3.1 Foot-initial Strengthening

Throughout this section we have assumed that IL feet in Gilbertese consist of a bimoraic iamb with a right adjunct ((μ_w μ_s)μ_w), but note that an alternative analysis with bimoraic trochees and a left adjunct (μ_w (μ_s μ_w)) could also be posited for the language (cf. 9a versus 9b). In these two IL feet, the second mora of the trimoraic sequence corresponds to the head of a foot, which is realized with stress, hence the two metrical representations are descriptively equivalent for Gilbertese stress. However, we will adopt representation (9a) over (9b) based on a phonological argument: the former allows a simpler and more straightforward account of the distribution of high pitch. In particular, we propose that high pitch placement in Gilbertese amounts to inserting a boundary tone in a foot-initial position as a means of strengthening the initial boundary of the metrical foot (Martínez-Paricio Reference Martínez-Paricio2013). That is, following Davis and Cho (Reference Davis and Cho2003), Rice (Reference Rice1992), and Bennett (Reference Bennett2012, Reference Bennett2013), among others, we assume that the initial constituent of a foot, similar to initial constituents in other prosodic categories, may exhibit particular strengthening effects. In Gilbertese, this greater strength is realized as a boundary H tone foot-initially.Footnote ³ Note that the representation in (9a) has only one foot-initial position, since the left edges of FtMin and FtNon-min coincide, while (9b) has two foot-initial positions, only one of which carries high pitch. If we had adopted the representation in (9b) instead, we would need to include an additional specification saying that strengthening only occurs at the left edge of some foot projections but not others (in this case, the left edge of a FtNon-min). This specification would be required to avoid a configuration like (9b) receiving two highs at the left edge of each foot projection: one on the first mora and another one on the second mora.

1. (9) Foot-initial strengthening in Gilbertese

To clearly determine which of the structures in (9) corresponds to this system, the particular pitch and stress patterns of bimoraic sequences in Gilbertese would be enlightening, since these could help us determine the exact shape of non-layered feet (i.e. feet that are minimal and maximal simultaneously); of particular interest would be the pitch/stress patterns of (i) bimoraic prosodic words and/or (ii) forms with a 3n+2 number of morae (e.g. (μ μ)_Ft (μ μ μ)_Ft). If in these cases the non-layered foot displays a H pitch on the first mora, this would be evidence for the representation in (9a), which exhibits a H on the left edge of the FtMin, which coincides with the FtNon-min. The representation in (9b) would be less adequate since again the rule (or constraint) for H pitch placement would need to specifically refer to FtMax. Unfortunately, bimoraic prosodic words are rare, since these generally undergo a lengthening process to fulfill a trimoraic minimal requirement (Section 3.3). Also, no acoustic measurements of the leftover bimoraic sequences in 3n+2 structures are available. When we asked Harrison (per.comm.) about the realization of bimoraic words and the first two morae in 3n+2 sequences, the author reported some variation/uncertainty: sometimes these sequences seemed to be realized as (μ^H ˈμ), with a high initial pitch and a stress on the second mora, giving support to the representation in (9a), whereas in other cases, there seemed to be a coincidence of pitch and intensity on the first mora, and something more neutral on the second, matching better with (9b) (Harrison, per.comm.; Blevins and Harrison Reference Blevins and Harrison1999: 218). Future experimental research will be needed to determine more specifically which of these two structures is instantiated in Gilbertese. For the time being and the sake of simplicity, we will assume (9a), but note that any of the two structures could in principle be posited for the language.

3.2 Binary Head versus Unary Head

As mentioned earlier in the introduction to this section, Blevins and Harrison (Reference Blevins and Harrison1999) had already proposed that Gilbertese metrical representations consist of trimoraic feet. However, the internal structure of these feet was slightly different from the ones proposed here: it consists of a bipartite head and a unary non-head (i.e. foot dependent) (Reference Blevins and Harrison1999: 217, based on Rehg Reference Rehg1993):

1. (10)

   Blevins and Harrison’s trimoraic feet with bipartite heads (Reference Blevins and Harrison1999: 217)

   I. Bipartite head

Although some studies have postulated the existence of feet with bipartite heads (Kager Reference Kager and Kleinhenz1996; Bye Reference Bye1996), we assume that the foot is a relational and rhythmic category that arises from combining a unique head (strong element) with one (or various, in the case of IL feet) optional non-head (weak element), this unique head being its defining property. This is why feet with only a head (i.e. unary feet) occasionally occur (but not degenerate feet with only a dependent) and, in case two adjacent positions are prominent, these have always been analyzed as each being the head of an independent foot. Another argument against positing a symmetrical bipartite head such as the one in (10) is that it predicts identical behavior for the two morae that compose the head. However, a typological survey on the relative strength of the elements contained in such binary structures (bipartite heads in Blevins and Harrison Reference Blevins and Harrison1999, minimal feet in our model) reveals that the two constituents in the head clearly behave differently, one of them being phonologically and phonetically stronger than the other (Martínez-Paricio Reference Martínez-Paricio2013). This is clearly the case in Gilbertese, where the morae in the binary head behave differently: the second mora is the true head of the foot and, hence, the stress-bearing unit, whereas the first mora is the tone-bearing unit. For all these reasons, we believe these structures are better analyzed via an IL foot, where the binary innermost constituent is not flat qua prominence, but it is actually a foot by itself.

3.3 Gilbertese Lengthening and the Minimality Requirement

A final piece of evidence which supports the idea that Gilbertese default feet are trimoraic is reported in Blevins and Harrison (Reference Blevins and Harrison1999: Section 4.1): the overwhelming majority of prosodic words in the language have at least three morae. Whenever this is not the case, there is strong evidence that bimoraic lexical words display lengthening to conform to this restriction. For instance, borrowed names display lengthening to conform to the trimoraic restriction (11a–c). Furthermore, even if it is not uncommon for lexical words (nouns and verbs) to have two morae, they hardly ever surface alone: they are either accompanied by a proclitic article or a possessive suffix (in the case of nouns) or a proclitic subject marker (in the case of verbs). But in bare verbal and nominal forms, like bare plurals (11e–g) or imperative forms (11h–j), the bimoraic lexical word undergoes lengthening. In (11) we indicate every moraic segment with the superscript <^μ>; even if we assume that long vowels consist of one vowel linked to two morae, in the following examples we represent each lengthened vowel as two independent vowels, to better illustrate the lengthening pattern.

1. (11)

   Lengthening (all data taken from Blevins and Harrison Reference Blevins and Harrison1999: 213–216)
   Borrowed proper names
   a.	taμaμmμ	‘Sam’
   b.	ti μiμmeμ	‘Jimmy’
   c.	biμiμtiμ	‘Fiji’
   In bare plural nouns			Noun phrases
   e.	bwaμaμtaμ	‘the/some huts’	cf.	teμ bwaμtaμ	‘the/a hut’
   f.	oμoμnμ	‘the/some turtles’	cf.	teμ oμnμ	‘the/a turtle’
   g.	baμaμiμ	‘the/some arms’	cf.	baμiμ-uμ	‘my arms’
   In imperative verbal forms			Verbs+subject marker
   h.	biμiμriμ!	‘run!’	cf.	eμ biμriμ	‘s/he ran’
   i.	niμiμmμ!	‘drink them!’	cf.	iμ niμmμ	‘I drank them’
   j.	aμmwμaraμkeμ!	‘eat!’	cf.	iμ aμmwaμraμkeμ	‘I ate’

Even if the trimoraic restriction is not completely inviolable, as there are a few exceptional forms that can surface with two morae (12), the lengthening patterns in (11) stand as strong support for feet with moraic terminal elements in Gilbertese. Note that the exceptions occur only in environments in which lengthening would have introduced an extra-long vowel, which are forbidden in the language (12a, b), or a geminate nasal in preconsonantal position, which are also illicit in prevocalic position (12c, d) (Blevins and Harrison Reference Blevins and Harrison1999: 215). The pitch/stress patterns of these forms are not indicated because these are precisely the ones that were unclear to the authors (see again Section 3.1).

1. (12)

   Bimoraic prosodic words (Blevins and Harrison Reference Blevins and Harrison1999: 215)
   Bare plural nouns
   a.	niμiμ	‘some coconut trees’	(cf. sing.: te nii)
   b.	baμaμ	‘some leaves’	(cf. sing.: te baa)
   c.	nnμeμ	‘some spots’	(cf. sing.: te nne)
   d.	nnμaμ	‘some fleets’	(cf. sing.: te nna)

3.4 Optimality Theory Analysis

We adopt here Martínez-Paricio and Kager’s (Reference Martínez-Paricio and Kager2015) Optimality Theory analysis of ternary rhythm. Under their view, the emergence of IL feet gives rise to ternary stress alternations. In (13) and (14) we present definitions of the relevant constraints: (13a–e) regulate the location and number of foot projections within a prosodic word, (13f) controls the location of unfooted syllables, banning them when the left and right versions of the constraint are both undominated. Finally, the constraints in (14) determine the location of foot heads and foot dependents within a foot. These metrical constraints were originally proposed to account for quantity-insensitive rhythmic stress, but they also control the location of feet in quantity-sensitive systems.Footnote ⁴

1. (13)

   Constraints regulating the location and number of foot projections and unfooted syllables within a prosodic word (Martínez-Paricio and Kager Reference Martínez-Paricio and Kager2015: 470)
   a.	Align-L/R([Ftmax]ω, Ft, ω) (Align-L/Rmax)
   	For every maximal foot Ftmax, assign a violation mark if some foot intervenes between Ftmax and the left/right edge of its containing ω.
   b.	Align-L/R([Ftmin]ω, Ft, ω) (Align-L/Rmin)
   	For every foot that is minimal and maximal ([Ftmin]ω), assign a violation mark if some foot intervenes between [Ftmin]ω and the left/right edge of its containing ω.
   c.	Align-L/R([Ftnon-min]ω, Ft, ω) (Align-L/Rnon-min)
   	For every non-minimal foot Ftnon-min, assign a violation mark if some foot intervenes between Ftnon-min and the left/right edge of its containing ω.
   d.	Align-L/R([Ftunary]ω, Ft, ω) (Align-L/Runary)
   	For every unary foot Ftunary, assign a violation mark if some foot intervenes between Ftunary and the left/right edge of its containing ω.
   e.	Align-L/R([Ftmain]ω, Ft, ω) (Align-L/Rmain)
   	For every head foot of a prosodic word (Ftmain), assign a violation mark if some foot intervenes between Ftmain and the left/right edge of its containing ω.
   	(Based on EndRule-L/R; Prince Reference Prince1983, McCarthy Reference McCarthy2003.)
   f.	Align-L/R([σ]ω, Ft, ω) (Chain-L/R)
   	For every unfooted syllable (σ)ω, assign a violation mark if some foot intervenes between (σ)ω and the left/right edge of its containing ω.

