On the Origins of Attic Reduplication*
Sam Zukoff
[email protected]
10/25/13
1. Introduction


In Ancient Greek, the default pattern of reduplication in the perfect tense is
o a CV reduplicant for consonant-initial roots:
κρίνω → κέκριμαι [ke-kri-mai]
o vowel lengthening for vowel-initial roots:
ἀγω → ἦχα [ kha]
For a subset of the (synchronically) vowel-initial roots, there is a different pattern, known
as Attic Reduplication (AR).
o AR forms show a reduplicant of the shape VC with concomitant lengthening of
the root-initial vowel.
(1) Some Attic Reduplication perfects and likely etymologies (van de Laar, 2000)1
Root (Greek < *PIE)
#*h1
(en-)eŋk < *h1nek
eleuth < *h1lewdh
#*h2
ager < *h2ger
ar < *h2er
#*h3
od < *h3ed
or < *h3er

Present Tense
‘bring’
‘g , c me’
n/a
n/a
Perfect Tense
ἐνήνοχα
ἐλήλουθα
[en n kh-]
h
[e
-]
‘ga her ge her’
‘j in, fi ge her’
ἀγείρω
ἀραρίσκω
[agēr-]
[arar-]
ἀγήγερμαι
ἄρηρα
[ag ger-]
[ar r-]
‘sme ’
‘inci e’
ὄζω
ὄρνυμι
[ozd-]
[or-]
ὄδωδα
ὄρωρα
[
-]
[ r r-]
The literature agrees that this pattern must in some way be tied to laryngeals (Winter,
1950; Beekes, 1969; Suzuki, 1994; Sihler, 1995: §444; Keydana, 2006, 2010).
o Virtually all AR forms are built to roots which can be securely reconstructed with
an initial laryngeal.2
o Only non-high-vowel initial roots participate in this pattern (implying necessary
connection with laryngeal coloration/vocalization).
*
Special thanks to Donca Steriade, Adam Albright, Jared Klein, Anya Lunden, and Brent Vine, and also to
audiences at MIT, Harvard, and UCLA. All mistakes are, of course, mine.
1
For a more exhaustive list of AR perfects, consult the Appendix (§8.3).
2
Only 2 of at least 20 AR roots, as enumerated in the appendix, are definitively not laryngeal-initial: (i) √or ‘keep
wa ch’ < *ser : perfect ὄρωρα [ r r-], which has clearly been attracted to the perfect of √or ‘inci e’ < *h3er, which is
also ὄρωρα; and (ii) √eme ‘v mi ’ < *wemh1 : perfect ἐμήμεκα [em me-], which after the loss of /*w/, has a root
shape equivalent to many other AR forms.
1
WeCIEC XXV, 10/25/13

Sam Zukoff
However, invoking laryngeals is not on its own sufficient to explain the phenomenon.
o Given that the default reduplication pattern is CV, we would expect laryngealinitial roots to reduplicate as HV-H(C)VC-.
o E.g., √*h2ger : perfect *h2e-h2ger- > Attic-Ionic **ἠγερ- [ ger-]

This paper will show that the Attic Reduplication pattern can be (independently)
motivated as the outcome of the regular phonology of a Pre-Greek system that still
contained laryngeals.
o This will be done by intersecting constraint rankings (framed within Optimality
Theory (OT; Prince & Smolensky, 1993)) that are independently necessary to
explain the default reduplication pattern of (Pre-)Greek and the phenomenon of
vowel prothesis.

These rankings will ultimately result in cluster-copying and reduplicant-internal
epenthesis exclusively for laryngeal-initial roots, yielding preforms as in (2).
(2) AR and non-AR pre-forms:
 Attic Reduplication pre-forms: *HVCV-HCVC- or *HVCV-HC Normal reduplication pre-forms: *CV-C(C)VC2. Motivating the prothetic vowel

In Greek, Armenian, and Phrygian, PIE word-initial *hC sequences ultimately surface as
the sequence VC. In Greek, the quality of the vowel corresponds to the quality of the
laryngeal.
o For example, Greek ἀνήρ [an r] ‘man.NOM.SG’ < PIE *h2 r (cf. Skt nar- )

This development can be motivated by a constraint militating against a sequence of
(consonantal) laryngeal + consonant in the onset of a syllable.
(3) *σ[HC – D n’ have a sequence of laryngeal + consonant at the left-edge of a syllable.

This is a constraint that would not have been present in the grammar of Proto-IndoEuropean, but becomes highly ranked at some subsequent point in Pre-Greek.

Given the proper ranking with respect to other constraints, *σ[HC will induce “ aryngea
v ca iza i n” in word-initial position.
2
WeCIEC XXV, 10/25/13

Sam Zukoff
I will be following the view in which laryngeal vocalization is not seen as direct
vocalization of the consonantal segment, but rather as epenthesis of a minimal vowel
(schwa) adjacent to the laryngeal.3
o /ph2tr-es/ ‘fa her-GEN.SG’ → [pah2tres] or [ph2atres], n

[ph2tr-es]
Since we are viewing it as epenthesis, the constraint violated is DEP-IO:
(4) DEP-IO – Segments in the output must have a corresponding segment in the input.
≈ Don’t Epenthesize!

