Prepublished draft, June 2013

Non-concatenative allomorphy is generalized prosodicaffixation: The case of Upriver Halkomelem∗

AbstractI argue that the continuative aspect in Upriver Halkomelem is marked by amorpheme consisting of a segmentally empty prosodic foot. Since there aredifferent strategies to fill a prosodic foot with segmental material, such an anal-ysis predicts the non-concatenative allomorphy between epenthesis, stress shift,vowel lengthening, and reduplication that we find in the continuative formationof Upriver Halkomelem. The analysis is situated in the line of research termed‘Generalized Nonlinear Affixation’ by Bermúdez-Otero (2012) that strives toderive all instances of non-concatenative morphology without any additionalassumptions simply from affixation of nonlinear phonological representationsthat are independently motivated.

Keywordsnon-concatenative allomorphy, segmentally empty foot, optimality theory, stress

1. Introduction

The formation of the continuative verbal aspect in Upriver Halkomelem canbe described as an instance of non-concatenative allomorphy, i.e. the phe-nomenon that different non-concatenative strategies mark one morpheme indifferent contexts. As can be seen in (1), the continuative form is derived fromits non-continuative counterpart through either stress shift (1-a), vowel length-ening (1-b), insertion of hE (1-c), or reduplication (1-d).

(1) Verbal aspect in Upriver Halkomelem (Galloway, 1993)

Non-Continuative Continuativea. ţ’EtE;m ‘crawl’ ţ’Et@m ‘crawling’b. Pím@ç ‘walk’ Pí;m@ç ‘walking’c. m@q@t ‘swallow’ h@mq@t ‘swallowing’d. q’ís@t ‘tie sth.’ q’íq’@s@t ‘tying sth.’

∗For helpful comments and discussions I am grateful to the audiences of the ConSOLE XVIIand the colloquium ‘Neuere Arbeiten zur Grammatiktheorie’ at the University of Leipzig whereearlier versions of this paper were presented. I am especially indebted to Jochen Trommerfor discussion and support. Finally, I also thank four anonymous reviewers for their valuablecomments and criticisms.

1

For one, the Upriver Halkomelem pattern is an example of the challenge thatnon-concatenative exponents pose for concatenative approaches to morphology:What is the phonological representation of a morpheme that manifests itself asstress shift? Secondly, the pattern is particularly interesting since it involves al-lomorphy between multiple non-concatenative exponents realizing a single mor-phological category. Is it possible that these different strategies result from asingle morpheme representation? And what governs the choice between thedifferent strategies to mark the continuative?

I argue that the continuative morpheme in Upriver Halkomelem consists ofa segmentally empty foot, henceforth a ‘morphemic foot’. If such a morphemicfoot is prefixed, it must not only be integrated into the prosodic structure ofits base, but must be filled with segmental material as well due to standardmarkedness demands about prosodic well-formedness. A version of morphemecontiguity prefers that the morphemic foot is only filled with segmental ma-terial that is not affiliated with the base and epenthesis and reduplication arepredicted to provide non-base material. Such an analysis is situated in the line ofresearch termed ‘Generalized Nonlinear Affixation’ (GNA) by Bermúdez-Otero(2012) that strives to derive all instances of non-concatenative morphology with-out any additional assumptions simply from affixation of nonlinear phonologicalrepresentations that are independently motivated (Lieber, 1992; Stonham, 1994;Akinlabi, 1996; Wolf, 2007; Finley, 2009; Saba Kirchner, 2010; Trommer, 2011;Bermúdez-Otero, 2012; Bye and Svenonius, 2012). A possible alternative view isthe assumption that (non-concatenative) morphemes may lack any underlyingphonological representation and that there is a constraint demanding that dif-ferent morphological forms must be phonologically distinct (e.g. Kurisu (2001)).It is concluded in section 4 that such an approach to non-concatenative mor-phology suffers from a serious overgeneration problem and predicts unattestedpatterns of non-concatenative allomorphy.

The paper is structured as follows: I begin with some relevant backgroundon Upriver Halkomelem phonology and especially its stress system in section2.1. In section 2.2, I introduce my theoretical background assumptions. Myanalysis of the allomorphy in the continuative of Upriver Halkomelem is givenin section 3. First, I draw the crucial generalizations about the contexts for thefour continuative allomorphs in section 3.1 and present my analysis of stressoverwriting in section 3.2 before I turn to the concrete analyses of the four con-tinuative allomorphs in section 3.3. Section 4 discusses the further predictionsof the proposed constraint system and the advantages of my GNA approach incontrast to alternative accounts. Section 5 broadens the picture of the allomor-phy in the continuative and discusses further irregular continuative allomorphs.I conclude in section 6.

2. Empirical and theoretical background

2.1. Phonological facts about Upriver HalkomelemSegmental inventory and syllable structure. Upriver Halkomelem is one of threedialects of Halkomelem, a Central Salishan language spoken in the south-eastern

2

end of Vancouver island and in British Columbia that had no more than twospeakers in 2004 (Brown, 2004, 1). As nearly all Salishan languages, UpriverHalkomelem makes extensive use of non-concatenative morphology, especiallyreduplication, as has been noted in several papers on Salishan morphology (foran overview see e.g. Czaykowska-Higgins and Kinkade (1998)).

In contrast to the description in Galloway (1993), I use standard IPA nota-tion throughout. In (2), the consonant and vowel inventory of Upriver Halkomelemis given together with Galloway’s original notation in brackets if it differs fromthe IPA notation.

(2) The inventory of Upriver Halkomelem1

ConsonantsLab. Dental Alveolar Palat. Velar Uvular Glott.p t [k] kw q qw P

Stops p’ t’ [k’] k’w q’ q’w

Nasals mFric. T s ç (xy) xw X (x

˙) Xw (x

˙w) h

ţ (c)Affric. tT’ (T’) ţ’ (c’)Lat.

ì (ł)fric.Lat. tì (ň)affric.Appr. w j (y)Lat. lappr.

Vowelsi uE @ o

a

The language has a relatively simple vowel system but a complex system ofconsonants. There are no voiced obstruents in Upriver Halkomelem and onlya single nasal, but phonemic distinctions between plain voiceless obstruents,glottalized obstruents (notated as C’), and labialized obstruents. There is ageneral onset requirement demanding that vowels can never occur initially or ina vowel cluster. Consonants, on the other hand, can occur in combinations ofup to four consonants medially or finally and of up to three consonants initially.

1While translating all symbols into IPA, I relied on the phonetic description given inGalloway (1993) for the Upriver Halkomelem sounds and on the description given in Suttles(2004) for the sound system of Musqueam that belongs to the Downriver dialect group ofHalkomelem. The sounds k and k’ given in brackets are very rare and surface only in loanwords(Galloway, 1993, 21+22).

3

Consonant clusters may not begin with a glottal sound and final clusters cannotend in sonorants or glottals. In a detailed study in Thompson (2005), it isconcluded that vowel length is contrastive in the language. Vowels can be short(V), long (V;), or superlong (V:). The latter, however, appear only in emphaticforms.

Stressed and unstressed vowels. Galloway (1993) describes three degrees of ‘phone-mic pitched stress or phonemic tone’, namely high-stressed (V), mid-stressed (V)and unstressed syllables (V). He lists many allophonic processes involving therealization of the three degrees of stress. I argue now that only the distinc-tion into stressed and unstressed vowels is made in the phonology. In Brown(2004), the functional load and predictability of the stress/tone patterns of vow-els in Upriver Halkomelem is investigated and he concludes that the typologyof possible acoustic cues for stress can be summarized as in (3).

(3) Acoustic properties of vowelsElmendorf and Suttles (1986), summarized in Brown (2004)

High Stress

Low Stress

loud, high, long

loud, high, short

loud, low, long

loud, low, short

Length present

Length absent

From a theoretical perspective, these combinatorial possibilities can easily besummarized with only two phonological distinctions, namely length and (main)stress. Secondary stress then lacks an independent phonological representation –an analysis that is highly plausible since there are actually very few such vowelsin the description. An exception are the suffixes -@m ‘passive’ and -@t ‘negativeor impossibility marker on a passive’ (P@w@tE skwíj@xt@m@t ‘nothing could bedone’ (Galloway, 1993, 62)). However, it is striking that all these ‘secondarystressed’ vowels are instances of @, which is described as the only vowel that cannever be phonologically long in Upriver Halkomelem. The reasonable hypothesisis therefore to assume that a ‘secondary stressed’ @, realized louder and higherthan an unstressed @, is simply an underlyingly long @. Full ‘mid-stressed’ vowelsare notated for only a few words and Galloway (1993) emphasizes that he foundthem for some speakers only in “sentences where morphological downsteppinghas changed the stress from its usual [high-stressed] citation form” (Galloway,1993, 40). The rare full ‘mid-stressed’ vowels are therefore taken as underlyinglystressed. In a recent description of the related Downriver dialect Musqueam,Suttles (2004) admits as well that he found the recording for secondary stressedvowels hard and that his ‘perception was probably influenced by the sentenceintonation pattern’ (Suttles, 2004, 14). Similarly, there is much variation inthe descriptions where some words can be found with different stress patterns.

4

In the examples in (4-a), secondary and main stress optionally alternate and(4-b) shows data where alternating stress patterns are found between differentspeakers for some words.

(4) Variation (Brown, 2004)a. hEm ∼ hEm ‘finished (of a story), over’

tEm ∼ tEm ‘shout, yell, holler’wEj;@l@s ∼ wEj;@l@s ∼ wEj;@l@s ‘tomorrow’

b. tT’@wE;lí ∼ tT’@wE;lì ∼ tT’@wE;li ‘Soowahlie’ (place name)qw@l;@s ∼ qw@l;@s ‘whale’

I therefore assume that the phonology only distinguishes between short andlong vowels (one mora/two moras) and stressed/unstressed vowels. What Gal-loway (1993) terms ‘mid-stressed’ vowels are taken to be either unstressed longor stressed short vowels.

Lexical Stress. There are some implications between length, stress and vowelquality: First, long vowels are always stressed in Upriver Halkomelem and sec-ond, unstressed vowels are often reduced to @ (cf., for example, Czaykowska-Higgins and Kinkade (1998), 15). If a long vowel becomes unstressed in thecourse of the derivation, it is often reduced to a short one or even to @. An ex-ample is the form T@tí;l ‘go/get dark’ (Galloway, 1993, 242) from the underlyingbase TE;t. Whenever a stressed suffix like –íl attaches to a base, the base be-comes unstressed, a phenomenon I discuss under the heading ‘stress overwriting’below. Consequently, the long base vowel E; becomes short and even reduces to@. Stress in Upriver Halkomelem is lexical and I therefore assume that stems andaffixes are underlyingly marked for being stressed or unstressed. The conclu-sion that stress is lexical is drawn in Elmendorf and Suttles (1986) and Brown(2004) (see also (Czaykowska-Higgins and Kinkade, 1998; Dyck, 2004)) as welland also adopted in the analysis in Kurisu (2001). In an insightful investigationby Shaw et al. (1999), stress in Musqueam (Coast Salishan) is analysed and itis concluded that many apparently lexical factors governing stress can indeedbe reduced to phonological demands (cf. also Suttles, 2004, 13). However, theyconclude that at least some affixes must be taken to be arbitrarily marked forstress.

At least two phonological tendencies can be found in the stress pattern. First,there is a strong tendency for stress to fall on the first syllable. For example,Urbanczyk (1998) citing investigations on Lushotseed, Cowichan, Squamish andMainland Comox concludes that initial stress is the default in Central Salishanlanguages in general. A quantitative left-aligned trochee is argued to exist inMusqueam as well (Urbanczyk, 1999; Shaw, 2004) and Suttles (2004) describesthat ‘primary stress is generally on the first vowel’ (Suttles, 2004, 13), too.Second, @ tend to remain unstressed. Consequently, the preference for initialstress is overridden if a word contains an initial @ and at least one following fullvowel – stress tends to fall on the full vowel.

5

The stress in morphologically complex forms is predictable given the stresspatterns of stems and affixes. Unstressed2 affixes simply attach to their stemwithout triggering any change in the stress pattern of their stem. Stressed af-fixes that are directly adjacent to a stem, however, make any stress on the stemimpossible (Galloway, 1993, 52). This instance of phonology that is sensitiveto morphological information is an instance of morphological dominance effects(Inkelas, 1998; Alderete, 2001a,b) that I term ‘stress overwriting’ in the follow-ing. Some examples of stress overwriting are given in (5) where several stressedaffixes attach to a stressed base that consequently becomes systematically un-stressed (cf. section 2.1). All Upriver Halkomelem data is taken from Galloway(1993); ‘Gxx’ refers to the page number in the grammar where the respectiveexample can be found.

