theillusionofmetathesis - goethe university frankfurtmoskal/docs/moskal... · 2019-02-14 ·...
TRANSCRIPT
The Illusion of Metathesis
Beata Moskal
A Thesis submitted to the Leiden University Centre of Linguistics (LUCL) in Partial Fulfillment of the Requirements for the Master of Philosophy's Degree
Supervisor: Prof. Dr M. van Oostendorp Second reader: Dr J.M. van de Weijer
June 2009
2
Table of Contents Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Part one: Theory 1 Diachronic Phonetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.1 Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.1.1 Perceptual Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.1.2 Compensatory Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.1.3 Coarticulatory Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.1.4 Auditory Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2 Phonology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.1 Correspondence Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.1.1 Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.2 Coloured Containment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.2.1 Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Part two: Evidence 3 Phonetic Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 3.1 Perceptual Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.1.1 Articulatory Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 3.2 The Failure of Phonetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 4 Phonological Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.1 Phonotactics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.1.1 Vowel deletion and concomitant CC metathesis . . . . . . . . . . . . . . . 33 4.1.2 CC metathesis without vowel deletion . . . . . . . . . . . . . . . . . . . . . . . . . . 41 4.2 Morphophonology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 5 Modeling Phonological Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 5.1 Phonetically driven Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 5.1.1 Correspondence Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 5.1.2 Coloured Containment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.1.2.1 Dorsal-‐labial Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 5.2 Morphophonologically driven Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 5.2.1 Correspondence Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 5.2.2 Coloured Containment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 5.3 Correspondence Theory versus Coloured Containment . . . . . . . . . . . . . . . 62
3
Part three: Conclusion 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 6.1 The Locus of Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 6.2 The Reality of Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 6.3 Beyond Metathesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 6.4 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4
Introduction
Metathesis is the phenomenon in which sound units reverse their order, as
*[kba] but as [kab]. It should be noted that the focus here does not lie on the
aspects that exactly define metathesis, but rather on the formalization of
metathesis; consequently, a working definition of metathesis as local reversal of
adjacent segments will be employed.1 Furthermore, while consonant-‐vowel and
consonant-‐consonant metatheses are widely attested, there are no cases
reported that involve vowel-‐vowel metathesis.2
Two kinds of metathesis can be identified: synchronic metathesis and
diachronic metathesis. The description of synchronic metathesis is fairly
straightforward; a language qualifies when it displays an alternation involving a
metathesized form (such as the Fur case above). Diachronic metathesis, though,
refers to metathesis over a period in time; that is, at a point B in the history of a
language a metathesized form is observed, while at an earlier point A in time of
that same language the form did not show metathesis. Evidence for diachronic
metathesis stems from sources within languages (such as in the transition from
Old English to Middle English) as well as from comparative data that show non-‐
metathesized and metathesized forms in related languages (such as the
metathesized forms in Southern Estonian, which correspond to non-‐
metathesized forms in Finnish). The status of diachronic metatheses is
disputable since any change that has taken place over time in the past lacks
records and there might be gaps in our knowledge concerning the pathways that
lead to the observed reversed patterns; hence, primarily, evidence from
synchronic metathesis will be considered.3 Nevertheless, diachronic metatheses
will be drawn on occasionally when these cases are relevant to the discussion at
hand. Additionally, cases of lexical metathesis will not be considered as examples
proper, because they do not constitute anything grammatical.
1 See also Hume (2001) for arguments supporting the claim that metathesis is limited to local movement. 2 The sole exception is Halle & Zeps (1966), who suggest VV metathesis for Latvian; however, no further sources for Latvian VV metathesis were found. 3 All examples of metathesis are synchronic unless explicitly indicated otherwise.
5
Metathesis is relatively rarely observed cross-‐linguistically (e.g. Montreuil
1981).4
a perceptual perspective and thus more noticeable than for instance deletion or
insertion; therefore speech communities are reluctant to incorporate the
innovation in their grammatical systems (Hume 2001). Although this contention
is not disputed, it will be argued here that the relative infrequency of metathesis
follows directly from its formalization as a by-‐product of other phonological
processes.
Overview
Part one will discuss the theoretical machinery that has been claimed to explain
metathesis; the first chapter summarizes a diachronic phonetic stance, while the
second chapter expands on metathesis as belonging to the phonology. Among
adherents of a phonological account there is discrepancy between those
regarding it as a separate phonological mechanism (§2.1) and those who claim it
to be a by-‐product of other phonological processes (§2.2).
The second part reviews evidence from a large number of languages
been consulted [http://metathesisinlanguage.osu.edu].5 Chapter three focuses
on phonetic aspects relating to metathesis and although it will be shown that
there is evidence to posit perceptually based metathesis in most cases a phonetic
account fails. The fourth chapter discusses phonologically based metathesis,
which is classified as driven by universal phonotactic tendencies (§4.1) or
morphophonological considerations (§4.2). Finally, the fifth chapter will return
to the formalization of metathesis; first phonotactically motivated metathesis
will be tested against frameworks representing metathesis as a process or a by-‐
product (§5.1), secondly a case study of a morphophonologically driven will be
tested against these same models (§5.2). At the end (§5.3) the viewpoints will be
compared and metathesis as a by-‐product will clearly be revealed as the superior
stance. 4 Pace Hume, who claims that it is regularly observed in a wide range of languages (see also §2.1.1). 5 All data on metathetic languages is from this database unless explicitly stated otherwise.
6
The last part summarizes the findings (§6.1 and §6.2) and broadens the
scope beyond metathesis; in particular, the effects of the current analysis on the
discipline of formalizing phonology in general (§6.3) and some thoughts on
future research (§6.4).
7
1 THEORY
8
1 Diachronic Phonetics
A diachronic phonetic perspective on linguistic sound patterns has been
advocated in particular by Blevins (2004: 5), who takes the radical view that
principled diachronic explanations for sound patterns replace, rather than
complement, synchronic explanations 6 This means that the main burden of
than synchronic systems discussing rules, constraints and such. Blevins (2004:
8) claims recurrent synchronic sound patterns have their origins in
recurrent phonetically motivated sound change Ohala (1981, 1993) and Blevins
(2004) argue that reanalysis lies at the heart of linguistic change; phonological
forms are re-‐interpreted due to mismatches between speaker and hearer in
terms of perception or interpretation: the acoustic signal can be either
incorrectly perceived or incorrectly parsed (phonological ambiguity or variation
leading to the same acoustic signal being mapped onto different underlying
forms), after which the sum of reanalysed items in the language results in
linguistic change. Synchronic alternations, then, stem from different pathways of
change.
1.1 Metathesis
In this view, metathesis is the result of diachronic phonetic changes, which
originally transformed a non-‐metathesized form into the currently observed
metathesized form over time. It is the reanalysis of a segment in a non-‐historical
position, resulting from phonetic effects; Blevins & Garrett (2004: 120) recognize
four pathways that lead to metathesis, each of which is grounded in the
characteristics of particular phonetic features:
6 unless independent evidence demonstrates, beyond reasonable doubt, that a separate synchronic account is warranted
9
(1) Type Phonetic feature Perceptual metathesis Elongated phonetic cues Compensatory metathesis Stress-‐induced temporal shifts Coarticulatory metathesis Consonant-‐consonant co-‐articulation Auditory metathesis Auditory-‐stream decoupling
Blevins & Garrett (2004) hasten to emphasize that the classification serves only
to distinguish the types rather than to reflect the precise nature of the particular
brands of metathesis; all four types involve co-‐articulation, perception and
audition.
1.1.1 Perceptual metathesis
Perceptual metathesis is caused by phonetic features which typically have a
domain larger than the segment, extending to a CV or VC sequence, a syllable or a
string of syllables. Crucially, these features do not necessarily cover a multi-‐
segmental domain, but in languages where metathesis is observed they display
this behavior. A list of phonetic cues that usually are prolonged is given below
(from Blevins & Garrett 2004: 123):
(2) Features with typically long durations Feature Segmental realisations Acoustic property with long duration rhoticity rhotics, rhotic Vs lowered F3 (LM: 244, 313) laterality laterals, lateral Vs lateral formants (LM: 193-‐7) rounding rounded Cs, rounded Gs, lowering of all formants (LM: round Vs 356-‐8) palatalisation palatalised Cs, palatal Gs, raised F2 (LM: 364) high front Vs velarisation velarised Cs velar Gs and lowered F2 (LM 361-‐2) high back Vs pharyngealisation pharyngealised Cs, Gs and lowered F3, raised F1 (LM: Vs, , 307) laryngealisation laryngealised Cs, Gs and more energy in F1, F2 more Vs, jitter (LMJ) aspiration aspirated/breathy Cs, Gs more energy in F0; more and Vs, , h noise (LMJ) retroflexion retroflex Cs and Vs lowered F3, F4; clustering of F2, F3, F4 (L: 203, LM: 28) nasalisation nasals, nasalized vowels spectral zero/nasal anti-‐ and glides resonance (LM: 116) (L = Ladefoged 1993; LM = Ladefoged & Maddieson 1996; LMJ = Ladefoged, Maddieson & Jackson 1988.)
