In WSCLA X: Proceedings of the 10th Workshop on Structure and Constituency in the Languages of the Americas,
ed. by S. Armoskaite and J. Thompson, pp. 109-122. (UBC Working Papers in Linguistics, 17.) Vancouver: Dept.
of Linguistics, University of British Columbia, 2005.
Latent segments in Yowlumne: an epiphenomenon of template satisfaction?
Gunnar Ólafur Hansson
University of British Columbia
Several affixes in Yowlumne (Yawelmani) contain initial consonants which exhibit otherwise unexpected C~Ø alternations,
so-called latent segments. These have typically been interpreted as underlyingly floating elements of some kind. An
alternative, representationally simpler and empirically superior
analysis is proposed, which takes latent consonants to be nondistinct from regular consonants. Latency is viewed as a consequence of differences between the phonologies associated
with the templatic and nontemplatic strata of morphology.
1
Introduction
’ lamni/, also known as Yawelmani), certain affixes
In Yowlumne (/yaw
contain so-called latent or “ghost” segments. These are either present or absent
depending on the prosodic and phonotactic context, in a way that normal
segments are not (Archangeli 1983, 1988, 1991; Noske 1985; Zoll 1993, 1998,
2001a, b). The distinction between full and latent consonants is illustrated in (1).
An otherwise unsyllabifiable consonant cluster is resolved by vowel epenthesis
when full consonants are involved (1a), but when one of the consonants is a
latent one, we instead find deletion of that consonant (1b).
(1)
Suffixes with full /h/ vs. latent /h/ (V-epenthesis vs. C-deletion):
a.
b.
//a˘ml-hat.'-/
//a˘ml-(h)atn-/
/a˘.mil.-hat.'-
‘call for help’
/am.l-_atn-
‘desire to help’
Generative analyses of Yowlumne phonology have generally taken
latent segments to be representationally distinct from full segments, by being
structurally “defective” in some way, e.g., as floating root nodes (Archangeli
1983, 1988; Noske 1985) or floating subsegmental elements (features or featuregeometric nodes; Zoll 1998, 2001a, b). This paper presents a radically different
analysis which is more compatible with the full range of facts on the distribution
of latent segments. The central thesis is that latent consonants are not special in
any representational sense (with the possible exception of the so-called glottal
quasi-segment), but are in fact just normal consonants. It is argued that “latency”
as such—that is, deletion through underparsing—is instead a predictable consequence of differences in the phonologies associated with the two layers of verb
morphology in Yowlumne: templatic and non-templatic affixation.
Section 2 gives relevant background information on Yowlumne
phonology and morphology. The basic facts of the patterning of latent conso-
nants are presented in section 3, and previous generative analyses are summarized. Section 4 presents some striking correlations between the distribution of
latent consonants and the templatic/nontemplatic affix distinction, correlations
which have gone unnoticed (or at any rate unappreciated) in earlier works. On
the basis of these facts, an alternative analysis is presented which renders any
representational distinction between full and latent consonants superfluous and
misleading. Section 5 deals with an apparent counterexample and briefly
addresses the status of latent vowels.
2
Some preliminaries
Yowlumne (better known as Yawelmani in the linguistic literature) is a
severely endangered Yokutsan language of south-central California, familiar to
most linguists from a vast number of analytical treatments in the theoretical
literature over the past half-century or so. The main description and data source
on the language is Newman (1944), from which all data in this paper are drawn.
2.1
Syllable structure and syllabification
The maximal syllable canon in Yowlumne is [CVX], i.e. syllables are
maximally either CV˘ or CVC, and this is enforced by shortening and epenthesis
wherever necessary. Vowel shortening in closed syllables is illustrated in (2).
Note that lowering of long /i u/ to [e o] (more accurately [E ç]), observed in (2b),
is conditioned by underlying (or template-induced) vowel length, not surface
length; lowering is not directly relevant to the issues addressed in this paper.
(2)
Closed-syllable shortening (to avoid CV˘C syllables)
a.
/la˘n-k’(a)/
lan.-k’a
‘hear!’
b.
/wu˘/y-i˘n/
wo/.y-en
‘will go to sleep’
Epenthesis of [i] (potentially [u] due to rounding harmony) is illustrated
in (3). Here and throughout the paper, epenthetic vowels are indicated by underlining, while deletion sites are indicated by an underscore as in (1b) above.
(3)
Vowel epenthesis (to avoid complex onsets/codas)
a.
b.
2.2
/bat.’n-k’a/
/logw-t/
ba.t.'in.k'a
log.w-it
‘fall down!’
‘was pulverized’
Templatic morphology
Regular verb roots (ones that participate fully in the templatic system)
contain either 2 or 3 consonants and one vowel quality. There are three basic
template shapes which can be imposed on verb roots, and these may be defined
in prosodic terms roughly as shown in (4) (see Archangeli 1983, 1988, 1991).