(5) Main stressed affix overwrites stress pattern of the stema. cákw –á;l@s cakwál@s G240

‘be distant’ ‘in the eye’ ‘goatsbeard plant’ (=bloomscan be seen from far away)

b. qá; –á;l@s q@Pál@s G240‘water’ ‘in the eye’ ‘tear’

c. TE;t –íl T@tí;l G242‘darkness’ ‘go, come’ ‘go dark, get dark’

d. Pít@t –lá;met Pit@tlám@t G257‘(to) sleep’ ‘reflexive, oneself’ ‘fall asleep’

The examples (5-c-d) illustrate the weakening of unstressed vowels in UpriverHalkomelem as well: the base vowels are realized as @ as soon as they areunstressed.

2.2. Theoretical backgroundIn this section, I present my theoretical background assumptions. My anal-

ysis is implemented in standard correspondence-theoretic OT (McCarthy andPrince, 1995) and I introduce all major constraint that are relevant in the anal-ysis of Upriver Halkomelem below.

A foot as a morpheme. The central assumption in the analysis I propose isthat the phonological representation of the continuative morpheme in UpriverHalkomelem is an empty prosodic foot. The analysis is therefore based on the as-sumption that prosodic nodes exist as entities in themselves and do not merelyarise through an emergence-of-the-unmarked effect (Prosodic Morphology, cf.McCarthy and Prince 1986/1996). That prosodic elements can be morphemeson their own is by now more or less uncontroversial at least for moras. Invarious analyses, morphological lengthening, insertion and even shortening isattributed to the affixation of a morphemic mora (e.g. Samek-Lodovici (1992);

2In Galloway’s terminology: or mid-stressed affixes. I take these as unstressed long orstressed short.

6

Davis and Ueda (2002, 2006), Grimes (2002a), Haugen and Kennard (2008),Seiler (2008); Saba Kirchner (2010) or Trommer and Zimmermann (2010)). Ifthe mora as the smallest prosodic entity is a possible morpheme, the assumptionthat other prosodic constituents are possible morphemes as well is a natural andperhaps even expected extension of already established mechanisms. One anal-ysis assuming the affixation of a morphemic foot was already proposed, namelyan account of the past tense formation in Modern Greek: “Tense in Greek is astress marker only, i.e. a segmentally empty foot” (van Oostendorp, 2012, 1173).At first glance, it seems rather peculiar to assume ‘floating feet’ since feet areonly projections and should be impossible on their own without any materialthey dominate. However, moras are projections as well. And if a theory al-lows floating moras to exist without any segment they dominate, nothing at allwould prohibit floating feet that do not dominate anything underlyingly (vanOostendorp, 2007b). That prosodic nodes as morphemes may have (phonolog-ical) effects simply follows from the fact that they must be integrated into theprosodic structure of their base and may not simply be deleted or float around.The two constraints in (6) are the relevant constraints ensuring that a footmust be prosodically integrated: (6-a) is a faithfulness constraint demandingthe preservation of feet that were present in the input and (6-b) is a standardmarkedness constraint against floating prosodic nodes (cf. *Float (Wolf, 2007;Saba Kirchner, 2010)) penalizing any foot that does not dominate a syllable.

(6) a. Max-FtAssign a violation mark for every input foot withouta correspondent in the output.

b. Ft⇒σAssign a violation mark for every foot that does notdominate a syllable.

Morphological colours and morpheme contiguity. Throughout the following, Iadopt the assumption of morphological colours (van Oostendorp, 2003, 2006,2007a, 2008; Revithiadou, 2007) stating that every morpheme has its own ‘colour’(=subscript indices) and that all elements in the specification of this morphemebear this colour, which cannot be changed by GEN. These colours are thereforesimply an explicit notation for the morphemic affiliation of elements, implicit inmany analyses. Elements without any colour do not belong to any morphemeand must be epenthetic. Dep constraints are consequently constraints againstcolourless elements in this framework.

The concept of morphological colours is important in the definition of at leastone constraint that is crucial in the analysis of Upriver Halkomelem, namely theconstraint about morpheme contiguity TierContig, given in (7). It demandsmorpheme contiguity across tiers and is violated as soon as a morphologicallycoloured element is dominated by a node of another morphological colour. Incontrast to the standard notion of morpheme contiguity that restricts the in-trusion into a sequence of segments affiliated to one morpheme (McCarthy andPrince, 1995; Landman, 2002), TierContig demands morpheme contiguityacross different tiers of the prosodic hierarchy.

7

(7) TierContig

Assign a violation mark for every coloured phonologicalelement that is dominated by an element (6=PrWd) of adifferent colour.

Several related concepts and constraints demanding morphemic contiguityacross tiers can be found in the literature. For one, TierContig is reminiscentof the constraint Morph-O-Contig (‘The tokens of output structure affiliatedwith a given morpheme collectively span an uninterrupted interval.’ (Wolf,2007, 59)) that refers to elements on different tiers as well, but is less restrictivethan TierContig since it refers to morphological spans, not local dominancerelations. And on the other side, there are constraints that predict similareffects but are sensitive to specific tiers of the prosodic hierarchy. Examples areMorphologicalSyllableIntegrity demanding that all elements integratedunder a syllable node should be in the morphological domain of that syllable(van Oostendorp, 2004) or Tautomorphemicity demanding that morphemeand syllable boundaries should coincide (Crowhurst, 1994; Bickel, 1998).

Lexical Stress. It was argued above that stress in Upriver Halkomelem is lexical.I assume that lexical stress is represented through placing metrical structure inthe underlying representation (Inkelas, 1999; Revithiadou, 1999). The choicebetween iambic and trochaic feet is implemented in OT through the rankingof RhT:T and RhT:I (Kager, 1999). If stress is lexical, the choice betweenanalysing it as a iambic or trochaic stress system is arbitrary in the first place:The foot structure of a word simply ensures that all underlyingly stressed syl-lables are in the head position of a foot. However, the existence of light initialstressed syllables is better analysable under a trochaic analysis (given that only(LH) and (LL) are possible iambic feet; cf., for example, Prince (1990); Hayes(1995)). In addition, I already argued that initial stress is the default in thelanguage. I therefore assume that high-ranked RhT:T demands that feet haveinitial prominence in Upriver Halkomelem. Stems are taken to be prosodifiedprior to concatenation, i.e. are integrated under a prosodic word node thatdominates at least one foot (Selkirk, 1986). This is either an underlying footthat specifies lexical stress or a foot that is assigned per default in an earlier cy-cle prior to concatenation (stratal OT; Bermúdez-Otero (2007, in preparation)).This is illustrated in (8) where the optimization of a stem without underlyingprosodic structure (8-a) is compared to a stem with underlying stress on thesecond syllable, i.e. an underlying foot structure where the second syllable isthe first in the foot (8-b). Crucially, the constraints preserving underlying footstructure (DepFt–σ ‘Assign a violation mark for every association line betweena foot Fo and a syllable σo in the output if the corresponding input nodes Fiand σi are not associated.’ and MaxFt–σ ‘Assign a violation mark for every un-derlying association line between a foot and a syllable in the input without anoutput correspondent.’) are ranked above the constraints demanding an initialtrochaic foot. The underlying foot is therefore preserved although it violatesAll-Ft-L. As will become apparent later, there are various constraints rankedhigher than the constraints preserving underlying stem foot structure and stemfeet remain unrealized in the presence of an affix foot.

8

(8) Underlying vs. assigned stress: optimizing of stems

MaxFt–σ

Dep

Ft–σ

RhT

:T

All

-Ft-

L

Prs-σ

A. Underlying stressFts

σs

ţ’sE

σs

tsE;sms

+ a.

PrWd

Fts

σs

ţ’sE

σs

tsE;sms

* *

b.

PrWd

Fts

σs

ţ’sE

σs

tsE;sms

*!

B. Default stressPsisms@sçs

a.

PrWd

Ft

σs

Psis

σs

ms@sçs

*! *

+ b.

PrWd

Ft

σs

Psis

σs

ms@sçs

9

Stress Faithfulness. I argue that the assignment of stress in complex forms inUpriver Halkomelem is restricted by the constraint on stress faithfulness (9).

(9) FtOverl

For every input foot Fi, assign one violation mark if itis not the case that a segment dominated by Fi has anoutput correspondent also dominated by a foot.

The constraint demands that at least one segment of every input foot must bein an output foot as well. It is a weak stress faithfulness constraints that unifieseffects from, first, faithfulness constraints demanding that an underlying stresspattern should be preserved and, second, faithfulness constraints demandingthat a prosodic constituent should be faithful to its form and content in theoutput. The first class of faithfulness constraints can be found in numerousversions in the literature, for example, Max-Stress or Max-Prom (Alderete,1995, 2001b; Smith, 2002; Gouskova and Roon, 2008).

(10) Max-Stress

No deletion of stress.

Another kind of prosodic faithfulness constraint refers to the content of theprosodic constituent as well, i.e. to the segmental material dominated by it. Anexample is the constraint (11)3 demanding that all material that was in a footunderlyingly should be in this foot at the surface as well.

(11) FootFaith

If a surface foot Fi corresponds to an underlying footFj, all segments in Fi should be present in Fj.

The constraint FtOverl in (9) naturally extends this typology of stress faith-fulness constraints. It differs from the stronger constraint (11) since it is satisfiedif there is a minimal overlap of at least one segment between underlying andoutput foot. Further predictions and evidence for this constraint are discussedin section 4.1.

Reduplication. Reduplication plays a crucial role in the non-concatenative mor-phology of Salishan languages. In this subsection, I present my background as-sumptions about the theoretical implementation of reduplication that are crucialin the analysis of the continuative allomorphy in section 3. Copying or ‘redu-plication’ is taken to be a general phonological repair process comparable toepenthesis or deletion, a view explicitly proposed and discussed in the theory ofMinimal Reduplication (Saba Kirchner, to appear, 2010). Copying of elementsis one strategy to avoid marked structures that can result from phonological op-erations or morphological concatenation. The latter case is what Kirchner calls‘morphological reduplication’, that applies in the presence of certain morphemes

3Discussed by van Oostendorp (1994) in an analysis of class I and class II affixes in Dutchand their different behaviour with respect to stress and syllabification. Although he rejects theconstraint for his analysis in the end, he actually interprets Max-Foot in a quite equivalentway, i.e. the constraint is violated if an underlying foot is associated with other segmentalmaterial in input and output.

10

and is triggered by the presence of empty prosodic nodes (=a marked structure).The main arguments for such a unified theory of copying are, first, its theoret-ical economy and elegance through abandoning any phonological mechanismsspecific to a certain morphological process4 and, second, the fact that there arelanguages like Upriver Halkomelem where copying alternates with other types ofnon-concatenative exponence. Other examples are Kwak’wala (Saba Kirchner,2010; Bermúdez-Otero, 2012) or Afar (Bye and Svenonius, 2012) where copy-ing of segments alternates predictably with epenthesis. The theory of MinimalReduplication straightforwardly predicts such a state of affairs since the emptyprosodic structure is not bound to a specific repair such as copying: it simplymust be filled with segmental content and the languages choose between thedifferent phonological strategies the phonology employs to provide additionalmaterial in a specific phonological context.

The theoretical implementation of copying adopted here is taken to be basedon string-internal correspondence argued for in Walker (2000a), Walker (2000c),and Hansson (2001). In their accounts of (long-distance) harmony processes,high-ranked Corr constraints demand correspondence between subsequent sim-ilar sounds in the output string. Once this correspondence is established, Ident-constraints demand featural identity between corresponding output elements.The ranking of Corr constraints is able to predict the typological finding thatsimilar segments are more likely to undergo agreement processes. I argue nowthat output-output correspondence relations between segments are not only es-tablished due to high-ranked Corr constraints but also to license segmentswithout a morphological colour.5 Copying introduces elements that were notpresent underlyingly, hence are colourless. However, this process is differentfrom epenthesis since the new colourless elements have at least an (identical orat least similar) output correspondent. The constraints penalizing epenthesisand copying are consequently constraints in a Paninian relationship: Colour!in (12-a) requires that segments bear a morphological colour (=they are neitherinserted nor copied) and License! in (12-b) requires that segments are eithermorphologically coloured or correspond to an output segment (=they are notinserted).

