Serial Reduplication Is Empirically Adequate and

Typologically Restrictive

Andrew Lamont

Linguistic Inquiry, Volume 54, Number 4, Fall 2023, pp. 797-839 (Article)

Published by The MIT Press

For additional information about this article

https://muse.jhu.edu/article/907959

[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

 Serial Reduplication Is
Empirically Adequate and
Typologically Restrictive
Andrew Lamont
Wei and Walker (2020) and Zymet (2018) claim that derivational
lookahead effects are attested in the interactions between reduplication
and other phonological processes in Mbe and Logoori, respectively.
On the basis of this evidence, they argue that reduplication in these
languages cannot be modeled by Serial Template Satisfaction (McCar-
thy, Kimper, and Mullin 2012), a theory of reduplication set in Har-
monic Serialism. This article refutes these claims and provides serial
analyses for both languages. It further identifies a novel prediction of
Base-Reduplicant Correspondence Theory (McCarthy and Prince
1994, 1995, 1999), a parallel theory of reduplication, that reduplicants
may surface with marked structures unattested elsewhere in the lan-
guage, and it demonstrates that these patterns are not replicated in
serial.
Keywords: derivational lookahead, reduplication, Mbe, Logoori, Har-
monic Serialism, Serial Template Satisfaction, Base-Reduplicant Cor-
respondence Theory, the emergence of the marked
[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

1 Introduction
In parallel Optimality Theory (pOT; Prince and Smolensky 1993), GEN produces candidates that
may be arbitrarily different from the input. In Harmonic Serialism (HS; Prince and Smolensky
1993, McCarthy 2000, 2006, 2008b, 2016), GEN only produces candidates that differ from the in-
put by at most one operation: no operation may be applied more than once to produce a candidate,
and no more than one operation may be applied to produce a candidate. Consequently, pOT can
evaluate interactions between operations that HS cannot. To give a schematic example, suppose
pOT maps an input /A/ onto an output [C] by making two changes. In HS, because [C] is not
in the candidate set generated from /A/, the derivation must pass through an intermediate step:
/A/ N B N C N [C]. However, if it is not optimal to map /A/ onto B in the first step, the
derivation may converge on an output other than [C]: /A/ N X N Y N [Y]. To favor mapping
/A/ onto B in anticipation of [C], the grammar would have to look ahead to future derivational
steps.

This article has greatly benefited from three anonymous reviewers for LI and discussions with Michael Becker,
Hossep Dolatian, Kuo-Chiao Lin, Dave Odden, Joe Pater, Brandon Prickett, and Sam Zukoff. All remaining errors are
of course my own.
Linguistic Inquiry, Volume 54, Number 4, Fall 2023
797–839
䉷 2021 by the Massachusetts Institute of Technology
https:/ /doi.org/10.1162/ling_a_00452 797
798 ANDREW LAMONT

Proponents of HS argue that its restrictiveness correctly expunges unattested mappings pre-
dicted by pOT (see McCarthy 2006, 2008b, 2010, 2016 for overviews of these results). For ex-
ample, outside of morphologically restricted environments, metathesis only targets adjacent seg-
ments (Buckley 2011). However, pOT predicts nonlocal metathesis as a possible repair of locally
defined constraints (Heinz 2005, McCarthy 2006, 2008b). The tableau in (1) illustrates nonlo-
cal metathesis motivated by the constraint *ORALCODA, which penalizes nonnasal segments in co-
da position (Walker 2000). This mapping is not possible in HS, as the candidates generated from
/ant/ by making one transposition, [nat] and [atn], do not improve on *ORALCODA (McCarthy
2006, 2008b; see Takahashi 2018, 2019 for an approach to metathesis in HS). For the grammar
to favor [atn] in anticipation of mapping it onto [tan], it would have to look ahead to the next
derivational step. However, because HS does not look ahead, operations only apply when their
application is locally optimal.
(1) Nonlocal metathesis in pOT (unattested, McCarthy 2006, 2008b)
/ant/ *ORALCODA LINEARITY

a. ant W1 L
b. nat W1 L1
c. atn W1 L1
d. tan 2

There is a growing body of literature arguing that there are natural language phenomena that
cannot be modeled in HS because they would require derivational lookahead (Adler 2017, Adler
and Zymet 2017, 2021, Wei 2018, Wei and Walker 2018, 2020, Zymet 2018, Stanton 2020).
These works draw on evidence from a diverse set of segmental processes, prosodification, and
reduplication to argue that pOT is a more viable model of phonology than HS. The project of
critically evaluating claims of derivational lookahead is necessarily beyond the scope of any one
work and must be broken down by empirical domain. The present article focuses on reduplication
and concludes that there is no evidence to support derivational lookahead.
Wei and Walker (2020) and Zymet (2018) argue that phonotactic restrictions in Mbe and
Logoori, respectively, cannot be satisfied unless reduplication occurs simultaneously with other
operations. Both works present analyses in Base-Reduplicant Correspondence Theory (BRCT;
McCarthy and Prince 1994, 1995, 1999), a theory of reduplication set within pOT, and argue that
Serial Template Satisfaction (STS; McCarthy, Kimper, and Mullin 2012), a theory of reduplication
set in HS, cannot model the data without derivational lookahead.
The tableaux in (2) illustrate an unattested mapping that requires derivational lookahead (see
Inkelas and Zoll 2005 for similar examples); constraints are based on Lombardi 1999, 2001. A
root / talab/ is the input to the first tableau (2a–c), and it takes a reduplicative prefix / RED-talab/
in the second (2d–g). In isolation, the root-final /b/ devoices, improving on *VOICEDOBSTRUENT
(2b). Devoicing the initial / t / would fatally violate IDENTONSET(voice)-IO (2c). When the root
takes a reduplicant, however, the final /b/ remains voiced (2d). Because it corresponds to the
 SERIAL REDUPLICATION 799

reduplicant-final [b], final devoicing would violate IDENT(voice)-BR (2e). The reduplicant [b]
cannot devoice without violating AGREE(voice) (2f ) or IDENTONSET(voice)-IO (2g). Thus, final
devoicing is blocked when roots with initial voiced obstruents undergo reduplication. In all other
contexts, final obstruents devoice.
(2) IDENT(voice)-BR blocks final devoicing (unattested)
IDENT AGREE IDONSET *VOICED IDENT
/talab/, /RED-talab/
(voi)-BR (voice) (voi)-IO OBS (voi)-IO

a. talab W2 L
b. talap 1 1
c. kalap W1 L W2

d. talab-talab 4
e. talab-talap W1 L3 W1
f. talap-talap W1 L2 W2
g. kalap-kalap W1 L W2

It is not surprising that faithfulness constraints evaluated along the base-reduplicant (BR)
correspondence dimension should produce nonlocal effects such as that in (2). After all, correspon-
dence relations hold between segments that are arbitrarily far apart. HS cannot reproduce this
mapping without an analogous nonlocal constraint or derivational lookahead. If copying and
devoicing must occur in separate steps, then there is no ordering that produces the surface form
[talab-talab] (3). If copying occurs first (3a), then nonlocal blocking must be incorporated into
the definition of final devoicing; otherwise, it cannot access the necessary information. If devoicing
occurs first (3b), then the grammar must be equipped with derivational lookahead: devoicing
applies unless it would create a disagreeing cluster later in the derivation.
(3) a. Underlying representation / RED-talab/
Copy root talab-talab
Final devoicing talab-talap
Surface representation *[talab-talap]
b. Underlying representation / RED-talab/
Final devoicing RED-talap
Copy root talap-talap
Surface representation *[talap-talap]
In a serial derivation, modeling mappings like (2) requires accessing information in the string
that may be arbitrarily far away, or accessing information in the derivation that may be arbitrarily
many steps in the future. Unless HS is equipped with these capacities, it cannot produce these
mappings. This is the correct prediction, as interactions like (2) are unattested. Wei and Walker
800 ANDREW LAMONT

(2020) and Zymet (2018) argue for lookahead effects in Mbe and Logoori, respectively. Their
claims of derivational lookahead are refuted by the lookahead-free STS analyses in sections 3 and
4. (STS is introduced in section 2.) Section 5 builds on (2), demonstrating unattested emergence-of-
the-marked effects in BRCT: structures surface in reduplicants despite being repaired elsewhere
in the language. It further argues that these predictions are unique to BRCT and that STS cannot
produce the emergence-of-the-marked effects.

2 Serial Template Satisfaction

As noted above, STS is a theory of reduplication set in HS. This section provides an overview
of STS; for illustrations, see Somerday 2015, Lin 2016, Anderson and Smith 2017, 2018, and
Lamont 2021.
STS treats reduplicative morphemes as underlying prosodic templates, such as feet, syllables,
and moras, without internal structure or segmental content. By lacking constituents, these tem-
plates violate constraints requiring prosodic heads (Selkirk 1995). Headedness constraints, short-
ened to HD(X), require prosodic nodes to contain nodes one level lower on the prosodic hierarchy
(4). Specifically relevant to this article, HD(F) penalizes feet that do not dominate syllables, HD( ␴ )
penalizes syllables that do not dominate moras, and HD( ␮ ) penalizes moras that do not dominate
segments.
(4) HD(X)
Assign one violation mark for every constituent of type X that does not contain a
constituent of type X ⳮ 1 as its head. (McCarthy, Kimper, and Mullin 2012:181)
GEN has two operations that install a constituent of type X ⳮ 1 in a constituent of type X.
One operation inserts an empty prosodic constituent. This trades a violation of HD(X) for one of
HD(X ⳮ 1), and, following McCarthy, Kimper, and Mullin (2012:180), does not violate any
faithfulness constraints. The other operation copies a contiguous string of constituents and inserts
the copy into the host constituent. Copying any length string of constituents of type X violates
the faithfulness constraint *COPY(X). Copying can target segments, prosodic constituents, or
morphological constituents; an illustration is given in (5). The input to GEN is /bata/, which is
syllabified and has been parsed into a bisyllabic foot. From this input, GEN can copy any contiguous
string of segments (5a), syllables (5b), or feet (5c). GEN cannot copy discontiguous elements like
the string [bt], at least not in one step.
 SERIAL REDUPLICATION 801

F
␴␴
␮␮
(5) Strings GEN can copy from input bata
a. b, a, t, ba, at, ta, bat, ata, bata
␴ ␴ ␴␴
␮ ␮ ␮␮
b. ba, ta, bata
F
␴␴
␮␮
c. bata
Copying cooccurs with the copy being incorporated into existing prosodic structure. If this
were broken into two steps, copying would not be harmonically improving: applying the copying
operation violates *COPY(X) and would fail to improve on HD(X). As an illustration, consider
the example in (6), where syllable strings are copied into a foot template. Excluded from (6) are
candidates where strings of segments or moras are copied into the foot template. Those candidates
do not improve on HD(F), which penalizes feet that do not dominate syllables. The input to (6)
contains three syllables, and any string of one (6a–c), two (6d–e), or three (6f ) syllables can be
copied into the foot template. In principle, arbitrarily many syllables can be copied into a foot
template in one step, with markedness constraints determining the optimal amount of material.
In this respect, copying is different from other structure-building operations available to GEN,
which are limited in how much material they parse in one step; for example, theories of footing
in HS allow at most two syllables to be footed in a single step (McCarthy 2008c, Pruitt 2010,
2012, Lamont to appear).
802 ANDREW LAMONT

F
⫹ ␴␴␴
␮␮␮
(6) Candidates produced by copying strings of syllables from input bataka
F F F
␴ ⫹ ␴␴␴ ␴ ⫹ ␴␴␴ ␴ ⫹ ␴␴␴
␮ ␮␮␮ ␮ ␮␮␮ ␮ ␮␮␮
a. ba bataka b. ta bataka c. ka bataka
F F F
␴␴ ⫹ ␴␴␴ ␴␴ ⫹ ␴␴␴ ␴␴␴ ⫹ ␴␴␴
␮␮ ␮␮␮ ␮␮ ␮␮␮ ␮␮␮ ␮␮␮
d. bata bataka e. taka bataka f. bataka bataka
The six candidates in (6) incur one violation of *COPY( ␴ ). Additionally, there is a pressure
for copies to be adjacent to their sources, formalized as the faithfulness constraint *COPY-LOCAL-
LY(X). This constraint assigns as many violations as there are intervening constituents of type X
between a copy and its source. Candidates (6b,e) each incur one violation of *COPY-LOCALLY( ␴ ),
and (6c) incurs two violations.

