Kattobase: The Linguistic Structure of Japanese Baseball Chants

This paper develops the first constraint-based analysis of Japanese baseball chants, whose intricate organization casts a very revealing sidelight on the prosodic organization of the Japanese language itself. * The source of most data and basic generalizations is the last author's dissertation (Tanaka 2008), supplemented by further empirical probing which clarified a number of unclear points. Since the beginnings of metrical phonology, the analysis of chants has played an important role in the development of the theory. Liberman (1975) uses the English vocative chant ( "Jo-ohn!" , with a High-Mid tune) to motivate basic properties of what came to be known as the "metrical theory of stress". He shows that, in order to formalize tune-to-text alignment, and to define what it means for a tune to be congruent with a text and its metrical pattern, a relational understanding of stress is necessary, as instantiated in metrical trees with their "strong-weak" labeling of all nodes. Japanese baseball chants, an obligatory and quasi-ritual part of virtually every baseball game, are delivered by fans each time their team is at bat. They are accompanied by a variety of musical instruments (drums, trumpets, etc.) and take the form in (1). They consist of two measures of four beats, each composed of three notes plus one pause. 'XXX' is a rhythmically adapted form of the player's name. At issue here is the form of the rhythmic adaptation, which is tightly regulated and grounded, as we will show, in the rhythmic structure of the language itself.

By way of introduction, we first summarize Tanaka's (2008) analysis, and at the same time present relevant examples. There are three separate rules, depending on the length of the input name, measured in moras (μ). Each CV-or V-unit is one mora, so ichiroo= i-chi-ro-o is a 4μ name. Syllable-final consonants (mostly nasals) are also one mora (son = so-n = 2μ). Rule 1 in (3) deals with names up to 3μ. The simplest case is names of exactly 3μ, which fit the X 1 X 2 X 3 template exactly, as seen in (4) (we write "L" for "light syllable", "H" for "heavy syllable", and "S" for "superheavy syllable"). (3) Rule 1a 3-mora names: Align the initial mora to the initial beat (X 1 ), the final mora to the final beat (X 3 ), the medial mora to the medial beat (X 2 ).

(5)
Rule 1b In 2μ names, there is no medial mora. Spread from the left to fill X 2 . In 1μ names, there is only an initial mora. Spread from the left to fill X 2 and X 3 .

(7)
Rule 2 4-mora names: Align the initial mora to X 1 , the final syllable to X 3 , medial moras to X 2 .
More examples illustrating the pattern are given in (8) As indicated, we understand the winning output as mapping the two o-moras to different beats (with an onset violation). Here we simply note the facts, and will return to their explanation in the next section. A special case are 4μ names like [ku 1 ]-[ra 2 -i 3 -n 4 ] (Klein) (9a), whose profile seems to be LS. If it is the case that the superheavy syllable is actually broken up into L+H [ku 1 ]-[ra 2 ]-[i 3 -n 4 ], as argued by several authors, including Vance (2008, 125-127), Kubozono (2015, 13-16), and Ito and Mester (2018, 212-216), the final syllable rule already covers this case. The long vowel in (9b) seems to resist this kind of splitting.
(9) Moras Profile Input Output 4 LS or LLH a. kurain Klein kuu-raa-in 3 b. kuruun Kroon kuu-ruu-nn SL or LHL c. baanzu Barnes baa-an-zuu d. joonzu Jones joo-on-zuu Names with 5 or more moras follow the same final syllable rule, but we need to distinguish two cases on the basis of the weight of the penultimate syllable. Names with penultimate H map this syllable to X 2 , as formulated in (10) and illustrated in (11).
(10) Rule 3a 5μ names with H penultimate syllable: Align the final syllable to X 3 , the penultimate H syllable to X 2 , the remainder (which can be of any length) to X 1 .

Analysis
The challenge any analysis must cope with is that we seem to be dealing with three separate patterns, as summarized in (15).
(15) 1. for 3-mora names: last mora goes to last beat 2. for 4-mora names: last syllable goes to last beat 3. for 5-mora names: last syllable goes to last beat, and special provisos for H and L penults Ideally we would like to unify everything into a single rule, but the distinction between the three patterns seems very well motivated, having to do with the length of the input. Since the overall goal is to split the input into three parts, which are then mapped to the three beats, it makes sense that what goes into the last beat is different for short names and for longer names. But "making sense" is not yet an explanation-our goal is now to give an analysis in Optimality Theory (OT, Prince and Smolensky 1993), in terms of ranked and violable constraints, and capture the different aspects of the pattern in one single and uniform constraint ranking, instead of having three distinct procedures. We first lay out our constraints, beginning in (16), where "K" stands for "kattobase form".