Using the *σ[HC constraint, and ranking it with respect to DEP-IO and another basic
faithfulness constraint, MAX-IO, we will be able to represent the development of the
prothetic vowel:
(5) MAX-IO – Segments in the input must have a corresponding segment in the output.
≈ Don’t Delete!
(6) Critical Ranking: *σ[HC, MAX-IO » DEP-IO4
*σ[HC
MAX-IO
DEP-IO
(7) V we “pr hesis”: PIE *h2ger-yō >> Greek ἀγείρω [agēr-ō] ‘I ga her-PRES.1SG’
 By hypothesis: Pre-Greek /h2ger-yō/ → [h2ageryō]5
/h2ger, yō/
a.
b.
c.
d.
h2geryō
geryō
h2eryō
 h2ageryō
*σ[HC MAX-IO DEP-IO
*!
*!
*!
*
3
Cf. Fortson (2004). See Byrd (2010, 2011) for recent accounts that make use of this view of laryngeal vocalization
in an OT framework.
4
This ranking also appears to be largely consistent with patterns of laryngeal vocalization word-internally,
especially in the *-VCHCV- environment. We might reasonably expect the post-consonantal *HC sequence to form
a complex onset (at the very least when C is a resonant); but instead, we do see vocalization: e.g., Pre-Greek
*/h2onh1-mos/ → *[h2on.h1e.mos] > Greek ἄνεμος [anem s] ‘brea h’; Pre-Greek */ e 1- ōr/ → *[ e 1e ōr] >
Greek γενέ ωρ [gene r] ‘bege er’ (examples from Sihler, 1995). See Byrd (2010) for a wider discussion of
laryngeal behavior in medial position in Greek and PIE.
5
Take the [a] to be a laryngeally-colored [ə]. This notation will be used throughout.
3
WeCIEC XXV, 10/25/13

Sam Zukoff
Here we see that the underlying h2 + g sequence, if it were to surface faithfully, would
violate our markedness constraint against laryngeal + consonant sequences at the
beginning of a syllable. Thus, *σ[HC rules out candidate (a).

The viable options for how to fix this problem are essentially deletion or epenthesis.

By having MAX » DEP, we are saying that epenthesis is preferable to deletion.
o This ranking chooses candidate (d) over (b) and (c).
 Candidate (d) straightforwardly derives into Greek as ἀγείρω [agērō], the attested form.
3. Reduplication in Greek
3.1. The reduplicative morpheme

Outside of Attic Reduplication forms, we never see any variation in the quality of the
reduplicative vowel in Greek.
o It is always [e] in the perfect and the aorist, and [i] in the present.
o This contrasts with Sanskrit, where, at least in certain contexts/categories, we do
see copying of a root [i,u].

Therefore, we should view the reduplicative vowel as deriving from the underlying form
of the reduplicative morpheme, not by copying.

I will represent the reduplicative morpheme for the perfect tense as /REDe/.6
o This is meant to show that there is a morphologically fixed [e] which is preceded
by a string of reduplicated segments.
3.2. Deriving normal reduplication

The normal pattern is illustrated by the data in (8).
(8) Normal reduplication in Ancient Greek
 Roots with initial singleton consonants
Root
Present Tense
Perfect Tense
d - ‘give’
δίδωμι [di-d -]
δέδωκα
[de-d -]
lu- ‘
λύω
λέλυκα
[le-lu-]
sen’
pemp-‘sen ’
6
[lu-]
πέμπω [pemp-]
For the reduplicated present, it would be /REDi/.
4
πέπεμπ αι [pe-pemp-]
WeCIEC XXV, 10/25/13
Sam Zukoff
 Roots with initial obstruent + resonant clusters
Root
kri(n)- ‘ eci e’
Present Tense
Perfect Tense
κρίνω [krī-n-]
κέκριμαι
[ke-kri-]
λάω [tla-]
έ ληκα
[te-tl -]
tla-/tl - ‘s ffer, are’
pneu- ‘brea he’
πνέω [pne-]
πέπνυμαι [pe-pnū-]
3.2.1. Generalizations and the necessary constraints7