(4)
a.
monomoraic
[σµ] ≈ CVC(C)
b.
bimoraic
c.
[σµµ] ≈ CV˘C(C)
iambic
[σµσµµ] ≈ CVCV˘(C)
Individual suffixes are either templatic or (more commonly) nontemplatic. Nontemplatic suffixes simply attach to the root in its underlying shape
and, unlike templatic ones, may attach not only to a bare regular root but also to
irregular roots or morphologically complex (already suffixed) bases. A templatic
suffix can attach only to an unsuffixed regular root, and superimposes on it one
of the three templates in (4). The contrast is illustrated in (5): in each of the two
pairs, the same root is followed by a templatic suffix in the first example and a
nontemplatic suffix in the second one. Throughout the paper, templatic affixes
are shown with a subscript index for the particular template they trigger.
(5)
a.
causative /-a˘la˘[σµ]/ (templatic) vs. passive /-t/ (nontemplatic)
/c'u˘m-a˘la˘[σµ]-k'(a)/
/c'u˘m-t/
b.
c'u.m-a˘.la-k' ‘make [him] destroy!’
c'o˘.m-ut
‘was devoured’
continuative /-(/)a˘[σµµ]/ (templatic) vs. future /-i˘n/ (nontempl.)
/hiwi˘t-(/)a˘[σµµ]-xo˘-hn/
/hiwi˘t-i˘n/
’ .t-a˘.-xo˘.-hin
hew
hi.we˘.t-en
‘walked about’
‘will walk’
As can be seen in (5), the phonological processes described in 2.1 often obscure
the prosodic shape of templates in actual surface forms, rendering the whole
templatic system fairly opaque. Another factor contributing to this opacity is the
fact that coda consonants do not count as moraic for the purpose of template
satisfaction, though they are obviously moraic in surface forms as such, as evidenced by the closed-syllable shortening in (2).
3
Latent segments in Yowlumne
3.1
The basic facts
Latent segments (also referred to as “ghost” segments) are affixal
segments which may or may not get realized on the surface, depending on the
prosodic or phonotactic context. Yowlumne is generally assumed to have not
only latent consonants but latent vowels as well. The factors governing the
distribution of the two are somewhat different, and the focus of this paper is
exclusively on latent consonants; latent vowels are addressed in §5.1. A
representative list of suffixes containing latent consonants is given in (6).
(6)
a.
/-(h)ni˘l/
/-(h)atn/
passive adjunctive
desiderative
b.
/-i(l)sa˘/
causative-repetitive
c.
/-(m)a˘m/
decedent (relative)
d. /-(n)/
/-(n)i˘t/
verbal noun
decedent (relative)
The overt manifestation of latency is alternation with zero. A normal
consonant is never deleted, even when it finds itself in an unsyllabifiable cluster;
instead vowel epenthesis is resorted to, cf. (3) above. A latent consonant, by
contrast, gets deleted under precisely the circumstances that normally condition
vowel epenthesis. The resulting C~Ø alternations are illustrated in (7)-(9).
(7)
/-(h)ni˘l/, passive adjunctive (triggers [σµσµµ] root template)
a.
b.
(8)
(9)
/t.'ik'-(h)ni˘l[σµσµµ]-/
t.'i.k'e-h.ne˘l- ‘instrument for being tied’
//a˘ml-(h)ni˘l[σµσµµ]-/ /a.mal.-_ne˘l- ‘place of being aided’
/-(h)atn/, desiderative (triggers [σµ] root template)
a.
/hud-(h)atn[σµ]-/
hud.-hatn-
‘want to know’
b.
//a˘ml-(h)atn[σµ]-/
/am.l-_atn-
‘desire to help’
/-i(l)sa˘/ causative-repetitive ([σµσµµ] template; /i/ docks onto σ2)
ni.ne˘-l.sa˘-
a.
/nini˘-i(l)sa˘[σµσµµ]-/
‘get to keep still several
times’
b.
//opo˘t-i(l)sa˘[σµσµµ]-/ /o.pe˘t.-_sa˘- ‘make get up repeatedly’
The contrast in behaviour between normal and latent consonants is seen
clearly in (10a) vs. (10b). Each form contains the same CC+C cluster at the rootsuffix boundary, the sole difference being whether the cluster-final and suffixinitial /h/ is a latent consonant or not. A normal /h/ triggers vowel epenthesis
(10a), while a latent /h/ undergoes deletion (10b).
(10)
a.
desiderative-reflexive /-hat.'/ (non-templatic)
//a˘ml-hat.'-/
/a˘.mil.-hat.'‘call for help’
b. desiderative /-(h)atn/ (templatic, triggers [σµ] root template)
//a˘ml-(h)atn[σµ]-/
/am.l-_atn‘desire to help’
One thing is perfectly clear and uncontroversial: the crucial factor controlling deletion vs. realization of a latent consonant is syllabification—the
availability of an onset or coda position for the segment in question. When unsyllabifiable, a latent consonant fails to surface (is deleted), whereas a normal
consonant does surface in the same circumstances (aided by vowel epenthesis).
