1.1 Outline of the article

In §2, I discuss the realisation of N in Tōkyō Japanese (TJ) in further detail. Following this, §3 examines the phonological facts of TJ relevant to the paper, namely the syllable template of Japanese, nasalisation, assimilation of N, and tone assignment. §4 then presents the facts of Ōsaka Japanese (OJ), while §5 examines Kagoshima Japanese (KJ).

Turning then to the revision of N, §6 focuses on the previous phonological representations of N in various frameworks, focusing on the syllable & mora representation of Vance (1987, 2008) and the variable representation proposed by Yoshida S. (1996, 2003). §7 introduces the Strict CV framework (Lowenstamm 1996), where the mora is equivalent to a CV unit. Here, N is reframed for each dialect in turn. §8 then discusses the way that the proposed representations can derive tone-bearing units (TBUs) in a straightforward manner by referring to the status of the V position in the representation of N, building on work from Yoshida Y. (1999). The paper concludes in §9 by outlining future avenues of research.

1.2 A note on the annotation of N and other special, deficient or dependent moras

I follow the tradition that annotates the second position within a heavy syllable with <R> or <H> for the second half of a long vowel, <N> for the moraic nasal, <Q> for the first portion of a geminate and <J> or <i> for the offglide of a diphthong. These labels are used in many structuralist and generative works including, but not limited to, McCawley (1968), Vance (1987, 2008), Labrune (2012a, b); Tanaka (2013), Kubozono (2015a, b); and Kawahara (2015).¹

As a set, these are called special, dependent or deficient moras. This is derived from their lack of free occurrence and their inability to support a lexical pitch accent or high tone (c.f. 3.2). I adopt the label ‘deficient moras’ following Labrune (2012a, b). In Japanese native terminology, R, N and Q are called respectively the chō’on, the hatsuon and the sokuon, while the set is called tokushu haku or ‘special beats’. See Labrune (2012b) and Otake (2015) for further discussion of Japanese terminology. While J or ‘i’ is not always included in the set of deficient moras, I include it in this set as {ai, oi, ui} behave as true diphthongs where /i/ fails to support an accent or tone on the high front off-glide (cf. Kubozono 2015b).² From here forward, I use R, N, J for the moras in isolation and CVN, CVR, CVJ and CVQ to denote heavy syllables.

2.1 Phonetic studies on N

Phonetic studies have clarified the production of N in further detail. The majority focus on the issue of oral closure. Vance (1987: 34–35, 2008: 96ff) provides review of instrumental and acoustic studies which show that uvulo-dorsal closure in intervocalic and pre-pausal position is variable (Nakano 1969: 220, Ushijima & Hirose 1974, Aoki 1976: 204, Kawakami 1977, Uemura & Takada 1990). More recently, closure of the oral tract is discussed by Hashi (2000), Yamane & Gick (2010), Yamane (2013), Mizoguchi & Whalen (2014), Nogita & Yamane (2015), Hashi et al. (2016) and Kochetov (2018). Work from Hashi and colleagues (Hashi et al. 1998, 2016) in particular shows that uvular closure during production of final N is in fact often absent, with other possible realisations including a labial nasal or a nasalised vocoid. Based on the variability of N finally and medially, it is clear that final and intervocalic N cannot be defined as sounds produced with reliable closure of the oral tract in Tōkyō Japanese. Maekawa (2019) has also shown that based on MRI evidence, there is no clear place of articulation for N following /a/ and he has also claimed that tongue movement is connected to the preceding vocoid.

With regards to the pre-consonantal position, it is mostly clear that N assimilates at some point in production to the following consonant. Recent studies include an electro-palatography study from Kochetov (2018) and ultrasound investigations from Mizoguchi (2019) examining N assimilation in L1 and L2 speakers.

In addition to these studies, it is also clear that N nasalises a preceding vowel (Bloch 1950, Vance 1987, 2008, Kawahara 2015). The accusative form /hoN-o/ ‘book-ACC’ has the surface form [hõɰ̃o] or [hõõo]. It must note that not all sources transcribe or discuss this phenomenon. Nasalisation preceding N is often ignored in descriptive overviews of Japanese phonetics (e.g. Shirota 1993, Akamatsu 1997), and in some phonological works (Yoshida S. 1996, 2003; Yoshida Y. 1999). With regards to instrumental phonetic studies, vowel nasalisation is empirically verified for the sequence <eN> by Ushijima & Hirose (1974), who note anticipatory velum lowering in an electromyography study. Nasalisation is also noted in descriptive phonetic studies by Nakano (1969) and Campbell (1999). In particular, Nakano (1969) proposes that acoustic spectrograms of Japanese show a transition of oral vowel to nasalised vocoid to nasal vocoid in the production of <aN> which is comparable to the production of French nasal vowels. Unfortunately, clear data is not provided beyond a figure showing the average nasalance curve for the production of the sequence <aN> in Japanese and the nasal vowel [ã] in French. While a reviewer notes that nasalisation could be interpreted as a regular low level phonetic effect, we will see that nasalisation is not regular in other dialects like Ōsaka Japanese.⁷

In sum, N is generally characterised as a homorganic consonant, but is best framed as a glide or vowel which assimilates only pre-consonantally. N then triggers nasalisation in the preceding vowel in all contexts. What remains unclear is whether to accept the supposition that N is a consonant, or to consider the rarer argument that N could be a nasal vowel. To consider this issue, I look at phonological behaviour and patterning of N in further depth in the following sections. When we consider all phonological facts together, it is clear that N patterning as a consonantal segment is not a universal truth.

3.1 Vowel nasalisation and nasal assimilation

Let us consider the analysis of nasalisation and assimilation. An example of vowel nasalisation has been demonstrated above; the orthographic word <ほん> or <hon> is realised as [hõɰ̃] or [hõõ]‘book’. The domain of nasalisation is defined as the syllable (Vance 2008, Kawahara 2016), and Vance (2008) proposes that the feature [nasal] spreads to the preceding vowel within a syllable.