The reduplicant is always the left-most element in the word.8
ALIGN-REDe-L

The default pattern for reduplication is CV.
ALIGN-ROOT-L

The initial segment of the reduplicant always corresponds to

the initial segment of the root.
ANCHOR-L-BR
The reduplicating syllable always has an onset consonant.
ONSET
3.2.2. Constraint definitions
(9) ALIGN-REDe-L– The left-edge of the reduplicant should coincide with the left-edge of the
prosodic word. Assign one violation mark * for each segment which intervenes between
the left-edge of the prosodic word and the left-edge of the reduplicant.
(10) ALIGN-ROOT-L– The left-edge of the root should coincide with the left-edge of the prosodic
word. Assign one violation mark * for each segment which intervenes between the leftedge of the prosodic word and the left-edge of the root.
(11) ANCHOR-L-BR – The segment at the left-edge of the reduplicant corresponds with the
segment at the left-edge of the base.
(12) ONSET – All syllables must have an onset.
3.2.3. Constraint ranking for reduplication in Ancient Greek
ANCHOR-L-BR
ONSET
ALIGN-REDe-L
ALIGN-ROOT-L
7
For a more thorough analysis of the basic pattern of reduplication in Ancient Greek, see Zukoff (2012: ch.2).
The reduplicant can be preceded by preverbs, and also the augment in the pluperfect and reduplicated aorist. This
order of morphemes can straightforwardly be achieved through higher ranking alignment constraints.
8
5
WeCIEC XXV, 10/25/13
Sam Zukoff
3.2.4. Reduplication in CRV roots
(13) Normal reduplication: κέκριμαι [ke.-kri-] ‘I have (been) j
/REDe, kri, mai/
a.  ke.-kri.-mai
b. kre.-kri.-mai
c. _e.-kri.-mai
d. re.-kri.-mai
e. kri.-ke.-mai

Can i a e (e), which
ALIGN-REDe-L
ge ’
ANCHOR-L-BR
ONSET
*!
*!
ALIGN-ROOT-L
k,e
k,r,e!
e
r,e
k!,r,i
esn’ have i s re p ican a he ef -edge of the word, is ruled out
by ALIGN-REDe-L.

Candidate (d), which has copied a consonant from non-root-initial position, is ruled out
by ANCHOR-L-BR.

Can i a e (c), which has ef he re p ican ’s fixe /e/ wi h
an nse , is r e
by
ONSET.

Lastly, the choice between (a) and (b) is made by ALIGN-ROOT-L.
o Candidate (a) has two segments preceding the root, but candidate (b) has three.
o By copying the [r] of the root, (b) has incurred a gratuitous violation of
ALIGN-ROOT-L, and is thus ruled out in favor of (a).
 In §4, we will see how this normal application of the reduplicative grammar goes wrong
in cases of Attic Reduplication, and comes to create an altogether different pattern.
4. Analysis of Attic Reduplication

The roots for which the AR derivation will be most straightforward are the *HCeC roots,
as the important interaction (between H and C1) is not dependent on ablaut.
o In order for the proper results to be derived from *HeC roots, we will be restricted
to using zero-grade forms (as discussed in the Appendix (§8.2)).

The exemplar *HCeC root will be √ager ‘ga her
ge her’ < *h2ger, which has a perfect
ἀγήγερμαι [a.g .ger.mai].

Ultimately, the form that our Pre-Greek grammar will produce is [h2a.ga-h2.ger.-mai].
o This form copies both members of the root-initial cluster, with an epenthetic
vowel between the copied segments in the reduplicant.
o This form hetero-syllabifies the root-initial cluster.
6
WeCIEC XXV, 10/25/13
Sam Zukoff
4.1. The state of affairs in PIE and very early Pre-Greek

Pre-Greek has inherited a grammar that produces a pattern of reduplication with
essentially the same properties as were displayed in Proto-Indo-European.
o The basic pattern of perfect-tense reduplication is a Ce- reduplicant.
o For example, the winning candidate in (14) was [ke.-kri.-mai].

Prior to the rise of *σ[HC (i.e., in the grammars of PIE and early Pre-Greek), there was
nothing to differentiate the treatment of laryngeal + consonant root-initial clusters from
all other root-initial clusters.

Applying this pattern to
r examp e r
√*h2ger, these grammars select a candidate
[h2a.h2ger.mai], which would have derived into Attic-Ionic as ** germai.
o It copies just C1, with a fixed vowel which has been colored to a.
o It has tauto-syllabified the root-initial cluster.9
4.2. Markedness forces a repair!

At some point in time, Pre-Greek deems the sequence laryngeal + consonant at the leftedge of a syllable to be unacceptable.
o The constraint *σ[HC rises to the top of the ranking.

This creates a problem for laryngeal-initial roots in reduplication.
o The output which conforms to the default pattern, [h2a.h2ger-], which was the
optimal form in the earlier grammars, fatally violates *σ[HC.
o This forces a repair.

What we might expect to be the simplest repair for a *σ[HC violation is just to syllabify
the laryngeal as a coda.
o This candidate would be [h2ah2.ger-].

This is not the form we get, but there are currently no constraints in our rankings that will
penalize this candidate.

I propose that there is a constraint that targets exactly these sorts of sequences:
(14) OCP-SYLLABLE (OCP-σ) – D n’ have i en ica segmen s within the same syllable.

