates as the head. Among them, what Shibatani and Kageyama (Reference Shibatani and Kageyama1988) call post-syntactic compounding (Row III in Table 11.2) exhibits by far the highest productivity, freely applying to any combination of a VN (Verbal Noun; nouns that have argument structure and case but do not inflect for tense by themselves) and its internal argument. Corresponding to the clausal structure in (8a), for example, a post-syntactic compound is formed as in (8b), where the compound is enclosed by square brackets, with the colon (:) inside signaling a short phonological break. The same kind of phonological break is observed with other types of complex words using Sino-Japanese elements as in prefixed words (e.g. datsu : genpatsu ‘de-nuclear power generation,’ zen : daitōryō ‘ex-president’) and compound nouns (e.g. bōeki-gaisha : shachō ‘president of a trading company’).

1. 1. a.

      daitōryō	ga	Howaito-Hausu	de	pātī	o	kaisai	no	sai
      president	nom	White-House	loc	party	acc	host	gen	time
      ‘when president hosted/hosts a party at the White House’

   2. b.

      daitōryō	ga	Howaito-Hausu	de	[pātī : kaisai]	no	sai
      president	nom	White-House	loc	[party : host]	gen	time

In Japanese, the order of arguments and adjuncts may be altered by scrambling as in (9a), with the consequence that the subject argument shows up in front of the predicate ‘host.’

1. 1. a.

      Howaito-Hausu	de	pātī	o	daitōryō	ga	kaisai	no	sai
      White-House	loc	party	acc	president	nom	host	gen	time
      (Same meaning as (8a), with the word order changed by scrambling)

   2. b.

      *Howaito-Hausu	de	pātī	o	[daitōryō : kaisai]	no	sai
      White-House	loc	party	acc	[president : host]	gen	time

Despite the linear adjacency, compounding the agent (president) with the following predicate yields an illegitimate result in (9b). This indicates that post-syntactic compounding is constrained by the composition of argument structure rather than by surface order.

3.2 Compounding of External Arguments in Property Predication

Surprisingly, the systematic exclusion of an external argument explained in Section 3.1 is flatly contradicted by a fairly productive process of compounding an agent (external argument) with a Sino-Japanese VN. The total ungrammaticality of the post-syntactic compound in (9b) should thus be compared with the well-formedness of compounds like those in (10).

1. 1. a.

      [daitōryō : shusai]	no	pātī
      [president : host]	gen	party
      ‘a/the party hosted by the president’

   2. b.

      [ichiryū-kenchikuka : sekkei]	no	bijutsukan
      [first.class-architect : design]	gen	museum
      ‘a/the museum designed by a/the first-class architect’

In (10), transitive VNs shusai ‘host’ and sekkei ‘design’ are combined with the external arguments ‘president’ and ‘first-class architect,’ while leaving the internal arguments ‘party’ and ‘museum’ outside of the combinations. By using diagnoses of “lexical integrity” showing that words make up an integral unit whose internal structure cannot be manipulated by syntactic means, it is easily confirmed that these combinations of agent nouns and VNs truly constitute morphological compounds. For example, their internal structure cannot be interrupted by insertion of a focus particle between the two members. Likewise, deletion of only the noun members results in total ungrammaticality, as in *shusai no pātī ‘Lit. a/the party hosted’ and *sekkei no bijutsukan ‘Lit. a/the museum designed.’ Such genuine agent compounds thus exhibit a striking contrast to spurious agent-VN combinations due to particle ellipsis, such as fan : taibō (no shinsaku) ‘(a new work) awaited by fans,’ which permit deletion of the agent noun, as in taibō (no shinsaku) ‘a long-awaited (new work).’

Given that agent compounds like those in (10) are robust morphological constructs, how could the asymmetry in argument realization between them and the post-syntactic compounds in (8b) be accounted for? A first conceivable analysis will be to hypothesize that the predicates in agent compounds are passives derived by an invisible passive morpheme. This suggestion sounds plausible in view of the fact that comparable compounds in English use adjectival passives, as in architect-designed (homes). Under this analysis, the Japanese agent compounds do not violate the universal ban on external argument compounding, because the erstwhile agents – ‘president’ and ‘first-class architect’ in (10) – are demoted to adjuncts. The passive analysis, however, does not seem to be supported empirically. First, postulation of an invisible passive morpheme is dubious because VNs themselves cannot be passivized due to their noun morphology. Second, the agent nouns in (10) refuse paraphrases using an adjunct agent phrase with niyotte ‘by’; instead, they are most plausibly paraphrased with a nominative marking, as shown in (11).

1. 1. a.

      [daitōryō	{ga/*niyoru}	shusai]	no	pātī
      [president	nom/by	host]	gen	party

   2. b.