(12) a. Colour!Assign a violation mark for every segment not li-censed by morphological colour

b. License!Assign a violation mark for every segment neitherlicensed by morphological colour nor by correspon-dence.

It is important to keep in mind that copying necessarily violates additional con-straints; either constraints against total identity (like OCP or *Repeat (Yip,

4An example for such an arbitrary mechanism is the standard correspondence-theoreticmysterious RED morpheme that triggers reduplication through establishing a new correspon-dence relation between base and reduplicant (McCarthy and Prince, 1995).

5I am grateful to an anonymous reviewer for pointing out this option.

11

1988, 1992, 1998)) or constraints against non-identical corresponding segments(Ident-OO). Without these additional constraint violation, epenthesis wouldalways be harmonically bound by copying.

In (13), a short example is given for illustration. An empty syllable affixesto a stem and must be realized, i.e. filled with segmental material. Candidate(13-a) inserts two epenthetic segments and candidate (13-b) and (13-c) copy thefirst two segment of the base to realize the syllable. Whereas (13-b) reduces thecopied segments to placeless sounds, candidate (13-c) realizes the copied seg-ments faithfully. The epenthesis candidate (13-a) violates both Colour! as wellas License! since the sequence P@ is neither licensed by colour nor by indexa-tion. In contrast, the copying candidates (13-b) and (13-c) only violate Colour!since the inserted colourless segments are coindexed with base segments. Dueto this indexation, the corresponding segments are subject to Ident-OO de-manding featural identity between corresponding elements. This constraint isviolated by (13-b) where non-identical segments correspond. Candidate (13-c)wins the competition where the copied segments are identical to their corre-sponding counterparts. In all following tableaux, I highlight colourless elementswith a grey background to make identification of epenthetic material easier.Morphological colours are notated as subscribed indices and OO-correspondencerelations with superscribed indices.

(13) Copying vs. Epenthesis

mα aα

µα

σασB

Lic

ense

!

Colo

ur!

Iden

t-O

O

a.

mα aα

µα

σασB

P @

µ *!* **

b.

1mα 2aα

µ α

σασB

1P 2@

µ ** *!*

+ c.

1mα 2aα

µα

σασB

1m 2a

µ **

Note that License! and Colour! are not ranked to each other in (13). Only

12

indirect ranking arguments are possible between constraints in a stringencyrelation since they never conflict with each other (McCarthy, 2008, 65-67).

3. Analysis: The non-continuative aspect

3.1. Four allomorphs in the continuativeThe central aspectual distinction for verbs in Salish is the one between the

‘continuative’6 and the ‘non-continuative’. The use of the former “implies thatthe action of the root continues for a moderate length of time after onset [...]or before conclusion [... or the] action is assumed to be continuing indefinitely”(Galloway, 1993, 261). In the following paragraphs, the four different strategiesto realize the continuative are discussed and the crucial generalizations abouttheir contexts are drawn. The context generalizations I discuss below are nearlyidentical to the ones that are the basis for the analyses of the four UpriverHalkomelem continuative allomorphs in Urbanczyk (1998) and Kurisu (2001).

Stress shift. If (main) stress is on a non-initial syllable in the non-continuative,the continuative is marked by shifting the stress of the base to the initial syllableas illustrated in (14). This often results in weakening the now unstressed base-vowel to @ (or even in optional deletion, as in the last example in (14)).

(14) Continuative allomorph: Stress shifting

Non-Continuative Continuativeţ’EtE;m ‘crawl’ ţ’Et@m ‘crawling’ G56ìElqí ‘soak’ ìElqi ‘soaking’ G56ì@xw@ìţE ‘spit’ ì@xw@ìţE ‘spitting’ G56cà;l@Xw@m ‘bleed’ cá;l(@)Xw@m ‘bleeding’ G56

Whenever the non-continuative form is stressed on the initial syllable, oneof the following three allomorphs marks the continuative.

Reduplication. Non-continuative forms that start with a non-glottal consonantand have a full vowel (=not @) in their first syllable, show reduplication in thecontinuative. A copied CV-sequence is infixed after the first syllable, accompa-nied with reduction of the copied vowel to @.7

6Under various terms: ‘actual’ in e.g Thompson and Thompson (1971), ‘progressive’ in,for example, Suttles (2004), or ‘imperfective’ in Turner (2011).

7That the copied CV sequence is infixed and not prefixed, follows in an analysis assumingthat the original base vowel remains unchanged and the copied vowel reduces to @.

13

(15) Continuative allomorph: Reduplication

Non-Continuative Continuativeq’ís@t ‘tie sth.’ q’íq’@s@t ‘tying sth.’ G68p’EtT’ ‘sew’ p’Ep’@tT’ ‘sewing’ G266t’Ej@q’ ‘get angry’ t’Et’@j@q’ ‘getting angry’ G136jíq ‘fall (of snow)’ jíj@q ‘falling (of snow)’ G135ìEw ‘run away, escape’ ìEì@w ‘running away’ G135mát’@s ‘point, aim’ mám@t’@s ‘pointing’ G135

Vowel lengthening. Whenever a full vowel is preceded by a glottal onset h or Pin the first syllable of the non-continuative form, vowel lengthening marks thecontinuative as in the examples in (16).

(16) Continuative allomorph: Vowel lengthening

Non-Continuative ContinuativePím@ç ‘walk’ Pí;m@ç ‘walking’ G66PíX@t ‘scrape sth./so.’ Pí;X@t ‘scraping sth./so.’ G67Pálm@ţ@l ‘wait’ Pá;lm@ţ@l ‘waiting’ G270hílt ‘roll sth. over’ hí;lt ‘roll sth. over’ G67hákw@ç ‘use sth.’ há;kw@ç ‘using sth.’ G270hEk’w@l@s ‘remember sth.’ hE;k’w@l@s ‘remembering sth.’ G270

Epenthesis. And finally, when the stressed first vowel of the non-continuativebase is a @ rather than a full vowel, insertion of epenthetic hE can be found in thecontinuative.8 Note that the vowel in this inserted sequence is realized on thesurface as either @ or E. This alternation is predictable: @ occurs before labialm and w (cf. Urbanczyk 1998, 659; Galloway 1993, 60). The @ that constitutesthe nucleus of the first base syllable is always lost in the continuative form.

8This context generalization about the hE inserting allomorph is actually different from thegeneralization given in the analysis of the Upriver Halkomelem non-concatenative allomorphyin Urbanczyk (1998). She assumes that hE insertion is only predicted for bases startingwith a sonorant followed by a @. That reduplication is impossible in those contexts followsin her optimality-theoretic analysis from the markedness constraint *SonOns. My simplergeneralization is possible since the only example of a non-continuative form starting withan obstruent and a @ (Urbanczyk 1998:659, fn.6) is neither counterevidence against my sim-pler generalization nor evidence for the stricter generalization in Urbanczyk (1998): it formsthe non-continuative with prefixing reduplication and hence is taken as a lexically markedexception in my analysis (cf. section 5).

14

(17) Continuative allomorph: hE-epenthesis

Non-Continuative Continuativem@q’@t ‘swallow sth.’ h@mq’@t ‘swallowing sth.’ G60w@q’w ‘drown, drift downstream’ h@wq’w ‘drowning’ G273j@qw ‘burn’ hEjqw ‘burning’ G60j@q’@s ‘file’ hEjq’@s ‘filing’ G61l@p’@ç ‘eat’ hElp’@ç ‘eating’ G61l@q@m ‘dive’ hElq@m ‘diving’ G61

The assumption that h@ is epenthetic follows the analysis in Urbanczyk(1998) but is not uncontroversial. Hukari (1977), for example, has argued thatthe h in reduplicated forms is rather a debuccalized sonorant than an epentheticsegment. However, there are good arguments in favor of the epenthesis anal-ysis. First, it is rather uncontroversial that (phonologically placeless) @ andglottal consonants are preferred epenthetic segments in the languages of theworld (Lombardi, 2002; de Lacy, 2006) and at least for the glottal consonant, wehave good evidence that it is the epenthetic consonant in Upriver Halkomelem.It is inserted for phonotactic reasons in other contexts; for example, to avoidvowel hiatus (Galloway, 1993, 118). Suttles (2004) describes an ‘intrusive /h/’in Musqueam as well that separates vowel-final stems from vowel-initial suffixes(Suttles, 2004, 18+19). Second, there is a striking observation with respectto the epenthetic status of h in the continuative. Resultative stems in UpriverHalkomelem are marked by the stative prefix s–. In (18), we see a context wherethis prefix is added to a continuative form h@lţ@t. Interestingly, the h does notappear at the surface although the cluster sh, that is expected after prefixation,is a well-formed sequence in the language. This deletion “makes sense if [h] isan epenthetic place holder when underlying segments are insufficient for thesatisfaction of the onset requirement” (Kurisu, 2001, 144).

(18) Stative prefix + Continuative form (Kurisu, 2001, 144)

underlying stative prefix+continuativel@ţ@t s@lţ@t ‘fill it/filed’

In fact, there are more continuative formations that I take as lexically storedexceptions. I discuss these irregular continuative forms in section 5 and focuson the four allomorphs that are analysed in the competing analyses by Kurisu(2001) and Urbanczyk (1998). The decision tree in (19) summarizes these dif-ferent continuative formation strategies and their context of appearance as itwas discussed above.

15

(19) Four allomorphs to form the continuative

The non-continuative has. . .

. . . stressed non-initial σ . . . stressed initial σ

. . . @ in first σ . . . full V in first σ

. . . initial glottal C . . . initial non-glottal C

stress shift

hE insertion

vowel lengthening reduplication

A close look at stem shapes in Upriver Halkomelem reveals the fact that veryfew stems start with a consonant cluster. Actually, Galloway (1993) summarizesthose together with other rare stem forms into a minor group of less then 7%of all stems in the language. The context description in (19) predicts thatstems starting with CC mark their continuative through reduplication or hEinsertion (depending on whether the base has a full vowel or a @ in the firstsyllable). However, no data backs up this claim simply because very few suchnon-continuative forms based on CC-initial bases are available in the grammar.The few CC-initial non-continuative forms that can be found in the descriptionare irregular and form their continuative through metathesis and/or ablaut (e.g.kwçE;t ‘count sth’ → kwEçt@s ‘he’s counting sth.’, G274). As is argued in detailin section 5, I take those as morphologically marked exceptions.

An important generalization about the four continuative allomorphs in Up-river Halkomelem is that all resulting continuative forms involve a main-stressedfirst syllable, regardless of the stress pattern in the non-continuative form.This prosodic contrast between the continuative and non-continuative formsis emphasized in the analysis of stress in Upriver Halkomelem given in Kurisu(2001) as well. It is predicted in his analysis by faithfulness constraints indexedfor different morphemes: ProsFaithNonCont � Align-L(Headσ,PrWd) �ProsFaithCont. The second important observation about the four continua-tive allomorphs is the fact that in two of the four patterns, the stressed syllablecontains segmental material that is not morphologically affiliated to the basebut is copied or epenthetic. These two generalizations of a fixed stress patternand the preference for inserted segments at the left edge of the continuativeform are the crucial intuitions that are captured by the analysis I propose. Itis based on the assumption that the phonological representation of the continu-ative morpheme is an empty prosodic foot. The three main ingredients for myanalysis of the continuative formation in Upriver Halkomelem as affixation of amorphemic foot can be summarized as:

16

1. Stress overwriting (cf. section 3.2)A morphological foot is integrated in the prosodic structure and ‘over-writes’ the underlying prosodic structure of the non-continuative base.

2. Preference for non-base material (cf. section 3.3)The foot must dominate segments but strives to dominate as few non-continuative base segments as possible: insertion of colourless segments ispreferred

3. Allomorphy (cf. section 3.3)Different strategies to fill the foot become optimal in different phonologicalcontexts due to standard markedness constraints.

In the following subsection, these three mechanisms are discussed in turns.

3.2. Analysis of stress overwritingI argue that there are at least two contexts where stress overwriting is rel-

evant in an analysis of Upriver Halkomelem: first, for the derivation of stressin morphological complex forms (cf. section 2.1) and, second, for the formationof the continuative where a morphemic foot overwrites the stress pattern of thebase. Overwriting of underlying stress follows from the assumption that derivedmultiple feet inside a prosodic word are avoided in Upriver Halkomelem. Thetheoretical OT implementation is based on the Dep constraint against colour-less (=epenthetic) association lines between the prosodic word node and thefoot, defined in (20).9

(20) DepPW—Ft

Assign a violation mark for every colourless associationline between a prosodic word node and a foot.