This constraint will eliminate the candidate [h2ah2.ger-], because it has two h2’s in he
reduplicating syllable.
9
It is possible that the output form in PIE may have hetero-syllabified the cluster.
7
WeCIEC XXV, 10/25/13
Sam Zukoff
 Due to the application of *σ[HC and OCP-σ in Pre-Greek, the default Ce- reduplication
process is blocked for *HCeC roots.
(15) Ruling out a Ce- reduplicant with *σ[HC and OCP-σ:10
/REDe, h2ger-/
a.
b.
c.
d.

h2a.-h2gerh2a-h2.ger h2a.ga-h2.gerh2a.ga.-h2ger-
*σ[HC OCP-σ DEP-V-IO ALIGN-ROOT-L
*!
*!
*
*
*!
**
**
****
****
With the normal Ce- reduplicant candidates blocked, an alternative copying pattern must
take over.

The characteristics of this alternative pattern will be determined by the ranking of the
lower-ranked constraints.
4.3. The alternative pattern

We have now ruled out the default pattern, but there are still a number of reasonable
possibilities of how the *σ[HC and OCP-σ pr b ems migh be av i e .

These possibilities largely coincide with those repairs modulated by the constraints
introduced independently in §2 and §3 in order to account for vowel prothesis and the
basic reduplication pattern.
(16) Potential repairs and their associated constraints

 Epenthesis: (a) [h2a.ga-h2.ger-]11
DEP-IO
 Unfilled onset: (b) [a-h2.ger-]
ONSET
 Improper anchoring: (c) [ga-h2.ger-]
ANCHOR-L-BR
 Deletion of a root consonant: (d) [ge.-ger-]
(e) [h2a.-h2ar-]
MAX-IO
As alluded to previously, the attested repair (i.e. the actual pre-form) is the epenthesis
repair candidate (a) [h2a.ga-h2.ger-].

All that is needed for our grammar to select this candidate is for DEP-IO (and also ALIGNROOT-L) to be crucially dominated by the three constraints militating against the other
repairs.
10
S perscrip e v we s are epen he ic, pres me
be [ə]’s which have been c re by a jacen aryngea s.
This candidate displays reduplicant-internal epenthesis. This will be contrasted with root-internal epenthesis
below.
11
8
WeCIEC XXV, 10/25/13
Sam Zukoff
(17) The alternative repair: low-ranking DEP-IO and ALIGN-ROOT-L
MAX-IO
/REDe, h2ger-/
a.
b.
c.
d.
e.

ANCHOR-L-BR
ONSET
DEP-IO
ALIGN-ROOT-L
*
****
*
**
**
**
a
 h2 .ga-h2.gera-h2.gerga-h2.gerge.-gerh2a.-h2ar-
*!
*!
*!
*!
It appears that this ranking may, in a certain sense, follow from the critical rankings
developed in §2 and §3, which are repeated here in (18):
(18) Critical rankings for vowel prothesis and default reduplication
Vowel Prothesis
MAX-IO
Default Reduplication
*σ[HC
ANCHOR-L-BR
DEP-IO

ONSET
ALIGN-REDe-L
ALIGN-ROOT-L
In the evaluation in the tableau in (17), the only constraints which receive less than
maximal satisfaction are DEP-IO and ALIGN-ROOT-L.
 These are exactly the two constraints which are crucially dominated – and thus can be
violated – in the two processes described by the grammar in (18), namely vowel prothesis
and default reduplication.

Therefore, by merging the rankings in (18) – and adding OCP-σ
he
p f he rankings
– we have a grammar that generates exactly the attested repair for laryngeal-initial roots.
(19) Total ranking for AR grammar
MAX-IO
*σ[HC
OCP-σ
DEP-IO
ANCHOR-L-BR
ALIGN-ROOT-L
9
ONSET
ALIGN-REDe-L
WeCIEC XXV, 10/25/13
Sam Zukoff
4.4. Epenthesis in the optimal alternative reduplication strategy

We have now shown that our independently established constraint ranking selects
epenthesis as the optimal alternative reduplication strategy to avoid *σ[HC and OCP-σ
violations.

There are actually two good options for where to epenthesize:
o Reduplicant-internal: [h2a.ga-h2.ger-]
o Root-internal:

a
[h2a.-h2 .ger-]
> Attic-Ionic ag ger-

> Attic-Ionic ** ger-

Root-internal epenthesis would actually allow for retention of a normal Ce- reduplicant,
but this is clearly not the attested outcome.

The choice of reduplicant-internal epenthesis can be modeled using CONTIGUITY
constraints.

CONTIGUITY constraints demand that adjacency relationships are maintained across
correspondence dimensions.

Since we are dealing with two correspondence dimensions, we can invoke two separate
constraints:
(20) CONTIGUITY-IO – Segments which are adjacent in the input must be adjacent in the output.
(21) CONTIGUITY-BR – Segments which are adjacent in the base must be adjacent in the
reduplicant.

Since the reduplicant is not directly subject to Input-Output faithfulness, it will be exempt
from CONTIG-IO violations.