What is far less clear is what the source of this dichotomy might be. In other
words, what property distinguishes latent segments (and/or the affix morphemes
in which they are contained) from normal segments (and/or the affix morphemes
containing them)?
3.2
Previous analyses
All previous treatments of latent segments in Yowlumne have shared
the same basic assumption. As we have just seen, when unsyllabifiable, these
segments trigger a different kind of repair than usual: deletion rather than epenthesis. Therefore (or so the conclusion goes) these segments must somehow be
special entities, different in kind from normal segments. In other words, latent
segments must be representationally distinct entities.
Different works advocate different ways of encoding the supposedly
special representational status of latent consonants, and it is impossible to give
justice to the full details of each proposal here. The passive adjunctive suffix
/-(h)ni˘l[σµσµµ]/ will serve as an example. Archangeli (1983) assumes that the
initial /h/ is marked diacritically as extraprosodic, and is thus left floating when
other consonants associate to available skeletal slots. The extraprosodic /h/ will
only surface when the preceding root happens not to saturate all three C slots of
its CVCVVC template (in effect: when that root is biconsonantal). Archangeli
(1988) does not appeal to extraprosodicity, instead assuming that a “hard-wired”
association of the following /n/ to the leftmost C slot of the affix forces the
initial /h/ to be left floating (other things being equal).1 Noske (1985) proposes
that latent consonants, unlike normal segments, are underlyingly unassociated to
their corresponding skeletal slot. The floating initial /h/ will associate to its C
slot, potentially triggering resyllabification, but only if such resyllabification
will not result in an increased number of empty nuclei (epenthetic vowels). In
other words, the floating /h/ will be parsed (and hence realized on the surface)
only when doing so will not increase the syllable count.
All of the abovementioned analyses take latent consonants to be, in
effect, floating root nodes, i.e. featurally complete segmental units. In a series of
works, Zoll (1998, 2001a, b) departs from this view, interpreting latent segments
as floating subsegments: features or feature-geometric nodes. The main hallmark
of a latent segment is then that it is representationally impoverished, lacking a
root node. For example, the abovementioned suffix would be represented as
/-(H)ni˘l[σµσµµ]/, where “(H)” stands for a floating Laryngeal node dominating a
[spread glottis] specification.
A fundamental aspect of Zoll’s analysis is that latent consonants, being
subsegmental units, are subject only to FAITH(F) constraints (such as MAX(F)),
while full consonants are also subject to FAITH(Seg) constraints (such as
MAX(Seg)). Constraints on possible segments (e.g. *CH; henceforth abbreviated
as SEG-STRUC) will generally prevent a floating affixal subsegment from docking onto a pre-existing segment (an exception being floating glottalization,
discussed in §3.3). Instead, the floating subsegment materializes only through
epenthesis of a root node. Zoll’s analysis is similar to that of Noske (1985) in
that a reluctance to increase the syllable count (by vowel epenthesis) counteracts
the need for realizing latent consonants in the output. Vowel epenthesis is constrained indirectly by *STRUC(σ); the fewer syllables in the surface form, the
fewer violations of *STRUC(σ) incurred.
Given that non-realization (deletion) of a subsegment violates only
MAX(F), while that of a full segment also violates MAX(Seg), the ranking of
each of the two MAX constraints relative to *STRUC(σ) will control the behaviour of latent vis-à-vis full consonants. This is illustrated in (11)-(12). The set of
constraints defining the surface segment inventory of Yowlumne is abbreviated
here as a cover constraint, SEG-STRUC, and those defining the maximal syllable
are similarly abbreviated as σ-STRUC; both are treated as undominated.
1
Archangeli (1991: 275-6) simply marks the initial /h/ diacritically as “not necessarily
syllabifying”, but hesitates to equate this with extraprosodicity and remains somewhat
agnostic on the exact nature of the characterization of latent vis-à-vis normal segments.
(11)

(12)

//a˘ml, -hat.'-/
SEG-
σ-
STRUC
STRUC
a. /aml.-hat.'b. /a.mil.-hat.'c. /am.l-_at.'//a˘ml, -(H)atn[σµ]-/
a.
b.
c.
d.
/aml.-ha.tin/amH.l-a.tin/a.mil.-ha.tin/am.l-_a.tin-
MAX(Seg)
*STRUC(σ)
MAX(F)
*!
**
***
**
*
MAX(Seg)
*STRUC(σ)
MAX(F)
***
***
****!
***
*
*!
SEG-
σ-
STRUC
STRUC
*!
*!
In cases such as these, where base-affix concatenation gives rise to an
otherwise unsyllabifiable consonant cluster, the choice of repair is determined
by the ranking of the relevant constraints. For a full segment, high-ranked
MAX(Seg) rules out deletion, instead forcing an extra violation of lower-ranked
*STRUC(σ) by epenthesis. For a latent segment, in this case a floating [spr. gl.]