3 Reduplication in Mbe Does Not Require Lookahead

Reduplication expones verbal imperatives and nominal diminutives in Mbe (Benue-Congo; Bam-
gbo⋅ se 1966, 1967a,b,c, 1971, Walker 1998a,b, 2000). The former surfaces as a syllable-sized pre-
fix (7a–c), and the latter surfaces as a coda on the Class 4 marker /k[-/ (7d–f ). Reduplicant
codas are limited to nasal consonants and must be homorganic to following onsets. Thus, verbs
without nasal consonants take a CV imperative (7a) and verbs with nasal consonants take a CVN
imperative (7b–c). Nouns without nasal consonants do not overtly realize the diminutive (7d); it
surfaces as a nasal coda only with nouns that contain a nasal consonant (7e–f ).
(7) Reduplication in the Mbe imperative and diminutive
a. sÛ-só.rò ‘descend-IMP’
b. tb  Û√m-tb  é.nò ‘collide-IMP’
c. jı̂«-jı̂:.nı̀ ‘forget-IMP’
d. kà--bà.rò ‘liver-DIM’
e. k[ˇ -n-nên ‘bird-DIM’
f. k[ˇ -n-t:̀.nı́ ‘earthworm-DIM’
This phonotactic restriction applies only to reduplicants (Walker 2000:95); roots take a wider
range of coda consonants word-finally (8) and at suffix boundaries (9).
 SERIAL REDUPLICATION 803

(8) Word-final codas in Mbe (Bamgbo⋅ se 1967a:7)

a. tém ‘bury’
b. tén ‘push’
c. tá√ ‘give in marriage’
d. tb ál ‘run’
e. tb ár ‘plaster’
f. táb ‘accompany’
(9) Codas at root-suffix junctures in Mbe (Bamgbo⋅ se 1967c:176, 181, 182)
a. jı̌[m-kı̀ ‘be singing’
b. lú:m-nı̂ ‘bite all over’
c. fùel-kı̂ ‘be blowing’
d. ts:̌r-kı̀ ‘be carrying’
e. jùab-kı̂ ‘be washing’
Wei and Walker (2018, 2020; see also Wei 2018) argue that derivational lookahead is required
to model the imperative, echoing an argument by McCarthy, Kimper, and Mullin (2012:212–213).
Given a CVC(V) stem, the decision to copy an open syllable CV-CVC(V) or a closed syllable
CVC-CVC(V) depends on whether the copied coda will later undergo place assimilation and be
licensed. Because copying cannot cooccur with assimilation, the grammar must look ahead to
avoid unlicensed codas. Wei and Walker further argue that stems with initial diphthongs pose
problems for analyzing the imperative as a foot, because there is no prosodic motivation to copy
a coda. Given a CVVN(V) stem, copying an open syllable CVV-CVVN(V) satisfies FTBIN, and
copying a closed syllable CVVN-CVVN(V) is unmotivated.
This section presents an analysis of Mbe reduplication in STS that refutes the claim of
lookahead.1 The imperative is analyzed as a foot template, which prefers that reduplicants surface
with codas. This is overridden by a phonotactic constraint banning nonnasal codas, deriving the
difference between (7a) and (7b). Roots with initial diphthongs, such as (7c), initially provide no
motivation to copy the nasal. However, copied diphthongs are obligatorily reduced, creating a
headless mora in the reduplicant, which motivates copying a nasal. The same mechanism derives
the diminutive, which is analyzed as an underlying mora.

3.1 The Mbe Imperative as a Foot Template

There are two verb classes in Mbe, Class I and Class II, which are distinguished by their conjuga-
tion and morphophonology (Bamgbo⋅ se 1967c). Both classes restrict reduplicant codas to nasals
homorganic to following onsets and impose class-specific restrictions on reduplicant vowels.

1
Lin (2021) has independently developed an alternative STS analysis of Mbe.
 804 ANDREW LAMONT

Class I verbs are all monosyllabic (Bamgbo⋅ se 1967c); examples are given in (10). The
reduplicant surfaces as an open syllable with open-syllable roots (10a,d) and roots closed by oral
consonants (10b,e). With roots that end in nasals, the reduplicant surfaces with a nasal coda
homorganic to the following consonant (10c,f ).
(10) Mbe Class I imperatives (Bamgbo⋅ se 1967c:183, 184)
a. bú bǔ-bù ‘grow’
b. káb kǎ-kàb ‘dig’
c. lám lǎn-làm ‘cook’
d. +ú[ +ǔ[-+ù[ ‘suck’
e. +ı́:r +ı̌-+ı̀:r ⬃ +ı́:-+ı̀:r ‘sneeze’
f. tú:m tǔ:n-tù:m ‘send’
Class I imperatives all take a tonal melody consisting of a rising tone on the reduplicant
followed by a low tone on the root. Tonal assignment does not otherwise interact with reduplication
and is not analyzed below. Reduplicant diphthongs are subject to lexical restrictions in Class I.
The second element of a diphthong is lost in an open syllable if it is /e/ or /o/: for example, [fǔ-
fùo] ‘blow’ (cf. (10d)). In closed syllables, diphthongs’ second elements are optional, as in (10e),
except when the syllable is closed by a labial, where they surface (10f ).
Class II verbs are all monosyllabic or bisyllabic (Bamgbo⋅ se 1967c); examples are given in
(11). The reduplicant surfaces as an open syllable with monosyllabic roots that lack codas (11a,f ),
monosyllabic roots with oral codas (11b,g), and bisyllabic roots with medial oral onsets (11c,h).
The reduplicant surfaces with a nasal coda homorganic to the following consonant with monosyl-
[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

labic roots closed by nasals (11d,i) and with bisyllabic roots with medial nasal onsets (11e,j).
(11) Mbe Class II imperatives (Bamgbo⋅ se 1967c:185, 186)
a. rû rû-rû ‘pull’
b. mâl mÛ-mâl ‘finish’
c. só.rô sÛ-só.rò ‘descend’
d. tâ√ tÛn-tâ√ ‘teach’
e. tb é.nô tb  Û√m-tb  é.nò ‘collide’
f. t+ûe t+û-t+ûe ‘bore (hole)’
g. fûel fû-fûel ‘blow’
h. kú[.lô kû-kú[.lò ‘nibble at’
i. dzû:√ dzûn-dzû:√ ‘be higher’
j. jı́:.nı̂ jı̂«-jı̂:.nı̀ ‘forget’
Class II imperatives all take a falling tone on the reduplicant and a falling tone on the root.
As with Class I verbs, the tones are not analyzed below. Reduplicant vowels are subject to very
stringent restrictions in Class II. High vowels are copied faithfully (11a), and all other monoph-
thongs are reduced to [U] (11b–e). Diphthongs are all reduced to their initial element (11f–j),
which, because Mbe only has falling diphthongs (Bamgbo⋅ se 1967a), results in a simple high
vowel.
 SERIAL REDUPLICATION 805

Class II verbs provide evidence that the imperative is an underlying foot. The reduplicant
surfaces with a coda with CV(V)NV roots, and there must be some motivation for it, formalized
as FTBIN (12). In addition to their different tonal contours and restrictions on vowels, Classes I
and II differ in how they fill in the foot template (see Lamont 2021 for the same lexical split in
another language). In Class I, the root syllable is copied and then nonnasal codas are deleted. In
Class II, an empty syllable is inserted and then a string of segments is copied to head it.
(12) FTBIN
Assign one violation to every foot that dominates fewer than two syllables and fewer
than two moras.
In addition to the HD(X) and *COPY(X) constraints introduced in section 2, constraints are
needed to model the vowel alternations in Class II verbs. Because diphthong reduction in Class
I is optional and does not feed other phonological processes, it is not discussed in the analysis.
The constraints in (13) motivate reduction.
(13) *MID /LOW ⬎⬎ *SCHWA ⬎⬎ *HIGH
Vowel reduction is modeled by an operation that maps full vowels onto schwa. Applying
this operation violates the faithfulness constraint MAX(V-place). Following McCarthy (2019),
GEN can delete schwa but not full vowels, and so vowel deletion is gradual: for example, /a/ N
U N , /u/ N U N . The analysis diverges from McCarthy’s in that mid and low vowels map
directly onto schwa without first raising.
Similarly, following McCarthy (2007, 2008a, 2016), consonantal deletion and place assimila-
tion are analyzed as gradual processes, both of which begin by deleting a segment’s place features.
GEN can only delete placeless consonants and spread place features onto placeless consonants.
Place deletion violates the faithfulness constraint MAX(C-place), and the resulting placeless conso-
nant violates the markedness constraint HAVEPLACE. CODACOND motivates debuccalization (14),
feeding consonantal deletion and place assimilation. Note that, as defined, this constraint penalizes
superficially homorganic clusters, requiring assimilation: for example, an.ta N aN.ta N an.t a.
(14) CODACOND
Assign one violation to every consonantal place feature not associated to an onset.
(McCarthy 2008a:279)
The differences between Class I verbs and Class II verbs are modeled with lexically indexed
constraints (Pater 2007, 2010). For example, the ranking MAX(V-place)I ⬎⬎ *MID /LOW ⬎⬎
MAX(V-place) derives the fact that low vowels surface faithfully in Class I reduplicants (10b)
but reduce to schwa in Class II reduplicants (11b).
The analysis of the Mbe imperative is presented in detail in sections 3.1.1 and 3.1.2; the full
set of tableaux is available in the online appendix (https://doi.org/10.1162/ling_a_00452). The
diminutive is discussed in section 3.2. Because they are irrelevant to the overall analysis, differ-
ences between the reduplicant and the root are not discussed. They may plausibly be analyzed
as faithfulness to roots or affix-specific markedness. Accordingly, only violations incurred by
806 ANDREW LAMONT

reduplicants are shown in tableaux. Pairwise relations between constraints reflect the overall
ranking. A Hasse diagram for Mbe is given at the end of the section (see figure 1).
3.1.1 Class I Imperatives Copy Syllables The restrictions on reduplicant codas are enforced via
deletion in Class I. The root syllable is copied, and then any place features in the coda are deleted.
If the coda is an oral consonant, it is then deleted entirely, and if it is a nasal consonant, it is
place-linked to the following onset. Thus, only place-linked nasal codas surface.
3.1.1.1 Copied Oral Codas Are Deleted The tableau in (15) shows the copying step of the
derivation mapping /F-káb/ onto [kǎ-kàb] ‘dig’. The root is syllabified, which implies earlier
derivational steps, assuming gradual syllabification (Elfner 2009, Pater 2012, Torres-Tamarit
2012, 2016, Moore-Cantwell 2016). The foot template is unfilled in the faithful candidate (15a),
which fatally violates HD(F). Copying the root syllable into the template (15c) is preferred to
inserting an empty syllable (15b), because HD( ␴ ) dominates *COPY( ␴ ). Other logically possible
candidates do not improve on HD(F).
(15) /F-kábI/ [ka-kàb
ˇ I], Step 1 (Mbe)

