2.1 Tone or accent

Gorka Elordieta & Lisa Selkirk (‘Notes on prosodic headedness and tone in Tokyo Japanese, Standard English and Northern Bizkaian Basque’) extend a proposal made by Ito & Mester (Reference Ito and Mester2016b) about unaccentedness in Tokyo Japanese in two respects. First, they propose to split the constraint WordAccent, which demands a prominence peak for every prosodic word, into two, one constraint demanding a head for a prosodic constituent and the other a tone associated with this head. Second, they show that headlessness is important beyond the level of the prosodic word. These claims are based on case studies from Tokyo Japanese, Northern Bizkaian Basque and English. Whereas the former two show truly unaccented prosodic words that influence the structure of the prosodic phrase, English shows epenthesis of a H tone on the syllable that is the head of a prosodic word, which in turn is the head of a phonological phrase. The authors argue that these patterns can only be predicted under a generalised constraint system in which the headedness of prosodic words and phrases is determined by independent constraints.

Anthi Revithiadou (‘Ancient Greek pitch accent’) modifies the analysis of Ancient Greek accent in Ito & Mester (Reference Ito and Mester2016a), where the tone melody HL and a boundary tone L% are linked to moras in such a way that the accent H tone is realised in antepenultimate position. Enclitics, however, alter this antepenult accent pattern in certain contexts. In addition, in certain words ending in a long and a short vowel the H is unexpectedly realised on the first mora of a heavy penult, effectively ignoring consonantal moras which contribute to syllable weight in other contexts. Two assumptions are made to explain these unexpected patterns. The first is that all phonological elements have a certain activity that can differ gradiently (gradient symbolic representations; Rosen Reference Rosen2016, Smolensky & Goldrick Reference Smolensky and Goldrick2016) and that coda consonants are weakly active in Ancient Greek, and consequently only project weakly active moras. The markedness constraint relevant for words ending in a long and a short vowel then only ‘sees’ fully active moras, which explains why coda consonants are irrelevant in those contexts. The second assumption is the layering of the prosodic word node: enclitics can only surface with a tone on their own if the preceding word ends in L%, but are integrated within the same prosodic word as their host if there is not enough space to realise the full HLL% melody on the host.

The concept of gradient activity in this account is crucially different from the one in the original proposal and most subsequent case studies (e.g. Zimmermann Reference Zimmermann2018, Reference Zimmermann2019), given that the activity difference is completely phonologically predictable: only coda consonants are weakly active. Activity differences that are phonologically predictable are discussed in Trommer (Reference Trommer2019) and Walker (Reference Walker2019), and are undoubtedly an interesting option arising for gradient symbolic representations. One still might feel uneasy about it in this context, since the relevant context – being a coda consonant – is derived in the phonology. It is hence an interesting challenge to explore how and where the predictable weak activity for coda consonants can be ensured. If this is in the general phonology, an apparent conundrum arises for a non-cyclic model, since the phonology also adds activity to strengthen weakly active elements to fully active output elements.

It is important in this account that the phonology can see the difference between inserted activity and underlying activity. This raises the question of what activity really is: it needs to be a phonological element, rather than a property of elements, to allow such an epenthetic–underlying distinction, making the system very powerful in allowing constraints to explicitly refer to activity.

2.2 Moras and syllables

Shigeto Kawahara & Jason A. Shaw (‘Persistence of prosody’) argue that Japanese has syllables headed by fricatives that result from the deletion of high vowels between voiceless obstruents. The claim that the resulting CCV structure is syllabified as C.CV, with a moraic fricative heading the first syllable, comes both from morphophonological (e.g. minimality requirements or pitch-accent assignment) and phonetic facts. Shaw & Kawahara (Reference Shaw and Kawahara2018) show in an ElectroMagnetic Articulography study that the temporal alignment of the initial consonant of these structures is timed with respect to the following consonant, not the following vowel, as would be expected if this consonant were part of a complex onset.

Haruo Kubozono (‘Mora sensitivity in Kagoshima Japanese: evidence from no contraction’) presents arguments that the phonology of Kagoshima Japanese is sensitive to moras. Several morphemes ending in /no/ show deletion of the final vowel in casual or colloquial speech. A phonological conspiracy ensures that the additional coda created by this deletion does not result in trimoraic syllables. Deletion can either be entirely blocked if an already heavy syllable precedes /no/, it can result in shortening of a preceding long vowel or it can result in syllabifying long vowels and diphthongs into two syllables. The active ban on trimoraic syllables in Kagoshima Japanese is more surprising than, for example, in Tokyo Japanese, where similar patterns are found in the loanword phonology: Kagoshima Japanese has previously been described as quantity-insensitive, since accent assignment only counts syllables.