Laryngeal node, MAX(Seg) is irrelevant (vacuously satisfied). The ranking
*STRUC(σ) >> MAX(F) ensures that deletion is preferred over epenthesis.
3.3
The glottal quasi-segment
One element in particular patterns in a way which suggests that it is indeed a
floating subsegment. This is the so-called glottal quasi-segment, which shows a
three-way alternation in its realization, shown in (13). The alternation between
(13b) and (13c) matches that found in other kinds of latent consonants. The
difference lies in the docking behaviour seen in (13a), which has no analogue in
latent /(H)/, for example.
(13)
Realization of glottal quasi-segment (/-(/)i˘xo˘/ consequent auxiliary)
a.
As glottalization on a sonorant in root-C2 position, if present:
/s. il-(/)i˘xo˘[σµ]-t/
s.i.l’e˘.xot
‘was being stared at’
/
’
/hulu˘s.-( )i˘xo˘[σµ]-hn/ hul .s.e˘.xo˘.hin
‘remain seated!’
b.
… otherwise as glottal stop if syllabifiable:
/go˘b-(/)i˘xo˘[σµ]-nt/
gob./e˘.xo˘.nit
/t.'ihi˘-(/)i˘xo˘[σµ]-//
c.
‘having been put in’
t.'ih./e˘.xo/
… otherwise not realized (i.e. deleted):
‘was lean’
/lihm-(/)i˘xo˘[σµ]-//
lih.me˘.xo/
‘has already run’
yit'.se˘.xo/
‘have five’
/yit's-(/)i˘xo˘[σµ]-//
The idea that the glottal quasi-segment is a floating [constricted glottis]
feature has long been the received opinion (see Archangeli 1983, 1988, 1991;
Archangeli & Pulleyblank 1994; Zoll 1998, 2001a, b). The central idea behind
Zoll’s proposal that all latent segments are floating subsegments is that [c.g.] is
the only one to show the docking behaviour in (13a) precisely because it is the
only one that can dock onto another segment, given independent limitations of
the Yowlumne segment inventory. For example, the latent /(H)/ = [spread glottis]
of /-(H)ni˘l[σµσµµ]/ or /-(H)atn[σµ]/ cannot dock onto a root sonorant in the form
of aspiration or devoicing, because Yowlumne does not allow aspirated or
voiceless sonorants; cf. candidate (12b) above, here taken to be ruled out by the
cover constraint “SEG-STRUC”.2 For an explicit demonstration of how the threeway alternation in (13) falls out from constraint interaction, the reader is referred
to Zoll (2001a) in particular.
A floating-subsegment account along these lines may well be the most
appropriate analysis for the glottal quasi-segment (though see §4.3). However,
the following section presents evidence suggesting that the other latent consonants—those which never show the docking behaviour of (13a)—are not to be
analyzed as subsegments, nor as floating elements of any other kind.
4
Latent segments and templatic morphology
4.1
Correlations between latency and templaticity
Previous analyses of Yowlumne latent segments have focused on
phonological factors to the exclusion of any morphological aspects of their
patterning. This section presents some striking generalizations about the
distribution of latent consonants across the affix inventory, generalizations
which are unexplained (indeed, unexplainable) by the analyses described above.
These in turn form the basis of the alternative proposal developed in §4.2.
4.1.1
Empirical observations
What previous works have failed to take note of—or at least failed to
recognize as a significant and revealing fact—is that the distribution of latent
consonants across affixes is far from random. In fact, there are strongly suggestive correlations between segmental latency and the templatic status of the affix
containing the segment in question, as well as the phonotactic shape of that affix
and of the (templatically modified) root it attaches to. Two fundamental empirical observations can be stated:
(14)
Latent Cs occur only in templatic suffixes, never in any of the (more
numerous and more productive) nontemplatic suffixes.
(15)
For each templatic suffix, the presence of a latent C (or rather, the
latent status of a given C) in that suffix is predictable, not idiosyncratic.
2
While Yowlumne does have a series of aspirated plosives (conventionally rendered as
/p/, /t/, /k/, etc.), in addition to its unaspirated and glottalized ones, plosives are consistently unaffected by any floating laryngeal feature, even the glottal quasi-segment. The
fact that /(H)/ fails to dock onto stops in the form of aspiration is thus unsurprising.
The import of (14) should be obvious. The generalization in (15)
requires clarification, however, as the sense of “predictable” here hinges on the
interaction between affix shape, template shape, and restrictions on syllable
structure. It is the interplay of these factors that determines when a given affix
consonant is syllabifiable vs. unsyllabifiable—which, if the consonant is a latent
one, translates into realization vs. deletion. This interplay is summarized in (16)
and (17) for affixes imposing disyllabic and monosyllabic (monomoraic or
bimoraic) templates, respectively. Recall that roots are either tri- or biconsonantal, and that the maximal syllable is [CVX], i.e. complex onsets and codas are
not allowed (making triconsonantal clusters impossible).