10
For instance, rhoticity has been shown to reach up to two syllables away from its
source (West 1999, 2000). Consequently it might be difficult for the listener to
identify the original location of the rhotic in the string. Blevins & Garrett (2004)
list examples of languages, which show metathesis involving some features
mentioned in the list above (2), viz. rhoticity, laterality, palatalization,
pharyngealization, laryngealization and aspiration.
1.1.2 Compensatory metathesis
Anticipatory co-‐articulation lies at the heart of compensatory metathesis; Blevins
movement for V2 in a V1CV2 sequence may begin during V1ingredient for compensatory metathesis is weak vowel loss, a commonly
observed process in various languages (Campbell 2004). Unstressed syllables
frequently lose their vowel articulation gradually; they start life as full vowels, go
through a stage of weakened pronunciation (usually surfacing as schwas) and
finally lose their existence altogether. Simultaneously, stressed vowels draw
material to themselves due to their prominent character, causing temporal shifts
of V2 into V1. Consequently, the material of V2 is at the same time pulled towards
the V1 position due to anticipatory co-‐articulation and stress, while at the same
time it loses its material in its historical V2 position due to apocope. This gradual
process would in the end lead to metathesis, as Blevins & Garrett (2004) argue is
discussed in detail in §3.2).
1.1.3 Coarticulatory metathesis
The third type also involves temporal shifts; gestures involved in speech
production are liable to overlap as a result of gestural retimings. Browman &
Goldstein (1989) show that retimings of gestures can lead to surface deletion,
epenthesis and assimilation; for instance, the allegedly inserted element [ ] in
something [ ] can be explained by timing adjustments and there is no
insertion of a segment. The labial gesture of the nasal is maintained for a longer
period of time, resulting in a situation in which it co-‐occurs with the gestural
settings for the dental fricative, which involve an open glottis to ensure
11
voicelessness and a narrowing of the tongue towards the teeth; crucially, though,
the dental friction is masked by the closed lips as long as the labial gesture is
maintained, resulting in the percept of a [ ].7
Metathesis is claimed to be an extreme case of gestural overlap; Blevins &
Garrett (2004) argue that when the consonants in a C1C2 sequences are (nearly)
simultaneously aligned, with C1 released after C2, the place of articulation of each
consonant is difficult to recover and the string might be perceived as C2C1.
Furthermore, Blevins & Garrett (2004) claim that the perceived consonant
cluster is biased towards particular directions of sound change; citing four
languages (Klamath, Wiyot, Aklanon and Mikolese) that show PK > KP
metathesis in labial-‐velar stop clusters but none that show the opposite pattern
KP > PK they argue that perceptual preferences stemming from the phonetics of
labial-‐velar stops play a crucial role. Acoustic and articulatory experiments
reveal that though the velar closure precedes or is synchronous with the labial
closure, the velar release always precedes the labial release (Connell 1994);
consequently, while the starting point of the sequence might be {K,P} the end
point will always be perceived as a labial, resulting in an overall perception of
the sequence as KP.8
1.1.4 Auditory metathesis
Auditory metathesis refers to stop-‐sibilant clusters, in which metathesis occurs
in both directions: Cs > sC as well as sC > Cs; for instance, in the West Saxon
dialect of Old English sk > ks is observed (Jordan 1974), while in colloquial
French ks > sk is reported (Grammont 1923). According to Bregman (1990), the
aperiodic noise involved in the production of sibilants somehow distracts the
listener, leading to high confusion rates of the relative order of the stop-‐sibilant
7 Although the perceived bilabial plosive in something is not considered to be part of the internal representation at this point, it might become internalized cross-‐generationally. Consider for example present-‐day English empty / /; its ancestor OE æmeti does not contain a <p> (The
f the word as containing a p as a separately identifiable unit. 8 instead, it will be claimed in §5.1.2.1 that labial-‐velar metathesis is unattested (one exceptional case of KP > PK is observed but this will be argued to involve a complex series of synchronic processes that result in stem allomorphy as well as suffix allomorphy, which leads to the impression of metathesis).
12
cluster, which makes such a sequence liable to perceptual confusion and
subsequent metathesis. Blevins & Garrett (2004) speculate that language specific
variation in directionality could be due to differences in the prosodic systems of
languages.
13
Phonological rules
Phonetic rules
2 Phonology
Most accounts of metathesis place the phenomenon solidly in the realm of
phonology (e.g. Brugmann 1902, Ultan 1978, Hock 1985, McCarthy 1995, Hume
2001). Phonology is grounded in the dichotomy between competence and
performance (Chomsky 1965). Competence refers to the cognitive linguistic
structure of the human mind and includes phonology; in contrast, phonetics is a
matter of performance and does not reside in the grammar proper. A classical
generative view of the grammatical architecture is a modular feedforward
model: each component of the grammar forms a separate module and input
progresses top down and sequentially.9 The lexicon is assumed to provide an
underlying form, which through phonological operations derives a surface
phonological form, which in turn is fed to the phonetic component (adapted from
Bermúdez-‐Otero 2007: 502):10
(3) Lexical representation Phonological representation Phonetic representation
correspond to the framework of Optimality Theory (Prince & Smolensky 1993).
Assuming that metathetic alternations do not spring from the lexicon, any
metathesis that is not phonetically based must reside in the phonological
component. However, placing metathesis in the phonological module does not
resolve the issue whether metathesis is a phonological process or a phonological
9 Note that the stance of a sequential feedforward model is not uncontroversial; some generative phonology assumes for instance bidrectional processing in parallel (e.g. Boersma 1998). 10
14
by-‐product. The former regards metathesis as a separate mechanism, which
exclusively deals with the transposition of segments. In a scenario in which
metathesis is a by-‐product, the segmental reversals are a surface result of (other)
interacting phonological processes but nothing actually switches place.
2.1 Correspondence Theory
Current OT mostly makes use of Correspondence Theory (McCarthy & Prince
1995). In Correspondence Theory (CT) there is a correspondence relationship
between the lexical input (S1) and the various generated phonetic outputs (S2).11
(4) Input (S1) | | Output (S2)
The ranking of a set of universal constraints decides the grammar of each
individual language; these constraints can be divided into two categories:
markedness constraints and faithfulness constraints. The former set serves to
reduce the complexity of output forms, prohibiting or requiring certain output
configurations. The latter set preserves the correspondence between input and
output structures; a typical faithfulness constraint is MAX-‐IO, which demands full
congruence between input and output (from McCarthy & Prince 1995):
(5) Max-‐IO: Every element of S1 has a correspondent in S2.
Crucially, the model refers to two structural representations: an input structure
and an output structure; between the two a relationship exists, which is
evaluated by the faithfulness constraints.
11 Formally, correspondence is defined as follows: Given two strings S1 and S2, correspondence is a relation from the elements of S1 to those of S2 S1 S2 are referred to as correspondents
15
2.1.1 Metathesis
In Correspondence Theory it is expected that metathesis is a separate
mechanism; notably, any change between input and output (i.e. violation of a
faithfulness constraint) is a process, since S1 S2 essentially means S1 S2 if
S1 2. Metathesis clearly involves a faithfulness violation and is consequently
expected to be a separate process. Indeed, it is formalized by way of a violation of
the LINEARITY constraint (6), which claims that any input ordering AB should be
reflected in the output in the same order AB (from McCarthy & Prince 1995).
(6) Linearity: S1 is consistent with the precedence structure of S2, and vice versa.
Since metathesis involves a violation of LINEARITY it must be driven by some
higher-‐ranked markedness constraint CON, which is more important than full
congruence between input and output; this is visualized in the tableau in (7).
(7)
The candidate in (7a) follows the underlying order; however, it violates the high-‐
ranking constraint CON against the surface sequence AB. Candidate (7b) satisfies
high-‐ranked CON but does this at the cost of violating LINEARITY; nevertheless, it is
the optimal candidate: the metathesized form wins.
Including metathesis as an accessible phonological process raises the issue of its
relative rarity. Although Hum
she nevertheless hastens to add
that the reason that it is not so widespread is partly connected to the fact that it
impedes word recognition; however, they do not offer a method how to measure
the ease of word recognition and it is not immediately apparent in what sense
metathesis would hinder word recognition more than for instance deletion.
16
Furthermore, an account including a LINEARITY constraint predicts no restrictions
on the segmental content of the metathesized material since it refers to segments
as a whole that can be transposed; nevertheless, patterns are observed: e.g.
liquids are overrepresented, velars and labials do not metathesize (see §5.1.2.1),
there are no cases of CV metathesis in which only obstruents partake. These
asymmetries are unexpected in an account of metathesis as put forward in CT.