In terms of assimilation, N assimilates to a following consonant and is realised as [n], [m], or [ŋ] preceding an alveolar, bilabial or velar consonant as in (1).⁹ As early as McCawley (1965: 95) and in autosegmental terms since Ito (1987, 1989), it has been assumed that N in Japanese is specified only for nasality.¹⁰ In the analysis of Vance (2008), manner and stricture features spread to N. A representation of nasalisation and assimilations for the nominative hon-ga is shown in (2), which draws on the analysis of Vance (2008). If no consonant follows, Vance (2008) claims that a default dorso-uvular place feature is added to the specification of N, giving a nasal glide [ɰ̃] as in (3). A similar approach is taken by Labrune (2012a, b), though the representation of N is a mora with a C and V position, with the C specified only for [+nasal],

(2)	Spreading in [hõŋga] ‘book-NOM’

(3)	Representation of <hoN> as [hõɰ̃]11 ‘book’

Recall that N in final and medial position is a vowel for Nakano (1969), Yoshida Y. (1999) and Yoshida S. (2003). Here, /hoN/ ‘book’ is realised as [hoɯ̃] ‘book’ in formal or careful pronunciation. Nakano and Yoshida S. further propose that in informal pronunciation, this word is realised as [hoõ]. For Yoshida Y (1999) and Yoshida S. (2003), formal [ɯ̃] is simply the realisation of a nasal feature with the quality of the epenthetic vowel [ɯ]. This requires no use of a non-local place feature absent in the consonantal phonology. However, both Nakano (1969) and Yoshida S. (2003) point out that an informal and assimilated pronunciation is the one evidenced in general speech, and [ɯ̃] is formal or hypercorrect with optional addition of glottal closure giving [ɴ].¹² I revisit this analysis formally in §6.2. However, the authors purporting that N is a vowel in final and medial position do not note nasalisation in the preceding vowel. Yet if there is a nasalised vowel preceding N, and if it assimilates as suggested by Nakano and Yoshida S., could it instead be that a sequence VN is in fact a long nasal vowel? Let us turn to evidence from tone assignment and spreading which suggests this is indeed the case.

3.2 Pitch accent/tone assignment and syllable structure

I now consider the general patterns of accent or tone assignment in TJ and the deficiency of N as a tone-bearing unit (TBU). I begin by discussing the general facts of TJ tone in 3.2.1, followed by the discussion of deficient moras as poor TBUs in 3.2.2. I then focus on the paradoxical ability of N to carry marginal tone finally in 3.2.3 and in pre-accenting contexts in 3.2.4.

Here and throughout the rest of this article, I adopt the assumption that ‘pitch accent’ is better understood as autosegmental tone assignment, following Haraguchi (1977). This allows for linked and floating High (H) and Low (L) tones as well as for a straightforward account of tone spreading. In contrast to Haraguchi (1977) but following Yoshida Y. (1999), I presume that TJ does not encode L. L is not phonologically active or contrastively marked, so a low pitch here is understood to be the absence of tone specification. For alternative formulations of Japanese pitch accent, see Haraguchi (1977), Labrune (2012a, b) and Kawahara (2015), who use both a H and L. There is also the approach of Pierrehumbert & Beckman (1988), who approach Japanese tone and pitch accent utilising accentual and phrasal H tones and L boundary tones, focusing on intonation, with later work discussed by Igarashi (2015). Finally, there is also the work of Yoshida Y. (1999), who assumes that high tone is the manifestation of an accent with no use of low tone and no true H. I treat TJ henceforth as a privative H system, with floating and linked tones which may spread. See Hyman (2001) for discussion of privative and binary tone system in Bantu as well as Hyman (2009) for insightful criticism regarding the use of ‘pitch accent’, which I take on board.

3.3 Tone spreading in Tōkyō Japanese

In addition to high tone assignment, TJ also exhibits regressive spreading. High tone spreads regressively up to and excluding the initial mora, as in the word /kagamí/ ‘mirror’, realised as [kagámí]. In informal TJ, it is clear that heavy syllables can disrupt this pattern and CVR, CVJ and CVN syllables attract tone to the initial mora (Haraguchi 1977). This has been noted by descriptive linguistics since at least Hattori (1951) and within generative quarters, has been treated by Haraguchi (1977), and Tanaka (2013).¹⁷ Consider the below data from Haraguchi (1977, marked with superscript a), Tanaka (2013, marked with superscript b), and from my informants (marked with superscript c). Data in (15a) exemplifies the case of initial light syllables, (15b–d) the case of CVR, CVN and CVJ syllables, and (15e) shows the case of initial CVQ syllable. Crucially, note that CVQ syllables do not trigger spreading to the initial mora.

What is clear here is that CVN syllables pattern with CVR and CVJ syllables in initial position attract a high tone to the initial mora, while CVQ syllables do not exhibit this behaviour.

The behaviour of heavy syllables as tone attractors is also evidenced elsewhere such as the Owari dialect (see Ebata 2013, Youngberg 2017) and the Matsue dialect (Poppe 2015). Matsue Japanese exhibits a similar pattern to Tokyo Japanese. In Owari Japanese however, tone spreading ignores both the initial and pen-initial mora unless a CV: or CVN syllable intervenes. Here too, CVQ syllables do not trigger tone spreading to the initial mora. The pattern is exemplified in (16).¹⁸

I must make a brief aside to point out why I do not follow the ToBI analysis of these tone patterns, initiated by Pierrehumbert & Beckman (1988) and reviewed by Igarashi (2015). This framework does not use regressive tone spreading. Here, accented moras are marked with accentual H* and the pen-initial mora is marked with phrasal H-, with boundary %L associated to the initial mora unless it is a part of a heavy sonorant syllable (or here, CVR, CVJ or CVN). ‘Tone spreading’ is rather the interpolation of high pitch between the phrasal and accentual H. However, it is unclear how such an account could account for dialects of Japanese such as Owari, where the initial unaffected stretch of no high pitch is two moras, nor is it clear how to prevent the %L associating to the initial syllable in TJ. I believe that regressive High tone spreading with a protected window accounts more elegantly for the variation in weight-sensitive tone spreading systems, as in Tōkyō and Owari Japanese and others such as Matsue Japanese (Poppe 2015). I return to a full analysis later.

With regards to the above process, one first questions why CVN syllables pattern with CV: and not CVQ for tone spreading. Furthermore, why doesn’t a heavy CVQ syllable pattern with other heavy syllables? These are questions considered below in light of previously proposed representations.

3.4 N as a TBU and vocalic segment in Tōkyō Japanese?

In the above subsections, we have seen that N is not simply a coda consonant which fails to be a TBU. It is surely not a homorganic consonant in all contexts, and is a nasal vowel or glide pre-pausally and intervocalically. We have also seen that a CVN syllable contains a nasalised or nasal vowel. While it is true that N is not a lexical TBU, it can bear a tone word-finally in some contexts along with R and J (namely citation form unaccented words and pre-accentuation contexts). Finally, we have seen that at CVN syllables trigger high tone spreading along with CVR and CVJ syllables and not CVQ syllables. In all cases, CVN patterns with long vowels and diphthongs. If we take this behaviour seriously, it is possible that VN does not contain a nasalised vowel but rather is itself an underlying nasal vowel. I take this proposal forward from here. The essential facts thus far are summarised in Table 1.