(16)
(17)
Disyllabic template [σµσµµ] (in effect CVCV˘C- or CVCV˘-)
a.
For affixes of shape /-CV…/, their C1 will always be syllabifiable.
b.
For affixes of shape /-CCV…/ their C1 will sometimes be
syllabifiable (i.e. only if root is biconsonantal).
Monosyllabic templates, [σµ] or [σµµ] (in effect CV(˘)CC- or CV(˘)C-)
a.
For affixes of shape /-CV…/, their C1 will sometimes be
syllabifiable (i.e. only if root is biconsonantal).
b.
For affixes of shape /-CCV…/ or /-C#/, their C1 will never be
syllabifiable.
With these simple facts in mind, the observation in (15) can be broken down
into two separate manifestations, stated in (18)-(19).
(18)
Among affixes taking a monosyllabic template, if an affix has the shape
/-CV…/ then its C1 is latent (will not trigger V-epenthesis).
(19)
Among affixes taking the disyllabic template, if an affix has the shape
/-CCV…/ (or /-C#/) then its C1 is latent (will not trigger V-epenthesis).
There is one notable apparent exception to (19), reflexive/reciprocal
consequent adjunctive /-wsi˘l[σµσµµ]/, which appears to have a non-latent C1
(cf. [lo.wo˘.n-iw.sel] ‘place where [they] attended each other’s feasts’ with
epenthesis rather than *[lo.won.-_sel] with deletion). See §5.2 for arguments
that this suffix is in fact vowel-initial and therefore irrelevant to (19).
4.1.2
Inadequacies of representational analyses
On each of the analyses described in §3, in which latency is encoded
representationally, the correlations just noted are entirely unexpected and
unaccounted for. In those analyses, latency is a phonological phenomenon with
phonology-internal causes. Firstly, latency is viewed as a pattern holding at the
level of the individual segment involved (rather than at the level of the affix as a
whole). Secondly, “latent behaviour” is seen as being triggered exclusively by
the phonological properties of that individual (sub)segment. The identity, shape,
or template subcategorization properties of the affix morpheme in which a latent
segment happens to be located are factors which do not (and cannot) play any
role.3 This leads to a series of failed expectations, stated here as (20)-(21).
(20)
Latent consonants should be (more or less) randomly distributed across
the affixal inventory, crossing the templatic/nontemplatic divide.
If a latent consonant is merely a specific type of segment (or subsegmental unit),
then containing such a consonant is roughly analogous to containing, say, a
coronal consonant, or a nasal consonant. It would be very surprising if coronalinitial or nasal-initial affixes were conspicuously limited to only one of the two
affix categories, and the same should hold for latent-consonant-initial affixes.
(21)
Within the category of templatic affixes, latency of consonants should
be a (more or less) random, idiosyncratic property of individual affixes.
The same reasoning applies here as before. Just as some affixes happen to
contain an initial /t/ and others an initial /n/, we should expect some affixes to
contain an initial full /h/ and others an initial latent /(h)/, with no particular
rhyme or reason as to which affixes happen to contain which type of segment.
(22)
Latency of a consonant should not be tied to (predictable from) the
shape of the affix containing it or of the template assigned by that affix.
As an example, consider templatic C-initial affixes assigning one of the monosyllabic templates. When these attach to a triconsonantal root, which are CCfinal due to the monosyllabic template, we should expect some such affixes to
delete their C1 (because that C happens to be latent) while others trigger vowel
epenthesis instead (because their C1 happens not to be latent). However, affixes
of the latter type are simply not found. Likewise, given that templatic CC-initial
affixes exist, we should expect some such affixes to delete their C1 (because that
C happens to be latent) and others to trigger vowel epenthesis (because their C1
happens not to be latent). Again, affixes of the latter type are not found.
In sum, none of the expectations in (20)-(22) are met in Yowlumne,
and this is a serious weakness of all previous accounts. The conspicuous correlations detailed earlier suggest that segmental latency may not be a purely phonological phenomenon, and that it is not defined and determined at the level of
individual segments and their representational properties after all.
4.2
A new proposal: segmental latency as epiphenomenon
4.2.1
Latency and the morphology-phonology interface
The nature of the correlations and regularities described above suggests
that it may be more appropriate to speak not of “latent consonants” as entities,
3
With one obvious exception (Archangeli 1991: 278f): latent segments cannot occur in
affixes where the combined effects of root, template and affix shape is such that the segment will never surface (making its existence non-recoverable), or such that it always
surfaces (making its latent status non-recoverable); cf. (17b) and (16a), respectively.
but of “consonant latency” as a behaviour exhibited by certain affixes. Consonant latency in this sense is a preference for (affix-)consonant deletion over
vowel epenthesis as the strategy for dealing with unsyllabifiable clusters arising
from base-affix concatenation. The main generalization is then that all affixes
displaying consonant latency belong to the templatic class; conversely, all
affixes in the templatic class display consonant latency whenever relevant.