Whenever two or more feet that belong to different morphemes are concate-nated under a single prosodic word node, violations of DepPW—Ftnecessarilyresult. High-ranking of DepPW—Ftconsequently predicts an output where thefoot structure of only one of the morphemes is preserved and the remainingsyllables are directly integrated under the prosodic word node. The fact thatthe affix foot is realized and not a stem foot follows in OT by the ranking of theMax constraints against deletion of feet (6-a) that are sensitive to the statusof a morpheme as being an affix or a stem. That the phonology is sensitiveto this morphological distinction is well established, cf. the overviews in Ur-banczyk (2011) or Trommer (2010) for discussion and literature. If the Maxconstraint demanding realization of affix feet is ranked above the Max con-straint demanding that feet affiliated with a stem must be realized, it followsthat an affix foot always ‘overwrites’ a stem foot. In the following tableaux, all

9Bermúdez-Otero (2001) argues that such faithfulness constraints preserving underlyingprosodic affiliation (as, for example, Depµ and in his terminology Dep-Linkµ) are problem-atic. Among other things, they predict that syllabification can be contrastive in a language.However, recent arguments as Elfner (2006) try to show that contrastive syllabification mightindeed exist.

17

new association lines (those that were not present underlyingly) are dashed forease of exposition.

Tableau (21) illustrates stress overwriting with the example (5)-c where thestressed suffix –íl attaches to the stem TE;t and T@tí;l results. Recall the as-sumption I made in section 2.2 that stems and affixes are prosodified prior toconcatenation. From this it follows that both stressed monosyllabic morphemesare already fully prosodified under a PrWd node in the input. DepPW—Ftnowpenalizes a faithful realization of both feet as in candidate (21-a): a colourlessassociation between the word node and the stem foot is inserted. Candidates(21-b) and (21-c) both integrate only one foot under the PrWd node, associatethe remaining syllable directly under the prosodic word node, and hence avoidthe violation of DepPW—Ft. In (21-b), the foot of the base is deleted and in(21-c), it is the affix foot that is deleted. The choice between both is decidedin favor of the latter since the parsing of affix prosody is more important thanparsing of stem prosody.

(21) Stress Overwriting

PrWdaPrWds

Fta

TsEsts

σa

iala

Fts

σs

Dep

PW—

Ft

Max-F

t Af

Max-F

t St

a.

PrWda

FtaFts

σs

TsEs

σa

tsiala

*! *

Û PrWddominatesaffix- andstem-foot

b.

PrWds

Fts

σs

TsEs

σa

tsiala

*! Û Affix Ftis deleted

+ c.

PrWda

Fta

σs

TsEs

σa

tsiala

* Û Stem Ftis deleted

18

I argue now that this very same mechanism applies in the formation of thecontinuative as well. DepPW—Ftensures that integration of multiple feet withdifferent morphological affiliation under a prosodic word node is impossible andsince affix prosody is more important than stem prosody, the former overwritesthe latter. The only additional mechanisms in the analysis are constraints en-suring that a floating affix foot must be integrated into the prosodic structureof its base in the first place. It follows from the interaction of Max-FtAf (6-a),penalizing the deletion of an affix foot, and Ft⇒σ (6-b), penalizing floatingfeet. In (22), foot overwriting in the context of the morphemic continuative Ftcis illustrated. A candidate that simply let the foot ‘float around’ (22-a) is ruledout by high-ranked Ft⇒σ and a candidate (22-b), that integrates the continu-ative foot and the underlying base foot into the structure, incurs a violation ofDepPW—Ft(quite parallel to the candidate a. in (21)). In contrast, candidate(22-c) only integrates the morphemic continuative foot under the prosodic wordnode and hence avoids the violation of DepPW—Ft: the winning candidateunder the given ranking.

(22) Morphological foot overwrites underlying prosodic structure

Fts

PrWdc

Ftc

σs

ţ’sE

σs

tsE;sms

PrWds

Ft⇒σ

Dep

PW—

Ft

Max-F

t Af

Max-F

t St

a.Fts

PrWdc

Ftc

σs

ţ’sE

σs

tsE;sms

PrWds

*!Û Cont. Ft

is notintegrated

b.Fts

PrWdc

Ftc

σs

ţ’sEs

σs

tsE;sms

*!Û Both feet

areintegrated

+ c.

PrWdc

Ftc

σs

ţ’sEs

σs

ts@sms

* Û Stem Ftis deleted

19

The independently motivated mechanism of stress overwriting in UpriverHalkomelem therefore ensures that the prefixed continuative foot overwritesall base feet, due to the ranking {Ft⇒σ, DepPW—Ft, Max-FtAf} � Max-FtSt. This derives the fixed initial stress in the continuative and is consequentlyall that is necessary to predict the first of the four allomorphs to realize thecontinuative: stress shifting for bases with stress on the non-initial syllable. Inthe following subsection I show why the additional phonological operations ofvowel lengthening, hE epenthesis, and copying apply to realize the continuativeas well and what governs the choice between them.

3.3. Predicting the four allomorphs in OTI claim that the insertion of hE, vowel lengthening, and copying in the contin-

uative take place to avoid a situation where the morphemic foot dominates toomuch segmental material of the base, penalized by the constraint TierContig.It was defined in (7) and demands morpheme contiguity across tiers, i.e. it isviolated as soon as a morphologically coloured element is dominated by a nodeof another morphological colour ( 6=PrWd). High-ranked TierContig thereforepredicts that morphologically coloured autosegmental nodes strive to dominatecolourless elements or elements with the same morphological colour.

There are two other constraints that are relevant for filling the foot with mor-phemic material. For one, the prefixed continuative foot must be ‘big enough’to form a good binary foot, due to the standard markedness constraint Ft-Bin (Prince and Smolensky, 1993). The strategy of inserting epenthetic hE orcopying base segments is then predicted in a quite straightforward way: bothoperations introduce material that has no morphological colour and thereforematerial that can be integrated under the continuative foot without any vio-lations of TierContig. On the other hand, filling the continuative foot withcolourless material is restricted by the faithfulness constraint FtOverl (9) de-manding that at least one segment that is part of the foot underlyingly must beintegrated into the foot in the output.

The four continuative allomorphs, their prosodic structure, and their vio-lations of the constraint TierContig that are crucial in the derivation of thecontinuative are given in (23). All the different structures of continuative formshave in common that the morphemic foot is realized at the left edge of theword, i.e. it always dominates the leftmost base segment. I assume that allstrategies realizing the prefixed foot further right are excluded by higher-rankedconstraints ensuring that the (suprasegmental) linear order of morphemes ispreserved (Yu, 2003; Fitzpatrick, to appear; Trommer, 2011).

The base foot is not integrated under the prosodic word node due to the con-straint DepPW—Ft (cf. tableau (22)). Undominated FtOverl is not violatedin any of these representations since at least one segment of the main stressedfoot in the base is integrated into the continuative foot.

20

(23) The continuative allomorphs

Input Output TierContig

I. Stress shift: ţ’EtE;m → ţ’Et@mPrWdsPrWdc

FtsFtc

σs

ţ’sEs

σs

tsE;sms

PrWdc

Ftc

σs

ţ’sEs

σs

ts@sms

*****

II. Vowel-lengthening: Pím@ç → Pí;m@çPrWdc PrWds

FtsFtc

σs

µs

isPs

σs

µs µs

ms @s çs

PrWdc

Ftc

σs

µs µ

i;sPs

σs

µs µs

ms @s çs

**

III. Reduplication: q’ís@t → q’íq’@s@t

PrWdc PrWds

FtsFtc

σs

q’sis

σs

ss@sts

PrWdc

Ftc

σs

1q’s2is

σ

1q’2@

σs

ss@sts

**

IV. hE epenthesis: m@q@t → h@mq@tPrWdc PrWds

FtsFtc

σs

ms@s

σs

qs@sts

PrWdc

Ftc

h E ms

σs σs

qs@sts

*

Given the constraint violations of TierContig in (23), epenthesis (23-IV)should always be predicted as optimal output for a continuative form sinceit integrates least base segments under the continuative foot. In the example(23-IV), it is only the morphologically coloured segment m that is dominatedby the morphemic foot affiliated with another morphological colour (recall thenotational convention to highlight all inserted – non-base – segments with agrey background). Stress shift (23-I) on the other side is an apparently very

21

bad strategy to realize the continuative foot since it invokes many violations ofTierContig.

That the apparent dispreferred strategies (23-I)–(23-III) nevertheless applyfor bases with certain phonological forms follows since other constraints are rel-evant in the derivation of the continuative foot as well. For one, there are theconstraints violated by the different phonological operations the language em-ploys to realize the foot: vowel lengthening violates *Vµµ (24), copying violatesColour! (12-a), and hE epenthesis violates Colour! and License! (12-b).The ranking of these three constraints in Upriver Halkomelem is taken to be theone in (25), which predicts that copying is the least marked strategy to providecolourless material. Epenthesis of hE violates higher-ranked License! in addi-tion to Colour! and is therefore a less preferred option. Vowel lengtheningviolates highest-ranked *Vµµ and should only surface if the other strategies areimpossible.

(24) *VµµAssign a violation mark for every vowel that is associ-ated with two moras.

(25) Preference hierarchy to realize the continuative*Vµµ � License! � Colour!

These constraints are always violated whenever a candidate undergoes vowellengthening, epenthesis, or copying. Another class of constraints that are crucialin the derivation of the continuative allomorphy are those that are only violatedif a base with a certain phonological shape undergoes one of the strategies torealize the continuative foot. One is the constraint Ident[σ-role] (Kenstow-icz, 2005; Rose and Walker, 2004; Bye and Svenonius, 2012) demanding thatsegments that correspond in the output have the same syllable role (onset, nu-cleus, coda). It restricts the possibilities to fill the continuative foot with copiedsegments and excludes a syllable with two corresponding consonants, one beingthe onset, the other the coda.

(26)Ident[σrole]

Assign a violation mark for every pair of OO-corresponding segments with different syllable roles.

Another relevant constraint is *P@ (Kurisu, 2001; Urbanczyk, 1998), a marked-ness constraint against placeless syllables, i.e. those only consisting of h or Pand @. That it is indeed active in Upriver Halkomelem is empirically supportedby a statistical examination about stem shapes discussed in Urbanczyk (1998),which led to the conclusion that placeless syllables are truly rare in the language.As will become clear in the discussion of the different continuative derivationsin (28)-(31), it excludes hE epenthesis and copying in some contexts since theseoperations create placeless syllables for bases with a certain shape.

(27) *P@ Assign a violation mark for every placeless syllable.

In the following concrete derivations of the different allomorphs, I abandonthe complex autosegmental illustrations I listed in (23) for ease of exposition.

22

In addition, I don’t include competing candidates not realizing the continuativefoot or integrating more than one foot into the prosodic word node since itwas already shown that those are excluded by undominated constraints (cf. thegeneral tableau on overwriting in (22)). The following tableaux therefore onlyillustrate how the constraint system chooses the optimal strategy to realize thecontinuative foot for different bases. The order of candidates is always the same:the continuative foot is filled with the first (a.), or the first two syllables (b.), aCV sequence is copied and infixed (c.), the first base syllable is parsed into thefoot and its vowel is lengthened (d.), epenthetic hE is inserted (e.), or epenthetichE is inserted and the first base vowel is deleted (f.).

I begin the discussion with the stress shifting allomorph. Recall from (23)that it is the allomorph with the worst constraint profile with respect to Tier-Contig. Nevertheless, stress shifting becomes optimal for non-continuativeforms with stress on a non-initial syllable. An example is ţ’EtE;m ‘crawl’ thatis realized as ţ’Et@m in the continuative. That the non-continuative base isstressed on the second syllable follows if only this second syllable is parsed intoa foot with initial prominence underlyingly. This can be seen in tableau (28)(cf. also (22)). If the base-foot is ‘overwritten’ by the continuative foot due toFt⇒σ, the faithfulness constraint FtOverl demands that at least one of thesegments of the syllable tEm must be integrated into this continuative foot inthe output. Integration of only the first base syllable into the morphemic foot(28-a), copying in candidate (28-c), vowel lengthening in candidate (28-d) andepenthesis in candidates (28-e)+(28-f) are all excluded by FtOverl since theyparse neither t, E nor m into the continuative foot.10 The crucial ranking in thecontext of a stress shifting allomorph is therefore the ranking of FtOverl overTierContig that necessarily demands that both base syllables are integratedinto the foot as in winning candidate (28-b).