F *ORALCODA

⫹␴

CODACOND
*MID/LOW
*COPY(␴)

␮␮
HD(␴)
HD(F)

FTBIN

kâb Class I

F
⫹␴
␮␮
a. kâb W1 L L L W1 L
F
␴⫹␴
␮␮
b. kâb W1 L L L W1 L
F
␴⫹␴
␮␮ ␮␮
c. kâb kâb 1 1 1 1
 SERIAL REDUPLICATION 807

In the next two steps of the derivation (16), the coda /b/ is deleted. First, its labial place
feature is deleted (16b); the exterior arrow indicates that this candidate is the input to the next
step (16d–f ). The diacritic distinguishes debuccalized stops derived from voiced obstruents, ≈,
ˇ
from those derived from voiceless obstruents, ≈. Copying a vowel into the reduplicant does not
improve on *ORALCODA or CODACOND, because the labial-vowel string cannot be resyllabified
as a CV syllable in one step. *MID /LOW cannot motivate vowel reduction because the lexically
indexed constraint MAX(V-place)I is dominant (16c). The placeless coda is then deleted in the
third step (16d–f ), satisfying HAVEPLACE and *ORALCODA (16e). Place assimilation also satisfies
HAVEPLACE, but fails to improve on *ORALCODA (16f ). Deleting the coda violates HD( ␮ ) because
the second reduplicant mora no longer dominates any segments.
808 ANDREW LAMONT

(16) /F-kábI/ ˇ
[ka-kàb I], Steps 2–3 (Mbe)

*LINK(place)
*CODACOND
*ORALCODA
MAX(V-Pl)I

HAVEPLACE
MAX(V-Pl)

MAX(C-Pl)
*MID/LOW
␴ ⫹␴

*SCHWA

HD(␮)
␮␮ ␮␮

MAX
kâb kâb Class I

F
␴ ⫹␴
␮␮ ␮␮
a. kâb kâb 1 1 W1 L L
F
␴ ⫹␴
␮␮ ␮␮
b. kâ≈´ kâb 1 1 1 1
F
␴ ⫹␴
␮␮ ␮␮
c. kUˆ b kâb W1 L W1 W1 1 W1 L L
F
␴ ⫹␴
␮␮ ␮␮
d. kâ≈´ kâb 1 W1 L W1 L
F
␴ ⫹␴
␮␮ ␮␮
e. kâ kâb 1 1 1
F
␴ ⫹␴
␮␮ ␮␮
f. kât kâb 1 W1 L W1 L

The headless mora is deleted in the fourth step (17e), as it cannot be retained without violating
higher-ranked constraints. Linking the mora to the reduplicant vowel fatally violates *LONGVOWEL
(17b), copying a stop to head the mora fatally violates *ORALCODA (17c), and copying a vowel
to head the mora introduces an additional violation of *MID /LOW (17d). The derivation converges
in the next step (17f–h). The faithful candidate violates both FTBIN and *MID /LOW (17f ), but
neither constraint can optimally motivate an operation to apply. Inserting an empty syllable (17g)
and copying a syllable into the reduplicant (17h) both create a binary foot, but violate constraints
that dominate FTBIN.
 SERIAL REDUPLICATION 809

(17) /F-kábI/ ˇ
[ka-kàb I], Step 4 and convergence (Mbe)

*ORALCODA

*COPY(seg)

CODACOND
*MID/LOW
␴ ⫹␴

*COPY(␴)

*LONGV

MAX(␮)
HD(␮)
␮␮ ␮␮

HD(␴)

FTBIN
kâ kâb Class I

F
␴ ⫹␴
␮␮ ␮␮
a. kâ kâb 1 W1 L L
F
␴ ⫹␴
␮␮ ␮␮
b. kâ kâb 1 W1 L L
F
␴ ⫹␴
␮␮ ␮␮
c. kâk kâb 1 W1 L W1 L W1
F
␴ ⫹␴
␮␮ ␮␮
d. kââ kâb W2 L W1 L
F
␴ ⫹␴
␮ ␮␮
e. kâ kâb 1 1 1
F
␴ ⫹␴
␮ ␮␮
f. kâ kâb 1 1
F
␴ ␴⫹␴
␮ ␮␮
g. kâ kâb W1 1 L
F
␴ ␴⫹␴
␮ ␮ ␮␮
h. kâ kâ kâb W1 W2 L
810 ANDREW LAMONT

3.1.1.2 Copied Nasal Codas Are Assimilated Nasal-final verbs like [tú:m] ‘send’ follow the
same first two steps, copying the root syllable and deleting the coda’s place features. In the third
step, the coda undergoes place assimilation instead of being deleted, as the tableau in (18) illus-
trates. Deletion violates HD( ␮ ) (18b) and, because the nasal does not violate *ORALCODA, is
dispreferred to assimilation (18c). As with [kǎ-kàb] *[kǓ-kàb] ‘dig’, MAX(V-place)I prevents the
reduction of the diphthong to [U:] or [uU]. The derivation converges in the next step on [tǔ:n-
tù:m].
(18) /F-tú:mI/ [tuˇ :n-tú:mI], Step 3 (Mbe)
F

*LINK(place)
␴ ⫹ ␴

HAVEPLACE
*MID/LOW

␮␮␮ ␮␮␮
HD(␮)

*HIGH

MAX
tú:N tú:m Class I

F
␴ ⫹ ␴
␮␮␮ ␮␮␮
a. tú:N tú:m 1 1 W1 L
F
␴ ⫹ ␴
␮␮␮ ␮␮␮
b. tú: tú:m 1 W1 1 L W1
F
␴ ⫹ ␴
␮␮␮ ␮␮␮
c. tú:n tú:m 1 1 1

3.1.2 Class II Imperatives Copy Segments Class II imperatives enforce the restrictions on re-
duplicant codas by only copying nasals into coda position. A nasal coda can be copied early in
the derivation, along with other segmental material copied to head the syllable. A nasal coda can
also be copied later in the derivation to head a mora whose constituent vowel was deleted. Once
copied, nasal codas are debuccalized and place-linked to the following segment.
3.1.2.1 Oral Consonants Are Not Copied as Codas In Class II, the foot template is headed by
inserting an empty syllable. The tableau in (19) illustrates this step of the derivation mapping
/F-só.rô/ onto [sÛ-só.rô] ‘descend’. As before, the syllabified root implies earlier derivational
 SERIAL REDUPLICATION 811

steps. The faithful candidate fatally violates HD(F) (19a) and loses to a candidate with an inserted
syllable (19b). Copying the root syllables into the foot template is preferred by HD( ␴ ) and FTBIN,
but is ruled out by the lexically indexed faithfulness constraint *COPY( ␴ )II (19c).
(19) /F-só.rô/ [sUˆ -só.rô], Step 1 (Mbe)
F
⫹ ␴␴

*COPY(␴)II

*MID/LOW
*COPY(␴)
␮␮

HD(␴)
HD(F)

FTBIN
sórô Class II

F
⫹ ␴␴
␮␮
a. sórô W1 L 1
F
␴ ⫹ ␴␴
␮␮
b. sórô 1 1
[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

F
␴␴ ⫹ ␴␴
␮␮ ␮␮
c. sórô sórô W1 L W1 W2 L

The rest of the derivation is illustrated in the tableaux in (20). A string of segments is copied
to head the syllable in the second step of the derivation (20b).
812 ANDREW LAMONT

(20) /F-só.rô/ [sUˆ -só.rô], Steps 2–4 (Mbe)

F

*ORALCODA

*COPY(seg)
MAX(V-Pl)

CODACOND
*MID/LOW
␴ ⫹ ␴␴

MAXHD␮

*SCHWA

HD(␮)
HD(␴)
␮␮

FTBIN

MAX
sórô Class II

F
␴ ⫹ ␴␴
␮␮
a. sórô W1 L L 1
F
␴ ⫹ ␴␴
␮ ␮␮
b. só sórô 1 1 1
F
␴ ⫹ ␴␴
␮␮ ␮ ␮
c. sór sórô 1 W1 1 L W1
F
␴ ⫹ ␴␴
␮ ␮␮
d. só sórô W1 L L 1
F
␴ ⫹ ␴␴
␮ ␮␮
e. sÚ sórô 1 1 1
F
␴ ⫹ ␴␴
␮ ␮␮
f. sÚ sórô 1 1
F
␴ ⫹ ␴␴
␮ ␮␮
g. s sórô W1 L W1 1 W1
 SERIAL REDUPLICATION 813

Copying the string [sór] with a coda is preferred by FTBIN, but fatally violates the higher-ranked
*ORALCODA (20c). In the third step (20d–e), the reduplicant vowel is reduced to [U], trading a
violation of *MID /LOW for a violation of *SCHWA (20e). Finally, the derivation converges on [sÛ-
só.rô] (20f–g). The reduplicant schwa is marked, but it is protected by MAXHD␮, a positional
faithfulness constraint that protects dependents of the head mora of a syllable ((21); Alderete
1995). In this example, the schwa is dominated by the syllable’s only mora, which is its head
mora. Throughout this analysis, the leftmost mora of a syllable is assumed to be the head. Thus,
MAXHD␮ protects vowels in CVC syllables and protects the high vowels in diphthongs. Note
that HD( ␴ ) is inactive here; as defined, HD( ␴ ) penalizes moraless syllables, not syllables without
moraic segments.
(21) MAXHD␮
Assign one violation to every application of the operation that deletes a segment domi-
nated by the head mora of a syllable.
3.1.2.2 Nasal Consonants Are Copied as Codas Derivations of nasal-final Class II verbs like
[tâ√] ‘teach’ also begin by inserting an empty syllable to head the foot template. Thereafter, a
CVN string is copied to head the syllable. The tableau in (22) illustrates this step. Because the
reduplicant is a monosyllabic foot, FTBIN prefers it to have a coda (22c) and dominates CODA-
COND, the only constraint that disprefers the coda.
After this copying step, the reduplicant vowel is reduced to [U]. As in (20f–g), the schwa
cannot be deleted because it is dominated by the head mora. The nasal then loses its place feature
and assimilates to the coronal stop: tÛ√-tâ√ N tÛN-tâ√ N [tÛn-tâ√].
814 ANDREW LAMONT

(22) /F-tâ√/ [tUˆ n-tâ√], Step 2 (Mbe)

F
␴ ⫹␴

*COPY(seg)

CODACOND
*MID/LOW
␮␮

HD(␴)

FTBIN
tâ√ Class II

F
␴ ⫹␴
␮␮
a. tâ√ W1 L L W1 L
F
␴ ⫹␴
␮ ␮␮
b. tâ tâ√ 1 1 W1 L
F
␴ ⫹␴
␮␮ ␮␮
c. tâ√ tâ√ 1 1 1

3.1.2.3 Vowel Deletion Feeds Nonlocal Nasal Copying When HD( ␴ ) motivates copying, nasal
codas are only copied to satisfy FTBIN —that is, when the root vowel is monomoraic. When the
initial root vowel is a diphthong, and thus bimoraic, there is no motivation to copy a coda in this
step. This is illustrated in the tableau in (23a–c) with [jı́:.nı̂] ‘forget’; by assumption, GEN cannot
copy only part of the diphthong. Because the copied diphthong satisfies FTBIN (23b), no constraint
prefers copying a coda. Doing so needlessly violates CODACOND (23c), even if the resulting cluster
is homorganic: place features in coda position are not licensed until they are linked to an onset,
which takes a step. While FTBIN does not motivate nasal copying in this step, HD( ␮ ) motivates
it later in the derivation. All diphthongs in Mbe begin with a high vowel and fall into a low or
mid vowel (Bamgbo⋅ se 1967a). In Class II imperatives, these second vowels reduce to schwa
(23f ) and then, because their mora is not the head mora, are deleted (23h), leaving their moras
headless.
 SERIAL REDUPLICATION 815

