Content uploaded by Iris Berent
Author content
All content in this area was uploaded by Iris Berent on Apr 14, 2014
Content may be subject to copyright.
Co-occurrence restrictions on identical
consonants in the Hebrew lexicon :
are they due to similarity ?
1
IRIS BERENT
Department of Psychology, Florida Atlantic University
JOSEPH SHIMRON
School of Education, University of Haifa
(Received 2 October 2001 ; revised 7 June 2002)
It is well known that Semitic languages restrict the co-occurrence of identical and
homorganic consonants in the root. The IDENTITY HYPOTHESIS attributes this pattern
to distinct constraints on identical and nonidentical homorganic consonants (e.g.
McCarthy 1986, 1994). Conversely, the SIMILARITY HYPOTHESIS captures these restric-
tions in terms of a single monotonic ban on perceived similarity (Pierrehumbert 1993 ;
Frisch, Broe & Pierrehumbert 1997). We compare these accounts by examining the
acceptability of roots with identical and homorganic consonants at their end. If well-
formedness is an inverse, monotonic function of similarity, then roots with identical
(fully similar) consonants should be less acceptable than roots with homorganic (par-
tially similar) consonants. Contrary to this prediction, Hebrew speakers prefer root
final identity to homorganicity. Our results suggest that speakers encode long-distance
identity among root radicals in a manner that is distinct from feature similarity.
It is well known that Semitic grammars constrain the structure of lexically
stored forms. To reveal these grammatical constraints, we examine here the
predictable regularities in a word’s consonantal melody – the sequence of
consonants obtained after removing nonreduplicative inflectional affixes and
vowels from the word. Because this consonantal melody coincides with the
unit listed as ‘root ’ in Semitic dictionaries, this terminology is often main-
tained even for melodies that exhibit predictable regularities. For instance,
the consonantal melody smm is dubbed ‘ root ’ by sources that clearly argue
against its lexical storage in this form (e.g. McCarthy 1981). For the sake of
simplicity, we follow here the same tradition. We wish to emphasize, however,
that the consonantal melodies we examine are strictly SURFACE forms. We
make no claims as to whether these strings are stored as such in the lexicon
nor do we argue that they correspond to a phonological or morphological
constituent.
[1] We wish to thank Diamandis Gafos, Dan Everett and Janet Pierrehumbert, and two
anonymous JL referees for helpful discussions of this research.
J. Linguistics 39 (2003), 31–55. f2003 Cambridge University Press
DOI: 10.1017/S0022226702001949 Printed in the United Kingdom
31
An inspection of triliteral Semitic roots reveals two patterns of constraints.
One concerns the co-occurrence of identical consonants : identical consonants
are frequent at the root’s end (e.g. skk), but rare at its beginning (e.g. ssk ;
Greenberg 1950 ; Bender & Fulass 1978 ; McCarthy 1986; Buckley 1997). A
second co-occurrence restriction concerns nonidentical homorganic con-
sonants. Nonidentical homorganic root consonants are underrepresented
(Greenberg 1950; Bender & Fulass 1978 ; Pierrehumbert 1993; Buckley 1997 ;
Frisch, Broe & Pierrehumbert 1997). Like identical consonants, homorganic
nonidentical consonants are rare root initially (e.g. kgs). Unlike identical
consonants, however, homorganic nonidentical consonants are infrequent
both root initially and root finally (Greenberg 1950).
These distributional patterns have received conflicting explanations in the
linguistic literature. According to the IDENTITY HYPOTHESIS, the restrictions
on identical consonants are distinct from those concerning nonidentical
homorganic consonants (McCarthy 1986, 1994). Conversely, the SIMILARITY
HYPOTHESIS offers a single explanation for the restrictions on identical and
nonidentical homorganic consonants (Pierrehumbert 1993 ; Frisch et al. 1997).
This view considers identity as a special case of similarity. The restrictions
on homorganic (partially similar) and identical (fully similar) segments are
therefore captured by a single principle – a statistical monotonic ban on
perceived similarity. Although there are various pieces of evidence supporting
the similarity hypothesis as an account for the distribution and acceptability
of nonidentical homorganic consonants, it is unclear whether this view can
specifically handle the restrictions on consonant-identity. The following in-
vestigation examines this question. Two experiments compare the ac-
ceptability of novel Hebrew roots with identical vs. nonidentical homorganic
consonants root finally. If acceptability is an inverse, monotonic function of
similarity, then identical (fully similar) segments should be less acceptable
than homorganic (partially similar) segments. We show that, despite their
greater similarity, identical consonants are in fact MORE acceptable than
homorganic nonidentical consonants, a result that is directly opposite to the
prediction of the monotonic similarity hypothesis. As we point out, our results
are moot with respect to the adequacy of the similarity hypothesis as an
account for the co-occurrence of nonidentical homorganic consonants. These
results nevertheless favor separate accounts for the co-occurrence of identical
and nonidentical homorganic consonants.
1. T HE IDENTITY HYPOTHESIS
According to the identity hypothesis, the restrictions on identical consonants
are distinct from those affecting nonidentical homorganic consonants. A
widely influential version of the identity hypothesis is offered by McCarthy
(1986, 1994). McCarthy attributes the restrictions on homorganic and identical
consonants to the Obligatory Contour Principle (OCP), a ban on adjacent
I.BERENT & J.SHIMRON
32
identical elements. Although identical and homorganic consonants are both
constrained by the OCP, the target of the OCP in the two cases is different.
The restrictions on identical root consonants reflect the application of the
OCP to the entire segment (the root node, hereafter, the OCP-total). The
OCP-total thus bans adjacent identical segments. Adjacency, however, is
determined by the placement of segments in autosegmental tiers, not by their
surface proximity. In particular, because the consonantal root morpheme is
represented on a single tier, segregated from vowels and affixes, root con-
sonants are psychologically adjacent, hence, subject to the OCP-total. Ac-
cordingly, the OCP-total bans the storage of ssm-orsmm-type forms in the
lexicon. Although root geminates may not be stored, they may be formed
productively by spreading an underlying biconsonantal representation (e.g.
sm), as shown in figure 1. The additional assumption that spreading proceeds
rightwards correctly predicts the frequency of identical consonants at the end
of the root (e.g. smm) but not at its beginning (e.g. ssm). In contrast, non-
identical homorganic consonants are rare both root initially and root finally
(Greenberg 1950). McCarthy (1994) attributes the restrictions on non-
identical homorganic consonants to the application of the OCP at the feature
level, specifically, the place node (hereafter, OCP-place). Because homorganic
consonants are adjacent on the place of articulation tier, they violate the
OCP-place. Furthermore, because, unlike the root node, the place node may
not branch to adjacent skeletal positions (McCarthy 1994), adjacent
homorganic consonants violate the OCP-place at either initial C1C2 or final
C2C3 root positions, as shown in figure 2. The identity hypothesis thus
captures the distinct distributional patterns of identical and nonidentical
homorganic root consonants by means of separate (albeit related) constraints.
s
CV
m
CVC
a
Root consonantal tier
Skeletal tier
Vowel tier
s
CV
s
CVC
a
Root consonantal tier
Skeletal tier
Vowel tier
*m
Figure 1
Violations of the OCP-total by roots in the verbs samam and sasam. Because the root
smm is formed by spreading, it does not violate the OCP. In contrast, the root ssm
includes adjacent identical consonants, which violates the OCP-total (OCP violations
are indicated by asterisks)
IDENTICAL CONSONANTS IN HEBREW
33
2. T HE SIMILARITY HYPOTHESIS
The similarity hypothesis attributes the restrictions on identical and non-
identical consonants to a single stochastic constraint. On this view, the
co-occurrence of consonants depends on their perceived similarity (Pierre-
humbert 1993 ; Frisch et al. 1997). Perceptual similarity, in turn, is defined by
two factors: feature overlap and distance. Feature overlap is a weighted
function of the number of shared vs. different features between two segments.
The distance between segments is not restricted to an underlying represen-
tation – Pierrehumbert (1993 : 379) notes ‘ there is no absolute impediment to
evaluating the homorganicity of the consonants after vowels are inserted ’.