Furthermore, besides the expectation that it would be a common strategy for
languages to use, the inclusion of metathesis as a process would increase the
machinery of grammar greatly and it is questionable whether such an elaborate
account is desirable. Clearly, the question whether a separate metathesizing
mechanism should be posited is warranted. In other words, the matter can be
rephrased as whether an account that does not need to appeal to a separate
mechanism can explain metathesis satisfactorily.
2.2 (Coloured) Containment
Various proposals have been made that assume metathesis to be a surface
manifestation of other phonological processes; metathesis has been argued to
emerge from successive application of rules of deletion and insertion (Besnier
1987, Hume 1991), single feature spreading (Rice 1992), planar segregation
(McCarthy 1989), template maximization (van der Hulst & van Engelenhoven
1994) and vowel epenthesis into degenerate syllables (Lyche 1995). Here the
idea of metathesis as a combination of deletion and insertion will be pursued.
The OT framework used for this approach will be Coloured Containment (Van
Oostendorp 1997) for reasons which will become apparent below.
Coloured Containment (CC) is a variant of Containment Theory (Prince &
Smolensky 1993), the predecessor of Correspondence Theory. The defining tenet
of Containment Theor o element may be literally removed from the
input form. The input is thus contained in every candidate form (McCarthy
1995). Consequently, no deletion in its absolute sense can occur; however,
they can remain unassociated
17
with the phonetic component and thus remain unpronounced. In contrast to CT,
(Coloured) Containment does not evaluate two representations to each other but
there is only one representation which is evaluated as a whole; there is no
derivation within the phonological module. Notably, phonology-‐internal
processes do not have a place in the model outlined here; surface deletions are
the non-‐parsing of lexical material but the material is crucially still contained at
the output level. In CC, the phonological (mono-‐stratal) representation is
connected to two structures: the morpho-‐syntactic structure and the
phonological structure. The morpho-‐syntactic structure of an item represents
the input to the phonological module; that is, after a lexical item has gone
through various modules of grammar it enters the phonology from the morpho-‐
syntactic module. The phonological structure of an item represents the state of
an item after phonological pressures have been exerted and forms the input to
the phonetic component.
(8)
Although the phonological representation is connected to both the morpho-‐
syntactic as well as the phonological structure, these need not necessarily mirror
each other. For instance, take a hypothetical lexical structure /klk/, which serves
as the input to the phonological component, but surfaces as [kuk] at the phonetic
representation.12
(9)
12 Note that this phonetic representation in turn is fed into the phonetic component, which is responsible for assigning the item detailed speech motor specifications, etc.
µ morpho-‐syntactic structure
Phonological representation
phonological structure
µ fed to the phonological component: klk k u l k kuk
18
On the surface, the liquid has been deleted and the back vowel [ ] has been
inserted. However, the liquid, while being connected to the morpho-‐syntactic
structure µ, is not associated with the phonological structure , which means
that it remains unparsed at the phonological level (and subsequent phonetic
levels), creating the impression of deletion; note, though, that strictly speaking it
is not removed from the input. Similarly, the back vowel is only connected to the
phonological structure but not to the morpho-‐syntax and thus on the surface it
seems to be inserted. The velar stops are connected to both the morphological
and the phonological structure, ensuring their uncontroversiality.
In CC, complexity reducing markedness constraints are maintained, but
the faithfulness constraints of CT are discarded. Instead of relating to
correspondences, the set of constraints employed in CC refers to the association
lines linking the phonological representation to the two structures (Van
Oostendorp 2007a):
(10) Parse-‐ the phonological structure ( no deletion ) (11) Parse-‐ morpho-‐syntactic structure ( no insertion )
Constraints as in (10) prohibit material that is linked to the morpho-‐syntactic
structure but is not linked to the phonological structure, while (11) requires that
phonological material also needs to be attached to the morpho-‐syntax. For
instance, (in a purely descriptive account) the constraints involved in (9) above
would be PARSE-‐ PARSE-‐µ(u), violations of which would result in surface
deletion of the consonant and surface insertion of the vowel, respectively.
2.2.1 Metathesis
CC does not refer to correspondences between elements and metathesis in its CT
sense of a process that results in the transposition of segments is unavailable by
virtue of the principle of containment; instead, metathesis is analyzed as
19
a lexical /CV/ sequence metathesizing to a [VC] sequence at the surface would
have a phonological representation {V}C_.13
(12)
The tableau below depicts the same situation by use of a hypothetical constraint
CON that does not allow for a CV structure and the constraints PARSE-‐µ(V), which
pro PARSE-‐
(13)
However, motivations for violations of both PARSE-‐ (xi) and PARSE-‐µ(xi) seem to
PARSE-‐ (x)) as a
strategy to resolve a problematic sequence would favor a violation of one
constraint over violating both constraints. This is illustrated in the tableau
below, which depicts the same situation as in (13) but a third candidate is added
that only violates PARSE-‐µ(V):
(14)
Additionally, the fact that the structure of the
resolved when taking into account that segments are a notation for a collection
of various features. The formulation of the constraints in (10-‐11) explicitly
13 The accolades indicate that the segment is not attached to the morpho-‐syntactic structure
µ fed to the phonological component: CV {V} C V fed to the phonetic component: VC
20
allows reference to sub-‐segmental material (note that CT faithfulness relations of
course also can refer to features); metathesis can then be understood as the
combination of a restructuring force that does not allow for the morpho-‐
syntactic input structure to be maintained in the phonological structure and is
resolved by not parsing a morpho-‐syntactic segment into the phonology
structure, together with the desire to realize the (sub-‐segmental) material from
the morpho-‐syntactic segment that remains unparsed. For instance, the desire to
preserve the place features of the unparsed segment, captured in a constraint
PARSE-‐µ(PLACE), drives the place feature to spread towards a position in which it
can be parsed, even at the cost of creating a vocalic node, which is not connected
to the morpho-‐syntax.14 This situation is schematically depicted in (15) and the
same situation is given in a tableau in (16):
(15)
(16)
14 The discussion focuses on CV metathesis in which the features of the vowel skip a consonant. Presumably, vowels and consonants are located on different tiers (e.g. Ewen & Van der Hulst 2001); while vocalic material can skip consonantal tiers as evidenced in numerous examples of
material. Note that this results in complications with regard to consonantal metathesis, which will be discussed in §5.1.2.
µ {V} C V ° ° ° {voc} voc place
21
Candidate (16c) is ruled out because it fails to satisfy the higher ranked
constraint requiring the preservation of place features. The winner, the second
candidate, satisfies both high-‐ranking constraints but at the cost of creating a
vocalic node that is disconnected from the morpho-‐syntax; nevertheless, the fact
that the sub-‐segmental elements of the unparsed segment can spread into the
position and can thus be realized at the phonological structure ensures that
(16b) is the optimal candidate. In CC metathesized items do not involve any
actual reversal: there is only the surface illusion of segmental transposition.
Instead, metathesis is the result of the requirement that some aspect of a
segment that remains unparsed at the phonological level needs to be
phonologically realized in any way possible.
22
2 EVIDENCE
23
3 Phonetic metathesis
The previous chapters discussed the place of metathesis in the grammar and
although most research assumes it to be located in the phonology, it has also
been argued to be purely phonetic. However, a particular problem for any
account based purely on diachronic phonetics is that it has difficulties accounting
for synchronic alternations. Indeed, Blevins & Garrett (1998: 549) largely ignore
the issue of synchronic alternations and restrict themselves to comment on only
-‐step change involved
in perceptual CV metathesis lends itself to phonological analyses whereby
-‐
metathesized form and one for the metathesized form. The existence of separate
lexical entries for morphologically related forms necessarily leads to a model in
which each morphological alternation is stored separately. Consequently, a view
adhering to diachronic phonetics naturally fits into an exemplar-‐like model, in
which the lexicon is regarded as a vast repository of highly detailed memory
traces of phonetic episodes experienced by the speaker, which are the exemplars
(Johnson 1997).15 However, usage-‐based frameworks of phonology have been
shown to have severe problems, notably they miss universal linguistic
generalizations and in their pure sense they cannot refer to abstract structural
constraints that operate on items.16 Hybrid models incorporating both the
notions of usage-‐based frameworks and phonological theories have been
proposed but these are also far from uncontroversial and will not be pursued
here.17
15 Note that exemplars are connected to each other based on their similarity, grouping together
the concept of similarity brings problems of its own as it is not clear what it entails or how to measure it. 16 Interestingly, Phillips (2006) adheres to a usage-‐based model and claims that an autonomous phonological component is superfluous, but she makes use of abstract structures such as syllable templates and phonotactic constraint. In particular, the structures that language users construct,
homorganic lengthening, are essentially phonological constraints. 17 A discussion on the problems connected with usage-‐based models falls outside the scope of this paper, but see e.g. Pierrehumbert (2002) for the need of a phonological component.