      [ichiryū- kenchikuka	{ga/*niyoru}	sekkei]	no	bijutsukan
      [first.class-architect	nom/*by	design]	gen	museum

While the passive analysis can be dismissed as infeasible, a second approach, suggested to me by Akira Watanabe (p.c. 2010), has some validity. Watanabe’s idea was that what we call agent compounds might be a special case of post-syntactic compounding applied to relative clause structure, where relativization of an object argument renders the transitive subject linearly adjacent to the predicate and hence compounding takes place. This analysis, it will be shown, is relevant only to one type of agent compound, that is, Class C in (12) below.

The two approaches mentioned above are based on the assumption that the relevant data involving external arguments constitute a uniform class with no internal variation. Careful scrutiny, however, reveals that the data are not monolithic but should be split into a few distinct classes. Kobayashi (Reference Kobayashi2004) gathered a large number of actually occurring examples from newspaper articles and attempted to confirm their compound status by applying certain diagnostics for lexical integrity such as modifiability of the noun members by external adjectives and deletability of the VN members on identity. Kageyama (Reference Kageyama2004) closely inspected Kobayashi’s as well as other data and proposed to tease apart three classes of expressions that share the formal appearance of an external argument combined with a transitive VN. The three classes (A to C) are exemplified in (12), with their essential features cataloged in Table 11.3.

1. 1. A. [Supirubāgu-kantoku : seisaku] no eiga ([Spielberg-director : produce] gen film) ‘a/the film(s) produced by Spielberg,’ [Andō-Tadao-shi : sekkei] no hoteru ([Ando-Tadao-Mr : design] gen hotel) ‘a/the hotel(s) designed by Architect Tadao Ando’

   2. B. [Warutā : shiki] no Koronbia-kōkyōgakudan ([Walter : conduct] gen Columbia-Symphony.Orchestra) ‘Columbia Symphony Orchestra conducted by Bruno Walter,’ [josei-untenshi : sōjū] no shinkansen ([female.driver : operate] gen Shinkansen) ‘a/the Shinkansen train operated by a female driver’

   3. C. [kishōchō : happyō] no taifū-jōhō ([meteorological.agency : release] gen typhoon-information) ‘typhoon information released by the Meteorological Agency,’ [Seki-Tsutomu-shi : hakken] no shin-shōwakusei ([Seki-Tsutomu : discover] gen new-asteroid) ‘a/the new asteroid discovered by Mr Tsutomu Seki’

Only Classes A and B qualify as full-fledged morphological compounds consisting of an external argument and a transitive VN. A typical diagnosis for testing the compound status is to see whether a given sequence is interruptible by a syntactic element. Classes A and B are equipped with the hallmark of lexical integrity, that is, non-interruptibility, while Class C appears to have a marginal status as compounds. Although the examples given in (12C) are collected from Japanese web pages, I conjecture that their acceptability will diverge among individual native speakers. Their questionable status can be ascribed to their irregular production by an aberrant operation. Recall that post-syntactic compounding rules out combinations of a transitive VP and an external argument, as shown by (13b).

1. 1. a.

      kishōchō	ga	[taifūjōhō : happyō]	chokugo
      meteorological.agency	nom	[typhoon.info : release]	right.after
      ‘right after the Meteorological Agency released typhoon information’

   2. b.

      *taifūjōhō	o	[kishōchō : happyō]	chokugo
      typhoon.info	acc	[meteorological.agency : announce]	right.after

However, if the accusative object is relativized, its object position becomes invisible (being occupied by a trace), as in (14).

1. (14)

   Kishōchō	ga	ti	happyō	no	taifūjōhō
   meteorological.agency	nom		release	gen	typhoon.info
   ‘the typhoon information released by the Meteorological Agency’

In this structure, the subject argument (Meteorological Agency) is linearly adjacent to the transitive VN ‘release’ (ignoring the trace of the relativized noun). Given this surface order, the condition on post-syntactic compounding may be relaxed (for some speakers) in such a way that it can work on linearly adjacent words to give rise to a compound-like unit. This rather irregular implementation of post- syntactic compounding, I suspect, accounts for the marginal status of Class C compounds. Their questionable status is confirmed by the fact that, unlike the full-fledged compounds of Classes A and B, Class C compounds are incapable of participating in an iterative application of compounding with the theme nouns they modify, as shown in (15).