1. (14)

   Constraints regulating the location of foot heads and foot dependents within a Ft (Martínez-Paricio and Kager Reference Martínez-Paricio and Kager2015: 473)
   a.	Trochee For every foot head, assign a violation mark if it is not initial in its containing Ft.
   b.	Iamb For every foot head, assign a violation mark if it is initial in its containing Ft.
   c.	Align-L/R(Ftmin, σ, Ftnon-min) (Trocheenon-min/Iambnon-min)
   	For every minimal foot Ftmin, assign a violation mark if some footed syllable intervenes between Ftmin and the left/right edge of its containing Ft.

The following tableaux show how the general constraints in (13–14) model the type and number of IL feet that arise in Gilbertese. Let us start with the simpler forms: those that only contain monomoraic syllables. Assuming that IL feet in Gilbertese consist of iambs with right-adjuncts (see the discussion in Section 3.2), we conclude that Trochee_non-min and Iamb must dominate Iamb_non-min and Trochee. In (15) we illustrate these ranking arguments with the evaluation of the trimoraic form a^Hrána ‘his/her name.’ (In this analysis and subsequent ones, we only derive the stress patterns. The placement of a high tone would arise by a high-ranking alignment constraint which aligns a high tone with the left edge of the foot, strengthening the foot-initial position, Section 3.2.)

1. (15)

   

To ensure that arána is parsed with an IL foot [((a.rá).na)] rather than a non-layered foot [(a.rá).na], the constraints on unfooted syllables, Chain-L and Chain-R, must both dominate Trochee_non-min and Iamb. As previously mentioned, the main role of these constraints is to chain unfooted syllables toward the right or left edge of the prosodic word, while favoring exhaustive parsings of syllables. This promotion of exhaustivity indirectly favors the emergence of IL feet as shown in (16), where candidate (16a) beats candidates (16b–c). Since some forms do occasionally leave some unfooted material at the left edge of the prosodic word, we will see that Chain-L must dominate Chain-R (see 18).

1. (16)

   

In longer words, alternating ternary rhythmic stresses occur on every third mora preceding the stressed penultimate mora. We propose this is the result of building adjacent IL ternary feet from right to left. The constraint ranking responsible for such parsings is illustrated in (17). Align-L/R_min must crucially dominate Align-L/R_non-min so that parsings with IL feet are preferred over parsings with non-layered feet. Here and henceforth, we assume the syllabification patterns reported in Blevins and Harrison (Reference Blevins and Harrison1999); remember that coda nasals and prevocalic long nasals are always moraic.

1. (17)

   

Chain-L must dominate Chain-R because forms with 3n+1 morae may leave an unfooted syllable/mora at the left edge of the prosodic word. Furthermore, to ensure that the first mora of these forms does not constitute a unary degenerate foot on its own, Align-R_unary must be ranked above Chain-R. This is demonstrated in (18).

1. (18)

   

Now that the general metrical constraints responsible for the emergence and distribution of IL feet in Gilbertese have been presented, we need to add a small set of constraints to derive the stress patterns of the language, where moraic information is crucial in the assignment of stress. The list of these constraints and their definition are presented in (19) and (20). First, to promote metrical parsings in which morae rather than syllables are terminal elements of feet, we need to add three high-ranked foot form constraints controlling the particular size of minimal feet (19a-b) and promoting monomoraic adjuncts (19c). The last constraint, which favours IL feet with monomoraic adjuncts, is typologically motivated by the cross-linguistic observation that in ternary stress systems, the syllable in between binary feet must be light (Hayes Reference Hayes1995).

1. (19)

   a.	Foot-MinμμMin	FtMin must be minimally bimoraic.
   b.	Foot-MinμμMax	FtMin must be maximally bimoraic.
   c.	Adjunct=μ	The adjunct of FtNonMin must be monomoraic

The constraints in (19a–b) ensure that the maximal and minimal size of FootMin equals two morae and, thus, they can be interpreted as being specific constraints on Binarity. To predict the ternary stress moraic alternations in Gilbertese, the constraints in (19) together with, Non-Finality (banning the head syllable from final position) and Anchor-R, two classical metrical constraints defined in (20a,b), must be high-ranked, crucially dominating the Weight-to-Stress Principle constraint (WSP) and two additional general constraints on prosodic organization (Parse-μ and Exhaustivity-Ft, 20d–e). These constraints are all defined in (20). Note that there is an important difference between Parse-μ (20d) and Exhaustivity-Ft (20e). On the one hand, Parse-μ promotes the footing of morae in general. This constraint is needed in addition to Chain-L/R because the latter exclusively refers to the chaining of unfooted syllables. That is, in comparison, Parse-μ is stricter than Chain-L/R, since it bans all unfooted morae, also when they belong to chained syllables. On the other hand, Exhaustivity-Ft bans feet that directly dominate morae without an intervening syllable. This last constraint is typologically motivated since feet preferably dominate syllables. In (20) we will see that Exhaustivity-Ft is only violated in a few cases in Gilbertese (it is violated by candidates containing a heavy syllable in which the two morae belong to different feet). In other forms, it will be assumed that there is a syllable mediating between morae and feet.

1. (20)

   Metrical constraints
   a.	Non-Finality	No head of FtMin is final in PrWd (restricts the head syllable).
   b.	Anchor-R	Every PrWd must end with a foot.
   c.	WSP	Every bimoraic syllable must be head of a foot.
   Constraints on prosodic organization, mora parsing constraints
   d.	Parse-μ	Every mora must be parsed by a foot.
   e.	Exhaustivity-Ft	No foot immediately dominates a mora. Assign a violation mark for every mora that is directly dominated by a foot.

Regarding (20e), we propose that the pressure to build feet on syllables – not morae – is not exclusive to stress systems, but enforced by a universal constraint that is active in metrical systems. The metrical system of Gilbertese strongly suggests the responsible constraint (Exhaustivity-Ft) to be violable (in the sense of Optimality Theory), and to interact with other constraints on metrical well-formedness, in particular foot shape constraints. In Section 4, we will support this hypothesis by an IL foot-based analysis of Tokyo Japanese loanword accentuation, showing that its metrical system can be accounted for by a straightforward re-ranking of constraints motivated for Gilbertese. For this analysis to succeed, it will be crucial that metrical feet are IL, not standard moraic trochees.

In (21) we summarize the particular ranking of these constraints in Gilbertese. Non-Finality and Anchor-R (20a–b) together with Adjunct=μ, FtMin=μμ_Max, and FtMin=μμ_Min (19a–c) all dominate WSP to ensure that bimoraic sequences do not always surface with stress – remember that heavy syllables only surface with stress if they contain a mora that is at a ternary interval from another stressed mora or the right edge of the prosodic word. These constraints must also dominate Parse-μ. Finally, Parse-μ must dominate Exhaustivity-Ft, the constraint that bans feet that directly dominate morae.

1. (21) Non-Finality, Adjunct=μ, FtMin=μμ_Max, FtMin=μμ_Min, Anchor-R » Parse-μ » Exhaustivity-Ft, WSP

To illustrate this particular constraint interaction, and demonstrate how it can occasionally give rise to IL moraic feet in Gilbertese, tableau (22) presents the evaluation of a form ending in a sequence of HLL syllables such as te^Htákaa^Hkáro ‘the/a leisure activity.’ This form, with stress on the penultimate and the peninitial mora, is parsed with two adjacent IL feet; for example, [((te^H.tá).ka) ((a^H.ká).ro)]. Crucially, under the current proposal, the antepenultimate heavy syllable incurs a violation of the Syllable Integrity Principle (henceforth SIP), since the second mora belongs to the final foot and the first mora to the initial. In terms of constraints, this syllable violates Exhaustivity-Ft, because its morae are immediately dominated by a foot. As for the other morae, we assume they are dominated by syllables, which at the same time are dominated by feet.

1. (22) Form ending in H … LL. SIP Violation: ((te^H. tá).ka) ((a^H.ká).ro))

   

In this tableau we see that candidates (22b) and (22d) are ruled out because they violate Anchor-R, but note also that they incur a violation of the previously presented constraint Chain-R, omitted here for the sake of simplicity.

In Gilbertese, violations of the Syllable Integrity Principle (and, consequently, of Exhaustivity-Ft) only occur to ensure purely moraic ternary rhythmic alternations. Hence, in a HL form like a^Hńti, there is no violation of it; for example, [((a^Hń).ti)]. This tableau is also enlightening because it illustrates the ranking argument for Anchor-R and Parse-μ » Exhaustivity-Ft, so that candidate (23a) is selected over candidate (23b).

1. (23)

   

To sum up, along the lines of Blevins and Harrison’s (Reference Blevins and Harrison1999) original description, we conclude that a ternary constituent is responsible for the iterative rhythmic patterns in the language and, as such, IL feet of the size of three morae arise as a valid option for their representation. Further support for this metrical layered constituent is presented in the next section.

4.1 General Properties of Accentuation

The accentual pattern of Tokyo Japanese has been the subject of numerous phonetic and phonological studies (McCawley Reference McCawley1968, Reference McCawley and Fromkin1978; Poser Reference Poser1984; Pierrehumbert and Beckman Reference Pierrehumbert and Beckman1988; Haraguchi Reference Haraguchi1977, Reference Haraguchi1991; Kubozono Reference Kubozono2006). Traditionally, it is referred to as a “pitch accent system,” in which the lexical accent marks the position where a pitch drop (phonetically realized by a High–Low tonal melody; Haraguchi Reference Haraguchi1977) occurs in a word. The accented syllable is the one just before the pitch drop.Footnote ⁵ See the trisyllabic nouns in (24), accented on their first, second, and third syllable, combined with the nominative suffix -ga (Kubozono Reference Kubozono2012: 1398), where the pair of syllables that shows the pitch drop is underlined.