24
Nevertheless, although phonology is a vital part of the grammar, there
seem to be cases of metathesis, which are strongly suggestive of a phonetic
process, in particular a process of perceptual driven confusion on the part of the
listeners. In answer to the need for both phonetically based as well as
phonologically based metathesis, Hume (2001) suggests a view in which low-‐
level effects (perception and production) are incorporated in the phonology; her
main argument for this model is that perceptual enhancement plays a major role
in metatheses. Although the integration of phonetics into phonology is
problematic and will not be considered here since it falls outside the scope of
this paper (but see e.g. Bermúdez-‐Otero 2007), the question is raised whether
metathesis (or any change for that matter) is perceptually optimizing. Showing
that most metatheses are not mono-‐directional, Blevins & Garrett (1998, 2004)
argue convincingly against the claim that metathesis serves to
emerging system.18 Note, though, that this does not mean that perceptual
enhancement cannot play a role at all; however, it is not the all-‐encompassing
mechanism behind metathesis.
Rather than trying to incorporate either phonetics into phonology or vice
versa, the division between phonetics and phonology will be maintained here;
instead, it is suggested that some cases of metathesis are phonetic, namely those
that involve acoustic and auditory similarity, even though the majority of
instances displaying metathesis have a strong phonological basis.19
Consequently, rigid restrictions are set on what will be considered phonetically
based metathesis; it has to be evidently marked by clearly identifiable acoustic
cues and is assumed to display variation between the metathesized and non-‐
metathesized forms since the signal is presumably ambiguous.20
18 Furthermore, it becomes increasingly difficult to maintain a teleological stance when considering processes other than metathesis, e.g. (vowel) chain shifts or Grasalso Ohala 1971, 1981, 1993). 19 It should be noted that phonetic and phonological metathesis do not necessarily exclude each other, but they can interact freely. 20 Additionally, lexically governed metatheses are likely to stem from phonetically driven metathesis.
25
3.1 Perceptual metathesis
Perceptual metathesis is an evident case of phonetic metathesis. The core idea is
the ambiguity of the signal, which has multiple interpretations; in the case of an
analysis that diverges from the originally intended analysis the lexical entry
gains a different phonological make-‐up from its original and metathesis takes
place.21 Blevins & Garrett (2004) argue that perceptually based metathesis
involves segments composed of features that usually have an impact across their
originating domain, which causes difficulties in identifying the historical position
of the segment (see §1.1.1). Clearly, variation should be observed, since the
ambiguous signal leads to difficulty in perception and as a direct consequence
people will not be able to construct a stable lexical form, which leads to variable
articulations as well.
The most clearly identifiable perceptually based metatheses involve
liquids and to a lesser degree the covering class of sonorants. Indeed,
consonantal metatheses (C1C2 > C2C1) involving sonorant consonants in both C1
and C2 position often display variation, in particular the rhotic leads to confusion
among listeners; Amharic (lr > rl), Twana (yl > ly), Ayacucho Quechua (rm > mr)
and Aymara (h/liq+w > w+h/liq) all show optional metathesis. Only two other
languages involving sonorant-‐
database: Chawchila (ln > nl; ly > yl) and Deg (m/w+r > r+m/w). Optional
metathesis between a rhotic and obstruent is registered in North-‐American
English (rt > tr).
In the case of CV metathesis the liquids also play a major role and again
the rhotic is central stage: North-‐American English (rV > Vr), Cajun French (rV >
Vr) and Avestan (Vr > rV) show variation. Bonggi (rV > Vr) is the only example of
rhotic-‐vowel interaction, which is not variable; interestingly, though, the
phonetic realization of the rhotic is a flap, which is perceptually far removed
from a vowel. Perceptual CV metatheses are not restricted to liquids, but include
all vowel and consonant sequences which are perceptually close, such as the
metathesis of a low vowel and a guttural consonant in Bedouin Arabic. 21 Note that auditory metathesis (§1.1.4), which involves sibilant metathesis, will be regarded as a subcategory of perceptual metathesis, since the acoustic properties of the sibilant lead to confusion on behalf of the speaker.
26
3.1.1 Articulatory metathesis
Although Blevins & Garrett (2004) argue for the existence of articulatory
metathesis, evidence for this type is insufficient to uphold its existence.22 In
general, articulatory based metathesis is unexpected, especially in CV sequences,
since the formations involved in vowels and consonants are so different that
articulatory considerations cannot play a role; in situations in which the vowel
and consonant gestures are very similar, such as in schwa-‐r (observed in Cajun
French and North-‐American English), articulatory characteristics might be
relevant. Similarly, most cases of articulatory based consonantal metathesis
would have to involve consonants which share gestures; this would allow for
retimings of particular gestures of C2 that do not interfere with the gestures
involved in forming C1. Consider for instance data from Hixkaryana in (17);
metathesis between a sibilant and a glottal fricative is observed (the two
segments are adjacent due to syncope):
(17) input syncope metathesis
Focusing on only the sibilant and / / sequence, the sibilant is created by a
critical (palato-‐)alveolar closure formed by the tongue moving towards the
alveolar ridge, while the glottis is open to ensure voicelesness. The formation of
the glottal only involves an open glottis through which air can pass.
Consequently, a delay in the tongue tip gesture results in the percept of a glottal
fricative preceding the sibilant; crucially, there is nothing that clashes with the
retiming of the tongue gesture. Similarly, a delay in the coronal closure in
coronal stop-‐fricative sequences would explain a change from ts to st, which is
observed in Ahtna, Hebrew and Udi.
Note, however, that the segments involved in the metatheses discussed so
far (schwa-‐r, ts and sh) are not only characterized by their articulatory similarity
but they are also perceptually close to each other. Since perceptual factors
leading to confusion have a solid basis (see above) and there is no independent
22 Blevins & Garrett (2004) cite labial-‐velar stop metathesis as an example of articulatory based metathesis; however, it will be shown that no such metathesis exists (§5.1.2.1).
27
evidence to posit that articulatory aspects play a role in metathesis, they will not
be considered a valid motivation for metatheses.23
3.2 The failure of phonetics
However, a large number of metatheses cannot be explained by referring to
maintained; they classify the metathesis observed in Rotuman as compensatory
metathesis. In Rotuman metathesis is recorded in the morphologically
conditioned Incomplete Phase:24
(18) Complete phase Incomplete phase
Blevins & Garrett (1998) list Rotuman metathesis in the incomplete phase as a
classical example of compensatory metathesis. Anticipatory co-‐articulation
combined with apocope results in a situation in which the movements for V2 in
V1CV2 sequences are formed in advance, i.e. before the consonant directly after
V1; the weak/zero articulation of the affected final segment is compensated for
by retaining its material in the preceding syllable. Hale & Kissock (1998) claim
that the origin of the phase contrasts in Rotuman stem from prosodically
different suffixes combined with the stress assignment in the language. Crucially,
Rotuman stress is regularly assigned to the penultimate syllable, resulting in the
final syllable to be unstressed and hence liable to reduction and concomitant
compensatory coarticulation on the vowel in the preceding syllable. Historically
different formats of the suffixes caused different reactions to the mechanism of
23 Further doubts on the justification of articulatory based metathesis arise in light of the lack of metathesis in homorganic stop-‐fricative clusters with a place of articulation other than coronal; kx and pf are expected to be observed if motor abilities of humans lie at the heart, yet they are unattested. In a purely perceptual account, metathesis in ts clusters is expected because of the presence of the sibilant (cf. ks and ps metatheses, attested in e.g. Faroese and Dutch, respectively). 24 All data of Rotuman are taken from McCarthy (1995), unless indicated otherwise.
28
compensatory coarticulation. Essentially, complete phase stems originally were
followed by monosyllabic suffixal morphemes, while incomplete phase stems
were composed of stems plus zero or disyllabic suffixes. This follows from the
regular penultimate stress system combined with apocope and concomitant
compensatory coarticulation as can be seen in the table below (adapted from
Blevins & Garrett 1998: 532):
(19) suffix stem stress changes stem phase none penultimate incomplete (Metathesis) monosyllabic ultimate -‐ complete disyllabic secondary -‐ incomplete (Metathesis)
Regular penultimate stress led to some final stem vowels ending up in strong
positions, resulting in their preservation (complete phase), while others
stranded in weak positions and were lost (incomplete phase). Preservation of
the material of the final syllable in CV1CV2 sequences results in a metathesized
form CV1V2C.