Therefore, if we rank CONTIG-IO over CONTIG-BR, our grammar will prefer reduplicantinternal epenthesis to root-internal epenthesis.
(22) The epenthesis candidates:
/REDe, h2ger-/
CONTIG-IO CONTIG-BR ALIGN-ROOT-L
a
a.  h2 .ga-h2.gerb. h2a.-h2a.ger
c. h2a.ga.-h2a.ger
*
*!
*!
****
**
****
CONTIG-IO must crucially dominate both CONTIG-BR and ALIGN-ROOT-L, as the winning
candidate (a) does worse on both of these constraints in order to satisfy CONTIG-IO.
10
WeCIEC XXV, 10/25/13
Sam Zukoff
(23) Critical Ranking: CONTIG-IO » CONTIG-BR, ALIGN-ROOT-L12
(24) Total ranking for AR grammar with CONTIGUITY
MAX-IO
*σ[HC
OCP-σ
ANCHOR-L-BR
ONSET ALIGN-REDe-L
CONTIGUITY-IO
DEP-IO
ALIGN-ROOT-L
CONTIGUITY-BR
4.5. Interim conclusions

In this section, we have seen that the high ranking of the constraints *σ[HC and OCP-σ in
Pre-Greek make it impossible for laryngeal-initial roots to reduplicate according to the
default reduplication pattern of the language

The constraint ranking, which is in large part independently necessary for vowel
prothesis and the normal reduplication pattern, ultimately selects reduplicant-internal
epenthesis as the optimal alternative reduplication pattern.
 A complete tableau for the candidates discussed in this section, plus a few additional
candidates, is included in the Appendix (§8.1).
5. The Obligatory Contour Principle and OCP-SYLLABLE

In §4, I introduced the constraint OCP-σ, repeated below, in order to rule out candidates
like [h2ah2.ger-], schematically: HV-H.CVC-.
(25) OCP-SYLLABLE (OCP-σ) – D n’ have i en ica segmen s wi hin he same sy ab e.

This is meant to be a (context-sensitive) version of the Obligatory Contour Principle
(OCP; Goldsmith, 1976).

The OCP dictates that elements within a given domain be different, and is typically used
to ban identical adjacent segments.
12
The use of CONTIGUITY-IO affects our analysis of the prothetic vowel, as our presumed optimal candidate
(/h2ger-yō/ → [h2əger-yō]) violates this constraint. In order to rule out the epenthesis candidate which fares better
on CONTIG-IO ([əh2.ger-yō]), all we need to do is critically rank ONSET above CONTIG-IO, meaning it is worse to
create an onsetless syllable than have imperfect contiguity. (We could also use a constraint like ANCHOR-L-IO to
achieve this result.)
11
WeCIEC XXV, 10/25/13
Sam Zukoff
o Byrd (2010) shows conclusively that this standard version of the OCP is at work
in PIE in banning hetero-morphemic geminates, playing a part in such phenomena
as he D b e Den a Law, Szemerenyi’s Law, an

he metron rule.
OCP-σ would be a more specific version of this constraint whose domain of application
is within the syllable.13
o Simi ar c ns rain s a e back a eas
Vennemann’s (1988) “She Law”.
5.1. OCP-SYLLABLE and the long-vowel preterites & perfects

We migh c nnec
his c ns rain wi h he “ ng-v we pre eri es/perfec s” seen in
numerous IE branches. Consider the following:
o Sanskri √pa ‘g ’: perfect pē r ( = pa-y.d-ur < *pa-p.d-ur)
o Gothic √gib ‘give’ : preterite gēbun ( < *geg.bun)
o O

Irish √gan ‘be b rn’: preterite gḗnair ( < *geg.nair)
In these cases, if reduplication was carried out in the otherwise expected manner, there
would be identical consonants within the reduplicating syllable.

Instead of allowing this to surface, one of the consonants is lenited (as in Sanskrit) or
deleted with compensatory lengthening (as in Gothic and Old Irish).

We may well be able to view this as a repair for an OCP-σ violation.
5.2. OCP- SYLLABLE and non-copying perfects in Ancient Greek

OCP-σ may have caused similar problems in Ancient Greek, but with a different
resolution.

As we’ve seen, the default reduplication pattern for consonant-initial roots in Ancient
Greek is to have a [Ce-] reduplicant.

But for roots with initial clusters which do not have rising sonority (i.e. all clusters which
are not stop + resonant clusters), the perfect displays a reduplicant in [e-].
o F r examp e, √kten- ‘ki ’ : perfect ἔκ ονα [e-kton-a] not **[ke-kton-a]
√re(w)- ‘f w’ : perfect ἐρρύηκα [e-rru- ka] not **[re-rru- ka]

If we assume that non-rising-sonority clusters had to be hetero-syllabified (due to highranking SONORITY SEQUENCE), but rising-sonority clusters could form themselves as
complex onsets, then OCP-σ would cause differential treatment of the two cluster types:14
13
We migh f r her refine his
say “f anking a v we wi hin a sy ab e.”
12
WeCIEC XXV, 10/25/13

Sam Zukoff
o [ke.kri.mai]

o [kek.to.na]

→
[ek.to.na]

o [rer.ru. .ka]

→
[er.ru. .ka]

By not copying the root-initial consonant, they escape the OCP-σ violation.
5.3. OCP(-SYLLABLE?) in the lexicon

We can notice that very few if any roots in the PIE lexicon have two identical segments.