(23) /F-jí:.nı/
ˆ [jıˆ«-jıˆ:.nì], Steps 2–4 (Mbe)
F

*COPY(seg)
MAX(V-Pl)

CODACOND
*MID/LOW
␴ ⫹␴ ␴

*SCHWA

HD(␮)
␮␮ ␮

HD(␴)

*HIGH

FTBIN

MAX
jí:.nı̂ Class II

F
␴ ⫹␴ ␴
␮␮ ␮
a. jí:.nı̂ W1 L L L W1
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
b. jí: jí:.nı̂ 1 1 1
F
␴ ⫹␴ ␴
␮␮␮ ␮␮ ␮
c. jí:n jí:.nı̂ 1 1 1 W1
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
d. jí: jí:.nı̂ W1 L L 1
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
e. jU:
´ jí:.nı̂ W1 1 1 L
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
f. jíU jí:.nı̂ 1 1 1
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
g. jíU jí:.nı̂ W1 L 1 L
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
h. jí jí:.nı̂ 1 1 1
816 ANDREW LAMONT

The headless mora motivates a second round of segmental copying (24). To satisfy HD( ␮ ),
a segment can be copied in as its head (24b–c), or the mora can be deleted (24d). When there
is no suitable segment to copy (as in (17a–e)), the latter is preferred. But, when a nasal is available,
it is copied (24c). Notably, this instance of copying is nonlocal—three segments separate the
original nasal from its copy. Copying the reduplicant [i] (24b) or the root [j] satisfies *COPY-
LOC(seg), but fatally violates *HIGH or *ORALCODA, respectively.
ˆ
(24) /F-jí:.nı/ [jıˆ«-jıˆ:.nì], Step 5 (Mbe)

*COPY-LOC(seg)
F
␴ ⫹␴ ␴

*COPY(seg)

CODACOND
MAX(␮)
␮␮ ␮␮ ␮
HD(␮)
*HIGH

FTBIN
jí jí:.nıˆ Class II

F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
a. jí jí:.nıˆ 1 W1 L L L
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
b. jíí jí:.nıˆ W2 L 1 L
F
␴ ⫹␴ ␴
␮␮ ␮␮ ␮
c. jín jí:.nı̂ 1 3 1 1
F
␴ ⫹␴ ␴
␮ ␮␮ ␮
d. jí jí:.nı̂ 1 W1 L L W1 L

Before the derivation converges, the copied nasal is debuccalized and then assimilates to
the onset: jı́n-jı́:.nı̂ N jı́N-jı́:.nı̂ N [jı̂«-jı̂:.nı̀].
 SERIAL REDUPLICATION 817

In Class II imperatives, nonlocal copying only occurs with roots of the shape CVVN(V). It
must be fed by diphthong reduction, and is motivated to preserve the second mora of the reduplicant
(see Kawu 2000, Zuraw 2002, and Elfner and Kimper 2008 for other cases of phonologically
driven nonlocal copying). It is crucial that this second round of copying is motivated by HD( ␮ )
and not by FTBIN. Otherwise, verbs of the shape NVC would be incorrectly predicted to take
nasal-final reduplicants. Consider the tableau in (25), which illustrates such a case with [mâl]
‘finish’. The derivation converges on [mÛ-mâl] (25a), with a monomoraic reduplicant. Copying
the root-initial nasal into the reduplicant satisfies FTBIN, but fatally violates *COPY(seg). Thus,
as in (17f–h), FTBIN is ranked too low to motivate operations to apply that do not improve on
another markedness constraint.
(25) /F-mâl/ [mUˆ -mâl], convergence (Mbe)
F
␴⫹␴
*COPY(seg)

CODACOND
*SCHWA

␮ ␮␮
FTBIN

mÛ mâl Class II

F
␴⫹␴
␮ ␮␮
a. mÛ mâl 1 1
F
␴⫹␴
␮␮ ␮␮
b. mUˆ m mâl 1 W1 L W1

3.1.3 Summary of Mbe Imperatives Class I and II imperatives differ in the structures they copy
and therefore how they enforce the phonotactic restrictions on reduplicant codas. In Class I, the
root syllable is copied, and then oral consonants are deleted from the coda. In Class II, codas are
only copied to fulfill a prosodic need, either as part of a larger segmental string to head an empty
syllable or nonlocally to head a mora. Both lexical classes effectively filter out nonnasal segments
from coda position. Nasals are preserved in codas, where they debuccalize and undergo place
assimilation.
This section has demonstrated that, despite their differences, Class I and II imperatives are
modeled with the same grammar. Lexically indexed constraints are necessary to derive the differ-
ences in vowel reduction and whether syllables or segments are copied, but otherwise the constraint
set is identical. The next section illustrates that this grammar also models diminutives.
 818 ANDREW LAMONT

3.2 The Mbe Diminutive as a Mora Template

Diminutives take the Class 4 marker /k[-/ (Bamgbo⋅ se 1966), but their primary exponent is a
nasal coda homorganic to a following consonant (Walker 1998a, 2000); examples are given in
(26). The nasal coda only surfaces with nouns that contain nasals (26d–f ); it is unrealized with
nouns that do not (26a–c). This indicates that the nasal is copied from the root and is not present
underlyingly.
(26) Mbe diminutives (Bamgbo⋅ se 1966:48, 49, 50)
a. lı́ k[ˇ -lı̂ ‘eye’
b. b[` l k[ˇ -b[´ l ‘wives’
c. lè-bà.rò kà-bà.rò ‘liver’
d. bù-mù k[` -m-mù ‘story’
e. k[` -nén k[ˇ -n-nên ‘bird’
f. bù-t:̀.nı́ k[ˇ -n-t:̀.nı́ ‘earthworm’
Because it surfaces as a single segment in coda position, the diminutive can be analyzed as
an underlying mora. Like Class II imperatives, this headless mora motivates (nonlocal) nasal
copying. When no nasal segment is available, the mora is deleted. This is illustrated in the tab-
leaux in (27)–(28) with the same constraint ranking as for the imperative. As before, tableaux
only show violations incurred by the reduplicant.
The tableau in (27) illustrates the derivation of [k[ˇ -lı̂] ‘eye’. As in (17), there is no suitable
segment that can be copied, and the mora deletes (27b). When a nasal is available, it is copied
to head the mora. This is illustrated in the tableau in (28) with [k[ˇ -n-t:̀.nı́] ‘earthworm’. Deleting
[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

the mora satisfies HD( ␮ ) (28c), but, because segment copying is preferred, the root-internal nasal
is copied as its head (28b). As in (24), copying is nonlocal: the copy surfaces two segments away
from its source.
(27) /k[-␮-lí/ [k[ˇ-lˆã], Step 1 (Mbe)
␴ ␴
MAX(␮)

␮⫹␮⫹ ␮
HD(␮)

k[ˇ lã̂

␴ ␴
␮⫹␮⫹ ␮
a. k[ˇ lã̂ W1 L
␴ ␴
␮⫹ ⫹ ␮
b. k[ˇ lã̂ 1
 SERIAL REDUPLICATION 819

(28) /k[-␮-t:` .ní/ [k[ˇ-n-t:` .ní], Step 1 (Mbe)

␴ ␴␴

*COPY(seg)
*COPY-LOC

CODACOND
MAX(␮)
␮ ⫹ ␮ ⫹ ␮␮

HD(␮)

(seg)
k[ˇ t:´ ní

␴ ␴␴
␮ ⫹ ␮ ⫹ ␮␮
a. k[ˇ t:´ ní W1 L L L
␴ ␴␴
␮ ⫹ ␮ ⫹ ␮␮
b. k[ˇ n t:´ ní 2 1 1
␴ ␴␴
␮⫹ ⫹ ␮␮
c. k[ˇ t:´ ní W1 L L L

Before the derivation converges, the copied nasal’s coronal place feature deletes, and the
onset’s coronal place feature spreads: k[ˇ -n-t:̀.nı́ N k[ˇ -N-t:̀.nı́ N [k[ˇ -n-t:̀.nı́].
As an underlying mora, the diminutive behaves exactly like headless moras created in the
derivation of imperatives. In the absence of a nasal consonant, the mora is deleted, as with oral-
consonant-final Class I verbs. Otherwise, nasal consonants are copied in as heads, as with Class
II verbs with initial diphthongs.

3.3 Summary
Mbe limits reduplicant codas to nasals homorganic to following consonants. This phonotactic
restriction applies to imperative verbs in both lexical classes and to diminutive nouns. Wei and
Walker (2020) argue that by requiring codas to be place-linked, Mbe cannot be modeled without
derivational lookahead. However, as this section has demonstrated, banning nonnasal segments
from coda position suffices. Oral segments are either deleted from coda position, as with Class
I imperatives, or are never copied into coda position, as with Class II imperatives and diminutives.
This leaves only nasals in coda position, which undergo place assimilation. Wei and Walker
(2020) also argue that a serial analysis cannot model imperatives of the form CVN-CVVN(V),
because there is no motivation to copy the nasal coda. This is true early in the derivation, but
reducing the diphthong to a single vowel creates a headless mora and, with it, the motivation for
a second round of segmental copying. The constraint ranking for Mbe is represented in the Hasse
diagram in figure 1; dashed lines represent disjunctive rankings.
820 ANDREW LAMONT

HD(F) *COPY(␴)II
(59a) (50b) ⴰ (59c)
HD(␴)
(50b) ⴰ (54b)

MAX(V-place)I (60a) *COPY(␴) (50b) ⴰ (54b)

(58b) ⴰ (61a)
*MID/LOW MAXHD␮
(61a)
(61a) (62f) ⴰ (72a)
MAX(V-place) *SCHWA *ORALCODA
(72c)
(72c)
(72a) (72a) (52c) ⴰ (57b) (52c) ⴰ (57b)

*HIGH HD(␮) *LONGVOWEL MAX

(53a)
(81e) ⴰ (82g) (53b)
MAX(␮)
(82h)
(82h)
*COPY-LOC(seg) *COPY(seg)
(64d) ⴰ (80e)

FTBIN
(64d)

CODACOND
(51a)
(51a)

MAX(C-place) HAVEPLACE
(57a)

*LINK(place)
Figure 1
Hasse diagram for Mbe; referenced tableaux are in the online appendix