Other constraints that are included in the ranking in (28) will become rel-evant for the optimization of other allomorphs: the markedness constraintsIdent[σ-role] and FtBin and the faithfulness constraints against deletion ofvowels MaxV[pl] and Max@.

10Given that three-syllabic feet are impossible due to FtBin (Prince and Smolensky, 1993)and a candidate like *(ţEtEm) is generally excluded.

23

(28) Stress shifting ţ’EtE;m → ţ’Et@m

(c ) + ţ’E(stE;m) FtO

ver

l

MaxV[pl]

Iden

t[σ-r

ole

]

Tie

rC

onti

g

Max

@

FtB

in

*Vµµ

Lic

ense

!

Colo

ur!

a. (cţ’E)t@m *! **

+ b. (cţ’Et@m) *****

c. (c1ţ’2E 1ţ’2E)t@m *! ** **

d. (cţ’E;)t@m *! ** *

e. (c hEţ’E)t@m *! ** ** **

f. (c hEţ’)t@m *! * * ** **

In contrast, it is possible to avoid more violations of TierContig if stressis on the first syllable in the non-continuative form. Only one – the first –base segment has to be integrated into the morphemic foot in order to avoida violation of FtOverl. Since more segmental material is required to form abinary foot (FtBin), epenthetic or copied segments are inserted to add syllabicweight to the morphemic foot.

Recall from (25) that copying, which violates only lowest-ranked Colour!,is taken to be the default strategy for providing non-base segments. This pref-erence for copying is crucial in tableau (29) where the continuative for the baseq’ís@t is optimized. Candidates (29-a) and (29-b) only integrate the continua-tive foot into the base without any additional phonological operation. However,integration of only the first light syllable results in a fatal violation of FtBin11

(29-a) and integration of two syllables results in too many violations of Tier-

11That FtBin is ranked high in Upriver Halkomelem and above the constraints penalizingrepair strategies to add prosodic weight (insertion, lengthening) predicts that words with astressed final short syllable are always neutralized into a form with a heavy stressed finalsyllable since (L) is an ill-formed trochaic foot. This prediction seems to be empiricallycorrect. For one, only a small percentage of stems are argued to be vowel-final (around 10%,cf. Galloway (1993)) and all of them are polysyllabic (which makes it more likely that the finalsyllable is stressed). Another crucial observation is the fact that many suffixes (derivationaland inflectional) are attached to bases, most of them C-final. Transitive verbs, for example,are obligatorily followed by a control suffix and all the various (allomorphs of) control suffixesare C-final. Consequently, there is rare evidence for words with a final light stressed syllable.A possible exception is the non-continuative form ìElqí ‘soak’ (Galloway, 1993, 56). If onerecalls the discussion of the phonetic manifestations of stress and length (3) and the difficultiesin eliciting them, it is a reasonable hypothesis that such rare final vowels that are notated asshort and stressed are indeed long.

24

Contig (29-b). In candidates c., d. and e., the continuative foot dominatestwo base segments and it is ensured through additional phonological operations(vowel lengthening, insertion, copying) that the foot is binary. The crucial rank-ing in (25) predicts that copying in candidate (29-c) is the optimal strategy tofill the foot with segmental material.

Note that candidate (29-f) fares best with respect to TierContig since itintegrates only one base segment under the morphemic foot. Avoidance of oneviolation of TierContig is possible since resyllabification after vowel deletionresults in a structure where only the first base segment is integrated underthe morphemic foot (as coda consonant of the first syllable). However, thisvowel deletion is excluded by high-ranked MaxV[pl] penalizing non-realizationof vowels with a place specification ( 6=@).

In sum, tableau (29) illustrates the ranking argument of MaxV[pl] aboveTierContig and of *Vµµ over Colour! in Upriver Halkomelem.

(29) Reduplication: q’ís@t → q’íq’@s@t

(c ) + (q’iq@s) FtO

ver

l

MaxV[pl]

Iden

t[σ-r

ole

]

Tie

rC

onti

g

Max

@

FtB

in

*Vµµ

Lic

ense

!

Colo

ur!

a. (cq’i)q@s ** *!

b. (cq’iq@s) ***!**

+ c. (c1q2i 1q2@)q@s ** **

d. (cq’i;)q@s ** *!

e. (c hEq’i)q@s ** *!* **

f. (c hEq’)q@s *! * ** **

It is clear that reduplication for a monosyllabic base like jíq does not minimizethe number of base segments that are dominated by the continuative foot: Inthe continuative form jíj@qCont, the same three base segments are integratedunder the continuative foot as in a possible non-reduplicated continuative form*jíqCont. The constraint TierContig is therefore not the trigger for providingnon-base segments in such a context. However, given the assumption that finalcodas are not moraic, every word consisting only of a CVC string violates FtBin.Reduplication and subsequent insertion of an additional mora avoids this markedstructure. The fact that reduplication only applies in the continuative andmonosyllabic words of the shape CVC are perfectly fine in other morphologicalcontexts follows from simple faithfulness demands: FtBin is lower-ranked thanthe constraint against the integration of epenthetic syllables into stem prosody

25

but higher-ranked than the constraint against epenthetic integration of syllablesinto affix prosody (DepFtSt—σ� FtBin� DepFtAf—σ). The expected repairof marked monosyllabic feet is hence only possible under an affix foot, neverunder a stem foot.

Whereas deletion of full vowels is prevented by high-ranked MaxV[pl], dele-tion of a placeless @ violates only lower-ranked Max@. Integration of only onebase segment is therefore possible for bases with a @ as first vowel. This isillustrated in tableau (30) for the non-continuative base m@q@t. The first basevowel @ is deletable and resulting resyllabification of the first base consonantm as coda of the new preceding syllable allows to integrate only a single basesegment under the continuative foot. As was argued above, the ranking ofColour! and License! favors copying as strategy to insert non-base segmentsas in (29). However, this default strategy is excluded in those resyllabificationcontexts due to Ident[σ-role]. The crucial candidate is (30-g), that inserts anepenthetic @ and copies the first base consonant. It avoids one of the violationsof high-ranked Colour! since colourless m is licensed through its correspon-dence with a morphologically coloured segment. However, this copying resultsin a situation where two consonants are in an OO-correspondence relation thatare syllabified in different syllable positions (as onset and as coda), excludedby Ident[σ-role]. Copying is therefore impossible and insertion of epentheticmaterial as second-best strategy to provide non-base material results as optimaloutput (30-f). This winning candidate violates TierContig only once. Allother candidates (no phonological operation as in (30-a+b), copying in (30-c)or vowel lengthening in (30-d)) incur two violations of TierContig since twobase segments m@ are integrated into the continuative foot.

(30) h@-epenthesis: m@q@t → h@mq@t

(c ) + (sm@q@t) FtO

ver

l

MaxV[pl]

Iden

t[σ-r

ole

]

Tie

rC

onti

g

Max

@

FtB

in

*Vµµ

Lic

ense

!

Colo

ur!

a. (cm@)q@t **! *

b. (cm@q@t) **!***

c. (c1m2@ 1m2@ )q@t **! **

d. (cm@;)q@t **! *

e. (c h@m@)q@t **! ** **

+ f. (c h@m)q@t * * ** **

g. (c1m @ m1)q@t *! * * * **

26

Finally, copying and epenthesis as strategies to fill the continuative footwith non-base segments are excluded for bases starting with a glottal conso-nant. This is due to the constraint I already introduced in (27), the markednessconstraint against placeless syllables. As can be seen in (31), copying (31-c) aswell as hE insertion (31-e+f) results in a placeless syllable whenever the non-continuative base starts with a glottal consonant. The optimal output for sucha non-continuative form is therefore the vowel lengthening candidate (31-d) thatintroduces an epenthetic mora under violation of *Vµµ. This mora is necessarysince FtBin demands that the foot must be bimoraic and excludes candidate(31-a), that only integrates the first syllable without vowel lengthening. Thetableau shows that the markedness constraint against placeless syllables *P@ isranked high in Upriver Halkomelem and at least in the same high stratum asFtOverl and MaxV[pl]. The position of the general markedness constraintagainst long vowels prohibits vowel lengthening in the continuative for all basesnot starting with a glottal consonant and a full vowel – it does, however, notcause any misprediction for underlying long vowels since the faithfulness con-straint against shortening is higher-ranked.

(31) Vowel lengthening: Pím@x → Pí;m@x

(c ) + (sPim@x) *P@

MaxV[pl]

Iden

t[σ-r

ole

]

Tie

rC

onti

g

Max

@

FtB

in

*Vµµ

Lic

ense

!

Colo

ur!

a. (cPi)m@x ** *!

b. (cPim@x) ***!**

c. (c1P2i 1P2@)m@x *! ** **

+ d. (cPi;)m@x ** *

e. (c hEPi)m@x *! ** ** **

f. (c hEP)m@x *! * * ** **

As in the reduplication contexts, TierContig cannot be the trigger for theadditional phonological operation in monosyllabic contexts: the continuativeform hí;lt includes as many base segments under the continuative foot as thepossible ungrammatical competitor *hílt. Again, it is FtBin that ensures theextension of the syllable into a bimoraic structure.

The four exemplifying optimizations in (28)–(31) illustrate how the contin-uative allomorphy in Upriver is predicted. In the next sections, the analysis,its theoretical implications and its advantages over alternative accounts are dis-cussed.

27

4. Discussion: Predictions and Alternatives

4.1. The constraint systemMy analysis for the continuative allomorphy is based on two main mecha-

nisms: First, the interaction of TierContig, FtOverl, and FtBin demandsthat the prefixed morphemic foot is filled with as much non-base material aspossible without being too unfaithful to the underlying foot structure. Second,the ranking of constraints that are violated whenever copying, epenthesis orvowel lengthening applies (TierContig, *Vµµ, License!, Colour!) predictsa preference order for phonological operations to supply the foot with additionalnon-base material (cf. (25)). The interaction with constraints that are only vi-olated if bases of a certain shape undergo one of these phonological strategiesrestricts these phonological operations to certain contexts. For example, *P@ isonly violated when epenthesis or copying applies to bases starting with a glottalsound. The same holds for FtBin and Ident[σ-role]– they exclude the morepreferred strategies to realize the continuative foot only for certain bases.

The complete constraint ranking for the Upriver Halkomelem continuativeallomorphy is given in (32).

(32) Full constraint ranking for Upriver Halkomelem{Ft⇒σ, DepPW—Ft, Max-FtAf}

�{Max-FtSt, *P@, FtOverl, MaxV[pl], Ident[σ-role] }

�TierContig

�{Max@, FtBin}

�*Vµµ�

{License!, Colour!}

I assumed four constraints that are not standard faithfulness or markedness con-straints in correspondence-theoretic OT: Colour!, License!, FtOverl, andTierContig. As was already discussed in section 2.2, the constraints Colour!(12-a) and License! (12-b) are the result of implementing copying as phono-logical operation that establishes a string-internal correspondence relation. Ihave argued that this modeling of copying and epenthesis is able to capturethe insights in, for example, Saba Kirchner (2010): copying is simply one ofother possible phonological operations to fill empty prosodic nodes. Imple-menting reduplication with these assumptions avoids the adoption of arbitrarymechanisms like the RED morpheme in standard correspondence-theoretic OT(McCarthy and Prince, 1995).

Another new constraint I proposed was FtOverl (9). I argued in section 2.2that this faithfulness constraint is just an expected extension in the typology ofstress faithfulness demands. Evidence for such a faithfulness mechanism comes

28

from languages with lexical stress in which some phonological process cruciallyrefers to the domain of the foot. An example for such a situation can be found ina well-known vowel reduction process in Russian, analysed in Alderete (1995).The stress system in Russian is far from trivial and for the sake of illustratingthe relevant point here, I will follow Revithiadou (1999) in assuming that stressin Russian is lexical stress. The relevant phonological process reduces vowelsto @ if they are not in the main foot. A reasonable analysis of such a patternis based on the assumption of general markedness constraints triggering vowelreduction to @. That vowels in the main foot are not affected follows thenfrom faithfulness demands preserving this prominent position. In the analysisof Alderete (1995), the constraint Head(P-Cat)-Ident(F) demands faithfulrealization of the feature specification for segments in prosodic heads. A crucialpoint in such an analysis is obviously some faithfulness demand that ensuresthat the elements that were in the main foot underlyingly remain in this mainfoot. Otherwise it is expected that the affiliation of segments into the outputfoot is changed and that only elements that do not invoke violations of themarkedness constraint in question are in the main foot.