Thus, the similarity proposal encompasses two logically distinct claims. One
claim is that a stochastic similarity constraint can better capture the co-
occurrence restrictions on nonidentical homorganic consonants compared to
the OCP-place constraint. A second claim is that the co-occurrence restric-
tions on identical segments are indistinguishable from those applying to
(nonidentical) homorganic consonants.
Consider first the co-occurrence of nonidentical homorganic consonants.
Pierrehumbert (1993) argued that the co-occurrence of homorganic root
consonants is inexplicable by a categorical constraint on place of articulation,
as their distribution is further sensitive to secondary place of articulation and
noncontrastive voicing (see also Frisch et al. 1997). Additional criticisms of the
OCP-place concerns locality. The OCP bans adjacent homorganic segments.
Because adjacency is defined relative to autosegmental representations that
encode homorganic consonants on a single tier, homorganicity effects should
be blind to intervening elements that are not specified for that place of ar-
ticulation. For instance, the co-occurrence of the coronal obstruents td should
be unaffected by the intervening labial b. The OCP-place thus predicts similar-
size dissimilation effects for adjacent (e.g. tdb) and nonadjacent (e.g. tbd)
homorganic consonants. Contrary to this prediction, however, the observed/
expected ratio for homorganic consonants is lower for adjacent consonants
relative to nonadjacent consonants (Pierrehumbert 1993 ; Buckley 1997 ;
s
CV
k
CVC
a
k
CV
g
CVC
a
s
*[velar]
g
[velar] *[velar] [velar]
Figure 2
OCP-place violations by the novel verbs sakag and kagas. Because the velar consonants
are adjacent on the place of articulation tier, these verbs each incur violations of
OCP-place (OCP-place violations are marked by asterisks)
I.BERENT & J.SHIMRON
34
Frisch et al. 1997). These observations suggest that the constraint on non-
identical root consonants is not limited to adjacent root radicals, nor is it
specific to homorganicity. Pierrehumbert (1993) proposed to capture the co-
occurrence of nonidentical root radicals by a stochastic constraint on per-
ceived similarity. Frisch et al. (1997) extended this proposal by providing a
detailed formal analysis of perceived similarity. They further demonstrated,
by means of a regression analysis, that a structured specification account of
similarity provides a superior fit for the distributional data compared to other
proposals, including McCarthy’s categorical account, a frequency account,
and Pierrehumbert’s (1993) proposal, which captures similarity in terms of
contrastive underspecification.
The stochastic account, however, not only offers an alternative for the OCP-
place, but further aspires to subsume the OCP-total. Similarity is presented as
a single unified explanation for the co-occurrence of homorganic (non-
identical) and identical consonants. Identical consonants, on this view, are
consonants that are maximally similar. The co-occurrence of identical con-
sonants thus presents the limiting case of a more general stochastic constraint
on perceived similarity. In support of this view, Pierrehumbert (1993), Frisch
et al. (1997) and Buckley (1997) demonstrate that the distribution of identical
consonants is constrained by perceived similarity and distance, the same
principles used to capture the distribution of nonidentical homorganic con-
sonants. For instance, in accord with the view of identical elements as
maximally similar, there is a larger gap in the co-occurrence of nonadjacent
identical consonants (e.g. tbt) compared to homorganic nonidentical con-
sonants (e.g. tbd). This finding is unexpected on McCarthy’s (1994) account :
because the OCP-total does not operate across an intervening C2 segment,
tbt and tbd should have incurred a comparable violation of the OCP-place
(Pierrehumbert 1993). Identical and homorganic nonidentical consonants
alike are further affected by distance : they are less likely to occur in adjacent
(C1C2 and C2C3) compared to nonadjacent (C1C3) positions ; and the mag-
nitude of the distance effect depends on similarity : nonadjacent homorganic
consonants are more frequent than nonadjacent identical consonants. None
of these findings is expected on McCarthy’s account. Pierrehumbert (1993)
and Frisch et al. (1997) thus conclude that the distribution of identical and
homorganic nonidentical root consonants is governed by a single constraint
on perceived similarity.
3. S OME CHALLENGES TO THE SIMILARITY ACCOUNT
Although there is some compelling evidence favoring the similarity hypoth-
esis as an account for the distribution of homorganic root segments, these
observations cannot unequivocally support its potential for capturing the
distribution of identical root radicals. The main problem is the failure to
fully account for the distribution of identical consonants. According to the
IDENTICAL CONSONANTS IN HEBREW
35
similarity hypothesis, the acceptability of root radicals is a monotonic func-
tion of their perceived similarity. Because identical consonants are maximally
similar, they should be less acceptable than nonidentical consonants. This
prediction could be evaluated at either root initial or root final position. The
evaluation of this prediction for root initial consonants is not entirely clear :
because root initial identical consonants violate the OCP-total, their greater
unacceptability may be due either to their greater similarity or to multiple
constraint violation (i.e. the violation of OCP-place and OCP-total). A greater
unacceptability of identical and homorganic consonants at root initial pos-
ition is thus consistent with either the similarit y or the identity hypotheses. The
predictions of these two views diverge, however, for root final consonants. The
monotonic similarity account predicts lower acceptability for identical (fully
similar) relative to homorganic (partially similar) consonants. Conversely,
the identity hypothesis allows for the constraints on identical consonants to
differ from those governing similar consonants. In fact, the predictions of
McCarthy’s (1986) account are clearly contradictory to the similarity hy-
pothesis. Because smm-type roots do not violate either the OCP-total or the
OCP-place, such roots should be clearly more acceptable than roots with final
homorganicity. Root final identity thus allows adjudicating between the
similarity and identity hypotheses. Unfortunately, existing tests of the simi-
larity hypothesis do not consider the distribution of smm-type roots. Recall
that the similarity hypothesis is supported by the observation of gaps between
the observed and expected co-occurrence of root radicals. These analyses,
however, exclude roots with final identical consonants on the grounds that
these are lexically represented in their biconsonantal form, as proposed by
McCarthy (1986).
On superficial examination, this decision appears unproblematic: if ‘ root ’
is a unit of lexical storage, and if lexical storage is contingent on identity
erasure, then smm-type forms are not truly roots. Accordingly, the distribution
of smm-type forms falls beyond the scope of an account of the lexicon. We see
several problems with this reasoning, however. To be sure, we do not dispute
the claim that verbs such as simem are subject to identity erasure, nor do we
argue that they are lexically stored as smm. As explained above, we use the
term ‘root ’ to refer to a surface unit – the string of consonants obtained after
removing nonreduplicative inflectional affixes and vowels. This unit may not
necessarily correspond to the unit of lexical storage, inferred after applying
grammatical constraints to surface representations. Nevertheless, we believe
that the exclusion of smm-type forms limits the evaluation of the similarity
hypothesis. First, the exclusion of smm-type roots on the grounds that they are
not lexically stored tacitly confines the similarity hypothesis to the lexicon.
This assumption does not follow from the similarity hypothesis. In fact,
proponents of this proposal specifically argue against such an approach
(Pierrehumbert 1993 ; Frisch et al. 1997). Second, the exclusion of smm-type
forms remains unjustified even if the lexical domain is explicitly assumed.
I.BERENT & J.SHIMRON
36
A principal motivation for the similarity hypothesis is the postulation of a
single explanation for the distribution of identical and homorganic root
consonants. Specifically, Pierrehumbert (1993 : 373) notes that
[i]n the standard model, the gradience of the dissimilarity requirements
in the Arabic verbal roots is disregarded, and the cooperation of several
grammatical mechanisms is invoked to cover those regularities which are
described. In the alternative approach to be developed here, the gradience is
exploited both to improve the descriptive coverage and to unify the most
categorical effect (the total OCP on adjacent consonants) with less cate-
gorical effects. _We propose that Arabic has an OCP effect of place alone.
It applies only to consonants which are perceived to be similar, and the
strength of the effect increases with perceived similarity. Identical con-
sonants are viewed as maximally similar, and so the total OCP arises as the
limiting case of maximally strong enforcement of the place OCP.