4 Reduplication in Logoori Does Not Require Lookahead

Logoori (also known as Maragoli; Bantu; Leung 1991, Glewwe and Aly 2016, Adler and Zymet
2017, 2021, Zymet 2018, Odden n.d.) does not permit vowels to surface in hiatus, and underlying
vowel clusters are subject to a number of different processes (see Glewwe 2016, Zymet 2018,
and Odden n.d. for discussion of hiatus resolution outside reduplication). Mid and high vowels
surface as glides, and trigger compensatory lengthening word-medially, but not word-finally:
/vi-a/ N [vja] *[vja] ‘of-CL.8’ (Zymet 2018:304). These processes are active in the formation
of second and third person singular possessives, which concatenate a (C)V noun class marker
 SERIAL REDUPLICATION 821

with a suffix vowel, /-: / in the second person and /-[ / in the third person. These possessives also
take a syllable-sized reduplicative prefix that shows similar, but distinct effects. Reduplicants
surface with a glide except when it would be adjacent to a homorganic high [ⳭATR] vowel: [tù-
tw-:́] *[twù-tw-:́] (29e).
(29) Logoori singular possessives (Leung 1991:24, Zymet 2018:309)
1.sg /-ant[ / 2.sg /-: / 3.sg /-[ / Noun class
/o-/ a. w-a√t[ b. w:̀-v-:́ c. w[` -v-[´ 1
/ to-/ d. tw-a√t[ e. tù-tw-:́ f. twı̀-tw-[´ 3
/e-/ g. j-a√t[ h. j:̀-j-:́ i. j[` -j-[´ 9
/ke-/ j. t+j-a√t[ k. t+ ı̀-t+j-:́ l. t+ ı̀-t+j-[´ 7
‘my’ ‘your (sg.)’ ‘his/her’
The phonotactic restriction on homorganic glides is not specific to reduplication. For example,
the indefinite future prefix /ri-/ surfaces as [rj-] before vowel-initial verbs (30a–c) and as [r-]
before those beginning with [i] (30d). There is no independent evidence for the avoidance of
[wu].
(30) [j] does not surface before [i] (Odden n.d.:chap. 3, 130)
a. /va-ri-ata/ [varjatá] ‘they may perform surgery’
b. /a-ri-erema/ [arjerémá] ‘he may float’
c. /a-ri-ëmba/ [arjëmbá] ‘he may sing’
d. /va-ri-ita/ [várı́tá] ‘they may kill’
Zymet (2018; see also Adler and Zymet 2017, 2021) argues that derivational lookahead is
required to model the possessive. The order between hiatus resolution and reduplication depends
on the shape of the stem. With /V-V/ stems, glide formation precedes copying, resulting in two
CV syllables: /V-V/ N V  -V N [V
 V-V  -V]. With /CV-V/ stems, however, that order would create
an illicit complex onset when the resulting vowel is high: /CV-V/ N CV  -V N *[CV  V-CV  -V],
and reduplication must precede hiatus resolution: /CV-V/ N CV-CV-V N [CV-CV  -V]. The
order between hiatus resolution and reduplication thus requires looking ahead to avoid glides
from surfacing before homorganic high vowels.
This section presents an analysis of Logoori reduplication in STS that does not require
lookahead. The noun class marker and suffix vowel are parsed into a single syllable, which is
copied. Glides are then formed independently in the reduplicant and in the stem. If, after raising
and tensing of the reduplicant vowel, the glide-vowel sequence is homorganic, the glide is deleted.
Thus, Logoori reduplication does not provide evidence for lookahead.

4.1 The Logoori Possessive as a Foot Template

Before arguing for the underlying form of the reduplicant, it is important to establish exactly
what is copied. The stem is transparently composed of a noun class marker and a suffix vowel:
/(C)V-V/. In order to derive the differences between the reduplicant and the stem, this entire
structure must be copied: /(C)V-V/ N (C)VV-(C)V-V. Hiatus resolution then proceeds indepen-
822 ANDREW LAMONT

dently in the reduplicant and in the stem. This is necessary to derive the length and quality of
the reduplicant vowel.
Underlying high vowels surface as vowels in the reduplicant and as glides in the stem. The
surface form of the stem is exactly what is expected from hiatus resolution: / ˜i-: / surfaces as [˜j-
:] with a complex onset and not as *[˜i-w] with a coda (31c). The form of the reduplicant results
from its preference to surface with a high [ⳭATR] vowel. The copied high vowel in / ˜i: / is thus
preserved at the expense of the mid vowel, resulting in an intermediate form with a coda [˜iw].
As Logoori does not allow consonants to surface in coda position, this glide is deleted.
(31) Noun class markers with high vowels (Zymet 2018:309)
2.sg /-: / 3.sg /-[ / Noun class
/dÇi-/ a. dÇı̀-dÇj-:́ b. dÇı̀-dÇj-[´ 4
/ ˜i-/ c. ˜ı̀-˜j-:́ d. ˜ı̀-˜j-[´ 5
/vi-/ e. vı̀-vj-:́ f. vı̀-vj-[´ 8
/zi-/ g. zı̀-zj-:́ h. zı̀-zj-[´ 10
/mu-/ i. mù-mw-:́ j. mwı̀-mw-[´ 18
‘your (sg.)’ ‘his/her’
The preference for high vowels to surface in the reduplicant triggers copied mid vowels to
raise and tense (32). In the third person, the reduplicant is exactly like the stem, except that the
copied suffix vowel / [ / raises to [i] (32b,d,f,h,j). In the second person, raising the copied suffix
vowel / : / to [u] creates a homorganic glide-vowel sequence, and the glide is deleted: / to: / N
tw: N twu N [tu] (32a). Just as in Class I imperatives in Mbe (section 3.1.1), some segments
that are copied are later deleted.
(32) Noun class markers with mid vowels (Zymet 2018:309)
2.sg /-: / 3.sg /-[ / Noun class
/ to-/ a. tù-tw-:́ b. twı̀-tw-[´ 3
/ ˜o-/ c. ˜ù-˜w-:́ d. ˜wı̀-˜w-[´ 11
/to-/ e. tù-tw-:́ f. twı̀-tw-[´ 13
/vo-/ g. vù-vw-:́ h. vwı̀-vw-[´ 14
/ko-/ i. kù-kw-:́ j. kwı̀-kw-[´ 15
‘your (sg.)’ ‘his/her’
The only exception to the pattern in (32) is the Class 7 marker, which Zymet (2018) analyzes
with an underlying mid vowel /ke-/. Its reduplicant always surfaces with [i]: [t+ı̀-t+j-:́] *[t+jù-
t+j-:́] (Zymet 2018:309). Leung (1991) and Odden (n.d.) analyze this morpheme as having an
underlying high vowel /kë-/, which is consistent with its patterning with other high vowel noun
class markers (31).
Vowel raising does not occur when the stem syllable has a simple onset, as with noun class
markers of the shape /Ca/ (33). Low vowels delete in hiatus contexts and do not leave behind a
glide: / ta-etu/ N [t-etu] ‘our-CL.6’ (Zymet 2018:305). Stems that are underlyingly /Ca-V/ thus
surface as /CV/, and the reduplicant vowel does not raise: [v:̀-v-:́] *[vù-v-:́] (33a).
 SERIAL REDUPLICATION 823

(33) Noun class markers with low vowels (Zymet 2018:309)

2.sg /-: / 3.sg /-[ / Noun class
/va-/ a. v:̀-v-:́ b. v[` -v-[´ 2
/ ta-/ c. t:̀-t-:́ d. t[` -t-[´ 6
/ka-/ e. k:̀-k-:́ f. k[` -k-[´ 12
/ha-/ g. h:̀-h-:́ h. h[` -h-[´ 16
‘your (sg.)’ ‘his/her’
The same is true of onsetless noun class markers, which surface as glides (34). There is no
consonant with which to form a complex onset in the stem, and the reduplicant surfaces with a
mid vowel: [j:̀-j-:́] *[jù-j-:́] (34a).
(34) Noun class markers without onsets (Zymet 2018:309)
2.sg /-: / 3.sg /-[ / Noun class
/e-/ a. j:̀-j-:́ b. j[` -j-[´ 9
/o-/ c. w:̀-v-:́ d. w[` -v-[´ 1
‘your (sg.)’ ‘his/her’
Simple onsets block vowel raising because they are transparent to height harmony. High
[ⳮATR] vowels lower to mid when the next syllable contains a mid vowel (35a–b). Harmony is
blocked by, among other things, consonant-glide clusters (35c–d). Because there is a general
pressure to avoid high-mid vowel sequences unless a complex onset intervenes, reduplicant vowels
raise only when the stem has a simple onset. Raising in words like [j:̀-j-:́] *[jù-j-:́] (34a) would
create a disharmonic high-mid sequence.
(35) Regressive [ⳮATR] vowel lowering (Odden 2019, n.d.:chap. 3, 93, 141)
a. / ë-kë-dete/ [e-ke-déte] ‘finger’
b. / ë-kë-dende/ [e-ke-dende] ‘swamp’
c. /F-kë-vo-e«a/ [Fkëvwe«á] ‘you are still wanting it’
d. /a-kë-vo-onot«a/ [akëvwonótó«á] ‘he is still messing it up’
In order to model Logoori reduplication, the noun class marker and the suffix vowel must
be present in both the stem and the reduplicant. There are a handful of underlying forms consistent
with this analysis: an underlying syllable filled by copying the morphological stem, an underlying
foot filled by inserting a syllable and then copying a string of segments, and an underlying foot
filled by copying a syllable. Without evidence to favor any analysis in particular, the latter is
chosen for its simplicity of presentation: the stem is syllabified into a .(C)VV. syllable, which is
copied into an underlying foot template. It is also possible to adopt a template-free analysis that
requires some overt realization of the reduplicant, as Anderson and Smith (2017, 2018) propose
for Tohono O’odham.
The analysis does not require a constraint on vowel hiatus per se. Instead, syllables that
dominate multiple vowels violate *NONHEADVOWEL ((36); roughly equivalent to de Lacy’s (2002,
2006) *ⳮ⌬␴ ⱖé). Adjacent underlying vowels are syllabified together to minimize violations of
ONSET; see the tableaux in the online appendix.
824 ANDREW LAMONT

(36) *NONHEADVOWEL
Assign one violation to every syllable that dominates more than one vowel.
To improve on *NONHEADVOWEL, GEN has an operation that maps vowels onto glides by
delinking their moras (cf. Jacobs 2019, where gliding is analyzed as mora deletion in HS). This
operation is posited primarily for analytical ease, but is consistent with the hypothesis that vowels
and glides differ in terms of weight and no other features (Rosenthall 1994). Two assumptions
with respect to glides are made for expository ease. First, vowels map directly onto the glides
[j,w]. This avoids intermediate raising steps like /e/ N e N [j]. Second, GEN deletes glides in
one step rather than first deleting their place features, suppressing unnecessary tableaux.
Vowel raising is motivated by the constraint *MID /LOWRED (37). The domain of this con-
straint must be specified as the reduplicant, because vowel raising is otherwise unattested in the
language. Raising a vowel violates the faithfulness constraint IDENT(high). For expository ease,
GEN raises and tenses vowels in one step.
(37) *MID /LOWRED
Assign one violation for every mid vowel and every low vowel in the reduplicant.
The pressure for vowel harmony is simplified as the constraint *HIGHMID (38). This constraint
suffices for the purposes of analyzing reduplication, but it may be stated too generally for the
language, as [ⳭATR] vowels are not subject to lowering: / ë-vi-dete/ N [ë-vi-déte] *[e-ve-déte]
‘finger-CL.8’ (cf. (35a)). Whatever mechanism is responsible for this distinction between /i/ and
/ ë / is beyond the scope of the article.
(38) *HIGHMID
Assign one violation for every high vowel followed by a mid vowel unless separated
by another vowel or a complex onset.
The analysis of Logoori reduplication is presented in detail in section 4.1.1. As in previous
analyses, tableaux represent pairwise relations between constraints that reflect the overall ranking.
That ranking is represented in the Hasse diagram in figure 2.
4.1.1 Logoori Possessives Copy Syllables The noun class marker and suffix vowel are sylla-
bified together and then copied into a foot template (see the online appendix for the syllabification
steps). Hiatus resolution proceeds independently in the two syllables, and produces different
outcomes to satisfy constraints on reduplicant vowels.
4.1.1.1 High Vowels Surface Faithfully in Reduplicants, as Glides in Stems The tableau in (39)
shows the first step after the stem has been syllabified, where the syllable is copied into the foot
template (39c). Inserting an empty syllable violates the higher-ranked HD( ␴ ) and ONSET (39b).
No other candidate improves on HD(F) (39a,d–e).
 SERIAL REDUPLICATION 825