Aside from the fact that synchronic grammars presumably do not have access to
the historical course of the incomplete stem phases (see §3 above), a diachronic
phonetic account misses the generalization that the metathesis observed in the
incomplete phase in Rotuman is only one of various ways to express the
incomplete phase. Although a diachronic phonetic account recognizes the
template since it is a direct consequence of stress assignment and apocope with
concomitant compensatory coarticulation (see (19) above), it fails to recognize
that the incomplete phase employs an assortment of strategies, which are united
when a phonological structure is allowed for. The variety of mechanisms
Rotuman employs in the incomplete phase are the following: metathesis,
incomplete phase are the same); these strategies all converge to one goal: to
satisfy the prosodic template required for the incomplete phase, namely a
bimoraic foot at the right edge of the word (more on this in §5.2, which discusses
Rotuman in detail). In contrast, in a diachronic phonetic account the templatic
29
coherence between the various strategies in the incomplete phase remains
unexplained and would have to be interpreted as accidental.25
Another problem with a phonetic account is that the incomplete phase
stems are not the only items which would be liable to compensatory
coarticulation; for instance, lima also has prefinal stress, yet, no metathesis
is observed. It seems to be the case that the combination of apocope and the
Rotuman stress system is not a correct representation of the prerequisite for
metathesis: it only affects the incomplete phase. Hence, a morphological
template is called for, which is not available in a diachronic phonetic perspective,
since it does not allow for modules but regards lexical items as being directly
linked to the phonetic event, rather than going through morphological and
phonological modules.
Morphology plays another role besides providing particular templates. In a
considerable number of languages metathesis is restricted to application across
morpheme boundaries only; these include among others Faroese, Hixkaryana,
Maori, Bedouin Arabic, Mikasuki, Hebrew and Lama. In these languages
metathesis only applies in derived contexts while allowing the same sequence to
remain unhindered in monomorphemic (stem) items.26 For instance, Bedouin
Arabic displays metathesis between a low vowel and guttural consonant. The
present perfective typically has a CVCCVC structure (e.g. taktib
nasbah esult in a CVlowCguttCVC sequence
metathesis is observed, as in akum t¿arif
uf
guttural appearing in coda position; however, forms such as
do not metathesize. 25 Additionally, Rotuman only allows for bimoraic syllables forms (i.e. closed syllables and long vowels) in two contexts: the final syllable of words with a long vowel (in either native monovocalic roots (ré , rí siká hané the final syllable of an incomplete-‐phase word, thus warranting the special status of the incomplete phase structure as a grammatical anomaly. 26 Although it will not be discussed in detail here, the model of Coloured Containment is particularly suitable to capture derived environment effects, while models based on Correspondence Theory are less effective (see van Oostendorp 2007b). The core idea of modeling derived environment effects in Coloured Containment is that spreading in monomorphemic items is restricted because the morpheme has the same color, which blocks spreading; in contrast, features in multi-‐morphemic items can spread across morphemes since they do not share the same color.
30
Similarly, in Faroese a coronal fricative and a velar stop metathesize when
they are followed by a (coronal) stop:27
(20) Underlying form Surface form -‐sg-‐ -‐sg-‐ -‐sg-‐ -‐sg-‐ -‐sg-‐
It seems that metathesis is a mechanism to avoid a sequence of two successive
stops, but the existence of words such as gikt gout insekt
the inclusion of a full ban on stop sequences.
Clearly, metathesis is sensitive to the morphological make-‐up of items;
however, this is only possible in a framework that allows for a morphological
module.
27 There are no stop-‐initial suffixes other than with a coronal place of articulation (Seo & Hume 2001).
31
4 Phonological metathesis
Clearly, not all cases of metathesis can be analyzed phonetically and, indeed,
most instances of metathesis will be shown to belong in the realm of phonology.
Phonologically driven metathesis works on the basis of non-‐phonetic restrictions
prescribed by the phonolog
metathesis was conditioned by a morphological template that was imposed by
the grammar. However, another motivation for metathesis is observed: the
avoidance of particular structures. An important aspect of grammars is that they
have the tendency to converge to simple structures, that is, reduce their
markedness.28 Note, however, that a reduction of complexity in one area of the
grammar usually leads to an increase in complexity of another area. A relevant
case at hand is the phenomenon of vowel deletions. Deletion of the medial vowel
in a trisyllabic word reduces the prosodic markedness of the item, since
trisyllabic words involve either a degenerate foot or contain an unfooted
syllable; after apocope, though, the word consists of a perfectly shaped foot. At
the same time that the prosodic markedness is reduced originally nonadjacent
consonants are forced to appear next to each other. These emerging consonant
clusters might not always agree with each other and sometimes the grammar
-‐ 29
4.1 Phonotactics
The ill-‐
prosodic assignment; syllables consist of a vowel-‐like nucleus with on both sides
28 Naturally, grammars are also conservative and want to preserve their current structures; from a cognitive perspective any change from an original form is strenuous, even changes involving the reduction of complexity. Other sources of increasing complexity stem from for instance grammaticalization processes. 29 It should be mentioned that the formulation of ill-‐formed refers to ill-‐formedness in a gradual sense: the cluster can be completely banned (no surface forms with the offensive cluster at all, as in e.g. Pawnee) or it is only repaired in derivative contexts, that is, in mono-‐morphemic words the ill-‐formed is observed but across morphological domains or due to phonological processes (in particular syncope) the segments are metathesized. In the latter scenario, the cluster is not so much illegal but dispreferred (see also §3.2).
32
optional consonantal material that can occupy an onset or coda node. Consider a
traditional representation of a simple CVC syllable structure in (21) below:
(21) /| OR |\ NC
The nucleus is the obligatory element as indicated by its head position in the
tree; words can exclude an onset or a coda but the nucleus is mandatory.
Furthermore, CV syllables are the norm, while onset-‐less syllables are marked
structures; languages allowing for onset-‐less syllables always also have CV
syllables but not vice versa.30 However, with regard to post-‐vocalic consonants
the situation is different: languages that ban codas altogether exist (e.g.
Hawaiian) and there are many languages which have restrictions on segmental
material in coda position (e.g. Standard Mandarin only allows for glides or nasals
in coda position).31 Word-‐internal codas also have limitations, e.g. in Misantla
Totonac the first consonant in coda-‐onset clusters is restricted to a homorganic
nasal after long vowels (Kaplan 2006); similarly, in most Germanic languages
word-‐internal codas are limited to sonorants (Howell 1991). Compared to nuclei
and onsets, codas have less freedom and are not always able to support all
ositions for a segment to be in.
However, most grammars make use of more complex structures than the
one in (21); they include complex clusters, such as onsets composed of two
consonants:
(22) O R |\ | CC V 30 This is captured in the Onset Maximization principle, which demands a consonant to be syllabified in onset position whenever this is possible, that is, a VCV sequence is parsed V.CV and not *VC.V. 31 Although the status of word-‐internal codas is indisputable, word-‐final codas are not uncontroversial. Based on the fact that the restrictions on word-‐final codas are much less strict compared to those on word-‐internal codas, Harris (1994) argues that word-‐final consonants are not codas at all but are situated in onsets of syllables with empty nuclei, i.e. nuclei that are not filled with segmental material; according to him true codas can only appear in word-‐internal consonant clusters.
33
The position of the second consonant in an onset cluster is in some sense similar
to that of codas: a number of languages bans complex clusters (e.g. Hawaiian)
and in most cases the dependent position of the cluster cannot support the full
gamut of consonants available in languages allowing for onset clusters; for
instance, in English the C2 position of a C1C2 onset is limited to liquids.32
nucleus, the segmental make-‐up of the positions is antagonistic: while nuclei
contain maximally sonorous content the traditional occupiers of syllable onsets
are non-‐sonorous obstruents.33 In contrast, the weaker positions, codas and
dependents in complex clusters, do not aim at maximizing contrast but instead
prefer to fit in the sonority pattern established by the strong positions;
consequently, word-‐final codas are merely restricted by having to be less
sonorous than its preceding nucleus and have full consonantal freedom. Word-‐
internal codas, however, are not only limited by the preceding nucleus but
additionally and much more importantly by their following onset.34 It is natural
then for weak positions to contain liquids since they satisfy both the
requirement of being more sonorous than the nucleus as well as being less
sonorous than the onset (see also Vennemann 1988).35
4.1.1 Vowel deletion with concomitant CC metathesis
As alluded to above, vowel deletions are expected to play a role in CC
metatheses.36 Deletion of the vowel results in a rearrangement of the prosodic
landscape of a word; in cases of syncope, two consonants that were originally not
adjacent are forced to stand next to each other: C1VC2 > C1C2. Apocope, the
deletion of a word-‐final vowel, is another case in which segments are
32 Coda clusters mirror the onset clusters: since it is a VC structure, sonority declines from the vowel to the consonant and the final consonant is less sonorous than the first element of a complex coda cluster. 33 It should be noted that sibilants are problematic (see e.g. Botma & Ewen 2008: section 4); they do not adhere to the general syllabic arch of sonority starting low (onset) rising until the nuclear position and declining afterwards (coda). 34 Restrictions in complex consonant clusters follow the same pattern; the dependent position frequently contains liquids. 35 It is important to realize, however, that in discussing phonotactic considerations tendencies rather than universals are expected; althouphonotactics of languages are determined by the conditions of individual grammars. 36 It should be mentioned that the conditions of vowel deletion will not be discussed here since they fall outside the scope of this paper.