1. (24)

   a.	í.no.ti-ga	‘life-nom’
   b.	ko.kó.ro-ga	‘heart-nom’
   c.	o.to.kó-ga	‘man-nom’Footnote 6

For our current purposes, the interest of Tokyo Japanese resides in its “blended” prosodic character. McCawley (Reference McCawley and Fromkin1978) introduced a distinction between “accent/tone-bearing unit” and “unit of counting,” where the former is the unit that bears the prominence peak, while the latter is relevant for measuring distances from either side of the word.Footnote ⁷ In Tokyo Japanese, the prosodic unit that bears the prominence (the “accent-bearing unit”) is the syllable. This is because no prominence contrasts occur within heavy syllables: in an accented heavy syllable, the H tone is invariably associated with the first mora, and the L tone with the second mora. McCawley argues that the mora is the “unit of counting.” One of his major arguments is that “accent in recent borrowings goes on the syllable containing the third-last mora.” This suggests a reference to morae regardless of the way morae are grouped into sequences of (light and/or heavy) syllables.

Loanwords are mostly accented, and display considerable variation in the position of the accent, partly lexical and partly phonologically conditioned (by patterns of epenthetic vowels, the position of stress in the source language, etc.). Nevertheless, there is a default accentual pattern for loanwords, which is metrically restricted. This pattern will be our focus.

As reported by Kubozono (Reference Kubozono2006), two patterns of loanword accentuation occur: one dominant among older speakers, which we will refer to as the “conservative pattern,” and another one dominant among younger speakers, which we refer to as the “innovative pattern.”

4.2 The Conservative Pattern

The default pattern for Tokyo loanword accentuation is captured by the well-known “antepenultimate rule” (McCawley Reference McCawley1968, Reference McCawley and Fromkin1978; Yamada Reference Yamada1990; Haraguchi Reference Haraguchi1991; Kubozono Reference Kubozono2006):

1. (25)

   Loanword Accent Rule (“antepenultimate rule”)

   “Put an accent on the syllable containing the antepenultimate mora.”

This pattern is reported to be “conservative,” being dominant among older speakers (Kubozono Reference Kubozono2006: 1142).

In the examples in (26), the first column lists the quantitative structure of the form as a sequence of L(ight) and/or H(eavy) syllables. The second column shows the place of the accent on the mora carrying the H tone, which is abbreviated as APU (antepenultimate mora) or pre-APU (pre-antepenultimate mora).

1. (26)

   a.i	LLL	gú.ra.su	APU	‘glass’	K1142
   	LLLL	su.tó.re.su	APU	‘stress’	K1142
   	LLLLL	ku.ri.sú.ma.su	APU	‘Christmas’	K1142
   a.ii	HL	páa.ku	APU	‘park’	K1142
   	LHL	gu.róo.bu	APU	‘glove’	Kubozono (Reference Kubozono, van de Weijer and Nishihara2001)
   	LLHL	ri.ba.púu.ru	APU	‘Liverpool’	K1162
   	HHL	man.hóo.ru	APU	‘manhole’	K1155
   a.iii	LH	há.wai	APU	‘Hawaii’	K1142
   	LLH	a.má.zoN	APU	‘Amazon’	K1154 (fluctuates)
   	LLLH	e.ne.rú.gii	APU	‘energy’	K1154 (fluctuates)
   a.iv	HLH	kuu.dé.taa	APU	‘coup d’état’	McCawley (Reference McCawley and Fromkin1978)
   	HLH	kaa.dí.gaN	APU	‘Cardigan’	K1154 (fluctuates)
   	LHLH	a.koo.dí.oN	APU	‘accordion’	K1154 (fluctuates)
   b.i	HLL	dóo.na.tu	pre-APU	‘donut’	K1158
   	LHLL	e.díN.ba.ra	pre-APU	‘Edinburgh’	K1162
   	HHLL	pai.náp.pu.ru	pre-APU	‘pineapple’	K1152
   b.ii	HH	róN.doN	pre-APU	‘London’	K1142
   	LHH	wa.síN.toN	pre-APU	‘Washington’	K1142
   	LLHH	ba.do.míN.toN	pre-APU	‘badminton’	K1142
   	HHH	aN.dáa.soN	pre-APU	‘Anderson’	K1142

The theoretical interest of the Loanword Accent Rule is its dual reference to the syllable and the mora: “the syllable containing the antepenultimate mora,” where morae are “counting units” and syllables are “accent-bearing units.” Arguably, its reference to “antepenultimate mora” implies a metrical restriction, highly similar to those found in antepenultimate stress systems (such as Latin), which involve metrical foot parsing. Then the question emerges: which prosodic units are such feet built on? The two options are: (a) feet are built immediately on morae (thus bypassing the syllable level, possibly disrespecting syllable integrity, as in Gilbertese); or (b) feet are built on syllables (respecting syllable integrity, as Hayes Reference Hayes1995 assumes for metrical stress systems).

Under the first view, since Tokyo Japanese is a (metrically restricted) tone system, not a stress system, the assumption of SBU=σ (together with implied syllable integrity) might simply be irrelevant. Hence, there would be no obstacle to assuming the mora to be the unit of parsing. However, this assumption predicts that Tokyo Japanese freely violates syllable integrity just like Gilbertese, with the mora being the accent-bearing unit, incorrectly predicting that prominence contrasts occur within heavy syllables.

Let us now turn to the second view, which we will eventually adopt, and consider its implications for standard analysis with bimoraic feet. Considering that the syllable is evidently the accent-bearing unit in Tokyo Japanese, the syllable might be hypothesized to be the unit of foot parsing as well. Although this view correctly predicts that syllable integrity is respected, it cannot properly account for reference to the antepenultimate mora in the Loanword Accent Rule (25). To show this, we eliminate reference to the mora in (25), re-stating the rule in terms of light and heavy syllables:

1. (27)

   Loanword Accent Rule (syllable-based version)

   “Put an accent on the antepenultimate syllable in case the word ends in two light syllables; else on the penultimate syllable.”

This statement, while giving the appearance of simplicity, turns out to resist a foot-based analysis into standard moraic trochees, respecting syllable integrity. Although the pattern is seemingly similar to the well-known Latin pattern (“Stress is on the antepenultimate syllable in case the penult is light; else on the penultimate syllable.”), the crucial difference resides in the Tokyo Japanese pattern’s reference to the weight of the final syllable. Unlike Latin, the final syllable cannot be ignored in the assignment of foot structure, that is, be analysed as extrametrical. Two challenges emerge for an analysis in terms of the moraic trochee plus extrametricality. One major challenge is how to parse the material containing the antepenultimate and penultimate mora so as to account for penultimate syllable accentuation in final quantitative sequences as distinct as … HL# and … LH# (e.g. [gu.róo.bu], [a.má.zoN]). Another major challenge is how to account for the relevance of the final syllable’s weight in sequences such as HLL# ([dóo.na.tu], with antepenultimate accent) versus HLH# ([kaa.dí.gaN], with penultimate accent). We will consider these challenges by going through the various weight patterns one by one.

A moraic trochee analysis cum extrametricality is straightforward for cases in which the final syllable is light (e.g. [gúrasu], [guróobu], [dóonatu], [manhóoru]). Note in particular how the assumption that metrical feet are built on syllables correctly rules out antepenultimate mora accentuation (*[doónatu]) in (28c):

1. (28)

   a.	LLL	(gú.ra).su	c.	HLL	(dóo).na.tu	*do(ó.na).tu
   b.	LHL	gu.(róo).bu	d.	HHL	man.(hóo).ru

Next, words ending in two heavy syllables ([wasíNtoN] and [aNdáasoN]) are also correctly predicted under a moraic trochee analysis, assuming extrametricality of either the syllable or mora. Observe how, once more, syllable integrity correctly prohibits a bimoraic trochee parsing the second mora of a heavy penult together with the first mora of the final syllable (*[wasiŃtoN], *[aNdaásoN]).

1. (29)

   		σ extrametricality	μ extrametricality
   			SIP respected	SIP violated
   a.	LHH	wa.(síN).toN	wa.(síN).toN	*wa.si(Ń.to)N
   b.	HHH	aN.(dáa).soN	aN.(dáa).soN	*aN.da(á.so)N

However, the moraic trochee analysis runs into serious problems when dealing with forms ending in a light-plus-heavy sequence, such as [amázoN] or [kaadígaN], regardless of whether the syllable or the mora is taken to be the extrametrical unit.

1. (30)

   		σ extrametricality	μ extrametricality
   			SIP respected	SIP violated
   a.	LLH	*(á.ma).zoN	*(á.ma).zoN	a.(má.zo)N
   b.	HLH	*(káa).di.gaN	*(káa).di.gaNFootnote 8	kaa.(dí.ga)N

Suddenly, it becomes evident that syllable extrametricality fails to derive the … LH# patterns (30a, b). The moraic trochee analysis combined with syllable extrametricality predicts that the penultimate or antepenultimate syllable are uniformly accented regardless of the weight of the extrametrical syllable: if the penult is light, accent falls on the antepenult; if the penult is heavy, this would attract the accent, as in Latin stress. This prediction evidently fails in light of comparing HLL# ([dóonatu]), with accent on the antepenult, and HLH# ([kaadígaN]), where the accent is shifted to the penult. Hence, the final syllable’s weight remains invisible under syllable extrametricality. Alternatively, when assuming mora extrametricality, the correct forms with accent on the penult ([amázoN], [kaadígaN],) are, once more, not generated because syllable integrity prohibits moraic trochees whose second mora constitutes the first half of a heavy syllable.Footnote ⁹

In sum, Tokyo Japanese loanword accentuation has the syllable as its accent-bearing unit, since heavy syllables show no accentual contrast regarding the position of the pitch drop. Yet the “antepenultimate rule” is difficult to capture while maintaining syllable integrity, at least when adopting an analysis of standard bimoraic feet. One crucial stumbling block is the (unexpected) relevance of the weight of the final syllable in quantitative sequences ending in … HLX#, where X is light versus heavy ([dóo.na.tu] versus [kaa.dí.gaN]). The other problem is how to account for penultimate accent in final three-mora sequences that are mirror-images quantitatively, viz. … LH# and … HL# ([amázoN] and [guróobu]).

These problems can be immediately solved by adopting IL feet. Under this analysis, feet are uniformly and strictly right-aligned with the PrWd. Whether an IL or non-IL foot will be selected depends on the possible satisfaction of foot well-formedness constraints on the size of FtMin and IL foot’s dependent. An IL foot will be selected as a default option, favoring maximal moraic parsing, yet under the strict condition that the dependent of FtMin be monomoraic. This limits its occurrence to cases in which the final syllable is light (31a–b; e–f). In case the final syllable is heavy, a disyllabic non-IL foot will occur (31c–d; g–h) in which FtMin deviates from its bimoraic ideal size. Note that FtMin has bimoraic (31a, b, f), trimoraic (31c, g), and even quadri-moraic (31d, h) shapes.