This further suggests that latency resides at a higher-order level still, as
a property characterizing not individual affix morphemes but an entire layer of
morphology: templatic affixation. The templatic/nontemplatic distinction is
undoubtedly a fundamental aspect of the morphology-phonology interface in
Yowlumne. With respect to ordering, templatic affixes are always closer to the
root than any nontemplatic affixes. Furthermore, templatic affixes may only
attach to a bare root and consequently cannot be stacked, unlike nontemplatic
ones. Finally, templatic affixes are obviously the only ones which place prosodic
restrictions on the base, potentially altering its phonological shape as a result.
The proposal is, then, to capitalize on this independently established
dichotomy by adding a further item to the list of properties differentiating the
templatic and nontemplatic layers of affixation morphology in Yowlumne. I
suggest that the two layers are associated with (slightly) different phonologies,
at least with respect to syllabification and prosodic parsing strategies. In the
templatic layer, the phonology is one which favours underparsing (deletion); in
the nontemplatic layer, the phonology instead favours overparsing (epenthesis).
4.2.2
Formal analysis — a tentative sketch
There are numerous possible ways of implementing the proposal just
suggested. The analysis outlined below is couched in terms of Optimality
Theory, but little hinges on that choice. Most sufficiently elaborate models of
the morphology-phonology interface will do (e.g. various versions of rule-based
Lexical Phonology). From an OT perspective, a difference in strategies or
priorities with respect to the treatment of otherwise unsyllabifiable segmental
material can be translated roughly into the terms in (23), as the relative ranking
of MAX vs. DEP constraints (or PARSE vs. FILL, in earlier versions of the theory).
(23)
a.
underparsing (templatic layer):
DEP-V >> MAX-C
b.
overparsing (nontemplatic layer):
MAX-C >> DEP-V
As for incorporating both ranking relations into a single grammar, two
main options exist. One of these, the stratal or co-phonology approach, would
take each layer of morphology to be associated with its own distinct (complete)
constraint ranking; in other words, the two layers correspond to separate strata,
or co-phonologies, within the grammar (see e.g. Kiparsky 2000). The templatic
stratum (perhaps to be identified with Kiparsky’s Stem level) then contains the
ranking DEP-V >> MAX-C, while the non-templatic stratum (perhaps equivalent
to the Word level) has the ranking MAX-C >> DEP-V. An alternative solution,
the indexed faithfulness approach, takes the grammar to contain only one unified
constraint ranking (a single stratum), but each of the two layers of morphology
would be associated with its own faithfulness constraints. As both constraints in
question are faithfulness constraints, there are two possibilities: [DEP-VTempl >>
MAX-C >> DEP-VNontempl] or [MAX-CNontempl >> DEP-V >> MAX-CTempl].
There is considerable ongoing debate in the OT literature as to which is
the more appropriate model of construction-specific phonology. In support of
the former see, e.g., Orgun (1996), Inkelas (1998), Kiparsky (2000), Anttila
(2002), Inkelas & Zoll (2003). In support of the latter see, e.g., Itô & Mester
(1999), Pater (2000), Alderete (2001). The Yowlumne facts appear compatible
with both models; the stratal approach is followed here for simplicity.
Tableaux (24)-(25) show the constraint interaction in the templatic
stratum. As before, σ-STRUC is a cover term for the constraints enforcing the
[CVX]σ maximum. When σ-STRUC can be satisfied (all consonantal material is
syllabifiable), then neither DEP-V nor MAX-C needs to be violated (24). When
high-ranked σ-STRUC compels some kind of unfaithfulness, as in (25), deletion
(25c) is preferred over epenthesis (25b). Note that final-consonant extrametricality is assumed to hold in the templatic stratum (indicated by square brackets),
leaving syllabification of such a consonant until the nontemplatic stratum.
(24)
/do˘s, -hatn[σµ]-/

(25)
a. dos.-hot.[n]b. do.si.-hot.[n]c. do.s-_ot.[n]/wu˘/y, -hatn[σµ]-/

σ-STRUC
a. wu/y.-hat.[n]b. wu./uy.-hat.[n]c. wu/.y-_at.[n]-
DEP-V
MAX-C
*!
*!
σ-STRUC
DEP-V
MAX-C
*!
*!
*
In the nontemplatic stratum, illustrated in (26), epenthesis is preferred
over deletion since DEP-V is here outranked by MAX-C rather than vice versa.
The epenthesis operating inside the aorist /-hn/ suffix itself is irrelevant (as the
suffix consistently surfaces as [-hin], it might as well be /-hin/ underlyingly).
(26)
/hogn, -hn/

a. hogn.-hin
b. ho.gin.-hin
c. hog.n-_ in
σ-STRUC
MAX-C
DEP-V
*!
*
**
*
*!