Finally, the constraint TierContig (7) was assumed that is reminiscentof well-established contiguity demands. Another situation predicted by Tier-Contig can be found in contexts where morpheme boundaries are relevant fora phonological operation. A possible example comes from Diola-Fogny (Kager,1999) where a deletion process seems to be guided by the demand to achievemaximal isomorphism between syllable boundaries and morpheme boundaries.Consider the data in (33), where deletion of the first consonant in word-internalCC-clusters can be observed.

(33) Consonant deletion in Diola-Fogny (Kager, 1999, 134)/let-ku-jaw/ [le.ku.jaw] ‘they won‘t go’/jaw-bu-Nar/ [ja.bu.Nar] ‘voyager’

Kager (1999) claims that the deletion is triggered by syllable well-formednessconditions (Coda-Condition) and that the choice for deleting the first conso-nant and not the second one follows from the constraint (34) demanding align-ment between morpheme and syllable boundaries. In a correct output formle.ku.jaw, all morpheme boundaries coincide with a syllable boundary whereasin an alternative candidate *le.tu.jaw, the coda of the first morpheme let issyllabified as onset of the following syllable and therefore misaligns the twoboundaries.

(34) Align-Morph-L

The left edge of every morpheme coincides with theleft edge of a syllable.

An alternative implementation that avoids reference to Alignment con-straints is a solution that refers to the prosodic structure itself. TierContigdemands that the segmental material affiliated to one morpheme must not besyllabified under the syllable node of another morpheme. For the Diola Fognydata, this predicts that *le.tu.jaw is excluded since t is syllabified under the

29

syllable node morphologically affiliated with another morpheme (–ku).Another straightforward prediction of high-ranked TierContig is the ab-

sence of compensatory lengthening in a language. A morphologically colouredmora that is left over after deletion of a segment causes a violation of Tier-Contig as soon as it dominates a vowel of another morphological colour.

4.2. Generalized Nonlinear Affixation vs. its alternativesThere are two major arguments for my account based on the GNA hypothesis

and against possible alternatives. In the following discussion, I concentrateon one class of alternative analyses, namely those based on some concept ofRealizeMorpheme. For one, the only two existing formal accounts of theUpriver Halkomelem non-continuative allomorphy are of this type. Secondly,the RealizeMorpheme approaches claim to have the same predictive poweras the approaches within the GNA program: they are able to predict all kindsof non-concatenative morphology. For reasons of space, I won’t discuss anyother alternatives that can be regarded as such general approaches to non-concatenative morphology as, for example, the transderivational antifaithfulnessapproach developed by Alderete (2001a) and Alderete (2001b).

The first argument for my account is an economy argument stating thatsuch an account is able to predict the same kinds of non-concatenative mor-phology as possible alternatives but introduces hardly any new mechanisms.The second argument is based on restrictiveness: alternative accounts sufferfrom a severe overgeneration problem. Before I discuss these two points in turn,I briefly sketch the existing alternative analyses of the Upriver Halkomelemnon-concatenative allomorphy.

Alternative accounts. An alternative to the assumption of GNA is to assumethat morphemes may lack any underlying phonological representation. This isthe view defended in the approach proposed in Kurisu (2001) where the (mor-phological) presence of a morpheme is sufficient to trigger some phonologicaloperation that makes the underlying form different from the resulting morpho-logically more complex form. This is ensured through the constraint RealizeMorpheme (RM).12 Given this architecture, a morpheme can be realized byany conceivable phonological operation the language’s phonology provides if seg-mental material is absent or if the realisation of available segmental materialresults in a phonologically marked structure. Instances of non-concatenativeallomorphy are taken as a strong argument for such an approach and an anal-ysis of the Upriver Halkomelem facts is given on pp. 142-152 in Kurisu (2001).The specific phonological instantiation of a morpheme is determined by the

12Note that there are various implementations of Realize Morpheme concepts. In itsoriginal formulation in, for example, Samek-Lodovici (1992) or Walker (2000b), it demands themapping of each morpheme to some phonological element in the output. Kurisu (2001) arguesthat such a version of RM is insufficient to account for certain types of non-concatenativemorphology (for example, subtractive morphology) and defines it as a constraint that demandsthat an underlying form is different from a morphologically more complex form.

30

ranking of faithfulness constraints ranked below RM. For example, the rank-ing in (35) predicts the continuative allomorphy in Upriver Halkomelem. Thecontinuative morpheme strives to be realized by hE epenthesis (=violating onlylowest-ranked Dep), and only if this allomorph is blocked by phonological de-mands, other allomorphs (e.g. reduplication violating Integ) are ‘tried’ in turn(cf. the preference hierarchy to realize the continuative in my analysis (25)).

(35) Ranking for the non-continuative allomorphy in Upriver Halkomelem(Kurisu, 2001, 150)

{Align-L(Headσ,PrWd), RM}� Ident-Length� Integ� Dep

Kurisu’s analysis therefore explicitly assumes that “a faithfulness constraintis subdivided into several indexed components” (Kurisu, 2001, 58), i.e. the factthat every faithfulness constraint must in principle be assumed to exist in anindexed version for every morpheme of the language.13 This follows from thesimple fact that one bare stem can serve as input to more than one morphosyn-tactic category. A hypothetical example may briefly illustrate this: In a hypo-thetical language L1, singular is marked through subtraction and first person ismarked through reduplication. The ranking in (36) with coindexed faithfulnessconstraints penalizing reduplication (Integ) and subtraction (Max) predictsexactly such a pattern.

(36) Indexed faithfulness constraints{IntegSg, Max1.Ps} � {Integ1.Ps, MaxSg}

Yet another alternative analysis of the Upriver Halkomelem facts is proposedin Urbanczyk (1998) where the continuative morpheme is analysed as reduplica-tive in nature, i.e. it is an abstract RED-morpheme. All other strategies to re-alize the continuative apply to avoid a marked reduplicant: “rather than havingunmarked structure emerge in the reduplicant, in the ‘continuative’, redupli-cation is avoided altogether” (Urbanczyk, 1998, 656). Although this approachseems quite different from the approach proposed in Kurisu (2001) since thereis ‘something’ representing the continuative morpheme underlyingly, there is anadditional RM ingredient that ensures that some phonological operation appliesin case reduplication is blocked: “The driving force behind the non-reduplicativeallomorphy is to produce a phonologically distinct ‘continuative’ stem.” (Ur-banczyk, 1998, 656). This is ensured by the constraint DistinctStem thatenforces an alternation between stem and morphologically related form.

The economy argument. Both approaches have in common that an additionaltheoretical ingredient is necessary: the assumption of RM or DistinctStem14,

13Cf. also the brief mentioning of the stress analysis in Kurisu (2001) in section 3 wherethe existence of ProsFaithNonCont and ProsFaithCont are crucial.

14It is emphasized in Kurisu (2001) that DistinctStem is conceptually different from RMin a way that makes it problematic: DistinctStem cannot capture the fact that the formsare phonologically different since a morpheme is present whereas “[b]y contrast, RM directly

31

two constraints demanding some phonological difference between two morpho-logically related forms. This crucially distinguishes these approaches from theanalysis I propose where independently motivated assumptions like the exis-tence of feet and prosodic well-formedness requirements about the integrationof prosodic nodes into the structure are sufficient to predict the effect of non-concatenative allomorphy. In addition, the indexation of all faithfulness con-straints to (potentially) all morphemes in a language that is crucial in Kurisu’sanalysis is definitely a non-trivial extension of the correspondence-theoretic con-straint system. It introduces a powerful mechanism into the phonological com-ponent of a language that crucially refers to morphological information. Suchtheories allowing direct access to morphological information in the phonologysuffer from “a severe lack of empirical content and heuristic power” (Bermúdez-Otero, 2012, 79) and an analysis is preferable that restricts the ability of thephonology to refer to morpho-syntactic information.

Of course, it is hard to weigh up the theoretical complexity of different ap-proaches, but the integration of morpheme-specific mechanisms is definitely apowerful and complex new mechanism. However, maybe it is motivated since itenables the theory to predict a wider range of relevant phenomena? What otherinteresting predictions are possible in an approach based on RM or Distinct-Stem? First, they allow a unified analysis of all kinds of non-concatenativemorphology. This is especially interesting for cases of non-concatenative mor-phology that are notoriously difficult to account for in an additive model ofmorphology, esp. subtractive morphology. Second, – and that is actually theargument from Upriver Halkomelem – they are able to predict instances of non-concatenative allomorphy.

I argue now that an approach based on GNA is able to predict the verysame range of attested patterns of non-concatenative morphology. There areproposals that apply the GNA idea to mutation (Akinlabi, 1996; Wolf, 2007;Finley, 2009), reduplication (Saba Kirchner, 2010), or even subtraction (Seiler,2008; Trommer and Zimmermann, 2010). As was argued in Trommer and Zim-mermann (2010), subtraction follows under an approach assuming nonlinearaffixation that makes full use of the predictions of autosegmental structures.More concretely, they argue that final coda deletion in Tohono O’odham followsfrom affixing a morphological mora that must dominate some segment (=thefinal consonant) but is itself not dominated by a higher prosodic node. Such adefective prosodic structure results in deletion since a prosodic node that is notdominated by the highest prosodic node is uninterpretable for the phonetics (andall it dominates) under standard assumptions on the phonetic interpretation ofautosegmental structure.

The second argument for the alternative approaches is the fact that they areable to predict instances of non-concatenative allomorphy. That an approachbased on GNA is able to account for such facts as well is the main argument

captures the idea that distinct forms are required by the presence of a new morpheme” (Kurisu,2001, 154).

32

of the present analysis for Upriver Halkomelem and has also been made else-where. Examples for approaches where the integration of a morphemic prosodicnode may result in non-concatenative allomorphy are Stonham (2007) or SabaKirchner (2010) where the affixation of a mora results in different phonologicaloperations to mark a morpheme.

The overgeneration argument. The second main argument for an approach inline with GNA is restrictiveness. I argue that RM-based approaches sufferfrom a severe overgeneration problem. Neither RM nor DistinctStem re-strict the phonological operations that are triggered by the presence of a (non-concatenative) morpheme and both approaches are therefore able to predictlanguages with any conceivable combination of non-concatenative allomorphs. Iclaim that this is a misprediction and that the examples of non-concatenative al-lomorphy found in the languages of the world are restricted in a way that allowsto predict them from some (abstract) phonological representation. One predic-tion that an approach based on GNA makes with respect to possible patternsof phonologically predictable allomorphs is that non-concatenative allomorphyis generally either additive or subtractive. Recall that subtraction under aGNA approach is based on a defective integration of a prosodic node. Addi-tive exponence, however, is based on the complete integration of an additionalprosodic node (e.g. the foot in UH). Whether an affixed prosodic node is com-pletely integrated into the structure or only partially integrated, results fromthe general ranking of well-formedness and faithfulness constraints referring toprosodic structure in a language. If the phonologically predictable allomorphsof a prosodic affix node now result from different phonological realisations ofone and the same new prosodic structure, it is clear that these allomorphs areeither realisations of a defective prosodic structure (=partial integration) or anenriched prosodic structure (=complete integration). No approach based onparadigmatic distinctness can make such a prediction.

Although negative generalizations about unattested patterns in the lan-guages of the world could only be regarded as tentative arguments, it is quitestriking that non-concatenative allomorphy is rather rare cross-linguisticallyand, most importantly, far fewer than all imaginable combinations of non-concatenative allomorphs are attested.

Other examples for non-concatenative allomorphy can be found in other Sal-ishan languages. An example is Saanich (Central Salishan) where CV-metathesis,CV-reduplication, or P epenthesis mark the continuative aspect. This allomor-phy is analysable as affixation of a mora since all three phonological opera-tion add syllabic weight (Davis and Ueda, 2006; Stonham, 2007; Zimmermann,2009). Something similar holds for the emphatic adjective formation in ShizuokaJapanese (marked through nasal insertion, vowel lengthening, or gemination;Davis and Ueda (2002)) or the imperfect formation in Alabama (vowel length-ening or consonant gemination; Lupardus (1982); Montler and Hardy (1988);Samek-Lodovici (1992)). There are also example of non-concatenative allomor-phy where the choice between the different allomorphs is lexically governed andcannot be predicted by phonological form alone. An example for such a pat-

33

tern is Hiaki (Molina, 1999; Haugen, 2008; Harley and Leyva, 2009) where thehabitual is marked through σ-reduplication, Ft-reduplication, vowel lengthen-ing, or consonant gemination. Again, the different phonological processes thatare exponent of a morpheme are additive: they add prosodic weight. Yet an-other type of non-concatenative allomorphy involves different mutation patternsas can be found in for example Irish consonant mutation (Ní Chiosáin, 1991).It has been argued that this pattern can be analysed as the affixation of anabstract (featural) morpheme as well (Rice, 1993; Trommer, 2009). However,imaginable non-concatenative allomorphy where, for example, additive phono-logical processes (e.g. vowel lengthening) and subtractive phonological processes(e.g. consonant subtraction) are combined seem cross-linguistically unattested.They are easily derivable under an analysis based on RM or DistinctStembut are apparently impossible under an analysis based on GNA. I take this asstrong argument against analyses assuming that non-concatenative morphologyresults from the demand that morphologically different forms must be phonolog-ically ‘different’. That patterns of non-concatenative allomorphy are restrictedfollows if all non-concatenative allomorphs are the realization of an (autoseg-mental) phonological element.