Upon consideration of further dialect data, it is clear that N should have varying representations beyond what is appropriate for TJ.

4.1 Tone in Ōsaka

Let us first examine the facts regarding tone assignment. Ōsaka Japanese (OJ) exhibits two general tone patterns or tone melodies (Haraguchi 1999) in addition to the accented and unaccented division seen in TJ. OJ exhibits a similar tone system to the Kyōto and Kōbe variants of Kansai Japanese.²⁰ Haraguchi (1999) proposes that OJ has two possible tone melodies, HL and LHL. Descriptively, words come with one of two sparse tonal specifications: words with only a High tone specification, and words with an initial Low and a High specification. I discard the second L in Haraguchi’s (2001) tonal melodies as it is best understood as a default fill-in L, which only ever follows the High-toned mora and which is not present lexically. I then divide the accent possibilities in OJ to four possible classes:

1. accented words with only lexical H

2. accented words with a lexical H preceded by L

3. unaccented words with a default final H

4. unaccented words with default final H preceded by L.

In accented category (i) words, H is associated to the lexically accented nucleus, with H spreading regressively to the initial vowel within a word. The L is filled in following the site of an accent, as in /otóko/ [ótóko] ‘man’. Turning to category (ii) words, these accented words exhibits a H assigned to a certain mora with the L tone assigned to the preceding vowel, as in /kàbúto/ [kàbúto] ‘helmet’. This L may also spread regressively to the initial mora.²¹ In both cases, the final surface L is not marked in transcriptions as it is not a lexical tone, but the default realization of a toneless mora.

Unaccented words in categories (iii) and (iv) in OJ behave differently to those of TJ. I have proposed that in TJ, a default H tone is assigned to the final mora in a word. Such an analysis is not feasible in this dialect, as unaccented words are found with both H only and LH possibilities. I analyse these words as having floating tones with either H or LH present in the underlying morpheme. These tones are analysed as floating as they are a property of the morpheme, but mobile under suffixation. In HL melody words such as /sakura H/ ‘cherry tree’, they are realised in citation form with final H as in [sákúrá] ‘cherry tree’ and also in the nominative as in [sákúrágá] ‘cherry tree-NOM’. For LH unaccented words like /suzume LH/, the L and H tones are assigned regressively from the final mora beginning with H, giving [sùzùmé] ‘swallow’ and [sùzùmègá] ‘swallow-NOM’.

These patterns are exemplified in (17). Like TJ, tones spread regressively where possible. Unlike TJ, there is no initial protected monomoraic domain. Once again, I assume the surface Low tone realization following the site of lexical H is a default fill-in and thus I do not mark it.

Alternative analyses are found in Yoshida & Zamma (2001), where words have either a High tone which spreads or which does not spread and Labrune (2006), who applies a high register and a low register analysis.

The major difference I would like to focus on here is that in OJ, the H of the tonal melody can be placed on N, as well as R and J. This is a constant truth regardless of whether the underlying form is accented, unaccented or specified with a H or LH tone. Consider the data given in (13) from Haraguchi (2001). This data exemplifies high tone on N in (18a), R in (18b) and minimal pairs in (18c). In the (19), I give data from Kōri (1997) which exemplifies further contexts.

Based on the data above, it is clear that N can bear a High tone with both attested tone patterns in medial and final position. It is also clear, based on a close examination of the data, that both R and J serve as TBUs for the principal H tone in this dialect. This lies in contrast to the facts of TJ, where deficient moras only serve as TBUs in limited contexts.

Let us now consider the case of longer words, namely either words of four or more moras and noun-noun compounds of more than four moras. Here too, R, N and J may support the principal high tone of the HL or LHL tonal melody to the antepenultimate mora. Some examples are given in (20). The exact assignment process of tone melodies in morphologically complex forms is discussed elsewhere by Haraguchi (1999).

The crucial point here is that in both short words and longer words OJ, N is a potential and independent TBU, not that N is always a TBU.

Reviewers have suggested that unexpected exceptions to the above pattern exist, but I suppose these are connected to strata and category effects, with a possible shift to a TJ-type pattern of deficient mora avoidance. Examining adjectives, one reviewer has pointed out that while adjectives in Kansai Japanese typically display antepenultimate accent, this is not the case in the adjective [ɕiNdói] ‘tired-NP’, *[ɕiŃdoi]. Another reviewer points out that loanwords exhibit avoidance of accent on special moras. However, it is the case that accent is permitted on N in nouns as we have seen above. Further consideration of gaps is worthy of investigation, but must be placed within a focused study on the OJ dialect itself.

In OJ, there is no universal word-medial avoidance of accent on N and it bears a tone in multiple environments in native words, suggesting that it is a fuller TBU than in TJ. To consider an appropriate representation in further detail, let us examine its phonetic realisation.

4.2 The phonetic realisation of N in OJ

Phonetic investigations of N in OJ are fewer in number, however Tronnier (1996, 1998, 1999) presents a number of studies looking at its realisation and assimilation. These studies shed light on N with regards to obstruency and nasalisation. A first study (Tronnier 1996) focused on the question of whether or not N was realised as an obstruent intervocalically, while a later study (Tronnier 1999) examined the nature of nasalisation and vowel formants preceding both the non-moraic nasal and the moraic nasal N.

The first investigation (Tronnier 1996) was driven by the author’s observations that in OJ, N was realised as an obstruent in intervocalic position, for examples as in <taN’i>, [taNi] ‘credit’. For his investigation, Tronnier recorded speakers using an oral microphone and a piezoelectric (accelerometer) microphone placed on the nasal cartilage. Eight Ōsaka natives were asked to read carrier sentences containing target tokens with intervocalic N, such as <daNatsu> ‘oppression’. In the results section, Tronnier (1996: 172–173) states that some speakers indeed produced N with a consonantal realisation. One speaker (Tooru) produced N with a consonantal realisation and not a vocalic realisation 33% of the time. However, some speakers never (n = 2) or rarely (n = 2, with N as consonant <10%) produced a consonantal realisation and three of these were younger speakers. All speakers did show a tendency to exhibit the vocalic realisation of N according to Tronnier, but crucially consonantal realisations of N are possible in this dialect of Japanese in intervocalic context. While the study is limited and consists only of a few tokens in elicited read speech from 8 speakers, it is essential to point out this difference from informal TJ where N is a glide or vowel.