Disregarding smm-type forms is inconsistent with this goal. If the similarity
hypothesis maintains that the distribution of lexical forms is fully explicable
by their perceived similarity, irrespective of identity, then it cannot simul-
taneously maintain that identity erasure is operative in the lexicon. It is the
theory’s task to explain how the language learner comes to acquire such a
lexical representation. There are two possible replies to this question. One is
that the erasure of the final radical from smm-type roots is due to C2C3
identity. This assumption indeed lies at the core of McCarthy’s (1986) OCP-
total proposal – an account that presupposes an innate grammatical con-
straint on segment identity. The identity hypothesis, however, is not limited
to McCarthy’s (1986) OCP-total : it includes any proposal that constrains
segment identity, regardless of whether it is violable or inviolable, innate or
learnable. The similarity account is clearly free to adopt any of these versions
of identity-avoidance, but this move would EMBRACE the identity hypothesis,
not subsume it. The other possibility is that the erasure of the final radical
from smm-type forms is due to C2C3 similarity, not specifically their identity.
This is indeed the only logical possibility available if the identity hypothesis
is rejected. The challenge to this account is to explain why root radicals are
routinely erased when they are fully similar, but never when they are only
partly similar. In the absence of an independent explanation for identity
erasure, the exclusion of smm-type roots from the lexicon remains unmo-
tivated.
The unmotivated exclusion of smm-type roots indeed compromises the
main empirical support for the similarity hypothesis, namely, its ability to
capture the under-representation of identity in the lexicons of Arabic and
Tigrinya. Indeed, if one were to apply these calculations to all triliteral roots –
not just to hypothesized lexical representations – then the observed frequency
of roots with identical elements would be likely to exceed that of roots having
homorganic elements at root final position (Greenberg 1950 ; Buckley 1997).
IDENTICAL CONSONANTS IN HEBREW
37
For Hebrew, our own count of the set of 1449 productive roots listed in the
Even-Shoshan (1993) dictionary corroborates these conclusions.
2
Specifically,
whereas root initially nonidentical homorganic consonants are somewhat
more frequent than identical consonants, at root final positions the pattern is
reversed:
3
root final identity is generally more frequent than homorganicity
C1C2 C2C3
Identical Homorganic Identical Homorganic
Coronal sonorants
(l, r, n)
0 1 38 2
Labials (b, p, m, v) 2 2 27 0
Velars (k, g) 1 0 17 0
Gutturals (?,@,h,x) 0 8 12 17
Coronal Obstruents
stops (d, t) 1 0 23 1
fricatives (z, s, s)0 3 21 0
affricates (c) 0 0 7 0
Table 1
The distribution of identical and homorganic root consonants in Hebrew at
initial (C1C2) and final (C2C3) positions
[2] Because of the historical changes in the Hebrew phonemic system, the classification of several
phonemes for place of articulation is uncertain. One case in point concerns the class of
gutturals. Biblical Hebrew included four gutturals : the pharyngeal fricatives /h/ and /@/, the
glottal stop /?/, and the glottal fricative /h/. Modern Hebrew has lost the phonemes /h/ and /@/.
The Biblical /@/ and /h/ are realized as /?/ and /x/, respectively, by most contemporary
speakers. Although the contemporary [x] is a velar, it nevertheless retains some of the
phonemic characteristics of a guttural (Bolozky 1978; Sandler 1994), and is therefore con-
sidered as such for the purposes of the present analysis. Another phonemic distinction lost
among most speakers of Modern Hebrew is the emphatic consonants /t
˚/, /s
˚/ and /q/, which
are realized as /t/, /c/ and /k/, respectively (Bolozky 1978). Although the distinction between
these emphatic consonants and their nonemphatic realizations is maintained in the ortho-
graphic transcription given in Even-Shoshan (1993), it does not affect the calculation of
consonant co-occurrence, as the emphatic and nonemphatic members never co-occur in
adjacent positions (C1C2 or C2C3). Hebrew also has a highly predictable spirantization rule
that realizes /b/ and /p/ as [v] and [f ], respectively (Bolozky 1978). Our analysis ignores these
predictable alternations and considers only the underlying representation, /b/ and /p/.
Finally, the homorganicity of a consonant combination with the affricate /c/ is computed
under the assumption that its manner is represented as a stop-fricative sequence. Accord-
ingly, affricate-fricative and stop-affricate sequences are considered homorganic. Our
analysis does not report the distribution of the phoneme /y/, as it is the only Hebrew glide.
[3] The one exception to the rule is the overrepresentation of homorganic relative to identical
gutturals at C2C3 positions. This observation, however, appears to be entirely due to the
frequency of the radical hat the root’s end in the Even-Shoshan (1993) dictionary. Even-
Shoshan (1993) uses the radical hto capture the root final radicals of weak roots. This
transcription, however, reflects an orthographic convention, as the hradical is never realized
in the stem. To maintain consistency with our source, we coded these roots as listed there. The
I.BERENT & J.SHIMRON
38
(see table 1). This observation is contrary to the predictions of the similarity
hypothesis.
Although the discrepancy between the predicted and the observed distri-
bution of root final identity is unexpected under the similarity hypothesis,
it does not necessarily falsify this view. The similarity hypothesis captures
speakers’ knowledge of root phonotactics. This knowledge, however, may not
be directly evidenced in the distribution of lexical forms. Distributional facts
may be shaped by nonlinguistic or diachronic factors that are not currently
active in a language ; hence they may not necessarily reflect the linguistic
competence of modern speakers (Everett & Berent 1998). Experimental work
allows for additional insights into speakers’ synchronic knowledge of root
structure. We now proceed to discuss the existing results.
4. B EHAVIORAL EVIDENCE FOR CO-OCCURRENCE RESTRICTIONS
ON ROOT STRUCTURE
4.1 Co-occurrence restrictions on root identity in Hebrew
Recent experimental evidence demonstrates that speakers of Modern Hebrew
are sensitive to the co-occurrence of identical root consonants. For instance,
novel Hebrew roots with identical consonants at their beginning (e.g. ssm) are
rated less acceptable than roots with adjacent identical consonants at their end
(e.g. smm, Berent & Shimron 1997 ; Berent, Everett & Shimron 2001(a)). The
sensitivity to the location of identical root radicals is also observed online.
Consider the lexical decision task, for instance. Participants in this task are
asked to quickly indicate whether or not a string of letters corresponds to an
existing word by pressing one of two computer keys. The speed and accuracy
of classifying a stimulus may be used to gauge its well-formedness : novel
words that violate co-occurrence restrictions should be classified as nonwords
more easily than novel words that do not violate such restrictions. Berent and
colleagues found that lexical decision is specifically sensitive to the location
of identical consonants in the root : novel words with ssm-type roots were
classified as nonwords faster and more accurately than smm-type controls,
a finding consistent with the view of ssm-type roots as ill-formed (Berent,
Shimron & Vaknin 2001(b) ; Berent, Marcus, Shimron & Gafos 2002). The ill-
formedness of ssm-type roots is evident even when no explicit response to the
word is required, as evidenced in the Stroop task (Berent, Bibi & Tzelgov
2000). Here, participants are asked to name the color of letter strings printed in
color (e.g. the word sisem printed in red) while ignoring the contents of the
printed word. The speed of color naming (e.g. saying ‘ red ’) is used to gauge
the internal ill-formedness of the root : because ill-formed stimuli are easier to
ubiquity of root final hwith homorganic radicals, however, should not be considered as
evidence for weakening of the OCP-place for root final gutturals.
IDENTICAL CONSONANTS IN HEBREW
39
ignore, they are expected to interfere with color naming less than well-formed
stimuli. In accord with the view that root initial gemination is ill-formed,
Berent et al. (2000) observed faster color naming latencies for words derived
from the ssm-type compared to smm-type controls. The sensitivity to the root
location of identical consonants in online tasks, including tasks that do not
require attention to root structure, suggests that it forms part of speakers’
linguistic competence.
Existing experimental investigations can further rule out several alternative
explanations for the co-occurrence restrictions on identical root consonants.