1. (31)

   a.	LLL	((gú.ra).su)	e.	HLL	((dóo.na).tu)
   b.	LHL	gu.((róo).bu)	f.	HHL	man.((hóo).ru)
   c.	LLH	a.(má.zoN)	g.	HLH	kaa.(dí.gaN)
   d.	LHH	wa.(síN.toN)	h.	HHH	aN.(dáa.soN)

Note how the forms with IL feet (31a, b, e, f) are approximately parallel to the moraic trochee analysis: the final light syllable is either extrametrical (in the moraic trochee analysis) or dependent of FtMin (in the IL foot analysis). However, the parallelism between the analyses breaks down for forms ending in a heavy syllable (31c, d, g, h). Here, as shown in (30), moraic trochees only give the correct result with mora extrametricality while disrespecting syllable integrity; in contrast, our analysis respects syllable integrity by parsing such forms with a disyllabic non-IL foot. This particular set of IL and non-IL feet, whose size of FtMin ranges from two to four morae, follows immediately from the principles of foot well-formedness and moraic parsing as outlined in the analysis of Gilbertese; a simple re-ranking of the same constraints suffices to generate the conservative Tokyo Japanese pattern. We assume that feet must be strictly right-anchored with the PrWd, while dependents of FtMin occur on the right of FtMin, due to undominated Anchor-R, Trochee_non-min, constraints which we omit below.

1. (32) Ranking for Tokyo Japanese loanword pattern (conservative pattern):

   Non-Finality, Exhaustivity-Ft, FtMin=μμ_Min, Adjunct=μ » WSP » FtMin=μμ_Max » Parse-μ

In the following tableaux, we show how this analysis predicts the correct accentuations for all eight cases. No candidates that violate undominated Anchor-R or Trochee_non-min will be shown.

1. (33) … LLL# Parse-μ prefers an IL foot ((LL)L) to maximize parsing

   

1. (34) … HLL# WSP » FtMin=μμ_Max enforces an IL foot ((HL)L) with a trimoraic FtMin

   

1. (35) … LHL# WSP » Parse-μ excludes ((LH)L) rendering a perfectly shaped IL foot ((H)L)

   

1. (36) … HHL# FtMin=μμ_Max » Parse-μ enforces a perfectly shaped IL foot ((H)L)

   

1. (37) … LLH# Adjunct=μ » FtMin=μμ_Max, Parse-μ prefers non-IL foot (LH) to IL ((LL)H)

   

1. (38) … HLH# Adjunct=μ » Parse-μ prefers non-IL foot (LH) to IL ((HL)H)

   

1. (39) … LHH# Adjunct=μ » Parse-μ prefers non-IL foot (HH) to IL ((LH)H)

   

1. (40) … HHH# Adjunct=μ » Parse-μ prefers non-IL foot (HH) to IL ((LH)H)

   

Crucially, it should be noted that this idea (differentiating the syllable weight of the dependent, while varying the size of FtMin) cannot be transposed to a “moraic-trochee-cum-extrametricality” analysis. On the analogous assumption that only final light syllables are extrametrical (41a–b, e–f), the parsing of the quantitative sequence LH becomes inconsistent. That is, LH is evidently metrified as L(H) – an unparsed light syllable plus bimoraic trochee – in L(H)L# (41b); however, in order to derive the correct output, the same sequence needs to be metrified as (LH) – trimoraic trochee (LH) – in L(LH)# (41c).

1. (41)

   a.	LLL	(gú.ra).su	e.	HLL	(dóo).na.tu
   b.	LHL	gu.(róo).bu	f.	HHL	man.(hóo).ru
   c.	LLH	a.(má.zoN)	g.	HLH	kaa.(dí.gaN)
   d.	LHH	wa.(síN.toN)	h.	HHH	aN.(dáa.soN)

The upshot of the IL foot analysis is that it derives the conservative pattern of loanword accentuation from constraints that are independently motivated for Gilbertese, by means of a re-ranking. Crucially, the IL foot analysis succeeds by imposing a strict condition on the maximal size of the dependent. We now turn to the innovative Tokyo pattern, which will provide further evidence for the IL foot analysis.

4.3 The Innovative Pattern

This pattern is innovative, dominant among younger speakers. It is identical to the conservative pattern except that all forms ending in a light–heavy sequence undergo an accent shift of one mora to the left.Footnote ¹⁰

1. (42)

   a.i	LLH	á.ma.zoN	pre-APU	‘Amazon’	K1154 (fluctuates)
   	LLLH	e.né.ru.gii	pre-APU	‘energy’	K1154 (fluctuates)
   a.ii	HLH	káa.di.gaN	pre-pre-APU	‘Cardigan’	K1154 (fluctuates)
   	LHLH	a.kóo.di.oN	pre-pre-APU	‘accordion’	K1154 (fluctuates)

According to Kubozono (Reference Kubozono2006: 1155), “this kind of age-related variation suggests that the following accent changes have been in progress in Tokyo Japanese and are actually almost complete now, with the new patterns overwhelming the old ones in statistical terms”.

Kubozono (Reference Kubozono1996, Reference Kubozono2006) observes that the innovative pattern is identical to the antepenultimate stress pattern of Latin. This well-known pattern can be captured straightforwardly by a moraic trochee plus syllable extrametricality:

1. (43)

   a.	LLL	(gú.ra).su	e.	HLL	(dóo).na.tu
   b.	LHL	gu.(róo).bu	f.	LHH	wa.(síN).toN
   c.	LLH	(á.ma).zoN	g.	HLH	(káa).di.gaN
   d.	HHL	man.(hóo).ru	h.	HHH	aN.(dáa).soN

Analogously to the classical extrametricality/non-finality analysis, we set up an IL foot with a bimoraic FtMin to which the final syllable adjoins into FtMax:

1. (44)

   a.	LLL	((gú.ra).su)	e.	HLL	((dóo.na).tu)
   b.	LHL	gu.((róo).bu)	f.	LHH	wa.((síN).toN)
   c.	LLH	((á.ma).zoN)	g.	HLH	((káa.di).gaN)
   d.	HHL	man.((hóo).ru)	h.	HHH	aN.((dáa).soN)

We may interpret the shift from the conservative pattern to the innovative pattern as a relaxation of the requirement that the dependent of FtMin be monomoraic, which now includes bimoraic dependents as well as monomoraic ones, while still respecting syllable integrity. A relaxation of the dependent’s size may be interpreted as bringing about consistency in the shape of feet in loanword accentuation; non-IL disyllabic feet, ranging in size from three to four morae, are transformed into IL feet in which FtMin is consistently bimoraic except in (44e, g), where bimoraic FtMin is unattainable due to syllable integrity.

The change from the conservative pattern to the innovative pattern is analysable as a radical demotion of Adjunct=μ to the bottom stratum of the hierarchy, increasing the influence of FtMin=μμ_Max and Parse-μ.

1. (45) Ranking for Tokyo Japanese loanword pattern (innovative pattern):

   Non-Finality, Exhaustivity-Ft, FtMin=μμ_Min » WSP » FtMin=μμ_Max » Parse-μ, Adjunct=μ

1. (46) … LLL#   Parse-μ prefers an IL foot ((LL)L) to maximize parsing

   

1. (47) … HLL# WSP » FtMin=μμ_Max enforces an IL foot ((HL)L) with a trimoraic FtMin

   

1. (48) … LHL# WSP » Parse-μ excludes ((LH)L) rendering a perfectly shaped IL foot ((H)L)

   

1. (49) … HHL# FtMin=μμ_Max » Parse-μ enforces a perfectly shaped IL foot ((H)L)

   

1. (50) … LLH#   FtMin=μμ_Max » Adjunct-μ enforces an IL foot ((LL)H)

   

1. (51) … HLH#   WSP » Adjunct-μ enforces an IL foot ((HL)H)

   

1. (52) … LHH#   FtMin=μμ_Max » Adjunct-μ enforces an IL foot ((H)H)

   

1. (53) … HHH FtMin=μμ_Max » Adjunct-μ enforces an IL foot ((H)H)

   

In sum, the two loanword accentuation patterns of Tokyo Japanese can be straightforwardly analysed by means of IL feet, involving a simple re-ranking of a single constraint set which can be understood to improve the consistency between the shapes of FtMin across the range of weight sequences.

4.4 Comparison with Moraic Trochee

Next, in order to facilitate a comparison with the standard bimoraic foot analysis of Tokyo Japanese, we briefly turn to the bimoraic foot analysis of loanword accentuation, as well as additional prosodic (morphological) phenomena: word minimality, loanword truncation, and lengthening and shortening patterns that are observed in a variety of contexts. We will show that these phenomena can be handled straightforwardly under the new analysis even when permitting IL feet with three or four morae. Note that the crucial element that the two analyses share is a bimoraic unit: the moraic trochee in the standard analysis, and the optimal shape of FtMin (confirming to FtMin=μμ_Min and FtMin=μμ_Max).

Tokyo Japanese loanword accentuation. The traditional strictly bimoraic foot structures for Tokyo loanword accentuation (e.g. Suzuki Reference Suzuki1995; Shinohara Reference Shinohara2000; Kubozono Reference Kubozono2006) were not included in the previous tableaux for the sake of simplicity. The following examples in (54) and (55) explicitly show how these forms are excluded in our analysis. The examples in (54) present the evaluation of forms in the conservative pattern with strictly bimoraic foot structures, whereas those in (55) illustrate the evaluation of bimoraic forms for the innovative pattern. Their fate is sealed by two undominated constraints, Anchor-R and Align-R_main, and occasionally by Exhaustivity-Ft:

1. (54)

   Strictly bimoraic analyses of the conservative pattern
   		Anchor-R	Align-Rmain	Exhaustivity-Ft
   a.	(gú.ra).su	*
   b.i	(dóo).(na.tu)		*
   b.ii	(dóo).na.tu	*
   c.	gu.(róo).bu	*
   d.	(maN).(hóo).ru	*
   e.	a.(má.zo)N	*		*
   f.	(kaa).(dí.ga)N	*		*
   g.i	wa.(síN).(toN)		*
   g.ii	wa.(síN).toN	*
   h.i	(aN).(dáa).(soN)		*
   h.ii	(aN).(dáa).soN	*

Note that in cases where the accent falls on the pre-antepenultimate mora (/dóo.na.tu/, /wa.síN.toN/, /aN.dáa.soN/), two strictly bimoraic analyses could be proposed: one in which the last two morae are unparsed, which are excluded by Anchor-R (54b.ii, 54g.ii, 54h.ii), and another in which the last two morae form a weak foot (which are excluded by Align-R_main, 54b.i, 54g.i, 54h.i). Furthermore, as we already argued in Section 4.2, forms ending in … LH# (/a.má.zoN/, /kaa.dí.gaN/) are intrinsically problematic for any strictly bimoraic analysis. For example, the analyses proposed in (54e, f) violate syllable integrity (Exhaustivity-Ft). Alternatives that satisfy syllable integrity must either fall below the bimoraic foot size, thus violating FtMin=μμMin (e.g. /a.(má).zoN/, /kaa.(dí).gaN/) or else, deviate from trochaic foot shape, thus violating trochee_non-min (e.g. /(a.má).zoN/, /(kaa.dí).gaN/).