Anya Lunden (‘Adjacent identical vowels: vowel length or hiatus?’) discusses the different syllabifications that are expected for adjacent identical vowels within one language, and argues that both Latin and Japanese allow the coexistence of long vowels and adjacent identical vowels. In both languages, /V₁.V₁/ only surfaces across morpheme boundaries and the long vowel /Vː/ elsewhere. Lunden presents an OT account of these patterns that is crucially based on Align(Morpheme,L; σ,L), which demands that the left edge of a morpheme be aligned with a syllable edge (Ito & Mester Reference Ito, Mester and Kubozono2015b), and hence blocks the repair of hiatus as a long vowel across a morpheme boundary.

It is highly relevant that Latin has underlying long vowels as well as long vowels derived from morphological lengthening. The analysis sketched in the paper seems to make an incorrect prediction in this context if vowel lengthening is indeed due to affixation of an unassociated mora (e.g. Samek-Lodovici Reference Samek-Lodovici, Ackema and Schoorlemmer1994). This affixed mora is affiliated with a morpheme other than the mora it precedes, and a lengthened structure (/me_μtu_μs + μ/ → [me_μtu_μμs] ‘fear (gen.sg)’) should therefore induce a violation of Align(Morpheme,L; σ,L). An obvious solution would be to assume that Align(Morpheme) is specified for different phonological tiers, which in turn predicts a language in which Align(Morpheme,μ) is ranked high but Align(Morpheme,Segment) low, and only the addition of a floating morphemic mora results in two adjacent identical vowels – long vowels surface everywhere else. A typological study could reveal whether this prediction is as odd as it seems at first glance.

2.3 Position, directionality and linear order

Jaye Padgett & Máire Ní Chiosáin (‘The perception of a secondary palatalization contrast: a preliminary comparison of Russian and Irish’) present a perceptual study on secondary palatalisation contrasts, and confirm the hypothesis that palatal contrasts are perceptually disfavoured in coda or final position. An additional interesting finding is a language difference: both Irish and Russian speakers perceived the contrast better if it was produced by a Russian speaker, and Russian listeners were overall better in perceiving the contrast. This might be an epiphenomenon from an independent difference between the Russian and Irish participants, but could also suggest that the palatalisation contrast is less stable in Irish.

Aaron Kaplan (‘Asymmetric Crisp Edge’) argues for a directional and hence asymmetric interpretation of the constraint CrispEdge (Ito & Mester Reference Ito, Mester, Kager, van der Hulst and Zonneveld1999b) that penalises spreading of a feature beyond the left edge of a stressed syllable, but not beyond the right edge. The empirical evidence comes from vowel harmony in Tudanca Montañes, where [―ATR] spreads from the final to the stressed vowel. Interestingly, it can ‘overshoot’, and also affect the pretonic vowel if this is preceded by a labial consonant. It is argued that the pattern follows from the ganging up of a positive licensing constraint rewarding every [―ATR] feature on a stressed and any additional vowel with a constraint against a [labial][+ATR] sequence. Crucially, CrispEdge must be directional, since [―ATR] can spread from the final to the stressed vowel, but preferably does not spread further to the left.

Rachel Walker (‘Feature identity and icy targets in Menominee vowel harmony’) proposes an Agreement by Correspondence (ABC) account of vowel harmony in Menominee. The language shows a regressive [+ATR] harmony that is parasitic for height and only affects non-low vowels. Low [+ATR] vowels are transparent for the harmony, whereas low [―ATR ] vowels block it. This falls out from ABC theory, since the constraints that enable vowel harmony by demanding surface correspondence prefer correspondence between segments that share certain features. In a sequence where a low [+ATR] vowel is interleaved between two non-low vowels, e.g. /ɪ-ə-i/, it will block vowel harmony, since it preferably corresponds to the preceding and following vowels, but can't correspond to both, given that it differs from them in a different feature dimension. This is a really neat account of icy targets, which basically reduces the complex pattern to an emergence of the unmarked effect: though there is no ‘active’ height harmony in the language, it still has the effect of blocking correspondence between too many vowels that don't agree in height. Such an argument raises the interesting question of how different vowel-harmony processes can interact within a language.