In a later report, Tronnier (1999) focuses on the effect of nasals on preceding vowels, namely nasalisation and formant changes. Focusing on the nasalisation portion of the study, Tronnier examined the onset of nasalisation in the vowels preceding nasal onsets and the moraic nasal in OJ. In question is whether the velum is lowered earlier preceding the moraic nasal (indicating nasalisation) than for the onset nasal, as evidenced in Ushijima & Hirose (1974) for TJ. This investigation consisted of oral microphone and nasal accelerometer recordings with six participants. The stimuli consisted of read speech, with target tokens containing both onset and intervocalic moraic nasals preceded by each of the five vowels of OJ, namely {a, i, u, e o}, comparing words like <tani> ‘valley’ with words like <daNatsu> ‘oppression’. While the results showed inter-speaker variation in the exact patterns of nasalisation, no categorical distinction was evidenced for nasalisation patterns between the onset nasal and the moraic nasal, though speakers implement individual nasalisation initiation timings which treat each category in a similar manner. Tronnier (1999: 394) points out that nasalisation cannot serve as a predictor for the following nasal category and its phonological status. For our purposes, we can point out that nasalisation is not contrastive preceding the onset or moraic nasal. This lies in contrast to TJ once again, where nasalisation occurs prior to the moraic nasal N, e.g. [hõN] ‘book’ and is absent prior to a nasal onset, e.g. [tani] ‘valley’.

4.3 Concluding in Ōsaka

In conclusion, OJ exhibits different phonological characteristics for N as it is a full TBU and can bear a lexical high. N in OJ exhibits a different phonetic character where it is realised potentially as a consonant and does not reliably nasalise the preceding vowel. For this reason, I propose that OJ has a syllabic and independent nasal segment. The facts are summarised in Table 2.

5.1 Tone in Kagoshima Japanese

Tone assignment in Kagoshima Japanese allows for only two possibilities (Kibe 1997b, 2010): high tone placed on the penultimate syllable or the ultimate syllable.²⁵ The tone system of Kagoshima is discussed at length by Kibe (1997b, 2000, 2010), Kibe & Hashimoto (2003), Kubozono (2004, 2007, 2015b) and Ito & Mester (2019) among others. According to the notations given in Kibe (1997b), a high tone is realised over the entire syllable, whether heavy or light.²⁶ The patterns are exemplified in (10).

Now let us consider the role of N as a TBU. According to Kibe (1997b), accent can never fall on the moras N, Q, R, or the second half of a diphthong. The author states explicitly:

The hatsuon [N], sokuon [Q] and chō’on [R] are one bunsetsu [segment] in length, however they are not independent and they cannot bear an accent. Furthermore, <i> in a vowel sequence is also not an independent mora. (Kibe 1997b: 10, translation mine).

For all words containing heavy syllables, only the syllable is counted in the assignment of penultimate or ultimate accent. This dialect is generally defined as a syllable-counting syllable dialect by McCawley (1978) and others, such as Kubozono (2012: 1403). Consider the data in (11) exemplifying words containing heavy syllables drawn from Kubozono (2012) and (2015b). These exemplify the fact that first, only the penultimate or ultimate syllable is referred to in accent assignment and second, that deficient moras do not bear the sole high tone in a word. Concretely, high tone is realised with a falling contour over the entire syllable in polysyllabic words, rather than an even high tone.

In the data, H is only assigned with reference to the syllable. N never behaves independently as a TBU. The existence or absence of a deficient mora is not relevant, and a deficient mora is never the sole bearer of a high tone (Kibe 1997b; Kubozono 2012, Ito & Mester 2019).

N as a non-TBU lies in contrast to its status in TJ and OJ. In KJ, N only bears a semblance of a high pitch if the preceding vowel carries H. A syllabic or vocalic representation of N would not be appropriate as this would allow partial or full TBU status. KJ could instead have a true nasal coda consonant. To support this possibility, let us turn to evidence from syllable weakening.

5.2 KJ syllable weakening and coda consonants

KJ has a process of syllable weakening or nisshōka (Kibe 1997b), where final high vowel apocope feeds consonant lenition. As a result. KJ exhibits a syllable template where N is one of a fuller set of WM and WF consonants. Crucially, the moraic nasal is the output of nasal onset lenition in word-final and word-medial contexts.

In native words in this variety of Japanese, word-final high vowels undergo apocope and the preceding consonant lenites. Stops and affricates are reduced as glottal stops, fricatives lose their voicing, nasals reduce to N and the liquid [r] is reduced to the glide [j], as discussed by Kibe (1997b) and Kaneko & Kawahara (2002). I transcribe the final lenited obstruent as a glottal stop, following Kaneko & Kawahara (2002), though Kibe (1997b) notes stylistic variation between [ʔ] in typical speech and [t˺]. Concretely, speakers I have consulted exhibit an unreleased glottal stop [ʔ˺] but I use the glottal stop for simplicity. This leads to cognates such as TJ [kaki] ‘persimmon’ and KJ [kaʔ] as well as TJ [kami] ‘paper’ and KJ [kaN]. Examples of the lenition patterns are given in (13).

The analysis of N is impacted by the surfacing of word-final and medial consonantal codas in this dialect. Possible WF codas are {ʔ, N, s, ʃ, j}.²⁸ While for nouns it is likely that this process is diachronic, it is active in verbal conjugations (Kibe 1997b, field notes). Consonant-final verb stems in the non-past such as /kak/ ‘write’ are realised with a lenited stem-final consonant, giving [kaʔ] ‘write-NP’.²⁹ Addition of the Negative suffix /-aN/ allows the underlying stem-final consonant to surface, giving [kakaN] ‘write-NEG’. This also occurs with m-final stems, as in /tanom/ ‘ask, request’, [tanoN] ‘ask-NP’, /tanom-aN/ [tanomaN] ‘ask-NEG’.