One explanation attributes the experimental results to a restriction on word,
rather than root structure. For instance, the ill-formedness of ssm-type roots
could be due to the ill-formedness of C
1
VC
1
combinations word initially
(Frisch & Zawaydeh 2001). The experiments reported above systematically
examined this account by manipulating the location of the root within the
word. The findings consistently revealed sensitivity to the location of identical
consonants across various word positions and despite intermediate vowels
and affixes. These results suggest that the restrictions on the location of
identical consonants are defined relative to a morphological constituent (the
root or stem), rather than to their word position. A second alternative ex-
planation attributes the restrictions on identical consonants to the distri-
bution of specific root tokens in speakers’ lexicons. On this account, the
unacceptability of the root ssm, for instance, is due to the rarity of its con-
sonant-combinations (especially the initial geminates, e.g. ss), rather than
to the violation of an abstract constraint on identity, a formal relationship
between any consonant pair (e.g. *XX, where X stands for any consonant).
Contrary to this view, the unacceptability of ssm-type roots was observed
even when ssm- and smm-type roots were equated for familiarity. Specifically,
Berent et al. (2002) observed the asymmetry in the location of identical con-
sonants for roots including novel geminate phonemes ; hence ssm and smm-
type roots were equally (un)familiar. Likewise, speakers discriminate between
novel smm-type roots and no-gemination controls where the materials are
equated for the co-occurrence of root radicals (Berent et al. 2001(a)). These
results demonstrate that speakers are sensitive to root structure, and that they
can discriminate between various root types depending on the presence of
identical consonants and their location in the root.
4.2 The source of the restrictions on identical consonants : identity vs.
homorganicity
Although the existing empirical findings clearly demonstrate sensitivity to the
location of identical consonants in the root, they cannot unequivocally reveal
its source. In particular, the existing results cannot determine whether this co-
occurrence restriction specifically concerns identity or homorganicity. Recent
experiments by Frisch & Zawaydeh (2001) report that speakers of Jordanian
I.BERENT & J.SHIMRON
40
Arabic consider novel CaCaCa verbs with homorganic consonants as less
acceptable than verbs with no homorganic consonants. Although the gen-
erality of homorganicity avoidance across morphological patterns remains
to be seen, these results suggest that speakers disfavor root homorganicity.
If nonidentical, homorganic consonants are actively disfavored, then it is
possible that, despite the greater frequency of roots with final identical con-
sonants in the lexicon (relative to homorganic consonants), identity may not
be fully desirable. In fact, Hebrew roots with final identity are rated as less
acceptable than no-gemination controls. For instance, in Berent et al.’s
(2001(a)) experiment 2, roots with final identity (e.g. smm) were assigned a
mean acceptability rating of 2.02, whereas the mean for roots with no ident-
ity was 2.58 (using a 1–3 scale, 1=sounds best, 3=sounds worst). Although
smm-type roots are certainly acceptable, their ratings were consistently lower
than roots with no identity (for similar findings, see also Berent & Shimron
1997). The relative unacceptability of smm-type roots could be due to their
similarity. More generally, the acceptability of Semitic roots may be cap-
tured by a single monotonic function of their similarity without assuming
separate restrictions on identity. Although Frisch & Zawaydeh (2001) have
raised some doubts regarding the adequacy of the similarity hypothesis
as an account for the restrictions on root-identity (see also Frisch in press),
4
the similarity account for the acceptability of root identity has not been
systematically investigated. In what follows, we examine the monotonic
similarity hypothesis experimentally by comparing the acceptability of roots
including homorganic and identical consonants. If acceptability is an in-
verse, monotonic function of similarity, then identical (fully similar) segments
should be less acceptable than homorganic (partially similar) segments.
5. E XPERIMENTAL DESIGN AND PREDICTIONS
The following experiments compare the acceptability of roots with identical
and homorganic, nonidentical consonants. The view of acceptability as a
monotonic, inverse function of similarity predicts that identical conson-
ants should be less acceptable than nonidentical homorganic consonants.
Although the lesser acceptability of identical compared to nonidentical
homorganic consonants is predicted at both root initial and root final
[4] Frisch & Zawaydeh (2001) reported higher ratings for verbs with identical root final con-
sonants compared to verbs with homorganic root consonants. This post-hoc observation,
however, was based on only three root tokens and was not evaluated statistically. Fur-
thermore, because these authors did not systematically control for the location of homor-
ganic consonants in the root or the word, the observed trend may be due to either of these
factors. For instance, because homorganic consonants may be either root final or root
initial, the preference for identical root final consonants over homorganic consonants may
be explained by a preference concerning the location of homorganic consonants in the
root (a preference for final over initial homorganic consonant), rather than a distinction
between identity and homorganicity.
IDENTICAL CONSONANTS IN HEBREW
41
positions, the interpretation of this effect is clearest root finally. As explained
above (see section 3), a greater acceptability of ssm- relative to szm-type roots
is explicable by either similarity or multiple constraint violation (OCP-place
and OCP-total). A greater unacceptability of identical compared to homor-
ganic consonants at root initial position is thus consistent with both the
similarity and the identity hypotheses. The predictions of these two views
differ, however, for root final consonants. The monotonic similarity account
predicts lower acceptability for identical (fully similar) relative to homorganic
(partially similar) consonants. Conversely, the identity hypothesis makes it
possible to dissociate the acceptability of identical consonants from their
similarity. Specifically, on McCarthy’s (1986) version of the identity hy-
pothesis, smm-type roots do not violate either the OCP-place or OCP-total.
Accordingly, smm-type roots should be more acceptable compared to roots
with C2C3 homorganic consonants, which violate OCP-place (see figure 3).
Roots with final identity thus make it possible to test whether the acceptability
of identical consonants is explicable by their similarity. Our experimental
investigation compares roots with identical and homorganic (nonidentical)
consonants at their final positions.
Our experiments elicit acceptability ratings for root trios comprised of three
members. One member had a root with adjacent nonidentical homorganic
consonants at the root’s end (e.g. nkg), a second member had adjacent
identical consonants at the root’s end (e.g. skk), and the third member had no
homorganic root radicals (e.g. nks ; see table 2). To assure that the distinction
between root types was not due to familiarity with specific consonant com-
binations, we evaluated the three root types (homorganic, identical and non-
homorganic controls) for their radical co-occurrence in the Hebrew lexicon.
To further secure the attribution of acceptability ratings to restrictions on the
structure of roots, rather than words, we systematically varied the position of
these roots within the word by conjugating each root in three classes of word
patterns. If the restrictions on homorganic and identical consonants truly
s
CV
k
CVC
a
s
CV
k
CVC
a
g
* [velar] [velar]
Figure 3
OCP violations by homorganic vs. identical c2c3 consonants for the novel roots skk and
skg. Identical consonants in the root skk incur no OCP violations, whereas the
homorganic consonants in sgk violate the OCP-place
I.BERENT & J.SHIMRON
42
concern their position in the root, then these restrictions should be observed
regardless of word position and irrespective of intermediate vowels. The first
class corresponded to unaffixed words in the verbal patterns known as
BINYANIM PI?EL and KAL, singular masculine past. The location of the C2C3
radicals in these verbs was invariably word final. In contrast, members of the
other two classes were preceded by a prefix and followed by a suffix ; hence, the
location of the root’s C2C3 radicals was word internal. The second word class
included the nominal patterns known as MISHKALIM MAF?IL and MIF?AL,
whereas the third word class included verbs in BINYAN HITPA?EL past tense.
Each of the three root members was conjugated in precisely the same word
pattern, and was presented in each of the three classes. Because the resulting
words were perfectly matched on all aspects other than root structure, ac-
ceptability differences between trio members should uniquely reflect the
effect of root structure. We obtained two separate acceptability measures.
Experiment 1 compared the acceptability of members of the root trios relative
to each other (hereafter: relative rating), whereas experiment 2 examined the
acceptability of each member on its own (hereafter : absolute rating).