In sum, my approach to the non-concatenative allomorphy first is more eco-nomic since it relies on the affixation of independently motivated phonologicalelements and avoids the introduction of morpheme-specific mechanisms, andsecond avoids an overgeneration problem that RM-base approaches to non-concatenative allomorphy have.

5. Broaden the view: More continuative formations

My analysis is another example for the successful implementation of theGNA idea. The starting point of the analysis are the analyses of the UpriverHalkomelem pattern by Urbanczyk (1998) and Kurisu (2001) that rely on ad-ditional theoretical machinery and suffer from an overgeneration problem as Iargued in section 4.2. As in these two alternative analyses, only four continuativeallomorphs in Upriver Halkomelem formed the empirical basis for my proposal.However, I already mentioned in section 3.1 that these four allomorphs are notthe only attested continuative formations in Upriver Halkomelem. There areother forms that are unexpected from the analyses in Urbanczyk (1998) andKurisu (2001) and from everything I said so far.

First, Galloway (1993) lists metathesis15 and ablaut or even combinationsof these as other allomorphs that mark the continuative. Some examples aregiven in (37-a) and (37-b) respectively. Note that some of the examples show

15There are good arguments that this continuative formation does not involve the actuallinear reordering of base segments: the stems undergoing ‘metathesis’ are underlyingly vowel-less and the @ in those examples is part of the following suffix resp. inserted due to phonotacticdemands (cf. also Leonard and Turner (2010)).

34

more than one phonological operation that distinguishes the continuative fromthe non-continuative; for example, some forms in (37-b) show ablaut as well aslength alternations and/or stress shift.

(37) More continuative formations

Non-Continuative Continuative

a. MetathesisTq’@t ‘spear sth./so.’ TEq’t ‘spearing sth./so.’ G274q’p@t ‘collect sth.’ q’Ept ‘collecting sth.’ G274TX@t ‘push so./sth.’ T@Xt ‘pushing so./sth.’ G274k’wçEt ‘count so./sth.’ k’wEçt ‘counting so./sth.’ G274

b. Ablaut. Tíjt ‘make sth.’ TE;jt ‘making sth.’ G68

Tij@qwE;ls ‘dig’ Táj@qw@ls ‘digging’ G70cakwí;lt ‘follow behind so.’ cakwú;lt ‘following behind so.’ G73q’@w@t ‘drum’ q’aw@t ‘drumming’ G113

With respect to the verbs using ablaut and/or metathesis, Galloway (1993)concludes that “one subgroup of these verbs drops the ablaut in the non--continuative but keeps the metathesis. The other subgroup uses both the ablautand the metathesis” (Galloway, 1993, 275). The affiliation of a verb to any ofthese ‘subgroups’ is clearly not predictable from their form or meaning alone.Some verbs even reverse the pattern that the continuative is derived from thenon-continuative: a “few small classes of verbs use metathesis to show non-continuative but lack the metathesis in the continuative” (Galloway, 1993, 262).

In addition, there is another reduplication pattern with stress on the secondsyllable that marks the continuative. Some examples of this reduplication withnon-initial stress that is unexpected under the analysis I propose are given in(38-d-f). Since this non-initially stressed reduplication allomorph cannot be pre-dicted from the phonological form alone, it is clear that only a morphologicallydetermined class shows this continuative formation. This is illustrated with theoverview in (38) where I confront non-continuative forms that are phonologi-cally very similar but show the reduplication pattern with initial stress in thecontinuative. In the rightmost column, I list forms that show the reduplicationpattern (15) that I take as regular reduplication allomorph in the continuative,and in the leftmost column, reduplicated continuative forms with non-initialstress are listed.

(38) Different reduplication patternStress on second syllable ‘Regular’ stress on first syllable

a. q’á;w@l q’@q’áw@l d. t’í;l@m t’ít’@l@m‘howl’ G154 ‘sing’ G135

b. pá;t p@pá;t e. q’á;j q’áq’@j‘blow on sth./so.’ G153 ‘dying’ G266

c. xwí xw@xwí f. Xíl@X XíX@l@X‘wake’ G264 ‘making war’ G136

35

However, for one class of stems the choice for this reduplication allomorph withnon-initial stress is fully predictable, namely for all stems with the shape C@Cas can be seen in (39).

(39) Continuative formation of C@C stemsa. t@s ‘get near, approach’ t@t@s ‘getting near, approaching’b. k’w@ì ‘spill over, tip over’ k’w@k’w@ì ‘spilling, tipping over’c. T@xw ‘disappear’ T@T@xw ‘disappearing’d. X@ì ‘get hurt’ X@X@ì ‘getting hurt’e. s@XwE ‘urinate’ s@s@XwE ‘urinating’

Although there is an apparently obvious phonological generalization about C@Cstems taking a reduplicative continuative form with stress on the second syllable,there is still good reason to believe that these stems are morphologically markedfor a specific continuative formation. The crucial observation is that thosestems show unpredictable continuative formations if one of the control suffixes–(@)t ‘do purposely to so.’ or –(@)m ‘intransitive’ attaches. Those controlsuffixes or transitivizer suffixes are obligatory for all transitive verbs in UpriverHalkomelem. As soon as a control suffix attaches to a (intransitive) C@C stem,they fall into two or even four subgroups, some showing ablaut, some metathesis,some both (Galloway, 1993, 274ff). It is therefore clear that the stems must belexically marked for their continuative formation in the context of a controlsuffix – it would then be not surprising at all that they show an unexpectedcontinuative formation in all contexts. This is exactly what I assume here:those stems do not form their continuative through foot affixation.

A final class of exceptional continuative forms are those where the contextgeneralizations given above predict a wrong continuative form. One exam-ple would be já(;)T@t ‘back oneself up’ (G60). Given the context restrictionsdiscussed above, we expect that the continuative is formed via reduplication(*já(;)j@T@t), quite parallel to the pair mát’@s + mám@t’@s. However, the con-tinuative form shows hE epenthesis and unexpected deletion of the first fullvowel: hEjT@t.

Given these various classes of exceptions, Galloway (1993) therefore con-cludes several times:

“The picture that emerges is a complex interplay of ablaut, reduplica-tion, stress shift, metathesis, and prefixing/infixing/suffixing used toexpress a number of verbal aspects, and nominal and verbal diminu-tives and plurals. Which type or combination of types processesor affixes is used is predictable, to a much greater degree than Ifirst thought, from the root and stem shapes. However, for manyof these conditioning shapes there are subgroups taking diver-gent processes, affixes, or combinations thereof, and at presentthose must be listed, since they seem morphologically de-termined.” (Galloway, 1993, 65; emphasis my own)

I therefore do not claim that the list in (19) is an exhaustive picture of thecontinuative formation in Upriver Halkomelem, but that it captures the class of

36

stems that form their continuative regularly. This is the largest class of stems,predictable “to a much greater degree” (Galloway, 1993, 65) than is apparent.Undoubtedly there are exceptional formations of the continuative that must bestored separately in the lexicon. A straightforward implementation of the dif-ferent factors governing the alternations between forms that are predictable byrules in a broad sense and those that are lexically stored is the assumption ofa dual-route model; cf. Bermúdez-Otero (2012) for a detailed discussion andliterature. For Upriver Halkomelem, this means that the grammar realizes thecontinuative by the empty prosodic foot affix and that other “patterns of expo-nence such as strong verb ablaut [. . . ] are encoded implicitly in the connectionsbetween nodes in a distributive associative memory” (Bermúdez-Otero, 2012,17). This is highly reminiscent of the famous example of English past tenseformation where no one would doubt the fact that the suffix –d marks the pastregularly (and is due to phonologically triggered allomorphy) and that otherirregular – ablaut – forms exist as well.

6. Conclusion

The non-concatenative allomorphy in the continuative formation of verbs inUpriver Halkomelem follows in an analysis where an empty prosodic foot is pre-fixed and must be filled with segmental material. Such an account attributes thedifferent non-concatenative strategies to form the continuative aspect in UpriverHalkomelem to one single abstract underlying representation that is indepen-dently motivated. In contrast to Realize Morpheme-based analyses like theone proposed in Kurisu (2001), such an analysis that adopts the hypothesis ofGeneralized Nonlinear Affixation does not suffer from an overgeneration prob-lem and predicts only the attested patterns of non-concatenative allomorphy inthe languages of the world. The proposal extends the typology of possible mor-phemes consisting only of a prosodic node and strengthens Oostendorp‘s (2012)claim that not only moras but also feet can serve as underlying representationof morphemes.

ReferencesAkinlabi, A., 1996. Featural affixation. Journal of Linguistics 32, 239–289.Alderete, J., 1995. Faithfulness to prosodic heads. In: Hermans, B., van Oostendorp,

M. (Eds.), The Derivational Residue in Phonological Optimality Theory. John Ben-jamins, Amsterdam, pp. 29–50.

Alderete, J., 2001a. Dominance effects as transderivational anti-faithfulness. Phonol-ogy 18, 201–253.

Alderete, J., 2001b. Morphologically governed accent in Optimality Theory. Outstand-ing dissertations in Linguistics, Routledge, New York.

Bermúdez-Otero, R., 2001. Underlyingly nonmoraic coda consonants, faithfulness, andsympathy. Ms. University of Manchester, online available at http://www.bermudez-otero.com/DEP-mora.pdf.

Bermúdez-Otero, R., 2007. Morphological structure and phonological domains inSpanish denominal derivation. In: Colins, S., Martínez-Gil, F. (Eds.), Optimality-theoretic studies in Spanish phonology. John Benjamins, Amsterdam, pp. 278–311.

37

Bermúdez-Otero, R., 2012. The architecture of grammar and the division of labour inexponence. In: Trommer, J. (Ed.), The morphology and phonology of exponence:The state of the art. Oxford University Press, Oxford, pp. 8–83.

Bermúdez-Otero, R., in preparation. Stratal Optimality Theory. Oxford UniversityPress, Oxford.

Bickel, B., 1998. Rhythm and feet in Belhare morphology. Ms. University of California,Berkeley, available online at ROA 287.

Brown, J., 2004. Some tonogenetic properties of Upriver Halkomelem. In: Harper, L.,Jany, C. (Eds.), Proceedings of the 7th Annual Workshop on American IndigenousLanguages. Santa Barbara Papers in Linguistics, vol.15, Santa Barbara, CA, pp.40–48.

Bye, P., Svenonius, P., 2012. Non-concatenative morphology as epiphenomenon. In:Trommer, J. (Ed.), The morphology and phonology of exponence: The state of theart. Oxford University Press, Oxford, pp. 426–495.

Crowhurst, M., 1994. Foot extrametricality and template mapping in Cupeño. NaturalLanguage and Linguistic Theory 12, 177–201.

Czaykowska-Higgins, E., Kinkade, M., 1998. Salish languages and linguistics. In:Czaykowski-Higgins, E., Kinkade, M. (Eds.), Salish languages and linguistics: the-oretical and descriptive perspectives. de Gruyter, Berlin, New York, pp. 1–68.

Davis, S., Ueda, I., 2002. The typology of mora augmentation. Proceedings of LP 2005.Davis, S., Ueda, I., 2006. Prosodic vs. morphological mora augmentation. Lexicon

Forum 2, 121–143.de Lacy, P., 2006. Markedness: reduction and preservation in phonology. Cambridge

University Press, Cambridge.Dyck, R. A., 2004. Prosodic and morphological factors in Squamish (Skwxwú7mesh)

stress assignment. Ph.D. thesis, University of Victoria.Elfner, E., 2006. Contrastive syllabification in Blackfoot. In: Baumer, D., Montero,

D., Scanlon, M. (Eds.), WCCFL 25. Cascadilla Proceedings Project, Somerville,MA, pp. 141–149.