(39) /F-vi-:/ [vì‰-vj-:´ ], Step 1 (Logoori)

F

*MID/LOWRED
*NONHEADV
⫹␴

*HIGHMID
*COPY(␴)
␮␮

*CODA
HD(␮)
HD(␴)

ONSET
HD(F)
vi:

F
⫹␴
␮␮
a. vi: W1 L L1 L1 L
F
␴ ⫹␴
␮␮
b. vi: W1 W1 L L1 L1 L
F
␴ ⫹␴
[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

␮␮ ␮␮
c. vi: vi: 1 2 2 1
F
⫹␴
␮␮
d. vj: W1 L L W1 L L
F
⫹␴
␮␮
e. viw W1 L L W1 L L W1

The input to the second step (40) contains two syllables that each dominate two vowels.
Forming a glide from any of the four vowels improves on *NONHEADVOWEL (40b–e), but only
targeting the mid vowel in the reduplicant satisfies *MID /LOWRED (40c). Raising the reduplicant
vowel to [u] has the same effect, but fails to improve on *NONHEADVOWEL (40f ).
826 ANDREW LAMONT

The high vowel of the stem becomes a glide in the next step (41), satisfying *NONHEADVOWEL
(41b). Because there is no pressure for the stem to surface with a high vowel, a complex onset
is preferred to a coda (41c). This step illustrates why hiatus resolution must proceed independently
in the two syllables: if hiatus resolution had preceded copying, the reduplicant would also have
a complex onset.
(40) /F-vi-:/ [vì‰-vj-´:], Step 2 (Logoori)
F

*MID/LOWRED
*NONHEADV

IDENT(high)
␴ ⫹␴

*HIGHMID

*CODA
␮␮ ␮␮
vi: vi: HD(␮)
F
␴ ⫹␴
␮␮ ␮␮
a. vi: vi: W2 L W2 W1 L
F
␴ ⫹ ␴
␮␮ ␮␮
b. vj: vi: 1 1 1 W1 L
F
␴ ⫹␴
␮␮ ␮␮
c. viw vi: 1 1 1 1
F
␴ ⫹␴
␮␮ ␮␮
d. vi: vj: 1 1 1 W1 L
F
␴ ⫹␴
␮␮ ␮␮
e. vi: viw 1 1 1 W1 1
F
␴ ⫹␴
␮␮ ␮␮
f. viu vi: W2 L 1 W1 L
 SERIAL REDUPLICATION 827

(41) /F-vi-:/ [vì‰-vj-:´ ], Step 3 (Logoori)

F

*NONHEADV
␴ ⫹␴

*HIGHMID
*LINK(␮)
MAX(␮)

*CODA
␮␮ ␮␮

HD(␮)

MAX
viw vi:
F
␴ ⫹␴
␮␮ ␮␮
a. viw vi: W1 L1 W1 1
F
␴ ⫹ ␴
␮␮ ␮␮
b. viw vj: 2 1
F
␴ ⫹␴
␮␮ ␮␮
c. viw viw 2 W2
F
␴ ⫹␴
␮ ␮␮
d. viw vi: W1 L W1 W1 1
F
␴ ⫹␴
␮␮ ␮␮
e. viw vi: W1 L W1 W1 1
F
␴ ⫹␴
␮␮ ␮␮
f. vi vi: W1 L1 W1 L W1
828 ANDREW LAMONT

(42) /F-vi-:/ [vì‰-vj-:´ ], Step 4 (Logoori)

*FINALLONGV
F
␴ ⫹␴

*LINK(␮)
MAX(␮)

*CODA
␮␮ ␮␮

HD(␮)

MAX
viw vj:
F
␴ ⫹ ␴
␮␮ ␮␮
a. viw vj: W2 L 1
F
␴ ⫹ ␴
␮ ␮␮
b. viw vj: 1 W1 L 1
F
␴ ⫹ ␴
␮␮ ␮␮
c. viw vj: 1 1 1
F
␴ ⫹␴
␮␮ ␮
d. viw vj: 1 W1 L 1
F
␴ ⫹ ␴
␮␮ ␮␮
e. viw vj: 1 W1 1 1
F
␴ ⫹␴
␮␮ ␮␮
f. vi vj: W2 L L W1
 SERIAL REDUPLICATION 829

(43) /F-vi-:/ [vì‰-vj-:´ ], Steps 5–6 (Logoori)

F
␴ ⫹␴

MAX(␮)

*CODA
␮␮ ␮␮

HD(␮)

MAX
viw vj:
F
␴ ⫹ ␴
␮␮ ␮␮
a. viw vj: W1 L 1
F
␴ ⫹ ␴
␮␮ ␮
b. viw vj: 1 1
F
␴ ⫹ ␴
␮␮ ␮␮
c. vi vj: W1 L L W1
F
␴ ⫹␴
␮␮ ␮
d. viw vj: W1 L
F
␴ ⫹ ␴
␮␮ ␮
e. vi vj: 1

Two vowels having been delinked from their moras, the input to the fourth step contains two
headless moras (42). Because MAX( ␮ ) dominates *LINK( ␮ ), deleting moras (42b,d) is generally
dispreferred to relinking them to vowels (42c,e). The reduplicant vowel is lengthened in this
step (42c). Linking to the stem vowel would create a word-final long vowel, in violation of
*FINALLONGVOWEL (42c).
In the next step, *FINALLONGVOWEL prevents the stem mora from relinking at all. It is
deleted, satisfying HD( ␮ ) (43b). In the next step, the reduplicant glide is deleted from coda
position, satisfying *CODA (43e).
The derivation converges in the next step on [vı̀-vj-:́], as its input does not violate any
active markedness constraints.
4.1.1.2 Mid Vowel Raising Feeds Homorganic Glide Deletion *MID /LOWRED dictates which
.
reduplicant vowel is targeted in hiatus contexts if one is high. When no constraint prefers a .CVV
830 ANDREW LAMONT

syllable, as when both reduplicant vowels are mid, *CODA chooses a .CV  V. syllable. The resulting
complex onset is opaque to height harmony, and vowel raising is not blocked. This is illustrated
in the tableau in (44a–b), the sixth step mapping /F-vo-: / onto [vù-vw-:́] ‘your (sg.)’. Raising
the reduplicant vowel to [u] satisfies *MID /LOWRED and creates a homorganic glide-vowel cluster,
violating *ji/wu (44b). The violation of *ji/wu is removed in the next step (44c–d) by deleting
the glide (44d).
(44) /F-vo-:/ [vù‰-vw-:´ ], Steps 6–7 (Logoori)
F
*MID/LOWRED
␴ ⫹ ␴
␮␮ ␮ IDENT(high)

*ji/wu

MAX
vw: vw:

F
␴ ⫹ ␴
␮␮ ␮
a. vw: vw: W1 L L
F
␴ ⫹ ␴
␮␮ ␮
b. vwu vw: 1 1

F
␴ ⫹ ␴
␮␮ ␮
c. vwu vw: W1 L
F
␴ ⫹␴
␮␮ ␮
d. vu vw: 1

The derivation converges in the next step on [vù-vw-:́].

4.1.1.3 Simple Onsets Block Mid Vowel Raising Because the stem vowel is always mid, it is
only possible to raise the reduplicant vowel when the stem has a complex onset. Thus, raising is
 SERIAL REDUPLICATION 831

blocked when the noun class marker vowel is low and deletes without leaving behind a glide, or
when the noun class marker does not have an onset. The latter is illustrated with the tableau in
(45), the convergent step of the derivation mapping /F-o-: / onto [w:̀-v-:́] ‘your’, ignoring the
intervocalic hardening of /w/ to [v]. Because the stem onset is simple and therefore transparent
to height harmony, the reduplicant mid vowel cannot raise without creating a disharmonic vowel
sequence and violating *HIGHMID (45b).
(45) /F-o-:/ ` -v-:´ ], convergence (Logoori)
[w:‰
F

*MID/LOWRED
␴ ⫹␴

IDENT(high)
*HIGHMID
␮␮ ␮

*ji/wu
w: w:

F
␴ ⫹␴
␮␮ ␮
a. w: w: 1
F
␴ ⫹␴
␮␮ ␮
b. wu w: W1 L W1 W1

4.1.2 Summary of Logoori Possessives Whether glides surface in the reduplicant depends on
how the reduplicant resolves hiatus, which is independent of how the stem resolves hiatus. Because
there is a preference for the reduplicant vowel to be high, high vowels surface as vowels, and no
glide surfaces. The preference for high vowels compels mid vowels to raise, which can suppress
glides from surfacing to avoid homorganic glide-vowel sequences.

4.2 Summary
Logoori bans glides from surfacing before homorganic high vowels. Zymet (2018) argues that
this restriction cannot be satisfied in reduplication contexts without derivational lookahead. How-
ever, as this section has demonstrated, this is not the case. Copied high vowels never surface as
glides in the reduplicant, satisfying their preference for high vowels. The same pressure motivates
mid vowel raising. Derived high vowels surface with heterorganic glides, but force homorganic
glides to delete. Thus, Logoori reduplication does not pose a challenge to STS. The constraint
ranking for Logoori is represented in the Hasse diagram in figure 2; dashed lines represent disjunc-
tive rankings.
 832 ANDREW LAMONT

PARSE(seg)
(86c) ⴰ (87a)

(87a)
HD(F) (87a)

ONSET (87b)

(86e) (87a)

*NONHEADVOWEL (87b) HD(␴)

(97a) (87b) (87b)
*FINALLONGVOWEL HD(␮) *HIGHMID *COPY(␴)
(91a)
(90b) ⴰ (91c) (99d) ⴰ (100a) (108a) ⴰ (117b)

MAX(␮) *MID/LOWRED
(108a) (100a)
(90b) (88b)

*LINK(␮) *ji/wu *CODA IDENT(high)

(109a) (92a)

MAX
Figure 2
Hasse diagram for Logoori; referenced tableaux are in the online appendix

5 The Emergence of the Marked in Parallel, but Not Serial, Reduplication

[35.149.200.162] Project MUSE (2024-06-19 13:59 GMT) the University of Connecticut

Wei and Walker (2020) and Zymet (2018) argue that BRCT is superior to STS on empirical
grounds. By arguing that reduplication in Mbe and Logoori, respectively, cannot be modeled
without derivational lookahead, they claim that STS is not expressive enough to model the attested
typology. As sections 3 and 4 have demonstrated, however, neither language evidences deriva-
tional lookahead, and STS is empirically adequate. This section identifies a novel class of unat-
tested patterns, the emergence of the marked, predicted by BRCT that cannot be replicated by
STS. On the basis of this evidence, STS is argued to be more typologically restrictive than BRCT.
It is not uncommon for reduplicants to admit only a strict subset of the marked structures
that surface elsewhere in a language (McCarthy and Prince 1994; see Becker and Flack Potts
2011 for an overview). In such cases, unmarked structures that are generally dispreferred to
faithful, marked structures emerge in restricted environments such as reduplicants. The converse
pattern, the emergence of the marked, where structures that are repaired elsewhere in the language
surface in reduplicants in order to satisfy faithfulness constraints, is unattested.2 There are mecha-