All accounts predict lower acceptability for roots with homorganic con-
sonants compared to controls. Of primary interest is the comparison of
identical and nonidentical homorganic consonants. If the co-occurrence of
homorganic and identical consonants is a monotonic function of their per-
ceived similarity, then roots with identical consonants (e.g. skk) should be
considered less acceptable compared to roots with homorganic, nonidentical
consonants (e.g. nkg). The predictions under the identity hypothesis are
reversed: because roots with final identity do not violate either the OCP-place
or OCP-total, they should be more acceptable compared to roots with homor-
ganic, nonidentical consonants. If the restrictions on homorganic and ident-
ical consonants concern their root position, then the same outcome should
emerge across word classes, regardless of the position of the root in the word.
Word class
SECOND THIRD
FIRST Affixed Affixed
Root Unaffixed adjacent nonadjacent
Homorganic nkg nikeg mankigot hitnakagta
Identical skk sikek maskikot histakakta
Control nks nikes mankisot hitnakasta
Table 2
The structure of the experimental materials used in experiments 1 and 2
IDENTICAL CONSONANTS IN HEBREW
43
6. M ETHOD
6.1 Participants
Participants were University of Haifa students who were native Hebrew
speakers. They were tested in two groups (for experiments 1 and 2, respec-
tively) in a class setting and received no compensation for participating in
the experiment. Twenty-four participants took part in experiment 1. An
additional group of twenty-two students participated in experiment 2.
6.2 Materials
The materials consisted of novel words, generated by conjugating 13 novel
roots in existing word templates. The roots were arranged in trios. One
member of the trio had adjacent homorganic consonants root finally, the
second root member had adjacent identical consonants root finally and a
third member included no homorganic consonants. The homorganic and
identical trio members were matched on the place of articulation of the
homorganic consonant pairs. Homorganic consonants were sampled from
one of five natural classes : nonsonorant coronals (t, z, s, c; a total of four
trios), labials (b, m, p ; a total of four trios), velar stops (g, k ; a total of
two trios), gutturals (x, ?,@; a total of two trios) and coronal sonorants (n, l ;
a total of one trio).
5
To assure that the distinction between members of the trio was not due to
familiarity with specific consonant combinations, we matched the three types
of roots for the co-occurrence of their radicals. Our calculation of the co-
occurrence of root radicals was based on a database of 1449 productive tri-
consonantal roots listed in the Even-Shoshan dictionary (1993). We consider
a root productive if it appears in at least one verbal form. The database lists
the orthographic transcription of all the triliteral productive roots in Even-
Shoshan (1993), including smm-type roots. Using this database, we deter-
mined the frequency of any two root-radicals (bigrams), including both
adjacent (i.e. C1C2, C2C3) and nonadjacent (i.e. C1C3) combinations. For
each member of the trio, we calculated the sum positional bigram frequency.
For instance, the novel root gkl has a summed bigram frequency of 14, be-
cause its C2C3 combination occurs in four roots (skl,tkl,skl,@kl), its C1C3
bigram occurs in ten roots (grl,gll,g?l,g@l,gyl,gbl,gdl,gml,gzl,gxl) and
its C1C2 bigram does not occur in any root. The mean summed positional
bigram frequencies were 11.5 (SD=2.0), 12.5 (SD=2.5) and 11.8 (SD=2.3) for
[5] The experiment also included three additional trios with the phoneme /r/. Many speakers of
Modern Hebrew encode this consonant as the uvular fricative /R/ (Sandler 1994). In view of
the great variability in the pronunciation of this phoneme, its classification for homor-
ganicity is uncertain. Accordingly, we excluded these items from all subsequent analyses.
I.BERENT & J.SHIMRON
44
the homorganic, identical and control roots, respectively. None of these
mean-differences approached significance by planned comparisons.
The experimental materials were generated by conjugating each root trio in
one of three classes of word patterns, as described above. Note that the
conjugation of roots could potentially lead to OCP violations due to identity
or homorganicity between the stem and suffixes. This problem is particularly
prevalent in the verbal system, which includes the suffixes ti,ta,t,nu,tem,ten.
To minimize such effects, we refrained from concatenating suffixes that share
the same manner and place of articulation with the root final radical. Accord-
ingly, roots ending with a coronal sonorant consonant were not paired with
the suffix-nu. Likewise, roots ending with a coronal obstruent stop were
never followed by suffixes beginning with a coronal obstruent stop. Further-
more, because roots with identical and homorganic C2C3 were matched for
place of articulation and suffix, any residual effects of suffix-stem similarity
must apply equally to both pair members. Accordingly, the differences be-
tween identical and homorganic roots must reflect their root structure. Each
root trio was conjugated in each of these three word classes, resulting in a
total of 39 word trios.
6.3 Procedure
Participants were presented with a printed list including 117 words, generated
by conjugating the 13 root trios described above in three word classes. The
words were typed and their vowels were encoded by the standard diacritic
marks. Participants were instructed to read each word carefully, while at-
tending to its pronunciation, and rate its acceptability as a possible Hebrew
word. The two rating experiments differed, however, in the precise rating
judgment that was required.
Participants in the relative rating procedure (experiment 1) were presented
with the experimental materials typed in trios. The order of the word trios
was random. Likewise, the order of the words within each word trio was
randomly determined. They were next instructed to compare the words in the
trio and assess the extent to which they sounded like a possible Hebrew
word. Participants were asked to assign the rating 1 to the word that sounded
the best, 2 to the word that sounded intermediate, and 3 to the word that
sounded the worst. To express high acceptability ratings by larger numbers,
we inverted the scale by subtracting each score from 4. Thus, in our report,
1 corresponds to the word that sounds worst and 3 indicates the word that
sounds best.
In the absolute rating experiment, the same words were presented in a single
randomized list, and participants were asked to provide a rating for each word
in isolation (rather than compare it to its matched trio members). Responses
were provided using a scale of 1–5 (1=worst, 5=best).
IDENTICAL CONSONANTS IN HEBREW
45
7. R ESULTS
7.1 Relative rating
Acceptability ratings were submitted to 2-way ANOVAs by participants
and items. All results referred to as significant have a p-value that is lower
than 0.05. Both of these analyses yielded a significant effect of root type
(F
1
(2, 46)=54.79, MSE=0.116; F
2
(2, 24)=7.03, MSE=0.497). The main ef-
fect of word class (F
1
(2, 46)=4.60, MSE=0.001 ; F
2
(2, 24)<1, MSE=0.000,
n.s.) and its interaction with root type (F
1
(4, 92)=4.00, MSE=0.057 ;
F
2
(4, 48)=1.32, MSE=0.098) were significant in the analysis by partici-
pants but not by items (‘ n.s. ’ is used throughout to indicate nonsignificant
effects). Figure 4 plots mean acceptability ratings as a function of root type
and word class.
The predictions of the similarity and identity hypotheses were further tested
using planned comparisons. Both accounts predict that homorganic roots
should be less acceptable than nonhomorganic controls. This prediction was
supported across word classes (F
1
(1, 46)=109.29; F
2
(1, 24)=14.05), as well as
within each of the three word classes separately (in the first word class :
(F
1
(1, 92)=49.54; F
2
(1, 48)=19.73); in the second word class : (F
1
(1, 92)=
100.45; F
2
(1, 48)=33.33); in the third word class : (F
1
(1, 92)=62.56;
F
2
(1, 48)=19.51)). The critical prediction that contrasts the two accounts,
however, concerns roots with homorganic vs. identical consonants. Contrary
to the prediction of the monotonic similarity hypothesis, roots with identical
consonants were more acceptable than the less similar, homorganic roots.
This trend reached significance in the analysis by participants across word
classes (F
1
(1, 92)=22.69; F
2
(1, 48)=3.12, p=0.09). Likewise, identical con-
sonants were significantly more acceptable compared to homorganic con-
sonants in the first (F
1
(1, 92)=26.96; F
2
(1, 48)=9.35) and third word classes
(F
1
(1, 92)=19.96; F
2
(1, 48)=6.16), as well as in the analysis by participants
(but not by items) in the second word class (F
1
(1, 92)=4.46; F
2
(1, 48)=1.84,
p=0.18, n.s.).
