The "Prefixing" Preference in Reduplication
University of California, Irvine
This paper argues in favor of a theory of reduplication in which the reduplicative morpheme (RED) is
positioned by the combined effects of two requirements. The first is ANCHOR, a positional faithfulness
constraint (expanding Beckman 1998's proposal), which demands that reduplication target only
"privileged positions" within the base. Crucially, whereas the psycholinguistically important left edge of
the base qualifies as a target for an anchoring constraint, the right edge does not (counter McCarthy &
Prince 1995). In addition, the reduplicant must obey locality. Locality is commonly understood to be a
requirement for adjacency between the portion of the base that is copied and the reduplicant (i.e.
(1) Constraints involved in reduplication
a. LEFT-ANCHOR (Base, Reduplicant): The left edge of the reduplicant corresponds to the left edge of the
base (McCarthy & Prince 1995).
b. "LOCALITY": The copied portion of base and the corresponding reduplicant must be adjacent (cf.
Marantz 1982, McCarthy & Prince 1993, 1995, Urbanczyk 1996, 2000).
Absent from this system of reduplication are the alignment constraints previously assumed to apply to all
affixes, ALIGN-LEFT/RIGHT. I assume that these may only position fixed-segment affixes.
(2) Constraints governing fixed-segment affix placement (McCarthy & Prince 1993)
a. ALIGN-L(affix, stem): Affix is a prefix.
b. ALIGN-R(affix, stem): Affix is a suffix.
This approach thus takes a step back from the proposal of Marantz (1982), largely followed in work on
reduplication since, that an empty RED morpheme is attached to the left or right edge of a stem, in the
same manner as fixed-segment affixes.
The theory promoted here has three main advantages. The first is that treating the two types of
affixes differently derives the result that reduplication behaves as an "internal" affix, attracted to the stem
even when it must be ordered after other affixes (Carrier-Duncan 1984). That is, positional faithfulness to
the root will draw the affix inward, regardless of when it is added in the word formation process. Table 1
illustrates the typology predicted in a system in which RED may be independently aligned.
Table 1 (not a tableau)
/badupi, mu, RED/ ALIGN-LEFT ALIGN-LEFT REDRoot L-ANCHOR
(RED, Stem) (mu, Stem)
a. mu-mu-badupi ** **
b. mu-ba-badupi **
c. ba-mu-badupi ** *
(a): Forced copying of non-stem material
ALIGN-L (RED, Stem) >> ALIGN-L mu >> RED Root or
ALIGN-L (RED, Stem), L-ANCHOR >> RED Root
(b): Local copying of stem material
ALIGN-L mu >> ALIGN-L (RED, Stem) or
RED Root, L-ANCHOR >> ALIGN-L (RED, Stem)
(c): Non-local copying across fixed segment affix
ALIGN-L (RED, Stem), RED Root >> ALIGN-L mu, L-ANCHOR
The combined effect of ALIGN-RED and the constraint RED Root (McCarthy & Prince 1993,
Urbanczyk 2000), which requires the reduplicant to copy material from the root, leads to the undesirable
potential optima whereby the reduplicant can copy a prefix (a), or the reduplicant can be separated from
Nicole Nelson The "Prefixing" Preference in Reduplication
the stem by an intervening fixed segment affix (c). Table 2 shows that in the proposed system, no
alignment constraint can pull the RED morpheme away from root material.
/badupi, mu, RED/ ALIGN-L mu "LOCALITY" L-ANCHOR
a.X ba-mu-badupi ** ** *
c.X mu-mu-badupi **
The only constraints acting on these forms are reduplicative anchoring constraints, alignment of fixed
segment affixes, and LOCALITY. When RED is present, it must copy adjacent root material. The other
candidates are harmonically bounded, as indicated by 'X'.
The second advantage is that this theory of reduplication can account for the tendency of RED to go
against the otherwise prevalent bias toward suffixing. Hawkins and Cutler (1988) for example note that
there is a strong preference for suffixing, even when head-initial ordering in a language would otherwise
lead to an expectation of prefixing. Thus, the odd behavior of reduplicants is even more striking in that
reduplicants overwhelmingly precede the stem cross-linguistically, regardless of the default head position.
I argue that whereas fixed segment affixes prefer final position for the sake of early access of the root
(Hawkins & Cutler 1988), reduplication has no such force pushing RED to word-final position. Rather,
since the reduplicant copies material from the root, placing it initially not only does not impede root
access, but it also allows for the additional information added by RED to be gained at the earliest point
possible in processing.
Finally, there is the advantage that an asymmetrical anchoring system, in the absence of alignment
constraints that explicitly position RED, beneficially eliminates the typological prediction that Marantz's
Generalization (MG) would ever be violated. Table 3 shows that "LOCALITY", ranked low enough, would
allow both ANCHOR and an opposite-edge alignment constraint to be satisfied.
/badupi, RED/ L-ANCHOR IO-CONTIGUITY ALIGN-R (RED, Stem) "LOCALITY"
a. badupi-pi *
b. badu-pi-pi * * **
c. ba-badupi ******
d. badupi-ba ****
If no alignment constraint is in fact working against L-ANCHOR, then the MG-violating candidate (d) is
harmonically bounded, shown in Table 4.
/badupi, RED/ L-ANCHOR IO-CONTIGUITY "LOCALITY"
a.X badupi-pi *
b.X badu-pi-pi **
d.X badupi-ba ****
Candidates (a) and (b) are harmonically bounded here as well. There is a chance that one of these
candidates could emerge as optimal if some other constraint, for example one requiring stressed syllable
copying, preferred one of these candidates over (c). It is clear however that no plausible constraint will be
able to salvage candidate (d) in this system.