Word-medially, possible codas are Q, N, and ʔ (Kibe 1997). Examples are given in (14). (20a) provides examples of medial ʔ, while (20b) simply exemplifies that as with TJ, N and Q are found word medially in KJ. (20c) exemplifies medial nasal geminates and voiced geminates, with [N] and [ʔ] undergoing assimilation, noted by Kibe (1997b: 10). KJ thus has NC clusters, glottal-consonant clusters, voiceless geminates and voiced geminates.³⁰

I claim that in KJ N is a true consonant. First, we have no evidence to show that N is a vowel or a syllabic consonant, unlike the other dialects of Japanese we have seen. N is never the site of H tone assignment. Second, there is no clear nasalisation in the preceding vowel that can be discerned observationally. Third, it is clear that N can be derived as in verb stems. There is no argument to be made that final N is a vowel if it is derived from a consonant, and there are no arguments to be made that internal N is a vowel either.

5.3 Interim summary and the goals for a revised account

Having now considered the facts of N in three different dialects of Japanese, I return to the broader comparison of N. I summarise the facts given thus far in Table 3.

The goal in the remainder of this article is to revise the representation of N. I consider first the previous representations most relevant to the data discussed in the following section. I then propose a set of new representations within Strict CV (Lowenstamm 1996) in §7 and I account for the TBU status of N in §8.

6.1 The syllable & mora account

As discussed in §3, the moras N, Q, R, and J are usually analysed as the second mora within a syllable, affording them a deficient or dependent status. The shape of a CVN syllable as well as underlying and surface forms of <hoN> ‘book’ are recalled in (21–23).

(21)	The template for the Japanese syllable containing N in the spirit of Vance (2008)

(22)	Underlying representation of <hoN> ‘book’

(23)	Surface representation of <hoN> ‘book’

The representation of N as a moraic consonant in the aforementioned works is able to account for the realisation of /N/ as a homorganic nasal consonant through feature spreading, as well as nasalisation. The non-accentuation of a deficient mora can be defined with reference to avoidance of accent on a special or non-head mora, as discussed in Kawahara (2015).

6.2 N as a variable segment – the Government Phonology account

The first treatment of N as a variable segment is proposed by Yoshida S. (2003), where N is a floating nasal element (or privative feature) N associated to either an Onset or a Nucleus. Yoshida’s account is framed within Government Phonology (Kaye, Lowenstamm & Vergnaud 1990, Charette 1991, Harris 1994, Gussmann 2002) and Element Theory (Kaye, Lowenstamm & Vergnaud 1985, Charette & Göksel 1998, Backley 2011). For recent review, see Scheer & Cyran (2018a, b) and Scheer & Kula (2018). This proposal is developed by Yoshida S. (1991, 1996, 2003), with a slightly alternative representation given by Yoshida Y. (1995, 1999, 2003). See also Nasukawa (1998, 2005).

6.2.1 A brief introduction to GP

I now consider the tenets of Government Phonology (GP) relevant for Japanese, focusing on N (Yoshida S. 2003); the TBU is discussed in detail later in §8. The main proposal here is that N is a floating feature which is associated to either a nucleus or an onset. Crucially, it represents a preceding view of N where N is treated as a vowel.

GP utilises Onsets, Nuclei and Rhymes and rejects the syllable, while applying government and licensing in a parallel with Government & Binding (Chomsky 1981) as well as principles and parameters. In this theory, government is a force which silences an empty constituent, while licensing supports the realization of a segment or branching constituent. (Kaye, Lowenstamm & Vergnaud 1990, Charette 1991). Yoshida Y. (1995, 1999) and Yoshida S. (2003) propose that Japanese permits only non-branching Onsets and Nuclei.³⁴ The representation of a word such as [murásaki] ‘violet’ is given in (24).

(24)	Structure of [murásaki] ‘violet’ (simplified from Yoshida Y. 1999: 79)

The mora in this framework is equivalent to an Onset-Nucleus pair, and the Nucleus is used as an accent site (or here, a TBU). In addition to the rejection of the syllable, a crucial difference to the syllable and mora framework (as used in Hyman 1985 and Hayes 1995) is that GP uses empty constituents.³⁵

Empty nuclei are used in the representation of NC clusters and geminates. They are captured as two Onsets surrounding an empty Nucleus. The onsets are partially assimilated (for NC clusters) or a segment is associated to both constituents (geminates). These Onsets surround an empty nucleus. I give the representation of N and Q in loanwords discussed earlier in (25) and (26), following the structures proposed by Yoshida Y. (1999).

(25)	Representation of [káNzasu] ‘Kansas’

(26)	Representation of [ták:uru] ‘tackle’

For Japanese, the empty nucleus accounts for pre-antepenult accent in words with CVN and CVQ syllables, where only the filled nuclei can participate in accent assignment (cf. Yoshida Y. 1999, Youngberg 2017). Accent is placed on the antepenultimate filled nucleus, and accent ‘shift’ is not a true process of the grammar.³⁶ We return to this filled and empty distinction in further detail in §8.2, but first let us consider how empty positions are restricted.

6.2.3 N as a floating element

In Yoshida Y. (2003) proposes that the underlying representation (UR) of N, shown in (29), is a floating nasal element to capture its vocalic and consonantal realisations. Following Yoshida S. (2003), nasality is represented as the element |N| in combination with the oral resonance elements |A|, |I| and |U|, framed within Element Theory (Backley 2011, Scheer & Kula 2018).

(29)	Representation of /N/ (Yoshida S. 2003)

Yoshida S. (2003) is in agreement with the earlier literature that N assimilates to following consonants. On the other hand, we have seen in §2 that for Nakano (1969) and Yoshida S. (2003), word-final and intervocalic N is a nasal vowel, with /hoN/ ‘book’ realised as [hoɯ̃] or [hoõ]. This variability drives the floating representation.

NC clusters are represented as adjacent onset consonants, as in /hoN-to/, [honto] ‘book-COM’ given in (30). The element |N| associates to O₂ as a governing relation can contract between O₃ and O₂. This satisfies the ECP, silencing N₂. Assimilation occurs as a result.

(30)	Representation of /hon-to/ [honto] ‘book-COM’ (Yoshida S. 2003: 540)

In other contexts, |N| associates to the nucleus, shown in (31–32). No following contentful Onset follows |N|, and inter-onset government cannot occur. |N| thus associates to the nucleus as the ECP cannot be satisfied. In formal pronunciation, N is given the place element |U|, the epenthetic element (e.g. kiss realised as [kisu], cf. Yoshida S. 1996, 2003). In informal representations, intervocalic and final N gain place elements from the preceding nucleus, shown in (31–32) for /hoN/ ‘book’ and /taNi/ ‘credit’.