The bimoraic analysis of the innovative pattern is identical to (54) except in forms ending in … LH#, which are given in (55).

1. (55) Strictly bimoraic analyses of the innovative pattern

   		Anchor-R	width-L/Rmain	Exhaust-Ft
   a.i	(á.ma).(zoN)		*
   a.ii	(á.ma).zoN	*
   b.i	(káa).di.(gaN)		*
   b.ii	(káa).di.gaN	*

We thus conclude that our IL foot analysis straightforwardly excludes the forms posited by the strictly bimoraic analysis, and restate our earlier conclusion that antepenultimate mora accentuation in … LH# forms in the conservative pattern are intrinsically problematic for the strictly bimoraic analysis.

Word minimality. Derived words in Japanese (such as loanword truncations and hypocoristics) must be at least two morae long (Itô Reference Itô1990; Poser Reference Poser1990; Kubozono Reference Kubozono and Tsujimura1999), for example /su.to.rai.ki/ → /su.to/ ‘strike.’ The bimoraic minimality effect is straightforwardly accounted for by our analysis due to undominated FtMin=μμ_Min.

HL versus LH asymmetry in loanword truncation. Loanwords are shortened to forms that are two to four morae in length. In the case of trimoraic outputs a well-known quantitative asymmetry holds: HL and LLL outputs are permitted, not LH outputs (Itô Reference Itô1990).

1. (56)

   a.	/roo.tee.syon/	→ /roo.te/	‘rotation’
   b.	/pan.fu.ret.to/	→ /pan.fu/	‘pamphlet’
   c.	/a.ni.mee.shoN/	→ /a.ni.me/	‘animation’
   d.	/te.re.bi.joN/	→ /te.re.bi/	‘television’
   e.	/ro.kee.syon/	→ /ro.ke/, */ro.kee/	‘location’
   f.	/de.mon.su.to.ree.syon/	→ /de.mo/, */de.mon/	‘demonstration’

The same HL versus LH asymmetry is found in various phenomena in Japanese including zuuzya-go formation and baby talk words (Kubozono Reference Kubozono, Féry and van de Vijver2003). The standard analysis of truncation states that the output must conform to the requirement of a PrWd with a left-aligned bimoraic trochee; this licences trimoraic forms [(H)L] and [(LL)L], while excluding *[L(H)], due to left-edge mis-alignment. Under our IL foot analysis, the HL versus LH asymmetry is straightforwardly accounted for. Crucially, licit trimoraic outputs of truncated loanwords are unmarked expansions of the IL foot; for example, HL [((roo).te)], LLL [((a.ni).me)], where “unmarked” means satisfying all major foot form and alignment constraints (FtMin=μμ_Min, FtMin=μμ_Max, Adjunct=μ, Trochee_non-min, Chain-L/R). In contrast, LH outputs are not unmarked IL feet, since on any possible analysis, one or more foot form or alignment constraints are violated; for example, *[(ro.kee)] (FtMin=μμ_Max), *[((ro).kee)] (FtMin=μμ_Min, Adjunct=μ), or *[ro.(kee)] (Chain-R). Given our constraint ranking, the least damaging way to parse LH is by means of a binary foot (LH), which only occurs under duress of foot form constraints, in particular due to the pressure to build a licit foot without deleting any segments or splitting bimoraic syllables (see tableaux 37–38). In loanword truncation, the factors that enforce a marked parsing (LH) are relaxed, as evidently full segmental faithfulness is not required. The “emergence of the unmarked” is a well-known phenomenon from prosodic morphology, including truncation (McCarthy and Prince Reference McCarthy, Prince, Beckman, Dickey and Urbanczyk1995). Without offering a complete analysis, it suffices to observe that Japanese loanword truncations can be viewed as an emergence of the unmarked, driven by markedness constraints in a specific situation in which segmental faithfulness constraints (penalizing deletion) are low-ranked. Hence we feel safe to conclude that the Japanese loanword truncation templates fall out of our grammar as unmarked expansions of the IL foot.

Lengthening and shortening patterns. Kubozono (Reference Kubozono, Féry and van de Vijver2003) reports several patterns of lengthening and shortening that are found in a variety of contexts, including loanword truncation, zuuzya-go formation, emphatic mimetics, and motherese (baby talk words). These patterns are summarized in (57–58):

1. (57) Shortenings

   a. LH → LL truncation /ro.kee.syon/ → /ro.ke/ ‘location’

   b. HH → HL zuuzya-go /koo.hii/ → /hii.ko/, /hii.koo/ ‘coffee’

   sporadic /tyoo.tyoo/ → /tyoo.tyo/ ‘butterfly’

1. (58) Lengthenings

   a.	LL → HL	zuuzya-go	/zya.zu/ → /zuu.zya/	‘jazz’
   		motherese	/ma.ma/ → /mam.ma/	‘food, to eat’
   		sporadic	/si.ka/ → /sii.ka/	‘poem’
   		mimetics	/pi.ka.pi.ka/ → /pik.ka.pi.ka/	‘shiny’
   b.	LH → HH	sporadic	/zyo.oo/ → /zyoo.oo/	‘woman king, queen’
   c.	L → HL	zuuzya-go	/hi/ → /ii.hi/	‘fire, cigarette light’
   		motherese	/ha/ → /pap.pa/	‘leaf’
   d.	L → LLL	motherese	/te/ → /o.te.te/	‘hand’
   e.	H → HL	zuuzya-go	/kii/ → /ii.ki/	‘love’
   f.	H → HH	motherese	/hau/ → /hai.hai/	‘crawling’
   g.	LH → HL	zuuzya-go	/hu.men/ → /men.hu/	‘taboo’
   		motherese	/o.buu/ → /om.bu/	‘a piggyback ride’

Interestingly, the target of a considerable number of changes happens to be HL, a perfectly shaped IL foot ((H)L); see (57b, 58a, 58c, 58e, 58g), and occasionally ((LL)L); (58d). Other changes seem to be driven by an avoidance of LH, a non-IL foot (LH), which occurs only under duress of foot form constraints in our analysis of loanword accentuation; see (57a, 58b, 58g), and (L), a degenerate FtMin; (58c, d). Occasionally, HH seems to be the target (58b, f), which in many instances provided involves segmental reduplication in motherese (Kubozono Reference Kubozono, Féry and van de Vijver2003: 107). Without giving a full analysis, we conclude that both the targets of these changes and the forms avoided are expected from our IL foot model.Footnote ¹¹

5.1 Double Accent in Altaic Languages

It has been argued that a word in Altaic languages has two types of accent: initial stress and right-oriented pitch accent (Matthews Reference Matthews1951: 60). Poppe’s (Reference Poppe1965: 180–181) observations are cited in (2).Footnote ⁹,Footnote ¹⁰

1. 1. a. The expiratory, dynamic stress is in all Altaic languages bound to the same syllable. It is fixed and, therefore, non-phonemic. Side by side with the dynamic stress there is also a musical tone.

   2. b. In most Turkic languages, for example, Turkish, Uzbek, Shor, Sagai, Yakut, etc., the expiratory stress falls on the first syllable … The musical tone is independent of the stress and falls on the last syllable, for example, Turkish áyàq ‘foot,’ áyaγìm ‘my foot.’

   3. c. In Mongolian the dynamic, expiratory stress, is usually on the first syllable … The musical tone is on the last syllable in di- and trisyllabic words, for example, Kh[alkha] írsèn ‘one who has come,’ írsendè ‘when he came.’

   4. d. Tungus has a dynamic stress and a musical tone. The first falls on the first syllable, the latter is on the second syllable. When a monosyllabic suffix is added the musical tone shifts upon the latter, for example, Udehe: tɑ́dɑ̀ ‘arrow,’ cf. tɑ́dɑȷı̀ ‘by means of an arrow.’ When a disyllabic suffix is added, the musical tone is on the second and the last syllable, the dynamic stress disappearing, for example, tadàtigı̀.

   5. e. In Korean there is an expiratory stress which is on the first syllable.

Sunik (Reference Sunik, Алпатов, Кормушин, Пюрбеев and Романова1997: 156) also makes some observations about Tungusic-Manchurian languages, as translated in (3).Footnote ¹¹

1. 1. a. In some languages, especially in Manchu, weak force stress falls on the first syllable of the word, often combined with the lengthening of the accented vowel (Tungus languages).

   2. b. Musical stress falls on the final syllable of the word (as can be judged according to the Nanai, Ulchi, Udehe languages) – a slight increase in tone, in some cases accompanied by emphatic lengthening of the final vowel, which is sometimes mistaken for a force stress (of long-tonic type).

Svantesson et al. (Reference Svantesson, Tsendina, Karlsson and Franzén2005: 94–97) discuss word stress in Mongolian, and list six different opinions in the literature about the position of Mongolian stress.Footnote ¹² The first opinion, which Svantesson et al. introduce as the opinion of almost all native Mongolian scholars, is that stress falls on the first syllable (see references cited by Svantesson et al. (Reference Svantesson, Tsendina, Karlsson and Franzén2005: 96–97)).Footnote ¹³

Similar observations are made about Turkic languages by Johanson (Reference Johanson, Johanson and Csató1998a: 34). He describes pitch accent and dynamic stress in Turkic, as shown in (4).

1. 1. a. Most Turkic languages have pitch accent, that is an increase of the tone height, on the last syllable of native lexical items.

   2. b. There is also an interesting changeable dynamic stress accent, characterized by more energy of articulation. It tends to fall on the first syllable, and seems to be the original factor of rhyme patterns in Old Turkic poetry. Being sensitive to phonetic factors such as weight, it often falls on heavy syllables, that is, closed syllables or syllables with a long vowel, for example Turkish evde [ˈεvdέ] ‘at home.’