Much is yet to be worked out with respect to the overall properties of
the two strata. Most importantly, how exactly is template association and satisfaction controlled and evaluated in the templatic stratum? Under what conditions
can affixal segments enter the template domain—e.g. in the case of an affixal V
“invading” the σ2 of [σµσµµ] as in (9)? Furthermore, every V-final templatic
suffix seem to end in a long vowel; is this a mere accident? Finally, what role, if
any, do cyclicity effects play? Such questions must await further research.
4.3
The glottal quasi-segment revisited
The glottal quasi-segment is likely different in kind, as it behaves quite
differently. Whenever possible, i.e. when C2 of the root is a sonorant, it is realized not as a full segment—even when eminently syllabifiable as such—but as
glottalization on that sonorant (cf. /s.il-(/)i˘xo˘[σµ]-t/ → [s.i.l’-e˘.xo-t] ‘was being
stared at’, not *[s.il.-/e˘.xo-t]). It is only when this realization is unavailable that
one finds it realized as a full segmental [/]. Furthermore, there appears to exist a
contrast between full /// and the quasi-segment, e.g. between durative present
/-/an[σµσµµ]/ and consequent auxiliary /-(/)i˘xo˘[σµ]/. This would appear to
support analyses positing a genuinely subsegmental element (Archangeli 1983,
1988, 1991; Archangeli & Pulleyblank 1994; Zoll 1993, 1998, 2001a).
However, things may not be quite so simple. The glottal quasi-segment
too is confined to templatic affixes, and its distribution is even more limited: it is
only found in affixes taking a monosyllabic template ([σµ] or [σµµ]). Templatic
affixes with initial full /// are in turn only found among ones taking disyllabic
[σµσµµ]. The C2 onto which a floating glottal potentially docks is thus always
located in the consonantal “interlude” at the root-suffix juncture, never further
back inside the root (e.g. between V1 and V2 of an iambic template). The quasisegment and /// may thus be in complementary distribution after all, inviting the
possibility that both reflect an underlying full ///. The docking behaviour would
result from some dispreference against intervocalic C/ or CC/ clusters, driving
fusion of /// with the cluster-initial C whenever possible (perhaps constrained by
stress). Whether a reanalysis along these lines is plausible remains to be seen.
5
Loose ends
5.1
Latent vowels
The above proposal is not meant to extend to latent vowels, ones which
alternate with Ø but which cannot be explained away as epenthetic for one or
both of the following reasons: (a) they are realized as [a] rather than the expected [i] (e.g., /tis-(a)l/ → [ti.s-al] ‘may come out’, not *[ti.s-il]); or (b) their
position is unexpected (e.g., /tis-m(i)/ → [tis.-mi] ‘after having come out’, not
*[ti.s-im]). For suffix-initial vowels, the diagnostic for latent vs. full status is
behaviour in hiatus resolution: a latent V yields to a preceding root V rather than
vice versa (e.g., /nini˘-(a)l/ → [ni.ne-l] ‘may get quiet’, not *[ni.n-al]).
Latent vowels occur in templatic and nontemplatic affixes alike, and
thus cannot be explained away in the same way as latent consonants. However,
there are also marked differences in the factors conditioning the appearance of
these two segment types (Zoll 2001b). Most importantly, while a latent C surfaces whenever possible (i.e. when it can be syllabified), a latent V surfaces only
when necessary (in order to syllabify surrounding Cs); cf. /nini˘-m(i)/ →
[ni.ne-m] ‘after having become quiet’, not *[ni.ne˘.-mi]).
It is worth noting that for vowels in suffix-initial position, the distribution of the V vs. Ø alternants of a latent vowel is identical to those of an epenthetic vowel. Compare dubitative /-(a)l/ → [-al]~[-l] to passive /-t/ → [-it]~[-t];
in each case, the vocalic form appears postconsonantally, the vowelless one
postvocalically. We could thus equally well interpret the passive as /-(i)t/ rather
than /-t/. More generally, as the epenthetic V is always a high /i/, suffixes with
initial [i]~Ø alternations are analytically ambiguous: they might be C-initial or
they might begin in a latent /(i)/. In practice, all such suffixes have been interpreted as C-initial, but there is no a priori reason to assume that no /(i)/-initial
suffixes exist in Yowlumne—a point to which I return in the following section.
5.2
Apparent exception: epenthesis with templatic /-wsi˘l/
There appears to be one glaring exception to the observation in (19):
reflexive/reciprocal consequent adjunctive /-wsi˘l[σµσµµ]/. This suffix is templatic, but seems to trigger V-epenthesis instead of C-deletion when added to a
triconsonantal root (creating an unsyllabifiable C+CC cluster), as shown in (27).
(27)
a.
/pi˘s.-wsi˘l[σµσµµ]/
pi.s.e-w.sel
‘payment for the performance of shamanistic services’
b.