While the domain for this effect need not necessarily be identified as the syllable, it
definitely points to some version of the OCP being active at the lexical level.15
6. Conclusions

The pattern of Attic Reduplication can be viewed as the outcome of the regular
phonology of a Pre-Greek system that still contained laryngeals.
o This pattern is induced by a new dispreference for the configuration of laryngeal
+ consonant at the left-edge of a syllable, as instantiated by the *σ[HC constraint.
o The interaction of this constraint with another markedness constraint – OCP-σ –
makes it impossible for laryngeal-initial roots to reduplicate according to the
default reduplication pattern of the language.

Contrary to previous accounts, neither analogy nor extrinsic stipulation is required in
order to account for the creation of the pattern.
o Many accounts (Sihler, 1995; Byrd, 2010, following Rix, 1992) have conceived
of the pattern as almost entirely analogical, treating the initial VC sequence as a
s r
f “re-app ica i n” f re p ica i n s bseq en

he
ss f he aryngea s.
√* 2ger : perfect *h2e-h2ger- > Pre-Greek *[ ger-] → (by ana gy)
Attic-Ionic ἀγήγερμαι [ag germai]
14
An OCP-σ-based account of the phenomena discussed in §5.1 and §5.2 presupposes a difference in syllabification
between rising sonority (i.e. obstruent + resonant) clusters and non-rising sonority clusters in the respective
languages: rising sonority clusters were tauto-syllabic (forming a complex onset), whereas non-rising sonority
clusters were hetero-syllabic (forming a coda + simplex onset). If all clusters were hetero-syllabified, OCP-σ w
be violated in all cluster-initial roots, thus requiring repair in all circumstances, not just those described here.
Cooper (2012) argues in favor of across-the-board hetero-syllabification in PIE, Greek, and Vedic, largely on the
basis of metrical evidence.
15
Lexical OCP effects are extremely common cross-linguistically, if not universal; cf. Pozdniakov & Segerer
(2007). However, this case seems unusual in that it refers specifically to complete identity rather than general
similarity (usually related to place of articulation). See also Cooper (2009) for discussion of these effects in PIE.
13
WeCIEC XXV, 10/25/13
Sam Zukoff
 However, any attempt to create a grammar that can formally erive his “ana gy”
makes incorrect predictions about the reduplicative grammar of the later
language.

It would wrongly force copying in vowel-initial and non-rising sonority
cluster-initial roots (like those discussed in §5.2).
o Winter (1950) (and, implicitly, Keydana, 2006, 2010) instead treats AR
phonologically,16 stipulating that *hC clusters exceptionally copied both elements
to create a reduplicant of shape *hCe-.

Once the form is fixed in such a way, it will derive correctly into Greek:
√* 2ger : perfect *h2ge-h2ger- ( > [* 2.ge-h2.ger-] ) >
ic-I nic
ἀγήγερμαι [a.g .ger.mai]
 However, there is no intrinsic reason to think that laryngeal clusters should have
had an exceptional copying pattern, i.e. C1C2V- vs. C1V-.

The account presented here produces a pre-f rm simi ar
Win er’s (1950), b
inc
es
a phonological motivation for the apparent exceptional behavior of laryngeal-initial roots.

The only way in which analogy must be invoked in this account is that there must have
been leveling of ablaut grades in *HeC roots.
o If an *HeC root displayed o-grade ablaut (which would be expected in the perfect
active singular) or e-grade, the problematic *-HC- sequence would not arise.
o However, if the root were in zero-grade (which would be expected in all other
categories of the perfect), we would have exactly that *-HC- sequence, inducing
the AR derivation as outlined in §4.
 This sort of intra-paradigmatic ablaut leveling is not exclusive to AR forms, but is
rather a common feature of the perfect in general.
 See the Appendix (§8.2) for fuller discussion of *HeC roots.