Nick Kalivoda & Jennifer Bellik (‘Prosodic recursion and pseudo-cyclicity in Danish compound stød’) argue that the prosodic structures of Danish compounds do not perfectly match syntactic structure, but do not entirely ignore it either. The important diagnostic is the distribution of stød, a lexical glottal accent that can only be realised at the right edge of a prosodic word (Ito & Mester Reference Ito and Mester2015a). Prosodic structure differences then explain why a short noun loses its stød only if it is the first member of the compound (e.g. [toːg + passageːˀr] ‘train passenger’ vs. [passageːˀr + toːˀg] ‘passenger train’). On the assumption that compounds are mapped to a recursive prosodic word structure with only binary branching, such a monosyllabic noun is simply too small to project a foot on its own, and is hence not at the right edge of a prosodic word if it is the first member of the compound ([(σ)N [(σσσˀ)N]_ω]), but only if it comes second ([[(σσσˀ)N]_ω (σˀ)N]_ω). An OT account based on three general constraints also makes the correct prediction that three-member compounds show pseudo-cyclic prosodic structure [[S Sˀ] Sˀ]_ω or [[S[Lˀ]_ω] Sˀ]_ω (‘S’ = word consisting of one foot; ‘L’ = word consisting of two feet).

A prediction that this model apparently makes is that words longer than five syllables in Danish should generally be mapped to pseudo-cyclic structures as well, to avoid non-binary branching. A six-syllable structure cannot be parsed as a single prosodic word without violating binary branching (*[(σσ)_φ(σσ)_φ(σσ)_φ]_ω) and a pseudo-cyclic structure with two prosodic words is hence predicted: *[[(σσ)_φ(σσ)_φ]_ω(σσ)_φ]_ω. If such long nouns existed in Danish, we would wrongly predict the possibility of two støds in longer words. Abstracting away from Danish, this proposal makes many interesting predictions about recursive word structures that are correlated with word length.

2.4 Alternatives to morphological/lexical listing accounts

Ryan Bennett & Robert Henderson (‘Nasal hardening and aspect allomorphy in Kaqchikel’) discuss the incompletive aspect prefix /n-/ in Kaqchikel, which surfaces as [nt-] or [nd-] before vowel-initial monosyllabic verbs, and argue that this alternation results from a cumulative constraint interaction that triggers coronal stop epenthesis. They acknowledge that a lexical listing account with suppletive allomorphs can easily account for this pattern, but that it misses important generalisations. If the aspect prefix /n-/ precedes a vowel-initial monosyllabic verb, a syllable with a dispreferred high-sonority onset is created which is prominent, in that it is both the initial and the stressed syllable. In addition, the affix would be syllabified within a stressed syllable that otherwise contains only root material. Stop epenthesis solves all these problems, since it pushes the nasal into its own syllable, thus allowing a syllabification that aligns syllable boundaries with morpheme boundaries and provides a low-sonority onset ([n.ˈtokʰ] instead of *[ˈnokʰ]).

Yranahan Traoré & Caroline Féry (‘Nominal classes and phonological agreement in Tagbana’) present a typologically unusual instance of ‘alliterative concord’ in the FròɁò dialect of Tagbana, where morphologically dependent morphemes within the nominal domain show agreement in their initial consonantal features. They argue that this is not an instance of phonological agreement, but the result of inserting phonologically underspecified featural morphemes that realise abstract class features. This is modelled within a Distributed Morphology account, where functional morphemes are combined with abstract inflectional features. Multiple vocabulary items then realise different features, each contributing a different partial piece of the full phonological specification of a dependent morpheme (e.g. articulatory and continuancy consonantal features specifying the nominal class and a fixed vowel determined by functional morpheme type and nominal class). The phonology then combines these elements, and fills potentially unspecified positions by epenthesis.

It is an interesting endeavour for future research to formally implement this floating feature intuition in a truly phonological account. The analysis in the paper relies on an Anchor constraint which ensures that the floating features are realised in initial position. This constraint is either morpheme-specific or specified as applying to only floating input features (cf. Wolf Reference Wolf2007). Both options imply a constraint type that does not evaluate only the phonological output. This might in fact be empirically problematic, since it excludes multiple floating features being realised in different positions in a language. A purely phonological solution to this locality problem is the assumption of radically underspecified segments that can fuse with other radically underspecified segments.

Another question involves the spell-out of features: many vocabulary items apparently realise the same features, but are nevertheless inserted. This seems to be incompatible with the assumption of feature discharge after spell-out in Distributed Morphology (e.g. Harley & Noyer Reference Harley and Noyer1999), but might be overcome with a more fine-grained morphological feature system.

2.5 Strata within a language

Jennifer L. Smith (‘Stratified Faithfulness in Harmonic Grammar and emergent core-periphery structure’) directly follows up on the research on the stratified lexicon in, for example, Ito & Mester (Reference Ito, Mester, Beckman, Dickey and Urbanczyk1995a, Reference Ito, Mester and Goldsmithb, Reference Ito, Mester, Miyagawa and Saito2008), showing that a language's vocabulary can be organised into different strata of ‘foreignness’, with a core that undergoes multiple repair processes and peripheral layers that undergo only subsets of these. It argues for a Harmonic Grammar implementation of a Stratified Faithfulness account in which faithfulness constraints are indexed to different strata. This correctly predicts that languages with a stratified lexicon (i) do not have to show the ‘core-periphery’ organisation, but (ii) show a strong preference for it. The main empirical argument is that the stratal organisation of the Japanese lexicon does not strictly obey the core-periphery organisation (Ito & Mester Reference Ito, Mester and Goldsmith1995b). Crucially, it is shown that constraint ganging predicts a preference for core-periphery structure, and hence excludes impossible new nativisations. This follows, since evidence during learning that a certain stratum obeys a certain faithfulness constraint results in an increase of the weight of both the indexed and the general faithfulness constraints.