/lown-wsi˘l[σµσµµ]/
lo.wo˘.n-iw.sel
‘place where [they] attended each other’s feasts’
However, there is reason to suspect that the initial high vowel appearing in (27b)
is in fact not epenthetic but underlying, in the form of a latent vowel; this suffix
thereby fills the gap mentioned at the end of §5.1. The suffix belongs to a series
of reflexive/reciprocal affixes, all containing /…w…s…/, which as a set have
somewhat confusing vocalism. The verbal suffix [-i˘sa˘-]~[-wsa˘-] shows a
similar [i]~Ø alternation and is similarly ambiguous between /-wsa˘-/ and
/-(i)wsa˘-/.4 Initial [i] is also manifest in the verbal noun suffix [-iws-]~[-iwis],
in which it cannot be epenthetic, cf. /pa/t.-(i)ws-/ → [pa/.t.-i.wis] ‘place for
shooting at each other’ (not *[ pa./it..-wis] as expected by epenthesis alone).
If the suffix in (27) is underlyingly /-(i)wsi˘l/, then (27b) is no longer
problematic for the analysis proposed here, as this suffix is no longer CC-initial.
The contrast between [-wsel]~[-iwsel] and passive adjunctive [-hnel]~[nel] (seen
in (7)) does not lie in their C1 being a full and a latent segment, respectively, but
in the former suffix containing an initial latent /(i)/ preceding that C1.
6
Summary
This paper has proposed an alternative analysis of so-called latent consonants in Yowlumne. The analysis is a radical departure from previous treatments and builds on certain observations, ignored in earlier works, about the
distribution of latent consonants across the affix inventory. Latency of suffix Cs
is in all cases argued to be the result of differences in the phonologies of the two
“strata” of templatic and nontemplatic morphology. The analysis is quite simple
and does not need to impute any special properties to the consonants exhibiting
latency; in fact, the very term “latent consonant” is misleading. A consonant’s
susceptibility to deletion is entirely predictable from three factors: (a) belonging
to a templatic vs. nontemplatic suffix; (b) the shape of the suffix itself; and (c)
the shape of any prosodic root template imposed by that suffix. The proposal is
superior in that it provides a straightforward explanation for the otherwise mys4
The contraction [iw] → [i˘] in this suffix is notable but irrelevant for the issue at hand.
terious limitation of latent consonants to templatic suffixes—and, within that set,
to those with specific pairings of suffix shape and template choice.
References
Alderete, John. 2001. Dominance effects as transderivational anti-faithfulness.
Phonology 18: 201-253.
Anttila, Arto. 2002. Morphologically conditioned phonological alternations.
Natural Language and Linguistic Theory 20: 1-42.
Archangeli, Diana. 1983. The root CV-template as a property of the affix: evidence from Yawelmani. Natural Language and Linguistic Theory 1:
348-384.
Archangeli, Diana. 1988. Underspecification in Yawelmani phonology and
morphology. New York: Garland.
Archangeli, Diana. 1991. Syllabification and prosodic templates in Yawelmani.
Natural Language and Linguistic Theory 9: 231-284.
Archangeli, Diana. & Douglas Pulleyblank. 1994. Grounded phonology.
Cambridge, MA: MIT Press.
Inkelas, Sharon. 1998. The theoretical status of morphologically conditioned
phonology: a case study from dominance. Yearbook of Morphology
1997: 121-155.
Inkelas, Sharon. & Cheryl Zoll. 2003. Is grammar dependence real? Ms.
University of California, Berkeley and MIT. [ROA-587]
Itô, Junko. & R. Armin Mester. 1999. The phonological lexicon. Handbook of
Japanese linguistics, ed. by N. Tsujimura, pp. 62-100. Oxford:
Blackwell.
Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17: 351-365.
Newman, Stanley. 1944. Yokuts language of California. New York: Viking
Fund.
Noske, Roland. 1985. Syllabification and syllable changing processes in
Yawelmani. Advances in nonlinear phonology, ed. by H. van der Hulst
& N. Smith, pp. 335-361. Dordrecht: Foris.
Orgun, C. Orhan. 1996. Sign-based morphology and phonology. PhD dissertation, University of California, Berkeley.
Pater, Joe. 2000. Non-uniformity in English secondary stress: the role of ranked
and lexically specific constraints. Phonology 17: 237-274.
Zoll, Cheryl. 1993. Directionless syllabification and ghosts in Yawelmani. Ms.
University of California, Berkeley. [ROA-28]
Zoll, Cheryl. 1998. Parsing below the segment in a constraint-based framework.
Stanford, CA: CSLI Publications.
Zoll, Cheryl. 2001a. Segmental phonology in Yawelmani. Ken Hale: a life in
language, ed. by M. Kenstowicz, pp. 427-457. Cambridge: MIT Press.
Zoll, Cheryl. 2001b. Constraints and representation in subsegmental phonology.
Segmental phonology in optimality theory, ed. by L. Lombardi, pp. 4678. Cambridge: Cambridge University Press.
[email protected]