Attic Reduplication comes about as an avoidance strategy so that reduplication can occur
without violation of the *σ[HC constraint.
o This creates a unique *HVCV reduplicant for laryngeal-initial roots, distinct from
the normal CV reduplicant of other consonant-initial roots.
16
Suzuki (1993) also takes a phonological approach. He generates a pre-form in *C2V-HC2VC by a rule of laryngeal
re-syllabification, but still requires analogical reintroduction of the initial vowel.
14
WeCIEC XXV, 10/25/13
Sam Zukoff
7. References
Beekes, Robert S. P. 1969. The Development of the Proto-Indo-European Laryngeals in Greek.
The Hague: Mouton.
Byrd, Andrew Miles. 2010. Reconstructing Indo-European Syllabification. PhD Dissertation,
University of California, Los Angeles.
Byrd, Andrew Miles. 2011. Laryngeal Vocalization as Schwa Epenthesis: Why does
Compensatory Lengthening not Occur? Paper presented at the 30th East Coast IndoEuropean Conference, Cambridge, MA.
Cooper, Adam. 2009. Similarity Avoidance in the Proto-Indo-European Root. University of
Pennsylvania Working Papers in Linguistics, 15(1), 8.
Cooper, Adam. 2012. Syllable Nucleus and Margin in Greek, Vedic, and Proto-Indo-European.
PhD Dissertation, Cornell University.
Goldsmith, John. 1976. Autosegmental Phonology. PhD Dissertation, MIT.
Fortson, Benjamin W. 2004. Indo-European Language and Culture: An Introduction (1st ed.).
Blackwell Publishing.
Keydana, Götz. 2006. Die indogermanische Perfektreduplikation. Folia Linguistica Historica 27.
61-116.
Keydana, Götz. 2010. Evidence for Non-linear Phonological Structure in Indo-European: The
Case of Fricative Clusters. Ms. http://www.keydana.de/writings.php.
McCarthy, John J. & Alan Prince. 1993. Generalized Alignment. In Yearbook of Morphology
1993, (eds) G. Booij & J.v. Marle. Dordrecht: Kluwer.
Pozdniakov, Konstantine & Guillaume Segerer. 2007. Similar Place Avoidance: A statistical
universal. Linguistic Typology, 11(2).
Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative
Grammar. Ms.
Rix, Helmut. 1992. Historische Grammatik des Griechischen: Laut und Formenlehre.
Darmstadt: Wissenschaftliche Buchgesellschaft.
Sihler, Andrew L. 1995. New Comparative Grammar of Greek and Latin. New York: Oxford
University Press.
Suzuki, Seiichi. 1994. Accounting for Attic Reduplication: A Synthesis. Journal of IndoEuropean Studies 22. 399-415.
15
WeCIEC XXV, 10/25/13
Sam Zukoff
van de Laar, Henri M.F.M. 2000. Description of the Greek Individual Verbal Systems.
Amsterdam: Rodopi.
Vennemann, Theo. 1988. Prefere ce Laws for Syllable S ruc ure : A d
e Expla a io of Sou d
Change with Special Reference to German, Germanic, Italian, and Latin. Berlin: Walter
de Gruyter.
Vine, Brent. 2011. On Dissimilatory r-loss in Greek. In T. Krisch & T. Lindner (eds.),
Indogermanistik und Linguistik im Dialog (pp. 1–17). Wiesbaden: Reichert.
Winter, Werner. 1950. On the Origin of the Saṁprasaraṇa Reduplication in Sanskrit. Language,
26 (3), 365-70.
Zukoff, Sam. 2012. The Phonology of Verbal Reduplication in Ancient Greek: an Optimality
Theory approach. MA Thesis, University of Georgia.
Available at: http://mit.academia.edu/SamZukoff
8. Appendix:
8.1. Complete tableau for §4
(26) Producing the AR pre-form: PIE √*h2ger >> Greek perfect ἀγήγερμαι [a.g .ger.mai]
 By hypothesis: Pre-Greek /REDe, h2ger-/ → [h2a.gah2.ger-]
/REDe, h2ger-/
a. h2a.-h2gerb. h2a-h2.gerc. h2ga.-h2gerd. h2ga-h2.gere. h2a.ga.-h2gerf.  h2a.ga-h2.gerg. h2ge.ra-h2.gerh. h2a.-h2a.geri. h2a.ge.ra-h2.gerj. h2a.ga.-h2a.gerk. a.-h2gerl. a-h2.germ. ga-h2.gern. ge.-gero. h2a.-h2ar-
*σ[HC OCP- MAX- ANCH- ONS CONT- DEP- CONT- ALIGNσ
IO
L-BR
IO
IO
BR
ROOT-L
*!
**
*!
**
*!*
***
*!
(*)
***
*!
*
*
****
*
*
****
*!
*****
*!
*
**
*
*
*****!*
*!
*(*)
****
*!
*
*
*!
*
*!
**
*!
**
*!
**
16
WeCIEC XXV, 10/25/13
Sam Zukoff
8.2. *HeC roots
8.2.1. The complication for *HeC roots

The expected ablaut grade for the perfect active singular is the o-grade. Therefore,
looking at a root like *√h1ed- ‘ea ’, he r
a
m rph which sh
be en ere in
he
derivation (for the perfect active singular) is /h1od/.
o Since the normal pattern for reduplication is C1 copying, the default candidate for
this allomorph would be [h1e.-h1od-].

Based on the formulation of the *σ[HC constraint, it is not simply the presence of a rootinitial laryngeal that induces the AR derivation, but specifically the configuration
laryngeal + consonant at the left-edge of a syllable.
o In this candidate, the laryngeal is intervocalic, and thus not in violation of *σ[HC.
 Therefore, there would be nothing to rule out this candidate, and it would be chosen as
the winner.