Elmendorf, W., Suttles, W., 1986. Pattern and change in Halkomelem Salish. Anthro-pological Linguistics 2, 1–35.

Finley, S., 2009. Morphemic harmony as featural correspondence. Lingua 119, 478–501.Fitzpatrick, J., to appear. A concatenative theory of possible affix types. Andrés

Salanova (ed.) Papers from EVELIN I, MIT Working Papers in Linguistics, on-line available as lingbuzz/000662.

Galloway, B., 1993. A Grammar of Upriver Halkomelem. University of California Press,Berkeley.

Gouskova, M., Roon, K., 2008. Interface constraints and frequency in Russian com-pound stress. In: Reich, J., Babyonyshev, M., Kavitskaya, D. (Eds.), Proceedingsof the 17th Meeting of Formal Approaches to Slavic Linguistics. Michigan SlavicPublications, Ann Arbor, MI, pp. 49–63.

Grimes, S., 2002a. Mora augmentation in the Alabama imperfective: an op-timality theoretic perspective. ms. Indiana University, online available athttp://pweb.ldc.upenn.edu/∼sgrimes/papers/alabama.pdf.

Hansson, G. O., 2001. Theoretical and typological issues in consonant Harmony. Ph.D.thesis, University of California, Berkeley.

Harley, H., Leyva, M. F., 2009. Form and meaning in Hiaki (Yaqui) verbal reduplica-tion. International Journal of American Linguistics 75, 235–74.

Haugen, J., 2008. Morphology at the interfaces. Reduplication and noun incorporationin Uto-Aztecan. John Benjamin.

Haugen, J., Kennard, C. H., 2008. Morphological moras and morphological doublingtheory. Paper, given at the LSA Annual Meeting, San Francisco.

38

Hayes, B., 1995. Metrical Stress Theory: principles and case studies. University ofChicago Press, Chicago.

Hukari, T., 1977. Resonant devoicing in Cowichan. Canadian Journal of Linguistics22, 46–61.

Inkelas, S., 1998. The theoretical status of morphologically conditioned phonology: acase study from dominance. In: Yearbook of Morphology 1997. Kluwer, Dordrecht,pp. 121–155.

Inkelas, S., 1999. Exceptional stress-attracting suffixes in Turkish: representation vs.the grammar. In: Kager, R., van der Hulst, H., Zonneveld, W. (Eds.), The Prosody-Morphology Interface. Cambridge University Press, Cambridge, pp. 134–187.

Kager, R., 1999. Optimality Theory. Cambridge University Press, Cambridge.Kenstowicz, M., 2005. Paradigmatic uniformity and contrast. In: Downing, L. J., Hall,

T. A., Raffelsiefen, R. (Eds.), Paradigms in Phonological Theory. Oxford UniversityPress, Oxford, pp. 145–169.

Kurisu, K., 2001. The phonology of morpheme realization. Ph.D. thesis, University ofCalifornia at Santa Cruz, available as ROA 490-0102.

Landman, M., 2002. Morphological contiguity. In: Carpenter, A., Coetzee, A., de Lacy,P. (Eds.), Papers in Optimality Theory II: University of Massachusetts-AmherstOccasional Papers in Linguistics. GLSA, Amherst, MA.

Leonard, J., Turner, C. K., 2010. Predicting the shape of Senćoten imperfectives. In: AFestschrift for Thomas M. Hess on the Occasion of his 70th Birthday. WA: WhatcomMuseum Publication, pp. 82–112.

Lieber, R., 1992. Deconstructing Morphology. Chicago: University of Chicago Press.Lombardi, L., 2002. Coronal epenthesis and markedness. Phonology 19, 219–251.Lupardus, K. J., 1982. The language of the Alabama indians. Ph.D. thesis, University

of Kansas.McCarthy, J., 2008. Doing Optimality Theory. Blackwell, Malden MA.McCarthy, J., Prince, A., 1986/1996. Prosodic morphology 1986. Technical Report

32, Rutgers University Center for Cognitive Science, 1996. Available online at:http://works.bepress.com/john_j_mccarthy/.

McCarthy, J., Prince, A., 1995. Faithfulness and reduplicative identity. In: Beckman,J., Dickey, L., Urbanczyk, S. (Eds.), University of Massachusetts Occasional Papersin Linguistics. GLSA, Amherst, MA, pp. 249–384.

Molina, F., 1999. Hippocrene standard dictionary Yoeme-English, English-Yoeme.Hippocrene books.

Montler, T., Hardy, H., 1988. Imperfective gemination in Alabama. International Jour-nal of American Linguistics 54, 399–415.

Ní Chiosáin, M., 1991. Topics in the phonology of Irish. Ph.D. thesis, University ofMassachusetts, Amherst.

Prince, A., 1990. Quantitative consequences of rhythmic organization. In: Ziolkowski,M., Noske, M., Deaton, K. (Eds.), Parasession on the Syllable in Phonetics andPhonology. Chicago Linguistic Society, Chicago, pp. 355–398.

Prince, A., Smolensky, P., 1993. Optimality theory: Constraint interaction in gen-erative grammar. Technical reports of the Rutgers University Center of CognitiveScience.

Revithiadou, A., 1999. Headmost accent wins: Head dominance and ideal prosodicform in lexical accent systems. Ph.D. thesis, LOT Dissertation Series 15 (HIL/LeidenUniversiteit), Holland Academic Graphics, The Hague.

Revithiadou, A., 2007. Colored turbid accents and containment: A case study fromlexical stress. In: Blaho, S., Bye, P., Krämer, M. (Eds.), Freedom of Analysis?Mouton De Gruyter, Berlin, New York, pp. 149–174.

Rice, K., 1993. A Reexamination of the Feature [sonorant]: The Status of Sonorant

39

Obstruents. Language 69, 308–344.Rose, S., Walker, R., 2004. A typology of consonant agreement as correspondence.

Language 80 (3), 475–532.Saba Kirchner, J., 2010. Minimal reduplication. Ph.D. thesis, University of California

at Santa Cruz, available as ROA 1078-0610.Saba Kirchner, J., to appear. Reduplicative exponence and minimal reduplication.

In: Trommer, J. (Ed.), New theoretical tools in the modeling of morphologicalexponence. Special issue of Morphology.

Samek-Lodovici, V., 1992. A unified analysis of crosslinguistic morphological gemi-nation. In: Ackema, P., Schoorlemmer, M. (Eds.), Proceedings of CONSOLE 1.Holland Academic Graphics, The Hague, Utrecht, pp. 265–283.

Seiler, G., 2008. How to do things with moras: variation and change of quantityalternations across Upper German dialects. Paper presented at the InternationalMorphology Meeting, Vienna.

Selkirk, E., 1986. On derived domains in sentence phonology. Phonology 3, 371–405.Shaw, P., 2004. Reduplicant order and identity: Never trust a Salish CVC either? In:

Gerdts, D. B., Matthewson, L. (Eds.), Studies in Salish linguistics in honor of M.Dale Kinkade. University of Montana Working Papers in Linguistics 17, Missoula,pp. 329–354.

Shaw, P., Blake, S., Campbell, J., Shepard, C., 1999. Stress in H@’n’q@mi’n@’m. In:WSCLA 4. UBCWPL, Vancouver, pp. 131–163.

Smith, J., 2002. Phonological augmentation in prominent positions. Ph.D. thesis, Uni-versity of Massachusetts, Amherst.

Stonham, J., 1994. Combinatorial morphology. John Benjamin, Amsterdam.Stonham, J., 2007. Metathesis as prosodic repair. Studies of Phonetics, Phonology and

Morphology, 13, 3–24.Suttles, W., 2004. Musqueam Reference Grammar. UBC Press, Vancouver.Thompson, J., 2005. Vowel duration in Upriver Halkomelem. In: Brown, J., Kiyota,

M., Peterson, T. (Eds.), Papers for ICSNL 40. UBCWPL, Vancouver, pp. 225–233.Thompson, L., Thompson, M. T., 1971. Clallam: a preview. Studies in American

Indian Languages 65, 251–294.Trommer, J., 2009. Chain-shifting mutation in Irish and multi-valued features, paper

presented at OCP 6, Edinburgh, January 2009.Trommer, J., 2010. Paradigmatic generalization of morphemes. Linguistische Arbeits

Berichte Leipzig 88, 227–246.Trommer, J., 2011. Phonological aspects of Western Nilotic mutation morphology.

Habil. University of Leipzig.Trommer, J., Zimmermann, E., 2010. Generalized mora affixation. talk given at the

18th Manchester Morphology Meeting, Manchester, 20th-22th May 2010.Turner, C. K., 2011. Representing events in Saanich (Northern Straits Salish): the

interaction of aspect and valence. Ph.D. thesis, University of Surrey.Urbanczyk, S., 1998. A-templatic reduplication in Halq‘eméylem. In: Shahin, K.,

Blake, S., Kim, E. (Eds.), WCCFL 17. CSLI Publications, Stanford, CA, pp. 655–669.

Urbanczyk, S., 1999. Echo vowels in Coast Salish. In: Caldicott, M., Gessner, S., Kim,E. (Eds.), WSCLA 4. UBCWPL, Vancouver, pp. 165–184.

Urbanczyk, S., 2011. Root-affix asymmetries. In: van Oostendorp, M., Ewen, C. J.,Hume, E., Rice, K. (Eds.), The Blackwell Companion to Phonology. Wiley Black-well, Malden MA, Ch. 104.

van Oostendorp, M., 1994. Affixation and integrity of syllable structure in Dutch.In: Bok-Bennema, R., Cremers, C. (Eds.), Linguistics in the Netherlands. JohnBenjamins, Amsterdam, pp. 151–162.

40

van Oostendorp, M., 2003. Comparative markedness and containment. TheoreticalLinguistics 29, 65–75.

van Oostendorp, M., 2004. Crossing morpheme boundaries in Dutch. Available onlineat ROA 655 0404.

van Oostendorp, M., 2006. A theory of morphosyntactic colours.Ms., Meertens Institute, Amsterdam, available online athttp://egg.auf.net/06/docs/Hdt%20Oostendorp%20coulours.pdf.

van Oostendorp, M., 2007a. Derived environment effects and consistency of expo-nence. In: Blaho, S., Bye, P., Krämer, M. (Eds.), Freedom of Analysis? Mouton deGruyter, Berlin, pp. 123–148.

van Oostendorp, M., 2007b. Stress as a prefix in Modern Greek. Talk given at OCP 4,online available at http://www.vanoostendorp.nl/pdf/moderngreek_ocp04.pdf.

van Oostendorp, M., 2008. Incomplete devoicing in formal phonology. Lingua 118,1362–1374.

van Oostendorp, M., 2012. Stress as a proclitic in Modern Greek. Lingua 122, 1165–1181.

Walker, R., 2000a. Long-distance consonantal identity effects. In: Billerey, R., Lille-haugen, B. (Eds.), WCCFL 19. Cascadilla Press, Somerville, MA, pp. 532–545.

Walker, R., 2000b. Nasal reduplication in Mbe affixation. Phonology 17, 65–115.Walker, R., 2000c. Yaka nasal harmony: spreading or segmental correspondence? In:

Conathan, L., Good, J., Kavitskaya, D., Wulf, A., Yu, A. (Eds.), BLS 26. BerkeleyLinguistic Society, Berkeley, pp. 321–332.

Wolf, M., 2007. For an autosegmental theory of mutation. In: Bateman, L., O’Keefe,M., Reilly, E., , Werle, A. (Eds.), University of Massachusetts Occasional Papers inLinguistics 32: Papers in Optimality Theory III. GLSA, Amherst, MA, pp. 315–404.

Yip, M., 1988. The Obligatory Contour Principle and phonological rules: A loss ofidentity. Linguistic Inquiry 19, 65–100.

Yip, M., 1992. Reduplication with fixed melodic material. In: NELS 22. GLSA Pub-lications, Amherst, MA, pp. 459–476.

Yip, M., 1998. Identity avoidance in phonology and morphology. In: Lapointe, S. G.,Brentari, D. K., Farrell, P. M. (Eds.), Morphology and its relation to morphologyand syntax. CSLI Publications, Stanford, CA, pp. 216–247.

Yu, A. C. L., 2003. The morphology and phonology of infixation. Ph.D. thesis, UCBerkeley.

Zimmermann, E., 2009. Metathesis without reordering. Master’s thesis, University ofLeipzig.

41