Johanson (Reference Johanson, Johanson and Csató1998a: 35) notes that pitch accent and dynamic stress “are distributed differently in the individual Turkic languages.” He observes that “Central Asian languages often tend to give more prominence to the initial syllable; languages of the Volga-Kama region to the last syllable.” Importantly, he argues that “[t]he differences between pitch and stress accent are usually ignored in studies of Turkic accent systems.”

As for Turkish, Csató and Johanson (Reference Csató, Johanson, Johanson and Csató1998: 207) observe pitch accent on the last syllable and stress on the first syllable as shown in (5).

1. 1. a. Pitch accent is normally on the last syllable of native lexical stems and expanded forms of them containing accentable suffixes, for example elmá ‘apple,’ elmalár ‘apples,’ elmalardán ‘from apples.’

   2. b. The distribution of dynamic stress, marked with ˈ in front of the syllable, is less predictable. It often falls on the first syllable, in particular if the latter is heavy (closed or with a long vowel), but can also move to other syllables.

This prosodic pattern is also seen in other Turkic languages. Schönig (Reference Schönig, Johanson and Csató1998: 249) observes that in Azerbaijani “[t]here is normally a pitch accent on the last syllable and often expiratory stress on the first one.” Similarly, Kirchner (Reference Kircher, Johanson and Csató1998: 320) observes that “[a]s in most Turkic languages,” Kazakh and Karakalpak have pitch accent (rising tone), which is normally on the last syllable, and stress accent, which normally occurs on the first syllable.Footnote ¹⁴

Based on these observations, I argue that a word in Altaic languages has a double accent, namely, initial stress and (near-)final pitch accent, and that the word-initial stress correlates with the head-final word order in Altaic languages (cf. Duanmu Reference Duanmu2008 for the view that Japanese has two types of accent). The question then arises as to why Altaic languages have double accents. The pitch accent in the final part of a word may well be phonemic: it distinguishes one word from another in a language. On the other hand, word-initial stress is neither lexical nor phonemic in the sense that it does not function to distinguish one word from another. Then, what kind of functions does this word-initial stress have? One possibility is that the stress marks the beginning of a word to show the word boundary to hearers. Perhaps word-initial stress has “culminative” and “delimitative” functions, in that it can indicate how many words there are and where the boundaries are (Trubetzkoy Reference Trubetzkoy and Baltaxe1969; Hyman Reference Hyman and Hyman1977; Hayes Reference Hayes1995; cf. Duanmu Reference Duanmu2008: 32). This seems to be a plausible idea because pitch accent may not be sufficient to show the word boundaries especially when pitch is flattened in connected speech. For example, the Japanese tongue twister sumomo-mo momo-mo momo-no uchi ‘plums and peaches are (in the same category of) peaches,’ which consists of unaccented words and is pronounced in a flat pitch LHHH …, can be easier to utter and to be understood if one uses word-initial stress as in sumomo-mo momo-mo momo-no uchi) even though there is no (falling) pitch accent.

It is also interesting to compare the double accent system of Altaic languages with other languages that have both stress and tone. For example, Kager and Martinez-Paricio (Chapter 5 this volume) discuss Gilbertese (a Micronesian language: Oceanic), which has both high pitch and stress. This example shows that Altaic languages are not a typologically isolated case. However, a comparison of the double accent systems in the world’s languages is well beyond the scope of this chapter and I will leave this matter open here. In the following sections, I present some phonological arguments for the word-initial stress in Altaic.Footnote ¹⁵

5.2 Vowel Reduction in Non-initial Syllables

First of all, in a number of Altaic languages, vowel reduction occurs in non-initial syllables in a word, but not in initial syllables. Given that vowel reduction universally occurs in unstressed syllables rather than in stressed syllables (e.g. policeman [pəlíːsmən]), vowel reduction in non-initial syllables is consistent with the hypothesis of word-initial stress in Altaic. For example, short vowels in some Altaic languages are reduced in non-initial position but not in word-initial position. This is generally the case in Mongolic languages. In Khalkha, the “short” vowels of non-initial syllables are reduced (centralized) versions of the vowel of the preceding syllable (e.g. xawar [xawər] ‘spring,’ mongol [mɔŋɢəɮ] ‘Mongol,’ guril [ɢurʲəɮ] ‘flour,’ ajil [aʧəɮ] ‘work’) (Svantesson Reference Svantesson and Janhunen2003: 158).Footnote ¹⁶ Bläsing (Reference Bläsing and Janhunen2003: 231) observes that in Kalmuck (Mongolic) the opposition between short and long vowels is valid only for the initial syllable and that “[i]n non-initial syllables, original short vowels disappear or are strongly reduced (retaining no phonemically relevant qualitative oppositions), for example pl. ek.nr ‘mothers’ < *eke.ner.” Birtalan (Reference Birtalan and Janhunen2003: 213) observes that vowel reduction in non-initial syllables is seen in spoken Oirat (Mongolic) as well as in Kalmuck and Mongol proper. She also points out that the long vowels of non-initial syllables in written Oirat become short (single) vowels in spoken Oirat; for example, *imaa/n ‘goat’ > Written yamaa/n > Spoken yama/n. Kuribayashi (Reference Kuribayashi1988: 210) discusses the development of reduced vowels in modern Mongolian languages such as Khalkha, Chakhar, Buriat, Kalmuck, and Dagur, and observes that in these languages the initial syllable of a word always receives stress (intensity accent) while the vowels in the other syllables are remarkably reduced. In Bonan (Bao’an), Hugjiltu (Reference Hugjiltu and Janhunen2003: 328) observes that “[i]n non-initial syllables, a regular reductive merger of the high vowels *i *u *ü into e can be observed, as in (*i:) gholer ‘flour’ < *gulir, (*u:) nase ‘age’ < *nasu/n, (*ü:) under ‘high’ < *öndür.” Weiers (Reference Weiers and Janhunen2003: 253) argues that the most important factor lying behind the absence of vowel harmony in Moghol is the diachronic change (neutralization) of the vowel *e into a in all non-initial syllables.

Vowel reduction in non-initial syllables is also seen in Tungusic languages. Ikegami (Reference Ikegami, Kamei, Kohno and Chino1989: 1070) observes that /a/ and /ə/ in Even may be reduced in non-initial syllables. He also observes that short vowels in Solon are pronounced ambiguously in non-initial syllables. Tsumagari (Reference Tsumagari2009: 3) notes that short vowels in non-initial syllables tend to be reduced and sometimes dropped in casual speech.

Turkic languages do not have vowel reduction in non-initial syllables. This seems to be due to their tendency to have vowel harmony instead of vowel reduction. I argue below that left-to-right vowel harmony is due to the stress in the initial syllable.

What is crucial here is the fact that vowel reduction usually does not occur in word-initial position in Altaic languages. Thus, the location of vowel reduction supports the idea that stress falls on the word-initial position rather than non-initial positions in Altaic languages.

One might argue that the word reduction phenomenon should not be attributed to stress but may be attributed to the general observation that word-initial (or root-initial) positions are universally privileged and phonologically more prominent positions than other positions, thus enhancing phonological contrasts (cf. Beckman Reference Beckman1998, among others). However, vowel reduction can readily occur in word-initial syllables if they are not stressed; for example, in languages with right-hand stress. For example, in English, which has right-oriented stress, vowel reduction occurs in unstressed word-initial position (e.g. able [éɪbl] → ability [əbɪ́ləti]), and in Romance, which has right-edge stress (e.g. tɔ́ro ‘bull’ → tor + ɛ́llo ‘bullock’ (the tensing of mid-lax vowels in Standard Italian, Miglio Reference Miglio2005: 56)). Thus, reduction is attributed to word-stress location instead of the alleged universal prominence of the word-initial position.

5.3 The Variety of Vowels in Initial versus Non-Initial Positions

The second argument for word-initial stress in Altaic, which is related to the first argument given in the previous section, is the variety of vowels appearing in the initial position versus the non-initial positions of words (cf. Hayata Reference Hayata and Hayata2005: 17). In some languages, the variety of vowels in stressed positions is larger than that in unstressed positions. For example, Italian has seven vowels in stressed positions (i, e, ε, u, o, ɔ, a) and five in unstressed positions (i, e, u, o, a) because of the neutralization of the contrast between close-mid (/e, o/) and open-mid (/ɛ, ɔ/) vowels (Rogers and d’Arcangeli Reference Rogers and d’Arcangeli2004: 119). Similarly, Russian has five (or six) vowels in stressed positions (i, e, u, o, a, (ɨ)) and three in unstressed positions (i, u, a). Given that this asymmetry of vowels in stressed/unstressed positions is universal (cf. Miglio Reference Miglio2005: 3, 189), comparing the varieties of vowels in a certain position with the other positions shows us the stress location in a language. In Altaic languages, the inventory of vowels is larger in initial position than in non-initial positions, which supports the hypothesis of initial stress in Altaic languages.

In Turkish, eight vowels appear in the word-initial position (i, y, e, a, ɯ, u, ø, o) while only six of them (i, y, e, a, ɯ, u) appear in non-initial positions, with two of them (ø, o) missing in non-initial positions. In Azerbaijani, nine-vowels (i, y, e, ø, ɯ, u, o, ɑ, æ) appear in the word-initial position but one of them (æ) does not appear in non-initial positions.

Mongolic languages also have a greater variety of vowels in the word-initial position than in non-initial positions. For example, in Dagur, five vowels (i, e, u, ɑ, o) occur in the word-initial position while only three of them (i, e, u) occur in non-initial positions (Tsumagari Reference Tsumagari and Janhunen2003: 131). Skribnik (Reference Skribnik and Janhunen2003: 107) describes how in Buryat, “[a]s a sign of incipient vowel reduction, the paradigm of short vowels in non-initial syllables has slightly diminished” and “the high rounded vowels *u *ü have merged with the low vowels *a *e, as in aba ‘father’ < *abu, *üder ‘day’ < *ödür.” Hugjiltu (Reference Hugjiltu and Janhunen2003: 328) observes that in Bonan (Bao’an), “[t]he long vowels are normally attested in the initial syllable only.”

Also in Tungusic languages, the variety of vowels is larger in the word-initial position than in non-initial positions because vowel reduction occurs in non-initial positions as we have just seen. For example, according to Tsumagari (Reference Tsumagari2009: 3), Solon has eight short vowels (e, ö, ü, a, o, u, i, ɛ), which tend to be reduced in non-initial positions.