2
The term the emergence of the marked has also been used in other contexts such as children idiosyncratically
deleting onsets (Velleman and Vihman 2002) or marked tones associating to reduplicants (Downing 2005). Blevins (2003,
2005) discusses cases of intervocalic consonant deletion and vowel syncope that create violations of ONSET and *CODA,
respectively, which only target reduplicants. However, as these processes conspire to eliminate phonologically predictable
material (Blevins 2005), they can be analyzed as satisfying a markedness constraint, distinguishing them from the cases
in this section.
 SERIAL REDUPLICATION 833

nisms that have been shown to generate these effects, such as constraints that penalize the lowest
end of a markedness hierarchy (Gouskova 2003) and constraints that penalize structures in roots
specifically (Albright 2004), but BRCT has been argued to avoid them (Spaelti 1997). This sec-
tion, however, dispels that: not only does BR-faithfulness prevent the elimination of marked struc-
tures, it creates them.
Consider the tableaux in (46), identical to those in section 1, but with a suffixal reduplicant;
this pattern too is unattested. Voiced obstruents that surface word-finally are generally devoiced
(46a–c). However, final voiced obstruents surface in the reduplicant whenever stems begin with
voiced obstruents (46d). The constraint IDENT(voice)-BR is crucial in preventing obstruents in
the reduplicant from devoicing (46e,g); with this constraint set, no markedness constraint prefers
the winner (46d) over these candidates.
(46) IDENT(voice)-BR blocks final devoicing (unattested)
IDENT AGREE IDONSET *VOICED IDENT
/talab/, /talab-RED/
(voi)-BR (voice) (voi)-IO OBS (voi)-IO

a. talab W2 L
b. talap 1 1
c. kalap W1 L W2

d. talab-talab 4
e. talab-talap W1 L3
f. talap-talap W1 L2 W1
g. talap-kalap W1 L1 W1
h. kalap-kalap W1 L W2

These tableaux demonstrate that a core component of BRCT can force marked structures to
surface faithfully that would otherwise be repaired. As argued in section 1, this cannot be replicated
in STS without introducing derivational lookahead or bizarre nonlocal constraints.
While IDENT-BR constraints can force marked structures to surface faithfully in the redupli-
cant, MAX-BR can create them, as the tableaux in (47) demonstrate with another unattested pattern.
In this example, consonants generally delete rather than surfacing in coda position (47a–b).
However, prefixal reduplicants not only surface with codas, but surface with arbitrarily complex
codas (47c–h). The base in this example is bisyllabic, and the constraint RED⳱␴ forces the redupli-
cant to be a single syllable (47c). This specific constraint is used for ease of presentation; the
analysis is consistent with multiple theories of CON. For example, if the base were footed, then
ALLFT-L would have the same minimizing effect. The optimal candidate satisfies MAX(C)-BR
by copying the stem’s onset cluster into the reduplicant coda (47d). Omitting one or more conso-
nants fatally violates MAX(C)-BR (47e–f ). The only way to maintain perfect correspondence
between the reduplicant and the base without violating *CODA is to delete consonants from the
834 ANDREW LAMONT

base (47g–h). However, the available options either create onsetless syllables (47g) or delete a
vowel (47h), violating higher-ranked constraints. As in the previous example, MAX(C)-BR is
crucial in ruling out candidates (47e–f ), which are not dispreferred by any markedness constraints.
(47) MAX(C)-BR creates arbitrarily complex codas (unattested)
MAX MAX MAX
/pakl/, /RED-pakla/ RED⫽␴ ONSET *CODA
(C)-BR (V)-IO (C)-IO

a. pakl W1 L
b. pa 2

c. pa.kla-pa.kla W1 L
d. pakl-pa.kla 1
e. pak-pa.kla W1 1
f. pa-pa.kla W2 L
g. pa-pa.a W1 L W2
h. pa-pa W1 L W2

The combined effect of the minimizing constraint RED⳱␴ and the maximizing constraint
MAX(C)-BR is to pack as many consonants into the reduplicant as possible. Notably, the redup-
licant is closed not to satisfy a prosodic requirement, but to be arbitrarily large. For every base
CV.CnV, MAX(C)-BR prefers the candidate CVCn-CV.CnV over every candidate CVCnⴑ1-
CV.CnV.
Because there is no mechanism in STS or HS generally that motivates arbitrary maximization,
STS cannot produce this mapping without significant modification. Segment copying cannot close
a syllable unless doing so satisfies a prosodic requirement like FTBIN. These constraints can only
motivate a fixed number of consonants to surface in the coda; FTBIN, for example, is satisfied
by a single coda consonant. In the example above, however, no such threshold can be set; MAX(C)-
BR always prefers copying n consonants instead of n ⳮ1 consonants. Because the CVCC string
is not a prosodic unit, it also cannot be copied directly as a string of prosodic constituents. Further,
if the two syllables of the stem were copied, the onset cluster would have to be resyllabified.
But, because this language parses underlying /VCCV/ as [V.CCV] and not as *[VC.CV] or
*[VCC.V], complex onsets are generally preferred to codas, and without reranking the constraints,
resyllabification would not be optimal. Copying a phonologically defined string is therefore infeas-
ible, but so is copying a morphological unit. In HS, reduplicants are not a distinguished class of
affixes, and differences in behavior reflect root-affix asymmetries, not base-reduplicant asymme-
tries (see Zukoff 2017 for discussion of a prediction of STS along these lines). Thus, even if the
CVCC string were a morpheme and copied into the reduplicant, it would be expected to behave
like affixes of the same shape, and delete the final consonants.
 SERIAL REDUPLICATION 835

In summary, the core components of BRCT, faithfulness constraints that regulate identity
between the reduplicant and its base, predict that marked structures should surface in the redupli-
cant, even if they do not surface elsewhere in the language. These unattested patterns, the emer-
gence of the marked, are not predicted by STS and represent a major difference in the typological
predictions between the two frameworks.

6 Conclusion
The limitations HS imposes on GEN restrict the interactions between operations it is able to
evaluate. Parallel Optimality Theory assumes no such restrictions and accesses a strict superset
of candidates at evaluation. Arguments favoring pOT over HS must identify mappings whose
outputs are inaccessible to HS. This article begins a project aimed at critically evaluating such
claims recently made in the literature, focusing on reduplication. Wei and Walker (2020) and
Zymet (2018) argue that phonotactic restrictions in Mbe and Logoori, respectively, cannot be
obeyed unless copying occurs simultaneously with other operations. Both works argue that BRCT
(McCarthy and Prince 1994, 1995, 1999), a theory of reduplication in pOT, is therefore superior
to STS (McCarthy, Kimper, and Mullin 2012), a theory of reduplication in HS. This article has
rebutted their arguments, providing serial analyses of the reduplication mappings in question.
Further, it has identified a class of unattested mappings, the emergence of the marked, that are
predicted only by BRCT. Therefore, not only is STS empirically adequate, it is appropriately
typologically restrictive.

References
Adler, Jeffrey. 2017. The nature of conspiracy: Implications for parallel versus serial derivation. Master’s
thesis, University of California, Santa Cruz. https://escholarship.org/uc/item/1d63t2g8.
Adler, Jeffrey, and Jesse Zymet. 2017. Irreducible parallelism in phonology. In NELS 47, ed. by Andrew
Lamont and Katerina Tetzloff, 1:31–44. Amherst: University of Massachusetts, Graduate Linguistics
Students Association.
Adler, Jeffrey, and Jesse Zymet. 2021. Irreducible parallelism in phonology: Evidence for lookahead from
Mohawk, Maragoli, Sino-Japanese, and Lithuanian. Natural Language and Linguistic Theory 39:367–
403.
Albright, Adam. 2004. The emergence of the marked: Root-domain markedness in Lakhota. Paper presented
at Linguistic Society of America 2004. http://web.mit.edu/albright/www/papers/Albright-Emergen
ceOfTheMarked-LSA04.pdf.
Alderete, John. 1995. Faithfulness to prosodic heads. Ms., University of Massachusetts Amherst. http://roa
.rutgers.edu/article/view/105.
Anderson, Skye, and Ryan Walter Smith. 2017. Non-templatic plural reduplication in Tohono O’odham. In
WCCFL 34, ed. by Aaron Kaplan, Abby Kaplan, Miranda K. McCarvel, and Edward J. Rubin, 42–51.
Somerville, MA: Cascadilla Proceedings Project.
Anderson, Skye, and Ryan Walter Smith. 2018. Plural reduplication in Tohono O’odham: An analysis in
Harmonic Serialism. In Proceedings of the Workshop on the Structure and Constituency of Languages
of the Americas 21, ed. by Megan Keough, Natalie Weber, Andrei Anghelescu, Sihwei Chen, Erin
Guntly, Khia Johnson, Daniel Reisinger, and Oksana Tkachman, 17–29. Vancouver, BC: University
of British Columbia, University of British Columbia Working Papers in Linguistics.
836 ANDREW LAMONT

Bamgbo⋅ se, Ayo⋅ . 1966. Nominal classes in Mbe. Afrika und Übersee 49:32–53.
Bamgbo⋅ se, Ayo⋅ . 1967a. Notes on the phonology of Mbe. Journal of West African Languages 4:5–11.
Bamgbo⋅ se, Ayo⋅ . 1967b. Tense/Aspect forms in Mbe. Research Notes, Department of Linguistics and Nige-
rian Languages, University of Ibadan 1:12–20.
Bamgbo⋅ se, Ayo⋅ . 1967c. Verbal classes in Mbe. Afrika und Übersee 50:173–193.
Bamgbo⋅ se, Ayo⋅ . 1971. Nasal harmony in Mbe. Annales de l’Université d’Abidjan 1971, Série H, Fascicule
hors série 1:101–107.
Becker, Michael, and Kathryn Flack Potts. 2011. The emergence of the unmarked. In The Blackwell compan-
ion to phonology, ed. by Marc van Oostendorp, Colin Ewen, Elizabeth Hume, and Keren Rice,
1363–1379. Malden, MA: Wiley-Blackwell.
Blevins, Juliette. 2003. A note on reduplication in Bugotu and Cheke Holo. Oceanic Linguistics 42:499–505.
Blevins, Juliette. 2005. The role of phonological predictability in sound change: Privileged reduction in
Oceanic reduplicated substrings. Oceanic Linguistics 44:517–526.
Buckley, Eugene. 2011. Metathesis. In The Blackwell companion to phonology, ed. by Marc van Oostendorp,
Colin Ewen, Elizabeth Hume, and Keren Rice, 1380–1407. Malden, MA: Wiley-Blackwell.
de Lacy, Paul. 2002. The formal expression of markedness. Doctoral dissertation, University of Massachusetts
Amherst.
de Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge Univer-
sity Press.
Downing, Laura J. 2005. The emergence of the marked: Tone in some African reduplicative systems. In
Studies on reduplication, ed. by Bernhard Hurch, 89–110. Berlin: De Gruyter Mouton.
Elfner, Emily. 2009. Syllabification and stress-epenthesis interactions in Harmonic Serialism. Ms., University
of Massachusetts Amherst. http://roa.rutgers.edu/article/view/1077.
Elfner, Emily, and Wendell Kimper. 2008. Reduplication without RED: Evidence from diddly-infixation.
In WCCFL 27, ed. by Natasha Abner and Jason Bishop, 150–158. Somerville, MA: Cascadilla
Proceedings Project.
Glewwe, Eleanor. 2016. Logoori hiatus resolution: A new analysis. Poster presented at Annual Conference
on African Linguistics 47. https://erglewwe.files.wordpress.com/2015/08/acal-47-poster-3.pdf.
Glewwe, Eleanor, and Ann M. Aly. 2016. Coronal palatalization in Logoori. In Diversity in African lan-
guages, ed. by Doris L. Payne, Sara Pacchiarotti, and Mokaya Bosire, 165–183. Berlin: Language
Science Press.
Gouskova, Maria. 2003. Deriving economy: Syncope in Optimality Theory. Doctoral dissertation, University
of Massachusetts Amherst.
Heinz, Jeffrey. 2005. Reconsidering linearity: Evidence from CV metathesis. In WCCFL 24, ed. by John
Alderete, Chung-hye Han, and Alexei Kochetov, 200–208. Somerville, MA: Cascadilla Proceedings
Project.
Inkelas, Sharon, and Cheryl Zoll. 2005. Reduplication: Doubling in morphology. Cambridge: Cambridge
University Press.
Jacobs, Haike. 2019. Cross-level interactions in Latin: Vowel shortening, vowel deletion and vowel gliding.
Catalan Journal of Linguistics 18:79–103.
Kawu, Ahmadu Ndanusa. 2000. Structural markedness and nonreduplicative copying. In NELS 30, ed. by
Masako Hirotani, Andries Coetzee, Nancy Hall, and Ji-yung Kim, 1:377–388. Amherst: University
of Massachusetts, Graduate Linguistics Students Association.
Lamont, Andrew. 2021. Mayo reduplication as a feeding Duke of York derivation. Virtual poster presented
at Linguistic Society of America 2021. https://aphonologist.github.io/presentations.
Lamont, Andrew. To appear. A restrictive, parsimonious theory of footing in directional Harmonic Serialism.
Phonology.
Leung, Elizabeth Woon-Yee. 1991. The tonal phonology of Llogoori: A study of Llogoori verbs. Ithaca,
NY: Cornell University, Department of Modern Languages and Linguistics, Phonetics Laboratory.
 SERIAL REDUPLICATION 837