Martin Krämer (‘The perfect prosodic word in Italian: or fruit salad matters’) shows that the concept of the perfect word (Ito & Mester Reference Ito and Mester2015a) captures various generalisations found in truncation in Italian. It gives a detailed empirical investigation of different Italian truncation patterns that show considerable variation (e.g. left-anchored bisyllabic or monosyllabic and stress-anchored trisyllabic nicknames). Krämer argues that the shape of truncated forms makes reference to the perfect prosodic word containing a layered trochaic foot, and that the perfect word requirement can be satisfied by the surfacing truncated portion or by the portion of the base that remains unrealised. He shows how this follows in an OT system based on base-truncation faithfulness relations that relate an existing base to the truncated form (Alber Reference Alber2010). Since different truncation patterns can coexist within one language, this system necessarily implies indexed constraints or strata (Ito & Mester Reference Ito, Mester and Tsujimura1999a, Pater Reference Pater2007). A welcome prediction is that language-specific variation about the shape of the perfect prosodic word simply emerges from a re-ranking of the constraints determining foot form, not from different versions of PerfectWord.

For cases where the unrealised portion of a base corresponds to the perfect word, Krämer assumes an explicit constraint, TrncPrfctWrd, which demands deletion of all elements that are in the main-stressed foot in the corresponding base. However, on this account (and most alternatives mentioned in the paper), it is a coincidence that only reference to stressed and unstressed portions of a base word can ever be relevant in these word formations. In principle, it should be possible for a portion sharing some other kind of phonological information (e.g. segmental features) to be realised or deleted. Whether these apparently unexpected predictions might in fact explain certain emergence of the unmarked effects in reduplication or templatic morphology more generally is an interesting question.

2.6 Optimality Theory and opacity

Alan Prince (‘Counting parses’) presents different mechanisms to determine how many and which parses are possible for a given structure, exemplified with different options to parse a string of syllables into feet and stressed syllables. The paper discusses different procedures for how to determine the absolute number of possible parses, including the ‘method of continuation’ and the ‘method of arrangements’. Based on these, equations are developed that allow the calculation of the absolute number of possible parses. Crucially, the paper also shows how this counting strategy can be turned into a procedure that produces all the possible parses. It is based on a simple copy-and-change procedure that applies different operations, which result in one of two possible forms for an element. After a list of inputs is copied, a certain operation applies only to the copied list. This procedure is iterated until all change operations have applied. This discussion of the derivation of all possible parses shows in particular that the paper is of more than mathematical interest with respect to the possible number of candidates; it forces the analyst to think about her concrete structural assumptions and the possible operations assumed in the grammar.

Eric Baković & Lev Blumenfeld (‘Overapplication conversion’) present an investigation of the formal relationship between different rule-interaction types, and demonstrate how they can be converted into one another. The primary focus is on overapplication opacity, cases where a process applies even though its context is not (or no longer) apparent on the surface. In order to investigate the formal similarities and relations between different rule interaction types, the paper identifies four basic formal relations that describe rule interactions. The excision or addition of one of these formal relations can then convert one rule interaction type into another. Any counterbleeding relation, for example, can be converted into a seeding relation if an input-removal mapping is excised, and vice versa. Another compelling result of the paper is that certain rule-interaction types are shown to be very closely related. This is most interesting in the case of seeding and cross-derivational feeding. That the empirical predictions of these two relations are so similar is highly relevant, since seeding analyses are impossible in a parallel OT system, whereas a cross-derivational feeding account is impossible under rule ordering.

The formal relations identified in the paper are rather general, which is desirable, since it results in a general typology of rule-interaction types. One might wonder, however, whether important generalisations about possible and impossible rule interactions are missed by the complete absence of any relation to the type of phonological processes that interact. One possible example is the absence of assimilation and epenthesis counterbleeding (e.g. Kiparsky Reference Kiparsky and Fujimura1973, Baković Reference Baković2007). This is a coincidence under the symmetrical description of possible interaction types presented here. A question which arises is whether a more fine-grained understanding of how input and outputs are provided or removed could lead to a more empirically adequate representation of possible process interaction types or whether this would destroy interesting results of the type presented here.