(31)	Representation of /hoN/ [hoõ] ‘book’, drawing on Yoshida S. (2003)

(32)	Representation of /taNi/ [taãi] ‘credit’ (Yoshida S. 2003: 539)

6.2.4 Problems with the GP representation of N

These representations must be considered in light of §3. First, it is clear the vowel preceding N is nasalised in all contexts. Yoshida S. does not discuss nasalisation, but not all representations could capture the process. In the proposed representation of NC clusters, one cannot account for nasalisation as there is no relation between the Onset containing N and the preceding Nucleus. If we attempt to say this is a low-level phonetic process, it cannot explain why nasalisation is constant in TJ and not in OJ. For N in intervocalic and final positions, there is progressive spread of elements for informal VNV and VN#. A slight adjustment to include |N| spreading can capture nasalisation here, but this would not account for NC. Finally, N is realised as an assimilated nasal vowel in word-final positions, but it is also preceded by a nasalised vowel. The surface identity would be a long nasal vowel, or [hõõ] ‘book’ for <hoN>. Why is this not the underlying representation?

Another issue is that the analysis of tone spreading remains problematic. Recall that CVN patterns with CVR and CVJ syllables. In (33), I provide the representation of CVQ from Yoshida Y. (1999) for <baQta> ‘grasshopper’, while the CVN word <keNka> ‘quarrel’ is shown in (34).

(33)	Geminate in CVQ initial word as in [bat:a] ‘grasshopper’

(34)	Geminate in CVQ initial word as in [keŋka] ‘quarrel’

N and Q have an identical constituent structure; the model incorrectly predicts that they should pattern together as their representation is identical. In this framework, the [±sonorant] feature cannot be invoked, nor the coda position. We could claim that that Onsets lacking |N| in the O₂ position block spreading, but this is a descriptive and unsatisfying solution.

A positive advance is made for N as a marginal TBU as final N is in the nuclear position, as in (31). If the TBU is a nucleus, Yoshida S.’s (2003) representation could explain how N supports a marginal high. The UR of N contains an empty nucleus and would not be a possible lexical TBU as it is empty, but could be a candidate post-lexically as the surface nucleus is filled.

This representation captures many, but not all, facets of TJ. While this variable representation of N could easily be adapted to the facts of OJ and KJ, variability is not clearly called for in these dialects. I turn to the possibility that the lexical representation of N in each dialect is entirely different, and not derived from one underlying form.

7.1 The basic foundations of Strict CV

Lowenstamm (1996) proposes that all words are composed of strictly repeating Consonant (C) and Vowel (V) positions and segments. Branching constituents are rejected, and the division between the constituent tier and the skeletal tier is collapsed (Scheer 2004). Segments are directly associated to the CV tier. Government is retained as the force which silences an empty nucleus (or V), while licensing supports the phonetic interpretation of a position (Scheer 2004). I do not provide a full overview of this framework, which is discussed in part by Scheer & Cyran (2018). For more on government and licensing in CV, see Scheer (1998, 2004, 2012), Dienes & Szigetvári (1999), Cyran (2010), Ségéral & Scheer (2008) & Ziková & Scheer (2010). CV units have also been used to investigate the relationship between morphosyntax and phonology; see Lowenstamm (1999) on the initial empty CV as a boundary marker, Bendjaballah & Haiden (2008) and Fathi & Lowenstamm (2016) on CV units as the spell-out of morphemes and syntactic terminals, and Scheer (2012) on CV markers at the interface.

The mora is redefined here as a CV pair (Scheer & Szigetvári 2005). This is in line with Yoshida Y. (1999), where an Onset-Nucleus pair is a ‘mora’. I return to the TBU in §8.2. In this section, I examine the structure of light and heavy syllables in Strict CV and I then turn to the structure of N.

7.1.1 Open syllables in a CV account of Japanese

First, compare the GP and CV representation of [murasaki] ‘violet’ in (35) and (36). In words with only open syllables, the difference is that the timing tier is no longer present – the timing tier and constituent tier are collapsed.

(35)	Structure of [murasaki] ‘violet’ in Standard GP (Yoshida Y. 1999)

(36)	Structure of [murasaki] ‘violet’ with a CV tier

7.1.2 Geminates, codas and long vowels in Strict CV

Long vowels and geminates are represented by the double association of a given segment to either two C or V positions. Revised representations for the Japanese words [ha] ‘tooth’, [to:] ‘tower’ and [bat:a] ‘grasshopper’ are given in (37a–c) Empty positions are underlined to represent a governed position.

(37)	Representation of CV, CV: and CVC:V words in Japanese
	a.	[ha] ‘tooth’
	b.	[bat:a] ‘grasshopper’
	c.	[to:] ‘tower’

A few comments must be made about the structures given in (37b) and (37c). In the representation of [bat:a] given in (38), I assume for the moment that here too, an inter-onset relation contracts between C₂ and C₃, governing and silencing empty V₂.³⁸

(38)	Representation of geminate in <baQta> ‘grasshopper’

Long vowels are formed of a segment associated to both vowel positions. The second position is filled, but it is externally licensed by the following vowel position when found word-medially (Yoshida S. 1993, Kaye 1995, Lowenstamm 1996, Scheer 2004). Final long vowels meanwhile are licensed by parameter (Youngberg 2017).³⁹ In relation to Japanese, this can account for the lack of medial *V:C: sequences in mono-morphemic Japanese words and the free occurrence of final long vowels. See the representation of [ko:ri] ‘ice’ (39) and [sato:] ‘sugar’ (40), where a dotted line represents licensing.

(39)	Medial long vowel in [ko:ri] ‘ice’

(40)	Final long vowel as in [sato:] ‘sugar’

7.2 Revised representation for N in Tōkyō

Now let us consider the structure of N in more detail. The suggested representations are recalled in Table 4. I begin the redefinition of N here in §7.2. focusing on TJ, where a <VN> sequence is a long nasal vowel. I then consider §OJ in 7.3, where N is a syllabic nasal, and conclude with §KJ in 7.4, where N is a ‘coda’.

To account for nasalisation and the patterning of long vowels and so-called vowel nasal sequences in TJ, I claim that /VN/ sequences are in fact /Ṽ:/. This representation is shown in (41) for <keNka> ‘quarrel’, which is phonologically /kẽ:ka/. The velar nasal, which is typically transcribed preceding /k/, is absent as this is no longer part of the phonological representation but rather an effect of phonetic transition between a nasal vowel and a following consonant.

(41)	Representation of N as a nasal vowel as in /kẽ:ka/ ‘quarrel’

First, this representation characterises the regular nasalisation found in TJ as a full lexical part of the CVN syllable. Second, CVR and CVN syllables are unified when we consider tone spreading. Compare the representations for CV, CVR, CVN and CVQ syllables in (42–45). Tone spreading is marked with an H.