Lin, Kuo-Chiao. 2016. The gradual path to variable reduplication: Kavalan. In AFLA 22, ed. by Henrison
Hsieh, 82–97. Canberra: Asia-Pacific Linguistics. https://openresearch-repository.anu.edu.au/handle
/1885/101155.
Lin, Kuo-Chiao. 2021. Some notes on derivational lookahead in Mbe reduplication. Ms., Kang Chiao Interna-
tional School. https://ling.auf.net/lingbuzz/005837.
Lombardi, Linda. 1999. Positional faithfulness and voicing assimilation in Optimality Theory. Natural
Language and Linguistic Theory 17:267–302.
Lombardi, Linda. 2001. Why place and voice are different: Constraint-specific alternations in Optimality
Theory. In Segmental phonology in Optimality Theory: Constraints and representations, ed. by Linda
Lombardi, 13–45. Cambridge: Cambridge University Press.
McCarthy, John J. 2000. Harmonic Serialism and Parallelism. In NELS 30, ed. by Masako Hirotani, Andries
Coetzee, Nancy Hall, and Ji-yung Kim, 2:501–524. Amherst: University of Massachusetts, Graduate
Linguistics Students Association.
McCarthy, John J. 2006. Restraint of analysis. In Wondering at the natural fecundity of things: Essays in
honor of Alan Prince, ed. by Eric Baković, Junko Ito, and John J. McCarthy, 195–219. Santa Cruz:
University of California, Linguistics Research Center.
McCarthy, John J. 2007. Slouching towards optimality: Coda reduction in OT-CC. In Phonological studies
10, ed. by Phonological Society of Japan, 89–104. Tokyo: Kaitakusha.
McCarthy, John J. 2008a. The gradual path to cluster simplification. Phonology 25:271–319.
McCarthy, John J. 2008b. Restraint of analysis. In Freedom of analysis?, ed. by Sylvia Blaho and Patrik
Bye, 203–231. Berlin: De Gruyter.
McCarthy, John J. 2008c. The serial interaction of stress and syncope. Natural Language and Linguistic
Theory 26:499–546.
McCarthy, John J. 2010. An introduction to Harmonic Serialism. Language and Linguistics Compass 4:
1001–1018.
McCarthy, John J. 2016. The theory and practice of Harmonic Serialism. In Harmonic Grammar and
Harmonic Serialism, ed. by John J. McCarthy and Joe Pater, 47–87. Sheffield: Equinox.
McCarthy, John J. 2019. How to delete. In Perspectives on Arabic linguistics XXX, ed. by Amel Khalfaoui
and Matthew Tucker, 7–32. Amsterdam: John Benjamins.
McCarthy, John J., Wendell Kimper, and Kevin Mullin. 2012. Reduplication in Harmonic Serialism. Mor-
phology 22:173–232.
McCarthy, John J., and Alan Prince. 1994. The emergence of the unmarked: Optimality in Prosodic Morphol-
ogy. In NELS 24, ed. by Mercè Gonzàlez, 2:333–379. Amherst: University of Massachusetts, Gradu-
ate Linguistics Students Association.
McCarthy, John J., and Alan Prince. 1995. Faithfulness and reduplicative identity. In Papers in Optimality
Theory, ed. by Jill Beckman, Suzanne Urbanczyk, and Laura Walsh Dickey, 249–384. Amherst:
University of Massachusetts, Graduate Linguistics Students Association.
McCarthy, John J., and Alan Prince. 1999. Faithfulness and identity in prosodic morphology. In The prosody-
morphology interface, ed. by René Kager, Harry van der Hulst, and Wim Zonneveld, 218–309.
Cambridge: Cambridge University Press.
Moore-Cantwell, Claire. 2016. Contexts for epenthesis in Harmonic Serialism. In Harmonic Grammar and
Harmonic Serialism, ed. by John J. McCarthy and Joe Pater, 236–260. Sheffield: Equinox.
Odden, David. 2019. Geminate blockage in Logoori harmony with no added machinery. Loquens 6(2), e064.
https://doi.org/10.3989/loquens.2019.064.
Odden, David. n.d. Introduction to Logoori phonology and morphology. Ms. https://languagedescriptions
.github.io/Logoori/.
Pater, Joe. 2007. The locus of exceptionality: Morpheme-specific phonology as constraint indexation. In
Papers in Optimality Theory III, ed. by Leah Bateman, Michael O’Keefe, Ehren Reilly, and Adam
838 ANDREW LAMONT

Werle, 259–296. University of Massachusetts Occasional Papers 32. Amherst: University of Massa-
chusetts, Graduate Linguistics Students Association.
Pater, Joe. 2010. Morpheme-specific phonology: Constraint indexation and inconsistency resolution. In Pho-
nological argumentation: Essays on evidence and motivation, ed. by Steve Parker, 123–154. Lon-
don: Equinox.
Pater, Joe. 2012. Serial Harmonic Grammar and Berber syllabification. In Prosody matters: Essays in honor
of Elisabeth Selkirk, ed. by Toni Borowsky, Shigeto Kawahara, Mariko Sugahara, and Takahito
Shinya, 43–72. Sheffield: Equinox.
Prince, Alan, and Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar.
Technical report RuCCs-TR2, CU-CS-696-93, Rutgers University and University of Colorado at
Boulder. Published, Malden, MA: Blackwell (2004).
Pruitt, Kathryn. 2010. Serialism and locality in constraint-based metrical parsing. Phonology 27:481–526.
Pruitt, Kathryn. 2012. Stress in Harmonic Serialism. Doctoral dissertation, University of Massachusetts Am-
herst.
Rosenthall, Sam. 1994. Vowel/Glide alternation in a theory of constraint interaction. Doctoral dissertation,
University of Massachusetts Amherst.
Selkirk, Elisabeth. 1995. The prosodic structure of function words. In Papers in Optimality Theory, ed. by
Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk, 439–470. Amherst: University of
Massachusetts, Graduate Linguistics Students Association.
Somerday, Megan. 2015. (Some) partial reduplication is full reduplication. In NELS 45, ed. by Thuy Bui
and Deniz Özyãldãz, 3:79–92. Amherst: University of Massachusetts, Graduate Linguistics Students
Association.
Spaelti, Philip. 1997. Dimensions of variation in multi-pattern reduplication. Doctoral dissertation, Univer-
sity of California, Santa Cruz.
Stanton, Juliet. 2020. Gurindji nasal cluster dissimilation as trigger deletion. Journal of Linguistics 56:
157–195.
Takahashi, Chikako. 2018. No metathesis in Harmonic Serialism. In Supplemental proceedings of the 2017
Annual Meeting on Phonology, ed. by Gillian Gallagher, Maria Gouskova, and Sora Heng Yin.
Washington, DC: Linguistic Society of America. https://journals.linguisticsociety.org/proceedings
/index.php/amphonology/article/view/4232/3905.
Takahashi, Chikako. 2019. No transposition in Harmonic Serialism. Phonology 36:695–726.
Torres-Tamarit, Francesc. 2012. Syllabification and opacity in Harmonic Serialism. Doctoral dissertation,
Centre de Lingüı́stica Teòrica.
Torres-Tamarit, Francesc. 2016. Compensatory and opaque vowel lengthening in Harmonic Serialism. In
Harmonic Grammar and Harmonic Serialism, ed. by John J. McCarthy and Joe Pater, 301–326.
Sheffield: Equinox.
Velleman, Shelley L., and Marilyn M. Vihman. 2002. The emergence of the marked unfaithful. In Papers
in Optimality Theory II, ed. by Angela C. Carpenter, Andries W. Coetzee, and Paul de Lacy, 397–419.
University of Massachusetts Occasional Papers 26. Amherst: University of Massachusetts, Graduate
Linguistics Students Association.
Walker, Rachel. 1998a. Minimizing RED: Nasal copy in Mbe. Ms., University of Southern California. http://
roa.rutgers.edu/article/view/274.
Walker, Rachel. 1998b. Nasalization, neutral segments, and opacity effects. Doctoral dissertation, University
of California, Santa Cruz. https://rucore.libraries.rutgers.edu/rutgers-lib/38466/.
Walker, Rachel. 2000. Nasal reduplication in Mbe affixation. Phonology 17:65–115.
Wei, Wei. 2018. Lookahead effect in Mbe reduplication. In Supplemental proceedings of the 2017 Annual
Meeting on Phonology, ed. by Gillian Gallagher, Maria Gouskova, and Sora Heng Yin. Washington,
DC: Linguistic Society of America. https://journals.linguisticsociety.org/proceedings/index.php/am
phonology/article/view/4243/3900.
 SERIAL REDUPLICATION 839

Wei, Wei, and Rachel Walker. 2018. Lookahead effects in the reduplication-phonology interface. In NELS
48, ed. by Sherry Hucklebridge and Max Nelson, 3:183–196. Amherst: University of Massachusetts,
Graduate Linguistics Students Association.
Wei, Wei, and Rachel Walker. 2020. A lookahead effect in Mbe reduplication: Implications for Harmonic
Serialism. Linguistic Inquiry 51:845–859.
Zukoff, Sam. 2017. Actually, Serial Template Satisfaction does predict medial coda skipping in reduplication.
In Supplemental proceedings of the 2016 Annual Meeting on Phonology, ed. by Karen Jesney, Charlie
O’Hara, Caitlin Smith, and Rachel Walker. Washington, DC: Linguistic Society of America. https://
journals.linguisticsociety.org/proceedings/index.php/amphonology/article/view/3983/3709.
Zuraw, Kie. 2002. Aggressive reduplication. Phonology 19:395–439.
Zymet, Jesse. 2018. A case for parallelism: Reduplication-repair interaction in Maragoli. In UC Berkeley
PhonLab Annual Report 14:302–342. https://escholarship.org/uc/item/9qg061pf.

Andrew Lamont
Department of Linguistics
University of Massachusetts Amherst
[email protected]