Note that we cannot show that the reverse is also true: if a language has right-hand stress, we cannot expect it to have a smaller variety of vowels in the word-initial position. For example, in a language with penultimate stress, stress is in fact realized on the word-initial syllable in two-syllable words and on the second syllable in three-syllable words (e.g. rózpraw ‘discussion [gen. pl]’/rozpráwa ‘discussion [nom. sg]’) in Polish (Goedemans, Heinz, and van der Hulst Reference Goedemans, Heinz and van der Hulst2014)). Then, we expect right-hand-stress languages to have the same variety of vowels in the word-initial position as in the non-initial positions. All we can say about these languages is that right-hand-stress languages have a larger variety of vowels in stressed positions than in unstressed positions. This is what Miglio (Reference Miglio2005: 56) observes about Romance languages, which have right-edge stress. She writes that “[i]n less prominent environments, such as unstressed syllables, the Romance languages show a restricted variety of vowel qualities compared to the vowels in stressed syllables.”

5.4 Vowel Harmony

The third argument is that progressive (left-to-right) vowel harmony in Altaic languages also shows that phonological strength is on the word-initial syllable. As is well known, in Turkish and Mongolian languages, vowels in non-initial syllables assimilate to the vowel in the word-initial syllable. For example, Turkish has vowel harmony with respect to front (e, i, ö, ü)/back (a, ı, o, u) distinction: köpek ‘dog,’ quɫak ‘ear,’ aːɫa-ma ‘cry-ing,’ çek-me ‘pull-ing.’ Similarly, Mongolic has pharyngeal vowel harmony with three pharyngeal vowels [ʊ, ɔ, a], three non-pharyngeal vowels [u, o, e], and one neutral vowel [i] (e.g. ugui ‘no,’ tʊɢʊi ‘circle’) (cf. Svantesson et al. Reference Svantesson, Tsendina, Karlsson and Franzén2005: 49). Ikegami (Reference Ikegami, Kamei, Kohno and Chino1989: 1071) observes that in Tungusic languages Class I vowels and Class II vowels usually do not co-occur in the same word (e.g. ice-ken ‘somewhat new,’ dacu-kan ‘somewhat sharp,’ in Written Manchu (cf. Zhang Reference Zhang1996: 49); ’ana-wa ‘boat-Acc,’ zolo-wo ‘stone-Acc’ in Udihe (cf. Nikolaeva and Tolskaya Reference Nikolaeva and Tolskaya2001: 75)).

Hyman (Reference Hyman2002: 14) argues that “a canonical trigger V occurs in a prominent syllable (e.g. root, stressed) with a contrastively specified feature” while “a canonical target V occurs in a non-prominent syllable (e.g. affix, unstressed) with a non-contrasting unspecified feature.” If this generalization holds, we can argue that Turkic and Mongolic languages have stress on the first syllable, which affects the following syllables in the word, including suffixes (for the possibility of non-canonical vowel height harmony initiated by unstressed vowels, see Walker (Reference Walker2005).

One may expect that the reverse happens in languages with right-hand stress. Right-hand-stress languages are expected to have vowel harmony from right to left in a word. This is the case in Ineseño (Chumashan, southern coastal California), which is reported to have penultimate stress (Gordon Reference Gordon2002: 542). Hyman (Reference Hyman2002: 7) gives the examples in (6), where the prefixes /a-/ and /u-/ undergo Vowel Harmony.

1. 1. a. /qal-wala-tepet/ → [qel-wele-tepet]

      of tying-with body/bulky obj.- roll ‘to roll up and tie a bundle’

   2. b. /aqpala-woyoc/ → [oqpolo-woyoc]

      of grinding-twist/be crooked ‘to wear down crookedly’

   3. c. /su-yul-c’ɨ/ → [sɨ-yɨ-c’ɨ]

      cause-of heat-be sharp ‘to heat’

The right-to-left vowel harmony can also be found in some Bantu languages (Southern Sotho, Northern Sotho, and Tswana) in “Zone S” (Gowlett Reference Gowlett, Nurse and Philippson2003: 612). In Sotho, which has stress on the penultimate syllable in a word, /ɛ/ and /ɔ/ are replaced by /e/ and /o/ when they immediately precede any higher vowel in a word stem, as shown in (7).

1. 1. a. mʊ̀≠rėk’-í ‘buyer’ < rɛ́k’-à ‘buy’

   2. b. bóf-úw-á ‘be tied’ < bɔ́f-à ‘tie’

The examples in (6) and (7) show that languages with right-hand stress may have vowel harmony from right to left in a word. If it is generally true that the direction of vowel harmony correlates with word-stress location, progressive (left-to-right) vowel harmony in Altaic languages supports the idea that phonological strength is on the word-initial syllable in Altaic languages.

Note here that Sotho also has left-to-right vowel harmony as well as right-to-left vowel harmony: -ɛ and -ɔ in a suffix are replaced by -e or -o when preceded by /e/ in a verb root, as shown in (8a) and (8b).

1. 1. a. i≠ʧw’éts’-è! ‘Tell yourself!’ < ì≠ʧw’éts’-à ‘tell oneself’

      cf. í≠tshɛ́p’-ɛ̀! ‘Trust yourself!’ < ì≠tshɛ́p’-à ‘trust oneself’

   2. b. k’èts’-ò ‘act, action’ < èts’-à ‘do’

      cf. k’ɛ̀p’-ɔ̀ ‘digging’ < ɛ̀p’-à ‘dig’

This left-to-right vowel harmony is triggered by the stress on the root, which changes the vowel in the affix. Thus, the left-to-right vowel harmony in Sotho goes well with the idea that the direction of vowel harmony correlates with word-stress location.

5.5 Alliteration in Verse and Wordplay

Next, let us consider rhythm in literature, which can be argued to be related to word-stress location. The change from alliterative to rhymed verse in the history of Germanic and Celtic languages has been attributed to the influence of Romance languages, and in discussing this change Donegan and Stampe (Reference Donegan and Stampe1983: 349) argue that alliteration and rhyme in poetry correspond to falling and rising rhythm in language, which roughly correspond to initial and final stress. They claim that Munda languages have alliterative verse and initial stress while Mon-Khmer languages have rhymed verse and final stress. Meid (Reference Meid1971: 105–106) also proposes a typological generalization shown in (9) (cited in Plank Reference Plank1998: 219).

1. (9) If verse is alliterative, then there is productive reduplication and word stress is initial.

In fact, alliteration can be seen in (Balto-) Finnic, Icelandic, Germanic, and some other languages, most of which have word-initial stress.Footnote ¹⁷ If this generalization is on the right track, the fact that (old) Turkic, Mongolic, and Tungusic languages have alliteration supports the idea that these Altaic languages have stress on the initial syllable of a word.Footnote ¹⁸

Johanson (Reference Johanson, Johanson and Csató1998b: 111) claims that “[t]he assumption of a certain accent prominence of the first syllable at early stages of Turkic would help to explain phenomena such as the reduction and partial loss of the final short vowels, the emergence of sound harmony and Old Turkic alliteration patterns.”

Examples of Turkic alliteration can be seen in the Epic of Manas in Kirghiz (taken from Shoolbraid (Reference Shoolbraid1975: 44)):

1. (10)

   a.	Altyn erden kashi ikan,	He was the bow of the golden saddle.
   	ata[ly] jurtnyn bashi ikan;	He was the father of all the people.
   b.	Kuk dunannyn basy bar,	The grey horse has a head.
   	Kukotaj-khannyn asy bar,	For Kukotaj-khan there is a wake.

In these lines, altyn er (the golden saddle) alliterates with ataly (father), and kuk dunnan (the grey horse) with Kukotaj-khan.Footnote ¹⁹

Mongolic languages have alliteration in their poetry, as pointed out by Kara (Reference Kara and Roper2011). He observes that the vowel of the alliterating syllable is usually stressed. Examples are shown in (11).

1. (11)

   namur-un saran metü ‖ naγar-tu buyan-iyar :	By the merit as bright as the autumn moon,
   naciyai eke ‖ nayiran ǰoqiǰu :	let Mother Earth be in order and harmony;
   naγarbai gür ulus ‖ naγadun čenggen-iyer :	let the wide empire enjoy forever
   nasu aburida ‖ nasulan ǰirγatuγai ❖	living in playful joy and pleasure.

In addition, Hashimoto (Reference Hashimoto and Eguchi1990: 288) observes that tongue-twisters in Mongolic use alliteration.

As for Tungusic, Sinor (Reference Sinor and Sinor1968) points out the alliteration in Manchu poetry. Tsumagari (Reference Tsumagari1987) shows that alliteration is found in verses and wordplay in Manchu. An example is shown in (12).

1. (12)

   oncohon tuheci obinngi tucime	Falling on her back, froth came out.
   umušhun tuheci silenngi eyeme	Falling on her stomach, saliva flowed.
   oforo niyaki be oton de waliyame	She blew her nose into a tub.
   yasai muke be yala bira de eyebume	Her tears flowed just into a river.

Tsumagari (Reference Tsumagari1995: 168) also shows that Nanai uses alliteration in wordplay (tongue twisters), as shown in (13).Footnote ^20, Footnote ²¹

1. (13)

   ǰəm ǰəm ǰəliski	Gem, gem, geliski
   daxoa gačin, darin morin	bought a coat, a pair of horses
   morin gačin, moktor əǰən	bought a horse, King Moktor
   əǰən gačin, əxən n’akan	bought the king, a foolish servant
   n’akan gačin, n’ombor boso	bought a servant, blue cloth
   bosoa gačin, bombor xoto	bought cloth, a bald head
   xotoa gačin, xoŋgori səǰən	bought a bald head, a jingling cart
   səǰən gačin, sənnə puǰin	bought a cart, a beautiful woman
   puǰin gačin, pudda mama	bought a beautiful woman, a trousseau old lady
   darawa gačin donsi	bought fodder things

Kazama (Reference Kazama2011: 43) gives examples of alliteration in Nanay, one of which is shown in (14).Footnote ²²

1. (14)

   ǰaarila	ǰapargoi,	xosiktala	xoǰii,	əriŋkulə	əsəligui
   the first star	grab (cloth)	star	finish	morning	lay (the clothes) away
   ‘grab cloth at the first star, finish (sawing) under the stars, lay (the clothes) away at the morning star’

These examples support the idea that Altaic languages have word-initial stress if we admit the correlation between word-stress location and rhythm in verse.

5.6 Emphatic Stress and Lengthening

Further evidence of word-initial stress in Altaic languages comes from the position of emphatic stress. Although this argument is not as strong as the ones in the sections above, I believe that emphasis is realized best on the stressed syllable. In languages with word stress, emphasis is expressed by putting stronger stress on the stressed syllable, which may well be lengthened, as shown in the examples in (15) in English (cf. Celce-Murcia, Brinton, and Goodwin Reference Celce-Murcia, Brinton and Goodwin1996: 177).