5
(16) a. K = X 1 X 2 X 3 A kattobase form consists of 3 beats, X 1 X 2 X 3 . b. X  FOOT A beat is minimally a foot (Ft). c. FOOTFORM(X 2 ) X 2 is a quantitative trochee (H, LL, or HL). d. MAX Every element of the input is present in K. e. ALIGN-LEFT(X 3 ,μ]) The left edge of X 3 corresponds to the left edge of (the content of) the last mora of the input. One violation when the edges do not coincide.
The left edge of X 3 corresponds to the left edge of (the content of) the last syllable of the input. One violation when the edges do not coincide.
For our purposes, the basic rhythmic structure of Japanese is the trochaic (strong-weak, sw) foot with the three forms in (17), which include the uneven trochee HL, for reasons we will return to.
ta ta taa taa ta (16c) raises an immediate question: Why is there a special constraint requiring X 2 to be exactly a trochee? Empirically speaking, the answer is clear: In long names, material exceeding the size of a trochee goes into X 1 , not into X 2 : MacDonald →makudo-naru-doo, not *maku-donaru-doo. X 3 is in any case restricted to the last syllable of the input because of the ALIGN-LEFT constraints (σ, μ): But what is the reason X 2 plays this special role, not X 1 or X 3 ? Our hypothesis is that the reason lies in what X 2 corresponds to in a Japanese word: It corresponds to the last, and most prominent, foot of the word, the foot which receives the default antepenultimate accent, as illustrated in (18).
In conjunction with (20a), the general DEP-μ constraint militating against any kind of lengthening, there is also another headfoot-specific positional faithfulness constraint (20b) preventing epenthesis in X 2 .
(20) a. DEP-μ Every mora in the output has a correspondent in the input-no epenthesis of a mora (i.e., no lengthening). b. DEP-μ (HDFT) Every mora in the output's head foot has a correspondent in the input-no epenthesis of a mora (i.e., no lengthening).
The remaining constraints cover familiar territory: There are two crisp edge constraints (Ito and Mester 1999) militating against spreading consonants or vowels across X-boundaries (21ab), and the familiar onset constraint (21c).
6 (21) a. CRISPEDGE-C The edges of X are crisp: no spreading of consonants across. One violation for every consonant linked to two different Xs. b. CRISPEDGE-V The edges of X are crisp: no spreading of vowels across. One violation for every vowel linked to two different Xs. c. ONSET A syllable has an onset.
The overall ranking of the constraints is as in (22). We will gradually justify all dominance relations. kake-ee-fuu 1 2 1 3 d.
ka-ke-fu 3 1 The winning candidate kaa-kee-fuu also shows three instances of mora epenthesis, and thus violates general DEP-μ three times. Since this constraint is bottom-ranked and does not contribute to the explanation in interesting ways, we will in general not include it in our tableaux. Turning next to short names, we give tableaux for 1μ-and 2μ-names in (24) and (25).

Summary and conclusion
We summarize by first assembling the evidence for all constraint rankings. The overall system is repeated in (41).
In (42), we present the evidence for each of the labeled ranking relation (produced with the help of OTWorkplace (Prince et al. 2015)). Crucial W(inner)~L(oser) pairs justifying a particular ranking are given in bold. "W" in a constraint column means that the constraint prefers the winner; "L" means that the constraint prefers the loser. What is important here is the core of OT ranking logic: In order for the winner to defeat a given loser, it must do better on the highest-ranking constraint that distinguishes the two. ONSET DEP-μ  ogasawara oga-sawa-raa gaa-sawa-raa W L W  son soo-oo-nn soo-nn-nn W L L W  ri rii-ii-ii rii-X-ii W L L L L  doddoson doo-oddo-son doo-ddo-son W L L L L  etoo ee-too-oo ee-ee-too W L W W  kurain kuu-raa-in kuu-rai-nn W L L  joojima joo-oji-maa joo-jii-maa W L  robaatoson roo-baato-son roba-ato-son W L In conclusion, the OT-analysis, with its ranked and violable constraints, has succeeded in folding what appeared to be a set of separate rules depending on the length of the input into a single unified constraint system with a single ranking, where the length of the input exerts its influence by resulting in different violation profiles in outputs, and does not require separate rules for inputs of different length. Besides the alignment constraints specific to the baseball chant, the other constraints are uncontroversial faithfulness (MAX/DEP) constraints and structural markedness constraints (on foot/syllable structure and their edges). Among the many remaining questions, however, the most important perhaps is why the desired foot form in X 2 is the quantitative trochee that admits also the trimoraic HL-foot, and not the bimoraic trochee otherwise firmly grounded in the phonology of Japanese. It is clear that much work remains to be done-in particular in grounding the constraints better in the prosodic system of the language.