The greater acceptability of fully similar, identical root radicals compared
to partially similar, homorganic radicals is incompatible with the view that
root structure is governed by a single monotonic constraint on perceived
similarity. Instead, this finding suggests that the constraint on total identity is
distinct from the restriction on place of articulation. Our findings, however,
suggest that root final identity is not entirely acceptable. Roots with identical
consonants were rated as less acceptable than controls across word classes
(F
1
(1, 46)=32.39; F
2
(1, 24)=3.93, p=0.06). This trend was significant in the
second (F
1
(1, 92)=62.56; F
2
(1, 48)=19.50) and it was significant by par-
ticipants in the third (F
1
(1, 92)=11.84; F
2
(1, 48)=3.74, p=0.06) and first
word classes (F
1
(1, 92)=7.38; F
2
(1, 48)=1.91, p=0.18, n.s.). Note that the
unacceptability of roots with identical consonants cannot be explained
simply by their homorganicity, as these roots were rated significantly higher
I.BERENT & J.SHIMRON
46
than the homorganic roots. We return to discuss the reasons for the partial
unacceptability of smm-type roots in section 8.
7.2 Results – absolute rating
Experiment 2 presents a more stringent test for the identity hypothesis by
introducing a simple modification to the rating procedure. Participants in
experiment 2 were presented with the same words as were used in experiment 1.
These words, however, were now arranged in a single randomized list, rather
than in matched trios. Participants were simply asked to determine the ac-
ceptability of each word on its own (instead of comparing it to its matched trio
members). The change in the rating procedure makes it possible to assess
the generality of the restrictions on identical consonants. Indeed, one may be
concerned that the explicit demand to compare words that differ solely in their
root structure (in experiment 1) could give rise to deliberate, nonlinguistic
discrimination strategies. In contrast, the absolute rating task does not call
attention to root structure, as words may be distinguished by a variety of other
factors, such as their word class, the place of articulation of root consonants,
etc. The absolute rating task thus minimizes the role of deliberate, non-
linguistic discrimination strategies relative to the comparative rating task used
1
2
3
First Second Third
Word class
Acceptability
Homorganic
Identical
Control
Relative rating
Figure 4
Mean acceptability ratings for roots with final homorganic consonants, final identical
consonants and controls as a function of root type and word class (experiment 1)
IDENTICAL CONSONANTS IN HEBREW
47
in experiment 1. If the restrictions on identical consonants form part of
speakers’ linguistic competence, and are routinely applied in the represen-
tation of novel words, then we expect to observe a distinction between identical
and homorganic consonants even when attention to root structure is not
highlighted.
The results of this experiment offer support for the identity hypothesis.
Participants’ ratings were analyzed by ANOVAs (3 root typesr3 word
classes) using both participants and items as random variables. The ANOVAs
yielded a significant main effect of root type (F
1
(2, 42)=33.14, MSE=0.121;
F
2
(2, 24)=7.65, MSE=0.319) and word class (F
1
(2, 42)=4.77, MSE=0.55;
F
2
(2, 24)=11.61, MSE=0.132), and a marginally significant interaction of
root typerword class (F
1
(4, 84)=2.42, MSE=0.087 ; F
2
(4, 48)=1.86, MSE=
0.068, p=0.13, n.s.). Figure 5 plots mean acceptability ratings as a function
of root type and word class.
Planned comparisons demonstrated that, across word classes, roots with
homorganic consonants were rated as less acceptable than controls, a finding
that emerged across word classes (F
1
(1, 42)=65.18; F
2
(1, 24)=15.06) as well
as within each of the word classes separately (in the first word class :
(F
1
(1, 84)=36.53; F
2
(1, 48)=27.90); in the second word class : (F
1
(1, 84)=
42.99; F
2
(1, 48)=33.18); in the third word class : (F
1
(1, 84)=15.57, F
2
(1, 48)=
12.09)). Replicating the findings of experiment 1, however, roots with identical
consonants root finally were significantly more acceptable than roots with
homorganic consonants (F
1
(1, 42)=24.49; F
2
(1, 24)=5.57). This finding was
significant for the first (F
1
(1, 84)=25.20; F
2
(1, 48)=19.63) and second
1
3
4
First Second Third
Word class
Acceptability
Homorganic
Identical
Control
Absolute rating
2
Figure 5
Mean acceptability ratings for roots with final homorganic consonants, final identical
consonants and controls as a function of root type and word class (experiment 2)
I.BERENT & J.SHIMRON
48
(F
1
(1, 84)=16.79; F
2
(1, 48)=12.09) word classes, but not for the third word
class (F
1
(1, 84)=1.05, p=0.31, n.s.; F
2
(1, 48)<1, p<0.3, n.s.). Although roots
with identical consonants were more acceptable than homorganic roots, they
were not considered perfectly acceptable. Across word classes, roots with
final identical consonants were rated less acceptable than controls, a finding
that was significant by participants but not by items (F
1
(1, 42)=9.76;
F
2
(1, 24)=2.31, p=0.14, n.s.). The relative unacceptability of smm-type roots
was significant in the second (F
1
(1, 84)=6.04; F
2
(1, 48)=5.21) and third
(F
1
(1, 84)=8.52; F
2
(1, 48)=6.54) word classes, but did not reach significance
in the first word class (F
1
(1, 84)=1.05, p=0.31; F
2
(1, 48)<1, n.s.).
7.3 Results – summary
The results of experiments 1 and 2 support the following conclusions.
(a) Homorganicity at the root’s end is undesirable. Speakers consider roots
with homorganic consonants to be less acceptable than controls. This
conclusion agrees with the previous results of Frisch & Zawaydeh (2001)
in Jordanian Arabic. Unlike these earlier studies, our experiments have
systematically manipulated the position of C2C3 homorganic consonants
within the word. Our Hebrew findings demonstrate the avoidance of
homorganic root consonants across a variety of morphological templates,
regardless of the position of the homorganic consonants in the word.
(b) The acceptability of identical consonants is inexplicable by similarity.
Contrary to the prediction of the monotonic similarity account, roots with
C2C3 identical consonants are more acceptable than roots with C2C3
homorganic, nonidentical consonants. In fact, these experiments each
yielded higher ratings for root final identical consonants compared to
homorganic nonidentical consonants, a trend that was significant in both
experiments for the first word class. The greater stability of the effect for
the first word class may be due either to the transparency of its mor-
phological structure or to its frequency in the language. These factors
may facilitate the extraction of the root, thereby increasing the salience
of its internal structure. Sensitivity to root structure, however, emerged in
other word classes as well. These results suggest that speakers encode the
identity of root consonants, and their representation distinguishes be-
tween identical and homorganic consonants.
(c) Identity is not entirely desirable. Replicating previous results (Berent &
Shimron 1997; Berent et al. 2001(a)), roots with identical consonants at
their end were rated lower than controls. This last result is unexpected
under the view that root final identity does not violate the OCP
(McCarthy 1986). Although, as we explain below, the source of the relative
unacceptability of smm-type roots remains uncertain, it is, nevertheless,
IDENTICAL CONSONANTS IN HEBREW
49
patent that this finding is due to the identity of the C2C3 radicals, rather
than to their similarity.
8. D ISCUSSION
Two experiments examined whether the acceptability of identical root con-
sonants is an inverse, monotonic function of their perceived similarity.
Although speakers are highly sensitive to similarity among root consonants,
their ratings of identical consonants clash with the monotonic similarity ex-
planation. We wish to make it perfectly clear that our experiments do not
address the adequacy of the stochastic similarity model as an account for the
restrictions on nonidentical root radicals. In fact, the similarity hypothesis
may easily be modified to account for our findings by incorporating a separate
set of constraints that specifically target identity. The merit of our conclusions,
however, should not be judged by the repair they impose on alternative
approaches but, instead, by their centrality to the domain of inquiry. The
existence of grammatical constraints that specifically target full-segment
identity is a question that is central to phonological theory. The phonological
literature includes countless case studies implicating a distinction between full
and partial identity in grammar. Our findings provide corroborative exper-
imental evidence for such a distinction. These results suggest that speakers
treat identity in a manner that is fundamentally different from partial simi-
larity. To the extent that Semitic roots are subject to a stochastic constraint on
perceived similarity, this constraint must be complemented by a separate set of
grammatical principles or constraints concerning identity.