It would become ** d-, which is not an attested outcome for this particular root.

Instead, we do want an AR form: ed d-.
8.2.2. Getting AR from the zero-grade

F r his r
, he bes /
es a es e perfec f rm is he par icip e ἐδηδ ς [e
- s].
o The participle is a proper zero-grade formation.
o Therefore, the input would be: /REDe, h1d, wōs/.

Plugging in the default candidate, we encounter our *σ[HC violation: [h1e.-h1d-wōs].

If we try to repair it by re-syllabification, we get our OCP-σ violation: [h1e-h1.d-wōs].
 This leads us down the same road as with the *HCeC roots, ultimately choosing a
candidate [h1e.de-h1. -wōs], which directly yields ἐδηδ ς [e

- s].
Therefore, while AR should not authentically arise on e-grade or o-grade formations for
*HeC roots, it should in the zero-grade formations, which include all categories in the
perfect other than the active singular.
8.2.3. Generalizing the AR stem

This predicts that, for a time, *HeC roots would have had normal reduplication in e/ograde categories but AR-type reduplication in zero-grade categories.
o Compare this situation to Sanskrit CeC roots:
√pad- ‘g ’ : perfect singular pap da
17
vs.
perfect plural pēdur
WeCIEC XXV, 10/25/13

Sam Zukoff
As ablaut distinctions collapsed, and as the transparency of the relationship between the
two reduplicative allomorphs was eroded by the loss of the laryngeals, speakers could
have generalized one or the other of the forms throughout the perfect paradigm.
o Might frequency effects have played a role in the direction of leveling?
o If a given root was used more frequently in zero-grade categories, would it be
more likely to have been leveled towards AR?
8.3. List of AR perfects
(27) Attested Attic Reduplication perfects and likely etymologies (van de Laar, 2000)
Root (Greek < *PIE)
Present Tense
#*h1
eger <*h1ger
ela(u) < *h1elh2
‘wake’
‘ rive’
eleuth < *h1lewdh
‘g , c me’
[egēr-]
[ela-],
[elau-]
n/a
(en-)eŋk < *h1nek
‘bring’
n/a
eme < *wemh1
ered < *h1reyd
ereip < *h1reyp
#*h2
ager < *h2ger
ako(u) < *h2kow(s)
ale < *h2elh1
ar < *h2er
aro < *h2erh3
#*h3
od < *h3ed
ol < *h3elh1
om < *h3emh3
op < *h3ekw
or < *h3er
‘v mi ’
‘ca se
ean’
‘ hr w wn’
ἐμέω
ἐρείδω
ἐρείπω
‘ga her ge her’
‘hear’
‘grin ’
‘j in, fi ge her’
‘p w’
ἀγείρω
ἀκούω
ἀλέω
ἀραρίσκω
ἀρόω
‘sme ’
‘ es r y’
‘swear’
‘see’
‘inci e’
ὄζω
ὄλλ μι
ὄμνυμι
or < *ser
oreg < *h3reg
orug < *h3ru-gh
‘keep wa ch’
‘s re ch, reach’
‘ ig’
ὄρομαι
ὀρέγω
ὀρύσσω
ἐγείρω
ἐλάω,
ἐλαύνω
ὄρνυμι
17
Perfect Tense
ἐγρήγορα17
ἐλήλαμαι
[egr g r-]
[e a-]
[eme-]
[ereid-]
[ereip-]
ἐλήλουθα
ἐλήλυθα
ἐνήνοχα
ἐνήνεγμαι
ἐμήμεκα
ἐρήρεισμαι
ἐρήριπα18
h
[e
-],
h
[e
-]
[en n kh-],
[en neg-]
[em me-]
[er reis-]
[er rip-]
[agēr-]
[akou-]
[ale]
[arar-]
[aro-]
ἀγήγερμαι
ἀκηκοα
ἀλήλεσμαι
ἄρηρα
ἀρήρομαι
[ag ger-]
[ak k -]
[a e(s)-]
[ar r-]
[ar r -]
ὄδωδα
ὄλωλα
ὀμ μομαι
ὄπωπα
ὄρωρα,
ὀρ ρε αι
ὄρωρα
ὀρ ρεγμαι
ὀρ ρυγμαι
[
[
[
[
[
[
[
[
[
[ozd-]
[ol-]
[om-]
n/a
[or-]
[or-o-]
[oreg-]
[orus-]
-]
-]
m m -]
p p-]
r r-],
r re-]
r r-]
r reg-]
r r g-]
The [r] in the reduplicant is secondary. Brent Vine (p.c.) has suggested that it is the result of hyper-corrective
r-insertion, along the lines of the phenomenon discussed in Vine (2011).
18
Beside this there is also ἐρέριμμαι [ererim-mai] wi h sh r [e] f r ng [ ].
18