34
prosodically rearranged; however, apocope is unlikely to lead to metathetic
behavior. There are two possible scenarios for consonant clusters involved in
apocope:
(23) pre-‐apocope post-‐apocope (a) VC1.C2V VC1C2 (b) V.C1C2V VC1C2
However, the apocope in (23a) would not lead the consonants to metathesize,
since the consonantal sequence before deletion of the vowel presumably adheres
to the principle that word-‐internal codas are less sonorous than their following
(onset) consonant. Consequently, even if the original sequence is ill-‐formed, the
situation after vowel deletion cannot be improved by metathesis.
The situation in (23b) is more interesting since, again assuming that the
sonority principle is obeyed, reversal of the segments does lead to an
improvement of the sonority pattern. However, while the change in (23a) is an
introduction from a simplex coda to a complex coda, the situation in (23b) is
much more drastic: a cluster that starts out initially as a complex onset is
relegated to a complex coda. Taking into account that this would be a
considerable change from the non-‐metathesized pronunciation form and that of
course other strategies such as deletion and insertion compete in the resolution
of phonotactically ill-‐formed clusters it is not expected to be widely observed
(see also Hume 2001, Steri
diachronic example, in Persian (diachronic) metathesis co-‐occurrs with
apocope:37
(24) > > >
37 Another example of apocope and metathesis is found in Southern Estonian (diachronic): Finnish kárhu vánha kahr and vahn, respectively. Note that this metathesis goes counter to the sonority curve; however, it seems that in this case segment reversal takes place to maintain the perceptual salience of the glottal, which overrules phonotactic considerations.
35
The final consonant cluster resulting from apocope is not optimal in terms of
sonority and Persian only allows for complex coda clusters which meet the
conditions that the coda head is at the right edge and is less sonorous than its
dependent. Consequently, the metathesized form is picked as the most suitable
solution.38
Syncope is more widely observed to co-‐occur with metathesis; although some
cases are restricted to lexical items and will thus not be included (Somali,
Hanunóo) there are a number of languages displaying synchronic metathesis
coupled with syncope, including Rendille, Elmolo, Deg, Hiliganyon, Cebuano,
Tagalog, Hixkaryana, Cherokee, Udi, Kabardian and Georgian.
The table on the next page in (25) shows a simplified overview of the
effects of syncope on CVC clusters. Notably, it is restricted to a set of forms which
are perfect from a phonotactic perspective; that is, while both sonorants and
obstruents are allowed for in simple onsets, in complex onsets it is assumed that
the first element can only be an obstruent and the second consonant is restricted
to a sonorant, and vice versa for complex coda clusters. Additionally, word-‐final
simple codas are not biased towards sonorants, since they have more freedom
and their status as weak positions is not indisputable (see also footnote 31);
however, word-‐internal codas are restricted to supporting sonorants only.
Furthermore, binary branching is assumed, i.e. both an onset and coda are
maximally composed of two segments. Finally, onset maximization is assumed as
well as a coda-‐onset parse over a complex onset parse, since languages allowing
for codas are less marked than those that do not prohibit complex onsets
(Blevins 1995).39
38 Note, though, that in Persian the preservation of material must play an important role, since deletion of the final consonant would also lead to a well-‐formed syllable. As explained above, most cases of apocope are expected to be differently resolved. 39 Although there are a few exceptional languages that have complex onsets but no codas.
36
(25)40 pre-‐syncope post-‐syncope C1 obstr, C2 son C1 son, C2 obstr (a) .C1V.C2V C1C2V no metathesis metathesis (1) (b) V.C1VC2. VC1C2 metathesis (0) no metathesis (c) V.C1V.C2V VC1.C2V metathesis (3) no metathesis (1)41 (d) V.C1V.C2CV VC1.C2CV -‐ no metathesis VC1C2.CV -‐ no metathesis V.C1VC2.CV VC1C2.CV metathesis (0) -‐ (e) VC.C1V.C2V VC.C1C2V no metathesis -‐ VCC1.C2V no metathesis -‐ V.CC1V.C2V VC.C1C2V -‐ metathesis (1)
Situations in which metathesis is expected are bold-‐faced, while the single
instance of unexpected metathesis is shaded grey; the bracketed number
indicates the number of languages for which that kind of metathesis is attested.42
Metathesis at word edges, (25a) and (25b), is rare; evidence for the type of
metathesis in (25a) is only found in Cherokee, in which a glide and glottal
fricative change their surface order in second singular forms: >
-‐ > -‐43 A probable reason for the scarcity of this type is that the cues for
the C1 sonorant in situation (25a) are presumably very low since they are
completely dependent on themselves; compare them for instance to their context
in (25c) where the sonorant is supported by their preceding vowel. Additionally,
metathesis introduces a complex onset, which might not always be permitted
(note that Cherokee has complex onsets as evidenced in e.g. dlayka
The pattern in (25b) is not attested at all; the reason might be related to
the one proposed for the rarity of apocope. Although the salience of the sonorant
in C2 position is better than its sonorant counterpart in (25a), metathesis
requires the introduction of a complex coda, which might not be allowed. All in
all, metathesis at word edges is unlikely, an assumption, which is supported by
40 Note that cases with C1 and C2 being equally sonorous do not trigger phonotactically-‐driven metathesis and thus are not included in the table. 41 Technically there are three languages reporting this kind of metathesis (Hiliganyon, Cebuano and Tagalog), but since they are closely related they are considered one example. 42 The table accounts for eight out of the eleven listed languages displaying metathesis after syncope; three languages remain: Hixkaryana and Udi involve sibilants, which are problematic for phonotactic accounts (see also footnote 33) and Kabardian displays CV metathesis. 43 According to Foley (1980) there are virtually no words in Cherokee which contain a glide followed by a non-‐vocalic segment; the exceptions involve a word-‐internal cluster (e.g. dlayka
37
(26b) metathesis by position in the word:
(26a)
(b)
The scenarios in (25d) and (25e) do not seem to be supported by the evidence
either; only Georgian metathesis patterns similarly to the situation in (25e):
38
(27) underlying form surface form
Ignoring the deletion of the root vowel, the consonants on both sides of the
deleted vowel of the suffix metathesize when the root ends in a sonorant.
The order of the sonorant and obstruent is reversed and results in a better
sonority contour.44 The infrequency of (25d) and (25e) might be related to the
fact that syncope leads to three consecutive consonants, which (i) might be
dispreferred to deletion in some languages, or (ii) when the language allows for
CCC sequences it presumably has more leeway with regard to the segmental
composition of codas and/or onsets, i.e. the restrictions on sonority would be
less strict.
The situation in (25c) is the most interesting, since although there are three
languages providing evidence for the expected metathesis pattern, one language
family shows an unpredicted metathesis. The anticipated obstruent-‐sonorant
transposition is observed in Rendille, Elmolo and Deg, while the reverse strategy
is observed in three Malayo-‐Polynesian languages (Hiliganyon, Cebuano and
Tagalog).
In Rendille, metathesis occurs when a liquid follows an obstruent or nasal
consonant as the result of apocope, as seen in nouns (28a) and verbs (28b):
(28) underlying form surface form (a) -‐ -‐ (b) -‐1sg/-‐3sg-‐ -‐1sg/-‐3sg-‐
In contrast to (28), no metathesis occurs in the reverse situation, in which the
liquid precedes the obstruent or nasal:
44 Although /v/ is phonologically a sonorant in Georgian, it nevertheless is less sonorous than the traditional class of sonorants.
39
(29) Underlying form Surface form -‐1sg/-‐3sg-‐ -‐1sg/-‐3sg-‐ -‐1sg/-‐3sg-‐
Elmolo patterns the same as Rendille; metathesis of obstruent-‐sonorant clusters
is observed (e.g. > -‐ -‐obstruent
sequences (e.g. > -‐ uation is slightly
different, since two sonorants are involved; /m,w/ metathesize with a following
/r/: e.g. > >
However, both m and w are more sonorous than the rhotic and thus the pattern
of coda-‐onset corresponding to decreasing sonority is adhered to.
Data from the three languages that opt for the reverse pattern and show
metathesis of a sonorant and obstruent are shown below in (30), while in (31)
data involving sonorant-‐sonorant and obstruent-‐obstruent clusters are listed.45
(30) Underlying form Surface form Hiliganyon Cebuano Tagalog
(31) Underlying form Surface form Hiliganyon Cebuano Tagalog 45 Interestingly, in the data in (30), stress falls on the second syllable in the root: Cebuano sulúd and Tagalog silíd (although final foot stress has not been confirmed in Hiliganyon it is assumed that it shares its stress system with its sister languages). Presumably, the fact that the sonorant is very prominent in the root would be an additional reason not to put it in a dependent coda, since it would perceptually be a considerable departure from the original.