1. 1. A:

      [[Supirubāgu-kantoku : seisaku]	bōken-eiga]	(OK as a compound)
      [[Spielberg-director : produce]	adventure-film]

   2. B:

      [[Warutā : shiki]	Koronbia-kōkyōgakudan]	(OK as a compound)
      [[Walter : conduct]	Columbia-Symphony.Orchestra]

   3. C:

      *[[kishōchō : happyō]	taifūjōhō]	(Bad as a compound)
      [[meteorological.agency : release]	typhoon.info]

3.3 Inherent versus Acquired Properties in Agent Compounding

Having delineated the range of relevant data, we will now grapple the vital question of why Classes A and B compounds are acceptable despite their ostensible violation of the universal constraint on external argument compounding. The key to the problem lies in the compatibility or incompatibility with spatiotemporal adverbials that designate a particular time or place when/where the event denoted by a compound takes/took place. This point is made clearer by comparing compounds of Classes A and B with Class C compounds. Observe first that Class C compounds, for those speakers who accept them at all, are fully compatible with such adverbials, as demonstrated in (16).

1. 1. a.

      gogo	san-ji	ni	[kishōchō : happyō]
      p.m.	three-o’clock	at	[meteorological.agency : release]
      no	taifūjōhō
      gen	typhoon.info
      ‘the typhoon information that the Meteorological Agency released at 3 p.m.’

   2. b.

      kyō	no	gikai	de	[yatō-giin : teishutsu]
      today	gen	Diet	loc	[opposition-dietmen : present]
      no	shitsumonsho
      gen	written.inquiry
      ‘the written inquiry the opposition party presented at today’s diet meeting’

The acceptability of (16) contrasts dramatically with the total incompatibility of such adverbials with compounds of Class A (17a) and Class B (17b).

1. 1. a.

      Sono	eiga	wa	(*2001-nen	ni	*Hariuddo	de)
      the	film	top	(*2001-year	in	Hollywood	loc)
      [Supirubāgu-kantoku : seisaku]				da.
      [Spielberg-director : produce]				cop
      ‘That film was produced by Spielberg in Hollywood in 2001.’

   2. b.

      (*1958-nen	ni	*Nyūyōku	no	sutajio	de)	[Warutā : shiki]
      (*1958-year	in	New.York	gen	studio	loc)	[Walter : conduct]
      no	Koronbia-kōkyōgakudan
      gen	Columbia-symphony.orchestra
      ‘the Columbia Symphony Orchestra conducted by Walter in 1958 in a studio of New York’

The disparity between (16) and (17) is exactly parallel to the one we saw in Section 2 between stage-level predication (as in My doctor is not available today) and individual-level predication (as in *My doctor is tall today). This is tantamount to saying that Class C compounds, compatible with spatiotemporal adverbials, are event predications describing the occurrence of a particular action or event along the time dimension, whereas the compounds of Classes A and B, which are incompatible with such adverbials, are property predications depicting a more or less constant quality attributed to the theme nouns they modify.

For example, (17a), with the adverbials left out, characterizes the particular film as one of superior quality on the basis of the pragmatic knowledge that Spielberg is a celebrated film director with outstanding expertise. Since, however, the quality of the film is a permanent attribute, the time and place adverbials that refer to the process of film-making are rendered irrelevant. Likewise, (17b) evaluates the performance of the Columbia Symphony Orchestra as having been excellent when it was under the baton of Bruno Walter, a conductor of eminent reputation in the early 1900s. Although the referential NPs do not necessarily refer to leading figures like Spielberg or Walter, they are indeed required to convey cogent information that helps to assign a noteworthy quality to the theme noun that the compound modifies. Nouns that do not convey information semantically or pragmatically sufficient to uniquely characterize the quality of a theme noun, such as ‘my son’ in (18a) or ‘a certain composer’ in (18b), are not suitable as the noun members in agent compounding.

1. 1. a.

      *[chōnan : seisaku]	no	ehon
      [first.son : create]	gen	picture.book

   2. b.

      *[bō-sakkyokuka : sakkyoku]	no	gasshō-kyoku
      [certain-composer : make]	gen	chorus-music

So far, it has been established that compounds of Class A and Class B are acceptable because of their semantic function as property predication. What distinguishes between the two classes then? They can be sharply differentiated from each other by the simple test of whether a given compound can fit into the copula construction ‘X (theme noun) wa COMPOUND da.’ Class A compounds comfortably fit into the copula construction, while Class B compounds do not.

1. (19) Class A compounds in the topic-copula construction

   1. a.

      Kono	eiga	wa	[Supirubāgu-kantoku : seisaku]	da.
      this	film	top	[Spielberg-director : produce]	cop
      ‘This film is one that was produced by Spielberg.’

   2. b.

      Kono	bijutsukan	wa	[Andoo-Tadao-shi : sekkei]	da.
      this	museum	top	[Ando-Tadao-Mr : design]	cop
      ‘This museum is one that was designed by Tadao Ando.’