1. 1. a. I’m NEVer eating clams again.

   2. b. I’m REALly enjoying it!

   3. c. I’m comPLETEly aGAINST the idea.

If this is the case in the world’s languages, then we would expect Altaic languages to have emphatic stress on the initial syllable. In fact, Udihe (Tungus) is one of the languages that has emphatic stress on the initial syllable (Nikolaeva and Tolskaya Reference Nikolaeva and Tolskaya2001: 95).

1. (16) Bi ‘o:no (<on’o) bagdi-ze-mi?

   me how live-Subj-1Sg

   ʻHOW should I live?ʼ

In (16), the initial syllable of ono ‘how’ receives emphatic stress and lengthening, although Nikolaeva and Tolskaya (Reference Nikolaeva and Tolskaya2001: 95) describe the unmarked stress location of the word as the second syllable (on’o).

Whether Korean is an Altaic language is also controversial (Polivanov Reference Polivanov1927; Poppe Reference Poppe1965; Greenberg Reference Greenberg, Hegedus, Michalove, Manaster and Ramer1997). However, word order in Korean is mostly head-final. Some scholars observe stress in word-initial syllables in Korean. In this relation, lengthening of the word-initial syllable in Korean may show that Korean has word-initial stress. Sohn (Reference Sohn1999: 196) observes that for emphatic purposes, length is attached to the first syllable of a word to be emphasized, even to an already long vowel (emphatic or connotative length represented by ::).

1. 1. a. cə.gi ʻover thereʼ

   2. b. cə::.gi ʻway over thereʼ

1. 1. a. k’ok ‘for sure’

   2. b. k’o::k ‘by all means’

1. 1. a. no.pha.jo ‘(It) is high.’

   2. b. no::.pha.jo ‘(It) is very high!’

If emphatic lengthening occurs on the stressed syllable cross-linguistically (e.g. very [ve::ri], really [ri::əli], and completely [kəmpı́::tli] in English), then lengthening of the word-initial syllable shows word-initial stress in Korean.

One might argue that emphatic stress can fall on a location different from the unmarked stress-location of a word. It has been pointed out that French changes stress-location for emphasis from the normal word-accented syllable (Tranel Reference Tranel1987: 201; Laver Reference Laver1994: 515). However, Polivanov (Reference Polivanov1927: 1198, fn. 2) argues that this shifted stress-location in French has a historical and phonetic reason: ancient Latin had the accent on the word-initial syllable until it adopted the right-oriented stress system. This stress location matches the fact that the basic word order of Latin is SOV. If this historical explanation of emphatic stress in French is on the right track, we can maintain the idea that emphatic stress falls on the unmarked stress-location (cf. Berg Reference Berg2008 for optional emphatic stress shift in German, which changes primary and secondary stress). Thus, emphatic stress in Udihe (16) supports our idea that word stress in Altaic falls on the initial syllable.

5.7 Emphatic Reduplication

Emphasis can also be expressed by reduplication in Altaic languages.Footnote ²³ Assuming that emphasis is realized mostly in the stressed syllable, reduplication provides evidence for the word-initial stress in Altaic languages. In Turkish, a reduplicative prefix is attached to some adjectives and adverbs for emphasis (Sebüktekin Reference Sebüktekin1971: 25).

1. (20)

   a.	incé ‘thin’	íp-ince ‘very thin’
   b.	temíz ‘clean’	tér-temiz ‘spotless’

I observe that a native speaker of Turkish pronounces the syllables marked with (´) in (20) with a high pitch. I argue that in Turkish, word-stress falls on the initial syllable in the non-reduplicated form as well as in the reduplicated form (ˈincé, ˈtemíz). Pitch accent falls on the word-initial syllable in the reduplicated form for emphasis.

According to Janhunen (Reference Janhunen2012: 120), alliterative reduplication is a very regular process in Mongolian.

1. (21)

   a.	oulaan ‘red’	oub-oulaan ‘bright red’
   b.	nogaon ‘green’	nob-nogaon ‘bright green’

Assuming that emphatic reduplication applies to the stressed syllable in the base form, alliterative reduplication provides evidence for the word-initial stress in Mongolian. Tsumagari (Reference Tsumagari and Janhunen2003: 135) also observes that a reduplicated prefix can be lengthened for emphasis in Dagur (Mongolic) as shown in (22b).

1. (22)

   a.	cigaang ‘white’	cim-cigaang ‘snow white’
   b.	xulaang ‘red’	xub-xulaang/xuub-xulaang ‘deep-red’

According to Tsumagari (Reference 223Tsumagari and Shoji1997: 180) and Li and Whaley (Reference Li and Whaley2000), emphatic reduplication can be found in a limited number of Tungusic languages including Kilen, Sibe-Manchu, Solon, and Oroqen (Orochen-Evenki). The following examples of Kilen are taken from Tsumagari (Reference 223Tsumagari and Shoji1997: 180).

1. (23)

   a.	tondo ‘straight’	tob tondo ‘extremely straight’
   b.	uyan ‘thin (liquid)’	ub uyan ‘very thin (liquid)’

Similarly, a demonstrative in Orok (Uilta) (Tungusic) tari ‘it, that’ has the emphatic form taatari ‘that (distal)’ (Ikegami Reference Ikegami1997).

As we have seen in Section 5.5, Meid (Reference Meid1971: 105–106) proposed a typological generalization (9), which claims the correlation between alliteration and productive reduplication and word-initial stress. If this generalization holds in the world’s languages, emphatic reduplication in Altaic languages supports the idea that Altaic languages have word-initial stress.

This idea predicts that languages with right-hand stress can have emphatic reduplication with a right-hand syllable. In fact, data from some Bantu languages show that this is a correct prediction. In Chichewa, which has penultimate stress (Downing Reference Downing, Goedemans, van der Hulst and van Zanten2010: 406), the final two syllables of a noun can be reduplicated to form a word meaning ‘a real X’ as in (24) (Myers and Carleton Reference Myers and Carleton1996: 39).

1. (24)

   a.	mnyamatá	‘boy’
   	mnyamatá-matá	‘a real boy’
   b.	chibwaná	‘childishness’
   	chibwaná-bwaná	‘real childishness’

In Kinande, which also has penultimate stress (Downing Reference Downing, Goedemans, van der Hulst and van Zanten2010: 406), the final two syllables of a noun can be reduplicated as in (24) (Mutaka and Hyman Reference Mutaka and Hyman1990: 76).

1. 1. a. o.ku-gulu ‘leg’

      o.ku-gulu-gulu ‘a real leg’

   2. b. o.mú-twe ‘head’

      o.mú-twé.mú-twe ‘a real head’

These data support our generalization that emphatic reduplication is related to stressed syllables.

5.8 Gemination

Another piece of evidence for word-initial stress in Altaic languages comes from the position of gemination. In Italian and other Romance languages, gemination often occurs on the consonant immediately after the stressed vowel (Anderson Reference Anderson and Baldi1984: 309; Borrelli Reference Borrelli2002: 21).Footnote ²⁴

1. 1. a. tútto ‘all’

   2. b. fulíggine ‘spot, soot’

Based on his survey of the Stanford Phonology Archive and additional sources, Thurgood (Reference Thurgood, Ashmore Nevis, McMenamin and Thurgood1993: 2) argues that geminates cross-linguistically tend to occur medially after a stressed vowel (cf. Blevins Reference Blevins2004; Dmitrieva Reference Dmitrieva2012). If gemination universally occurs on the consonant immediately after the stressed vowel, emphatic gemination in Altaic languages supports the idea that Altaic languages have word-initial stress. Although it is controversial to include Korean and Japanese with Altaic languages, these languages have emphatic gemination on the consonant immediately after the first vowel. Cho and Inkelas (Reference Cho, Inkelas and Kim-Renaud1994: 51) give the Korean data in (27) as examples of optional emphatic speech gemination.

1. (27)

   a.	ap’a → app’a	‘dad’
   b.	apha → appha	‘be sick’
   c.	tok’i → tokk’i	‘ax’

Nasu (Reference Nasu1999) found that in Japanese mimetic words such as (28a), emphatic gemination occurs on the consonant immediately after the first mora as in (28b) rather than after the second mora as in (28c) and (28d).Footnote ²⁵

1. (28)

   a.	pika-pika	‘glittering’
   b.	pikka-pika	‘very glittering’
   c.	?pikap-pika	‘very glittering’
   d.	*pika-pikka	‘very glittering’

The fact that Korean and Japanese have emphatic gemination on the consonant immediately after the first syllable or mora matches the idea that Altaic languages have word-initial stress even though the status of these languages as Altaic is controversial.

5.9 Summary of the Arguments for Word-Initial Stress in Altaic

In this section, I have shown nine arguments for word-initial stress in Altaic: native linguists’ observations, vowel reduction, the paradigm of vowels, vowel harmony, alliteration in poetry, emphatic stress, lengthening, reduplication, and gemination. These facts and observations show that Altaic languages have word stress on the first syllable or mora in addition to pitch accent on a right-hand syllable.

There are different opinions in the literature about the word-stress location in Altaic languages. Some studies argue for word-final stress in Turkic (Kabak and Vogel Reference Kabak and Vogel2001; cf. Fukumori Reference Fukumori2010), Mongolic (cf. Svantesson et al. Reference Svantesson, Tsendina, Karlsson and Franzén2005: 97), and Tungusic (e.g. Nikolaeva and Tolskaya Reference Nikolaeva and Tolskaya2001 for Udehe). However, for Mongolian, Svantesson et al. (Reference Svantesson, Tsendina, Karlsson and Franzén2005: 97) remark that “[t]he reason for this situation is apparently that different criteria and different phonetic correlates have been used for defining stress.” I have shown that this is the case for Altaic languages in general. This study shows that it is important to divide stress and pitch accent in order to understand the accent system of these languages (cf. van Heuven (Chapter 1 this volume) for acoustic correlates and perceptual cues of stress).

To summarize, I show my analysis of stress location as Rev(ised) in Table 6.12, where I+U shows initial stress and final pitch accent, OB observation, BS bibliographic survey, VR vowel reduction, VV varieties of vowels, VH vowel harmony, AL alliteration, EL emphatic lengthening, and GE gemination.Footnote ²⁶

Table 6.12 shows that a number of Altaic languages have word-initial stress as well as word-final pitch accent.