(42)	[sakúrá] ‘cherry tree’

(43)	[kó:rí] ‘ice’

(44)	[kẽ́:ŋká] ‘quarrel

(45)	[bat:á] ‘grasshopper’

Here, tone spreads regressively from the site of the default final H placed on the final vowel in these lexically unaccented words. The analysis of tone spreading is as follows. Following Yoshida Y. (1999), I assume that the initial CV is protected in Tokyo Japanese, though in other dialects this protection window may be 0 or 2 CV units, with the former represented by OJ and the latter represented by Owari Japanese (Youngberg 2017). Initial syllable protection is evidenced in other languages such as French (Charette 1991), Tonkawa (Yoshida Y. 1990) and Turkish (Charette 2004); see also Becker, Nevins & Levine (2012). In CV and CVQ words, V₁ is protected and not affected by spreading. However, this window only holds unless branching occurs between the initial V₁ position (within the protection window) and V₂ (within the tone spreading domain). The initial V position in CVR and CVN syllables have an identical status, with V₁ and V₂ being associated to the same segment, breaching the barrier between V₁ and V₂. It is not ‘weight’ alone which is relevant for spreading, but also the association between positions. No reference to features or a subset of syllables is necessary. CVQ, having no such relation, fails to trigger spreading to V₁. I revisit tonal behaviour of N in §8. Let us turn now to OJ.

7.3 Revised representation of N in OJ

I propose that in OJ, N should be represented as an independent syllabic nasal. Within Strict CV, this can be represented as the association of |N| to both the C and V positions, with the V position hosting the element. A similar representation for syllabic consonants in other languages can be found in Scheer (2004), with sonorant sensitive stress systems such as Kwak’wala (Northern Wakashan, Western Canada) finding a similar analysis in Scheer & Szigétvari (2005). See the representation of the word [oŃna] ‘woman’ in (46).

(46)	CV representation of OJ [oŃna] ‘woman’

Here, |N| is associated to the V position and it is independent from the preceding and following V positions. The spreading to the C position also accommodates possible glide or obstruent-like interpretations of N and excludes regular nasalisation, following discussion of the phonetic realisation in OJ given previously in §4.

7.4 Revised representation of N in KJ

Finally, for KJ I propose that |N| is a ‘coda’, which in CV is a filled C position preceding an empty V position. This representation follows from cross-linguistic evidence that final consonants precede an empty nucleus (Kaye 1990, Harris & Gussmann 1998, 2002, Scheer 2004) and language internal evidence from syllable weakening (cf. §5). The structure is given in (47).

(47)	Kagoshima Japanese N as a ‘coda’, as in [kaN] ‘spirit

Above, V₂ is empty. It is governed through the domain-final parameter of the ECP, which is set to [ON] and thus represented with underlining. This correctly predicts that C₂ may also contain other ‘final’ consonants, as in [kaʔ] ‘persimmon’. Unlike TJ where empty V positions are only permitted medially under strict conditions (now, only in geminates) and the WF parameter of the ECP is [OFF], in KJ we have WF empty V positions as the domain-final parameter of the ECP is [ON].

This representation also allows for a simple analysis of synchronic weakening where the non-past /kam-u/ ‘chew’ is realised as [kaN]. Descriptively, the final high vowel is suppressed, and the preceding consonant lenites. How does lenition function here? Elements in an onset are interpreted based on the filled or empty status of the following nucleus (Harris 1992, 1997, 2000, Cyran 2010). In GP, a ‘coda’ is a final onset or C followed by an empty nucleus or V. Following Harris (1992, 1997), lenition is derived from the diminished ability of an empty nuclear position to license its preceding onset, and the interpretation of subsegmental content is suppressed. Turning to KJ, C₂ is lenited in words affected by apocope which elides final high vowels, with the emptied V then unable to fully license the full segmental content in C₂ cannot be expressed. This gives rise to [kaN] ‘chew-NP’ from the UR /kam-u/. Stem final consonants are fully realised when the V position is filled by a suffixal mid or low vowel, as in the negative verbal form [kamaN] ‘chew-NEG’. Fuller analysis of weakening is presented in Youngberg (2020). Crucially, no resyllabification is necessary from the viewpoint taken here, and N is unified with other coda consonants.

8.1 Previous approaches to dialect typology in Generative Phonology

Since McCawley (1978), Japanese varieties are often divided into syllable dialects (Kagoshima), syllable-and-mora dialects (Tōkyō) and mora dialects (Kansai) based on what unit is relied on for accent assignment.⁴⁰ These divisions are used also in recent typologies and analyses (e.g. Kubozono 2012, Ito & Mester 2019). Optimality Theory (Prince & Smolensky 1993) analyses refer to the syllable and mora directly in constraints, capturing typology through ranking. Labrune (2012a, b) proposes that certain moras are incomplete in their featural makeup, with the ideal mora having both a consonant and a vowel. Deficient moras are penalised by the markedness constraint NADM (non-accentuation of deficient moras) and the TBU is also ranked against a mora hierarchy, where R, J and N are ranked low and Q is ranked at the bottom. Dialects which permit accent on deficient moras have NADM ranked low (e.g. as in OJ), while TJ would have NADM ranked high to prevent output forms with H on deficient moras. Ito & Mester (2019) also utilise markedness constraints such as NOCONTOUR-SYLL, NOCONTOUR-MORA and H-TO-HEAD (among others) to derive the tone patterns of various dialects in Kagoshima prefecture, including that of KJ, and separates syllable dialects like KJ from others by high ranking of H-TO-HEAD and NOCONTOUR-SYLL. Such analyses indeed derive the correct typology, but they rely on a universal syllabic structure and a universal structure of N with constraints deriving the desired output. Additional constraints and a hierarchy are not necessary when the input involves adjusted representations.

As we have seen earlier, features can be used to derive variation as well. Haraguchi (1975: 138, 441ff; 1977) is an early example, altering the target features in tone assignment and spreading rules to derive accent differences. The definition of a TBU, for example, relies on features such as [±syl], with N being [+syl] in OJ and thus being a TBU. N is [-syl] in TJ and cannot support a tone. These features are assigned only to derive the possible tone patterns. The author does not appeal to further evidence beyond tone to justify such an assignment, but it must be noted that he also focuses only on tonal typology and ignores segmental processes or discussions of phonetic regularity. I replace both constraints and feature sensitivities here with simple reference to the V position.