1. (20) Class B compounds in the topic-copula construction

   1. a.

      *Kono	kōkyōgakudan	wa	[Warutā : shiki]	da.
      this	symphony.orchestra	top	[Walter : conduct]	cop
      ‘The symphony orchestra is one that was conducted by Walter.’

   2. b.

      *Kono	shinkansen	wa	[josei-untenshi : sōjū]	da.
      this	Shinkansen.train	top	[female-driver : operate]	cop
      ‘This Shinkansen train is one that is operated by a female driver.’

The discrepancy between (19) and (20) can be attributed to the archetypical function of the copula construction of the form “X wa Y da,” which serves to describe an inherent property of the theme noun realized as the topic.

As briefly noted in Section 2, Masuoka (Reference Masuoka and Masuoka2004) distinguishes two types of property: (i) property inherent in a category and (ii) property acquired from a past experience. The former refers to an inherent property, and the latter to an acquired or derived property. Since the copula construction prototypically denotes an inherent, classificatory property, the Class A compounds are construed as representing inherent properties of a topic noun. Strong evidence for this construal derives from the lexical meaning of VNs employed in these compounds. Specifically, the VNs qualified for Class A compounds, such as shusai ‘host,’ sekkei ‘design,’ and sakusei ‘create,’ are all accomplishment verbs that converge on the core meaning of production and creation. With the aid of the agent nouns, which mostly represent eminent figures in a particular field of activity, the VNs of creation and production assign an inborn quality to the topic phrase. The meaning of inborn properties is also activated by compounds like those in (21), where the noun members denote producing centers for artifacts, food, and other commodities.

1. 1. a.

      Kono	kagu	wa	[Itaria : chokuyunyū]	da.
      this	furniture	top	[Italy : direct.import]	cop
      ‘This furniture is imported directly from Italy.’

   2. b.

      Kono	ise-ebi	wa	[sanchi : chokusō]	da.
      this	lobster	top	[producing.district : direct-send]	cop
      ‘This lobster is sent directly from the producing center.’

By metonymic extension, a noun designating a famous place of production is naturally conceived of as its “creator” or agent. Notice that the VNs in (21) express ‘acquisition’ or ‘arrival’ of the goods, which are broadly equivalent to the lexical meaning of ‘production.’

The delineation of the semantic range of VNs in Class A compounds as having the meaning of production and creation strongly indicates that these compounds bear a classificatory function, assigning an inherent property to the topic phrase they are predicated of. By contrast, Class B compounds, which are unable to occur in the topic-copula construction, represent the kind of property that has been acquired through the action or event denoted by its modifying compound. Since past experiences that motivate the assignment of an acquired property are variegated, the VNs that may participate in this class of compounds exhibit diverse lexical meanings in diverse aspectual classes.

This section has shown that Masuoka’s distinction between inherent property and acquired property, which is not known in Western linguistics, is neatly reflected in the Japanese agent compounds of Class A and Class B. It remains to be seen how the distinction is formally represented and exploited in theories of event semantics. A plausible proposal is to attribute the difference of the two classes to different modules of grammar in which each of them is generated. Class A compounds denoting inborn properties are formed in the lexicon, whereas Class B compounds representing acquired properties are formed in syntax (Kageyama Reference Kageyama and Masuoka2006b). Telling evidence for this distinction is found in the availability or unavailability of the honorific prefix go-, which normally cannot be embedded inside lexical words. As predicted, Class B compounds in syntax permit this prefix on the VNs inside compounds, while Class A compounds in the lexicon do not.

1. 1. A:

      Kono	kyoku	wa	[Takada-Saburō-sensei : (*go-)sakkyoku]	desu.
      this	music	top	[Takada-Saburo-Mr : (*hon-)compose]	cop
      ‘This music was composed by Mr. Saburo Takada.’

   2. B:

      [Heika : go-hōmon]	no	chi
      [emperor : hon-visit]	gen	place
      ‘the place(s) visited by the emperor’

To recapitulate, agent compounding in Japanese is a robust rule of creating compounds that depict a characteristic property/quality of the theme noun they are predicated of by combining a transitive VN with its external argument. Given that agent compounding is allowed only in property predication, we must ask whether the convergence of its morphological peculiarity and its property-predication function is a mere coincidence or a logical consequence of some abstract principle relating them to each other. Probably it is not an accidental idiosyncrasy because there are many other phenomena in which property predication is associated with the violation of otherwise valid constraints in syntax and morphology, as reported by Kageyama (Reference Kageyama, Tsunoda and Kageyama2006a, Reference Kageyama2009). Those phenomena that look disorderly when viewed from event predication become orderly when regarded as belonging to the realm of property predication that is distinct from the more familiar realm of event predication.