The role of identity in the grammar of Semitic has been the subject of some
debate in the linguistic literature. McCarthy (1986) attributes the co-occur-
rence restrictions on identical root consonants to a constraint that specifically
bans identity. Identity avoidance operates at the root domain and concerns
adjacent radicals. The perceived adjacency of root radicals, in turn, is captured
by segregating them from vowels and affixes. Each of these assumptions,
however, has been met with some criticism. Pierrehumbert (1993) observed
that identical root radicals are underrepresented at nonadjacent (C1C3) po-
sitions in Arabic roots, a finding that is inconsistent with the local definition
of the OCP-total by McCarthy (see also Buckley 1997, for Tigrinya). Bat-El
(1994) further rejected the OCP-total as an account for the distribution of
adjacent identical consonants, suggesting that samam-type verbs are formed
from monosyllabic nouns (e.g. sam) by means of reduplicative stem modi-
fication (see also Ussishkin 2000). More generally, Gafos (1998) criticized
long-distance spreading on the grounds that it requires the segregation of
consonants and vowels, in general, and, in the case of Semitic roots, the
segregation of root consonants from vowels and affixes. Plane segregation
incorrectly predicts unattested partial spreading of consonantal features
across vowels (Gafos 1998). The proposals of Bat-El (1994), Gafos (1998) and
I.BERENT & J.SHIMRON
50
Ussishkin (2000) each capture the asymmetry in the distribution of identical
consonants in Semitic verbs without appealing to either the root or identity
avoidance. Additional evidence against the consonantal root challenges its
role as a morphological constituent. Such observations suggest that Hebrew
word formation is sensitive to the interdigitation of consonants and vowels.
For instance, Bat-El (1994) observes that Hebrew verbs formed from nouns
maintain the noun’s clusters (e.g. praklitppriklet). Likewise, the shape of
roots in denominal Hebrew verbs is predictable from the vowel in the base
(Ussishkin 1999, 2000).
These observations raise three questions with respect to the interpretation
of our findings: (a) Do speakers of Semitic languages encode the identity of
consonants relative to a morphological domain that coincides with the root?
(b) How is this domain encoded ? (c) Is identity banned by the grammar ?
Our results clearly demonstrate that speakers can discriminate between
novel words that differ solely in terms of the identity of root final consonants.
Speakers are further sensitive to the presence of identical consonants re-
gardless of their position in the word, and despite intermediate vowels and
affixes (Berent & Shimron 1997; Berent et al. 2001(a)). For instance, ssm-type
roots are unacceptable when the identical consonants are either word initial
(e.g. sisem) or word medial (histasem), and when they are separated by various
vowels (e.g. ior a) or an infix (e.g. -ta- ; see Berent et al. 2002). The general-
ization of the restrictions on identical consonants to novel words is theo-
retically significant. Several existing accounts consider the asymmetry in the
location of identical stem consonants in existing denominal verbs as an
emerging property of anchoring the biconsonantal nominal base with the
output (e.g. Bat-El 1994 ; Gafos 1998 ; Ussishkin 2000). These proposals
successfully capture the distribution of identical root consonants without
invoking grammatical constraints on identity. The asymmetry in the location
of identical consonants for novel verbs, for which a biconsonantal base is
not independently motivated, challenges this approach. It is, of course,
perfectly possible that speakers infer a biconsonantal base for novel
smm-type forms as well. The erasure of identity in novel bases calls for an
explanation, however. Such an explanation may well require a grammatical
constraint on full-segment identity. To capture the restriction on identical
consonants across intermediate vowels and affixes, and irrespective of word
position, the identity constraint must further operate within a morphological
domain.
The nature of this morphological domain, however, is more difficult to
ascertain. The perceived adjacency of identical C1C2 and C2C3 root radicals
is handled naturally by the view of identity as constrained within the root
domain, a phonological constituent that renders root consonants adjacent
by segregating them from vowels. The segregation of the consonantal root
in Semitic can further account for the prevalence of Arabic root-consonant
metathesis errors in the output of a bilingual Arabic-French aphasic despite
IDENTICAL CONSONANTS IN HEBREW
51
their rarity in his French output (Prunet, Be
´land & Idrissi 2000). As noted
above, however, this approach predicts unattested effects of long-distance
spreading (Gafos 1998). Furthermore, the restriction on consonant-identity
is not entirely blind to intervening vowels. Previous experiments consistently
demonstrate that root initial identical consonants are less acceptable when
they are locally adjacent (e.g. massimim ; the initial ma- and final -im are
affixes), compared to when they are separated by vowels (e.g. sisem; Berent
& Shimron 1997; Berent et al. 2001(a)). This finding suggests that the rep-
resentation available to speakers specifies the interdigitation of consonants
and vowels, an observation consistent with the proposals of Bat-El (1994) and
Gafos (1998). To address these concerns, the identity of stem consonants must
be captured without segregating the consonantal root from vowels.
Several existing accounts encode the identity of consonants in this fashion.
Although these proposals are designed to capture distinct roles of identity in
the grammar (i.e. identity avoidance vs. identity enforcement), they involve
several strategies for encoding the identity between disjoint segments. One
approach restricts identity between both adjacent and disjoint stem con-
sonants by means of a family of identity-avoidance constraints (Everett &
Berent 1998).
6
Although this approach captures the acceptability of identical
root consonants in Hebrew (homorganicity is not specifically addressed), it
does so by invoking numerous related constraints, each of which targets a
distinct type of identity (the stem’s C1C2 consonants, adjacent identical
consonants and any identical stem consonants). An alternative approach
assumes a single identity constraint that can be selectively applied to a pair
of segments by virtue of their correspondence (Rose & Walker 2001). This
approach enforces correspondence between (disjoint) similar consonants and
favors full identity among these corresponding segments. Extending this
general approach to Hebrew C2C3 consonants would correctly predict an
advantage for full identity over similarity (provided that C1C2 identity is
independently banned). A correspondence relation between identical stem
consonants is also assumed by Bat-El (2002). This proposal accounts for
speakers’ ability to generalize the restrictions on long-distance identity to
novel forms (i.e. without prior knowledge of the base). Bat-El (2002) suggests
that the correspondence between identical stem consonants favors an in-
terpretation such that one is a copy of the other, provided that the copied
segment occupies the rightmost position in both the base and the stem. In
summary, each of these three proposals offers a mechanism for restricting
the co-occurrence of disjoint identical consonants without segregating them
from vowels. Whether speakers encode the domain of identity-avoidance
[6] Although Everett & Berent’s (1998) proposal invokes the root, there is nothing in their
analysis that requires the representation of the root as a separate phonological constituent.
Indeed, they specifically assume that identity formation is achieved by reduplication, rather
than by long-distance spreading.
I.BERENT & J.SHIMRON
52
as the stem or the root – a phonological constituent that is segregated from
vowels – is a question that falls beyond the scope of this discussion. Likewise,
our discussion cannot address the role of the root in the morphology of
Semitic, a question that is largely orthogonal to its role as a phonological
constituent. Regardless of whether the co-occurrence restrictions on identical
consonants operate over a root or a stem, however, speakers are clearly
sensitive to long-distance identity and discriminate it from nonidentity.
Although the grammar of Hebrew appears to encode long-distance identity,
it is uncertain whether it bans it. The experimental results demonstrate that
roots with final identity are rated as relatively unacceptable compared to
nonhomorganic controls. These results replicate our previous rating experi-
ments (Berent & Shimron 1997 ; Berent et al. 2001(a)) indicating that root
final identity is not entirely desirable. These findings also converge with the
observations of Rose (2000), who documented long-distance avoidance of
root final identical consonants (specifically, gutturals) in Tigrinya. Unlike
the aversion to identical gutturals observed by Rose (2000), the unaccept-
ability of root final identical consonants in Hebrew cannot be explained
by homorganicity : if the unacceptability of smm-type roots was due to
homorganicity, then this rejection should have been at least comparable to
(in fact, larger than) the rejection of homorganic controls. Contrary to this
prediction, smm-type roots are more acceptable than homorganic controls.