40
These data are in contradiction to the pattern discussed above, in which the
preferred strategy seems to be to create a coda-‐onset cluster rather than a
complex onset; note that the language does allow for codas in general. Although
the surface clusters in (30) look like onset clusters, most of the data in (31) seem
to contradict the hypothesis that the outputs of metathesis are complex clusters:
for instance, the consonant clusters in Cebuano imna and Tagalog aptan would
form very poor onsets. Indeed, none of CC clusters in the metathesized forms are
permissible clusters in Cebuano (Bunye & Yap 1971). Nevertheless, the fact that
these sequences are coda-‐onset clusters rather than complex onsets does not
explain why they do not adhere to the established sonority pattern that requires
the coda to be more sonorous than the onset. However, the data in (31) reveal
that metathesis in these languages is subject to another restriction: all of the
examples contain a coronal in onset position; it seems that this metathesis is
governed by the prerequisite of a coronal in C2 position.46 Indeed, no metathesis
is observed if there already is a coronal in onset position (Hiliganyon, Tagalog)
or when no coronals are involved (Cebuano):
(32) Underlying form Surface form Hiliganyon Cebuano Tagalog
There is an apparent counterexample to the claim that these languages require a
coronal in onset position: the Tagalog form >
form and the forms in (31) is that in the case of the metathesis would
occur word-‐medially, while in (31) the metatheses occur at the right edge of the
word and are in direct contact with the suffix; compare the forms below:
46 It should be noted that in situations in which there is a choice which coronal segment should occupy the onset position surprisingly enough the liquid is favored over the obstruent.
41
(33) Metathesis No metathesis input syncope _ _ output
They have the same prosodic composition at the starting point but syncope is
executed in different syllables: the third syllable in the metathesized case but the
second syllable in the non-‐metathesized case.47 In the first case, metathesis thus
occurs at the border between the stem and the suffix, while in the latter situation
metathesis would have to take place within the stem; consequently, a ban against
metathesis within stems is posited for Malayo-‐Polynesian languages.48
Although Hiliganyon, Cebuano and Tagalog diverge from universal phonotactic
tendencies, there is a grammatical pattern identifiable. The obligatory C2
position for coronals is not phonotactically driven but presumably is a particular
restriction set by the grammar; accordingly, the explanation for these types of
metatheses shifts towards another aspect of the phonology (see §4.2 below).
4.1.2 CC metathesis without vowel deletion
Clearly, CC metathesis not originating from syncope should follow the same
tendencies described in the previous section. Excluding vowel deletions the same
pattern can be observed in Sidamo (CN > NC)49, Wichita (kr > rk), Kota (Cy > yC),
Oromo ( )50, Tubatulabal (h+son > son+h) and Hungarian (h+liq > liq+h),
which adhere to the sonorant-‐obstruent reordering.
Despite the majority of languages behaving as expected, two languages do
not conform to the generalization identified above: Pawnee and Zoque. Pawnee
matathesizes in the opposite direction, creating an obstruent-‐sonorant cluster:
e.g. >
47 The difference between the forms might be stress-‐related; habílin is explicitly recorded with stress on the second syllable, while kalabit does not have a stress mark and might have a different stress pattern (see also footnote 45). 48 The observation that monomorphemic items are resistant to metathesis is confirmed in more languages and might be relevant in metatheses in general (see also §3.2). 49 mbata with different,
50 Metathesis in Oromo is restricted to homorganic clusters, which might reflect the perceptual origin of Oromo metathesis, although by now the process has been phonologized.
42
language specific ban against *rh sequences in Pawnee; similarly to the situation
in Malayo-‐Polynesian languages, a restriction imposed by the grammar is active,
honotactic tendencies to surface. Finally, Zoque
displays metathesis of a sonorant followed by a glottal stop (e.g. >
51
As shown above, in consonantal metathesis there is a clear preference for
sonorants, and liquids in particular, to be drawn to postvocalic positions. In CV
metatheses this pattern is confirmed:52
(34) Class Languages Nasals Lama (NV > VN), Maori (NV > VN), Alsea (NV > VN) Glides Kabardian (jV > Vj) Liquids Alsea (lV > Vl), Bonggi (rV > Vr), North-‐American English (rV > Vr), Cajun French (rV > Vr)
Sonorant-‐vowel interactions show a clear bias towards a SV > VS pattern, which
results in the sonorant ending up in coda position. Only Avestan (Vr > rV) shows
the opposite pattern, which might be due to the realization of the rhotic as a
(relatively non-‐sonorous) flap. CV metatheses that are clearly phonotactically
motivated are exemplified by Fur (35), Mikasuki (36) and Mutsun (37):
(35) verb root with C prefix C C (36) V initial root C initial root (37) Stem Locative suffix 51 It seems that again a particular ban is active: the San Miguel dialect of Zoque does not allow for prevocalic glottals when they are preceded by a consonant, that is [C V] structures are illegal (Johnson 2000); it seems likely that such a constraint operates in Zoque as well. 52 Furthermore, there are no examples of languages displaying CV metathesis, which exclusively metathesize obstruents.
43
Fur metathesizes the root to prevent complex onsets, which do not occur in the
language (Jakobi 1990). Mikasuki clearly aims for CV syllables; the motivation is
to avoid two-‐vowel clusters in Mikasuki nuclei (Derrick-‐Mescua 1980). Mutsun
avoids a sequence of three successive consonants by metathesizing the second
consonant of the locative suffix with its vowel; although the language does not
categorically ban word-‐internal CC clusters they are extremely rare (Goddard
1912).
4.2 Morphophonology
Nevertheless, there are some phonological metatheses in which phonotactics do
not play a role at all. Here the drive for metathesis is not a strategy to reduce
markedness, but to conform to particular (morpho-‐)phonological restrictions
dictated by the grammar.53 Leti presents a clear case of this type of metathesis;
(Bonthuis 2001):
(38) Underlying form Surface form (a)
(b)
(c)
The forms in (38a) show metathesis, the items in (38b) maintain the order of the
underlying form and (38c) displays deletion of the stem vowel. The fact that the
second element in (38c) starts with a vowel cannot be the decisive factor, since
in (38b), while being vowel-‐initial, does not trigger syncope but
maintains the underlying form. Similarly, the character of the consonant of the
second element cannot be relevant: obstruent-‐initial and 53 Cf. the Malayo-‐Polynesian requirement of C2 being a coronal.
44
trigger metathesis but fails to metathesize;
sonorant-‐initial maintains the status quo but
surfaces with metathesis. Clearly, the conditions governing metathesis in Leti are
unrelated to phonotactics.
Similarly, metathesis in Rotuman is not phonetically driven; rather, it
even goes against the usual phonotactic constraints of the language. Not only
does the segmental structure play no role since all segments are involved,
Rotuman does not allow for closed syllables or diphthongs, except in the
incomplete phase. A full analysis of Rotuman will be presented in §5.2.
45
5 Modeling phonological metathesis
Having identified that some metathesis is clearly phonological in nature raises
the question of how to model it. As mentioned previously, two views can be
distinguished: metathesis as a phonological process or as a by-‐product of other
phonological processes. In the second chapter, it was shown that mainstream
Correspondence Theory (CT) predicts metathesis to be a mechanism (§2.1). In
contrast, the architecture of Coloured Containment (CC) necessarily leads to the
view that metathesis is a by-‐product (§2.2). In this chapter the two approaches
will be tested against the types of phonological metathesis identified in the
previous section, namely phonotactic metathesis (§4.1) and morphophonological
metathesis (§4.2).
5.1 Phonotactic metathesis
Rendille will represent a straightforward obstruent-‐sonorant reversal as a result
of phonotactic tendencies; underlying surfaces metathesized as
while retains the same order surfacing as (more data in (28-‐
29) above). The circumstances governing syncope will not be pursued here;
instead a constraint BALANCE will be employed, which ensures that the vowel is
deleted in the required context (Butskhrikidze & Van de Weijer 2001).
Additionally, the markedness constraints *CCONSET and ONSETMAX are active,
prohibiting complex onsets and unnatural syllabification, respectively.
5.1.1 Correspondence Theory
In a CT based model, the remaining constraints are MAX-‐IO to express the notion
that input segments should not be deleted in the output and LINEARITY, which
needs to be violated to allow for the metathesized candidate to be the optimal
output. Consider the tableau below, depicting a metathesizing input
-‐
46
(39)
The first candidate violates the syncope requirement and is therefore ruled out.
The candidates (39b) and (39c) are suboptimal in terms of prosodic
requirements; (39b) does not adhere to the ban on complex onsets, while
candidate (39c) violates onset maximization. The remaining two candidates
violate MAX-‐IO, but in (39e) incurs one more violation than the winning
candidate (39d), which saves a violation of MAX-‐IO at the cost of violating low-‐
ranking LINEARITY. The tableau below shows how a non-‐metathesizing input
-‐1sg/-‐3sg-‐
(40)
The completely faithful output (40a) again fatally violates BALANCE. Aside from
violating LINEARITY, the metathesized forms in (40c) and (40d) receive fatal
violations from ONSETMAX and *CCONSET, respectively. Since the consonantal
sequence resulting from syncope is harmonious there is no need for a repair
strategy employing the reversal of segments; candidate (40b) is optimal.