8.2 A Strict CV view of the TBU

Turning to Strict CV, I claim that the consideration of the V position status leads to a clear division of well-formed, marginal and malformed TBUs. I build on the analysis of Yoshida Y. (1999) for defining an accentable site, and I develop this model by including dialects besides TJ and by incorporating Strict CV structures. Previous work on stress and parameters in CV began in Scheer & Szigetvári (2005), who examine stress assignment in Latin, Kwak’wala, Malayalam and other languages. The proposal here complements the parameter approach developed by Ulfsbjorninn (2014) and Faust & Ulfsbjorninn (2018), who focus on the role of empty nuclei in stress systems.⁴¹ The crucial ingredient introduced here is that full, marginal and impossible TBU status above the CV tier is derived from the status of the V position below the CV tier.

8.2.1 The well-formed lexical TBU as a projected position

In this section, I claim that a full TBU in the lexicon is a filled and independent V position (N in OJ), while a marginal TBU on the surface is a filled V position which is not independent but externally licensed (N in TJ), and a malformed TBU is an empty and governed position (N in KJ).

Yoshida Y. (1999) has defined an accentable position as a filled nuclear position which is unlicensed and ungoverned (or here, V position). Empty nuclei are governed where possible, or they are phonetically interpreted. At this point, filled nuclei remain unlicensed; they project in order to find a licenser at a higher level (Charette 1991, Harris 1994, Kaye 1995).⁴² Empty nuclei are governed at the constituent tier; to go a little further than we have, licensing is a inter-constituent force which supports a segment while government is understood to be a sub-type of licensing which supports a constituent’s silence (Charette 1991).⁴³ As empty nuclei are governed at the constituent tier, they do not project.⁴⁴ A possible TBU can be initially defined as the filled nucleus, while an impossible TBU is one which is empty and governed.

Translating this into Strict CV, any CV which is independent and which is filled projects. Governed V positions do not. Consider the structure of the words [murásaki] ‘violet’ demonstrated in (48). A high tone is assigned to the antepenultimate projected nucleus, as in the earlier work of Yoshida Y. (1999).⁴⁵ A similar analysis of Latin stress is given by Scheer & Szigetvári (2005).

(48)	Projection in TJ [murásaki] ‘violet’

8.2.2 Defining an unaccentable site – government of empty V

Yoshida Y. (1999) proposes that unaccentable positions are empty nuclei. In Japanese, these are found in geminates as in /bat:a/ ‘grasshopper’. Empty nuclei are not projected as they are already governed at the constituent tier. Adapted to CV, geminates and ‘coda’ consonants contain a governed empty V which fails to project, and in Japanese these positions are governed in the formation of a geminate. This is shown in (49).

(49)	Projection in TJ [ták:uru] ‘tackle’

The definition of an accentable and unaccentable site can be directly transposed to OJ and KJ. Firstly, the representation I have proposed for N in OJ is a full TBU as the V position is filled. The representation of [oŃna] ‘woman’ is given (50). The V₂ position is filled with the element |N|, which then projects to serve as a potential lexical TBU as it is unlicensed and independent.⁴⁶

(50)	CV representation of OJ [oŃna] ‘woman’

In KJ, N is never a TBU. Here, |N| is associated to a C position which precedes an empty V position. The final V is governed by the WF parameter of the ECP and does not project.⁴⁷ The structure of [kaN] ‘spirit’ is recalled in (51).

(51)	KJ projection in [kaN] ‘spirit

In each of these dialects, N as a lexical TBU or an impossible TBU is easily captured with little adjustment: N cannot serve as a TBU in KJ as it has no contentful nucleus, while in OJ it is independent and has a contentful nucleus through being ‘syllabic’.⁴⁸ An issue remains for TJ – the V position is filled but still fails to serve as a TBU.

8.2.3 Refining an unaccentable site –long vowels and external licensing

For TJ long vowels, we must consider the status of the second V position. We have seen earlier in §7 that the second position is licensed externally at the CV tier. We must exclude the possible of lexical accent on V₂ in a long vowel. The representation of [ko:ri] ‘ice’ is given again in (52).

(52)	Licensing in V: as in [ko:ri] ‘ice’

Deficiency based on the filled or empty status of a V position alone does not suffice. Yoshida Y. (1999) relies on the post-lexical formation of branching nuclei to deal with the unaccentability of the second half of a long vowel, but this is an ad hoc solution and not possible in Strict CV.

I claim that the deficient nature of a long vowel (and N) falls out from the original proposal from Yoshida Y. (1999) in combination with the CV structure of long vowels: only those positions which are unlicensed at the constituent tier project and serve as accentable units. In the configuration in (53) showing the representations of [ko:ri] ‘ice’ and [kẽ:^ŋka] ‘quarrel’, V₂ is externally licensed by V₃. V₂ does not project as it licensed at the CV tier and is unable to serve as a TBU. This is the same for CVV and CVN syllables, identical in structure but not in segmental content.

(53)	Projections in [ko:ri] ‘ice’ and [kẽ:ŋka] ‘quarrel’

This proposal easily accounts for cases of accent shift in long words and compounds where antepenult H is expected, but found on the pre-antepenult. Consider the structure for the word <kaNzasu> ‘Kansas’. The representation is given in (54). High tone is assigned to the pre-antepenult V position, which does project. V₁ is the antepenultimate projected nucleus.⁴⁹

(54)	CV representation of [kã:nzasu] ‘Kansas’

8.2.4 Marginal accentuation in TJ revisited

Now what of marginal accentuation in TJ? In final position, the second V position for long vowels is filled. As discussed in §7, they are licensed by virtue of being domain-final. However, note that we are not discussing lexical highs but rather tones placed in a post-lexical context. R, J and N are never good lexical TBUs and there is no underlyingly accented word such as */tokoó/ in TJ. This is because the second V position in a long vowel does not project. I propose, however, that final N, R and J are marginal TBUs due to their filled nature. Upon spell-out of an unaccented form or under pressure from a pre-accenting suffix, a default H or suffixal floating H can be placed and interpreted on the final filled V position (55). It does not matter whether this is projected or not as this is not a lexical tone, which relies on projected nuclei.

(55)	Surface representation of <udoN>, [udõ:] ‘udoN noodles’ in TJ

Why are CVQ syllables never TBUs in TJ, even marginally? While no final consonants means the structure forming Q can never take a marginal tone, it is also clear that Q lacks the correct configuration to become a marginal TBU – it has no filled vowel position.