The relative unacceptability of smm-type roots must then be due to their total
identity.
Why are smm-type roots relatively unacceptable ? One explanation attri-
butes this finding to a grammatical constraint that bans long-distance identity
(Everett & Berent 1998 ; Rose 2000). Smm-type forms are nevertheless attested
because identity avoidance is dominated by higher-ranked constraints. For
instance, Everett & Berent (1998) attribute the emergence of simem to the
preference to fully align its biconsonantal base with the three consonant slots
in the word pattern (i.e. MAXIO, see Everett & Berent 1998). The location
of identity, in turn, is explained by the lower ranking of identity violation
at the left edge of the root (Everett & Berent 1998). Conversely, according
to Bat-El (1994, 2003), the emergence of long-distance identity in verbs such
as simem is due to a constraint on the minimal word requiring bisyllabicity,
whereas its location is explained by the leftward alignment of the base
and the reduplicated output (but see Bat-El 2002 for a different approach).
Despite their optimality with respect to higher ranked constraints, the ident-
ical consonants in simem violate an identity constraint (Everett & Berent 1998 ;
Rose 2000; Bat-El 2003). The relative unacceptability of smm-type roots in
our experiments may reflect such a violation of a grammatical constraint. How-
ever, the empirical evidence for identity violation in these forms is currently
limited. To the extent that this experimental finding is not corroborated by
converging linguistic evidence, an alternative explanation may be considered.
This account attributes the relative unacceptability of smm-type roots to
IDENTICAL CONSONANTS IN HEBREW
53
the rarity of XYY root types (relative to the XYZ type) in the lexicon, rather
than to an active grammatical constraint.
7
The question of whether long-distance identity is banned by the grammar
must await further linguistic analysis. Our present goal, however, was to
examine whether identity of root final consonants is represented and dis-
tinguished from homorganicity. The experimental evidence on this question
is rather clear : the co-occurrence restrictions on identical consonants are
inexplicable as a monotonic function of their similarity. Long-distance con-
sonant identity, a formal relationship among segments, is thus irreducible to
partial similarity at the level of individual features.
REFERENCES
Bat-El, O. (1994). Stem modification and cluster transfer in Modern Hebrew. Natural Language &
Linguistic Theory 12. 571–596.
Bat-El, O. (2002). Hebrew reduplication: the interpretation of forms with identical consonants.
Paper presented at a colloquium, Department of Linguistics, University of Southern Cali-
fornia.
Bat-El, O. (2003). Semitic verb structure within a universal perspective. In Shimron, J. (ed.),
Language processing and language acquisition in a root-based morphology. Amsterdam: John
Benjamins. 29–59.
Bender, M. L. & Fulass, H. (1978). Amharic verb morphology, a generative approach. East Lansing,
MI: African Studies Center, Michigan State University.
Berent, I., Bibi, U. & Tzelgov, J. (2000). The autonomous computation of linguistic structure in
reading: evidence from the Stroop task. A paper submitted to the 41st meeting of the Psycho-
nomic Society.
Berent, I. & Shimron, J. (1997). The representation of Hebrew words: evidence from the
Obligatory Contour Principle. Cognition 64. 39–72.
Berent, I., Everett, D. & Shimron, J. (2001(a)). Do phonological representations specify formal
variables? Evidence from the Obligatory Contour Principle. Cognitive Psychology 42. 1–60.
Berent, I., Shimron, J. & Vaknin, V. (2001(b)). Phonological constraints on reading: evidence from
the Obligatory Contour Principle. Journal of Memory and Language 44. 644–665.
Berent, I., Marcus, G., Shimron, J. & Gafos, A. (2002). The scope of linguistic generalizations :
evidence from Hebrew word formation. Cognition 83. 113–139.
Bolozky, S. (1978). Some aspects of Modern Hebrew phonology. In Berman, R. (ed.), Modern
Hebrew structure. Tel Aviv: Universities Publishing Projects. 11–67.
Buckley, E. (1997). Tigrinya root morphology and the OCP. University of Pennsylvania Working
Papers in Linguistics 4.3. 19–51.
[7] The sensitivity of our participants to type frequency could further be prompted by the explicit
demand to provide acceptability ratings. Indeed, when well-formedness is assessed implicitly,
there is no evidence that smm-type roots are unacceptable. For instance, consider again the
lexical decision task (Berent et al. 2001(b)). Participants in this task are asked to determine
whether or not a string of letters corresponds to an existing word by pressing one of two
computer keys. Of particular interest are responses to novel words: the greater the ill-
formedness of a novel word, the easier it should be for participants to classify it as a nonword.
Specifically, if novel words with smm-type roots were ill-formed, then it should have been
easier to classify them as nonwords, compared to matched controls with no identical con-
sonants. Contrary to this expectation, nonword classification is significantly MORE difficult
for smm-type roots relative to no-identity controls. Berent et al. (2001(b)) suggested that this
difficulty reflects the perception of identity as a product of a grammatical operation, an
attribute that renders novel stimuli more word-like. Clearly, however, the implicit measure of
lexical decision yields no evidence for the unacceptability of smm-type roots.
I.BERENT & J.SHIMRON
54
Even-Shoshan, A. (1993). Ha’milon ha’xadash [The new dictionary]. Jerusalem: Kiryat Sefer.
Everett, D. L. & Berent, I. (1998). An experimental approach to the OCP: evidence for violable
identity constraints in Hebrew roots. Rutgers Optimality Archive (ROA-235).
Frisch, S. (in press). Language processing and segmental OCP effects. In Hayes, B., Kirchner, R. &
Steriade, D. (eds.), Phonetic bases of markedness. Cambridge: Cambridge University Press.
Frisch, S., Broe, M. & Pierrehumbert, J. (1997). Similarity and phonotactics in Arabic. Ms.,
University of Michigan, Ann Arbor.
Frisch, S. A. & Zawaydeh, B. A. (2001). The psychological reality of OCP-place in Arabic.
Language 77. 91–106.
Gafos, A. (1998). Eliminating long-distance consonantal spreading. Natural Language &
Linguistic Theory 16. 223–278.
Greenberg, J. (1950). The patterning of root morphemes in Semitic. Word 6. 162–181.
McCarthy, J. (1981). A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12.
373–418.
McCarthy, J. (1986). OCP effects : gemination and antigemination. Linguistic Inquiry 17. 207–263.
McCarthy, J. (1994). The phonetics and phonology of Semitic pharyngeals. In Keating, P. (ed.),
Papers in laboratory phonology III. Cambridge: Cambridge University Press. 191–283.
Pierrehumbert, J. (1993). Dissimilarity in Arabic verbal roots. In Proceedings of NELS 23, GLSA.
Amherst, MA: Department of Linguistics, University of Massachusetts. 367–381.
Prunet, J. F., Be
´land, R. & Idrissi, A. (2000). The mental representation of Semitic words.
Linguistic Inquiry 31. 609–648.
Rose, S. (2000). Rethinking geminates, long-distance geminates and the OCP. Linguistic Inquiry
31. 85–122.
Rose, S. & Walker, R. (2001). A typology of consonant agreement as correspondence. Ms., Uni-
versity of California at San Diego & University of Southern California.
Sandler, W. (1994). Suboral articulation. Ms., Haifa University.
Ussishkin, A. (1999). The inadequacy of the consonantal root: Modern Hebrew denominal verbs
and output-output correspondence. Phonology 16. 401–442.
Ussishkin, A. (2000). The emergence of fixed prosody. Ph.D. dissertation, University of California
at Santa Cruz.
Authors’ address: Department of Psychology, Florida Atlantic University,
777 Glades Road, Boca Raton, FL 33431-0991, U.S.A.
E-mail: Iberent@fau.edu
IDENTICAL CONSONANTS IN HEBREW
55