5.1.2 Coloured Containment
An account based on Coloured Containment takes metathesis to be driven by the
requirement of preserving material, which spreads to a position newly formed at
the phonological structure. However, the situation concerning CC metathesis is
47
slightly different from the CV metathesis discussed in §2.2; while vocalic material
l and can spread across consonants, the position
of consonants themselves is not so luxurious (see also footnote 14). Since
consonants are situated on the same tier and Coloured Containment considers
any metathesis as the spreading of material into a newly created position, CC
metathesis would lead to the crossing of association lines. However, this is not
allowed based on the no-‐crossing principle (e.g. Goldsmith 1976, Coleman 1998),
which prohibits the crossing of association lines. Consider for instance the
representation below:
(41)
The spreading of the coronal segment is blocked by the presence of a
consonantal dorsal segment. However, it has been claimed that coronality is
actually unspecified for place underlyingly and rather a default implementation
(e.g. Avery & Rice 1989). In such a view coronal segments can be crossed since
they cannot obstruct any segments by the very fact they do not have content;
consider the revised representation of metathesized (< ) in (42):
(42)
The dorsal segment can move unhindered towards a pre-‐vocalic position. Note
that this approach also predicts that non-‐coronal segments are the ones that
u {r} g _ o ° ° ° ° ° {cons} cons cons | dorsal coronal
u _ r {g} o ° ° ° ° ° cons cons {cons} dorsal
48
metathesize, since PARSE-‐
question how coronal-‐coronal metathesis is possible; to avoid an analysis in
which the reversal in > has a crucially different motivation from
the metathesis in > it might be assumed that in some cases
coronality is not underspecified, notably when they would serve a contrastive
function, e.g. stops can be dorsal, labial and coronal in Rendille, but liquids are
always (i.e. by default) coronal and only these latter are considered unspecified
for coronality.54
Formalizing metathesis in Coloured Containment is fairly straightforward; the
constraints BALANCE, *CCONSET and ONSETMAX remain the same and MAX-‐IO is
replaced by PARSE-‐ (CONS
PARSE-‐µ(CONS) and the fact that the deleting segment does not delete altogether
but retains its features is captures in PARSE-‐ (C-‐PLACE); the tableau below
visualizes the situation for an input :
(43)
5.1.2.1 Dorsal-‐labial metathesis
A crucial prediction of Coloured Containment is the impossibility of dorsal-‐labial
metathesis. However, Blevins & Garrett claim the existence of a monodirectional
change PK > KP (see also §1.1.3). They cite four languages that presumably
display PK > KP metathesis (Klamath, Wiyot, Aklanon and Mikolese), but the
evidence they provide for it is dubious at best. Three of the languages involve
(Blevins &
Garrett 2004: 136). Evidence from Klamath seems to be limited to a few words 54 Cf. Contrastive Specification Theory (see e.g. Steriade 1987, Clements 1988, Mester & Itô 1989).
49
denoting close family relationships such as bqolyb
variable realizations and (Barker 1964); Wiyot diachronic
metathesis is only supported by a single comparison of Wiyot kbad
pkenc -‐Ritwan *pkanc (Berman 1990).
Mikolese shows variation in the realization of clusters:
(44) Underlying form Surface form ~ ~ ~
However, the process is not productive (Harrison cited as p.c.) and there are only
two more words in Mikolese that show metathesis of a labial and a velar stop:
(45) ~ ~
Clearly, lexicalized metatheses should not be regarded as stemming from the
phonology.55 The final case, Aklanon, also displays diachronic metathesis: e.g.
Cebuano libgus palibga
ligbus paligba
instance of metathesis involving labial-‐dorsal clusters, but since diachronic
processes are not completely reliable the language will not be regarded as a valid
argument against the ban on dorsal-‐labial metathesis.
PK > KP exists, but cites two South Dravidian languages, Kui and Kuvi, which
display KP > PK metathesis. Consider the data from Kui, which show metathesis
of stem-‐final k and morpheme initial p in the present participle (-‐pi) and the
infinitive (-‐pa) in (46) and non-‐metathesizing forms in (47):
55 Moreover, the stops in the left column of (45) are interrupted by a labial glide, violating the locality principle, although the [w] might also reflect liprounding of the labial, i.e. [pw].
50
(46) Stem Future Past Pres.part. Infinitive
(47) Stem Future Past Pres.part. Infinitive
Blevins & Garrett (2009), however, claim that the surface metathesis in Kui and
Kuvi is not a case of reversal at all. They argue that an opaque causative
formation process led, through reinterpretations of the forms, to a situation in
which some stems had two allomorphs: for instance,
an alternative allomorph . Furthermore, the suffixes had other basic
variants, which were dorsal-‐initial and operated on these alternative stem forms
resulting in PK sequences. Crucially, the alternations in the South Dravidian
languages do not stem from segmental reversals but result from restructuring
leading to stem allomorphy and subsequently, the choice of a different allomorph
of the suffix.56
To conclude the section on phonotactically driven metathesis, an analysis that
regards metathesis as a by-‐product fares better than a model based on a
separate mechanism exemplified by the LINEARITY constraint. Whereas the
LINEARITY constraint does not make any predictions with regard to the
composition of metathesized segments, in Coloured Containment it is expected
that metathesis of dorsals and labials is impossible by its incorporation of
feature spreading; this prediction is confirmed by the data.
56 Further evidence against dorsal-‐labial metathesis stems from the Chinna Kindi dialect of Kui, in which the following variation is observed: or
and and (Friend-‐Pereira 1909); this can be interpreted as language users analyzing the sequences as metathesized forms and avoiding dorsal-‐labial metathesis by deleting the labial segment.
51
5.2 Morpho-‐phonological metathesis
Metathesis that is strongly rooted in phonotactics apparently does not require
any particular mechanism and can be easily explained in Coloured Containment.
However, morpho-‐phonological metathesis is by its very nature deeper
entrenched in the grammar and might be more challenging for an analysis that
regards metathesis as a by-‐product of other phonological processes. Rotuman
will represent a metathesis that is templatic in nature; rather than alluding to
phonotactics it takes a morphologically based prosodic prescription as its
motivation. As mentioned earlier (§3.2), Rotuman makes use of an array of
strategies to indicate phase distinctions; these strategies are exemplified in (48):
(48) Complete phase Incomplete phase (a) Metathesis (b) Umlaut ö (c) Deletion (d) Diphthongization (e) Nothing é é é é
McCarthy (1995) identifies the template alluded to in §3.2; in the incomplete
phase words need to adhere to the restriction of containing a bimoraic foot at the
right edge of the word. (49) shows that all incomplete phase items fit this
prosodic structure (note that codas are moraic in Rotuman).
52
(49) Incomplete Phase (prosodic structure) Ft | / \ / \ | p u e r metathesis57 m ö s umlaut t i deletion v a o diphthongization r i i nothing
This template is regarded as the morpheme of the incomplete phase; an
incomplete phase stem contains the (complete phase) stem plus an affix, which
captures the requirement of the incomplete phase to pattern as a bimoraic foot.
Hence, the notation / +IncPh/ represents a structure as in (50):
(50) /pure + Ft / | / \
The incomplete phase morpheme introduces a structure that most complete
phase forms do not fit to; however, the constraint demanding the realization of
the morpheme is undominated, resulting in the strategies observed in (48). In
(51) below an informal definition of Realize-‐morpheme (RM) is given (Van
Oostendorp 2007a):
(51) Realize-‐morpheme (RM): every morpheme in the input has to contribute to the phonological surface structure
Since the focus lies on metathesis, the structures in (48d) and (48e) will not be
considered here (but see McCarthy 1995). The remaining incomplete phase
57 Evidence for analyzing the nucleus of metathesized forms as monomoraic diphthongs stems from stress assignment as well as the fact that some of the light diphthongs in Rotuman have been transcribed as sequences of an onglide and vowel, e.g. the incomplete phase of rito been recorded as ryot (Besnier 1987). Furthermore, diphthongs in closed syllables can only rise in sonority, which is a particular characteristic of light diphthongs (Kaye 1983, Kaye & Lowenstamm 1984, Rosenthall 1994).
53
strategies (48abc) all have the same structure in the complete phase, that is, they
end in a VCV sequence. Comparing the strategies of metathesis, umlauting and
deletion, there seems to be an intuitive notion that metathesis is the most
conservative solution to satisfy the incomplete phase requirement, while
deletion is the most rigorous deviance from the original, i.e. the complete phase
counterpart. Both the view that takes metathesis to be a separate process
(§5.2.1) as well as a model that regards metathesis as a surface result of other
processes (§5.2.2) conform to this intuition, but the analyses differ crucially in
their expression of the relationship between the strategies listed in (48a-‐c).
5.2.1 Correspondence Theory
Starting with the analysis of metathesis, LINEARITY must be low-‐ranked to allow
the metathesized candidate to be the winner. Furthermore, the fact that in the
metathesizing contexts deletion does not apply while in other contexts the final
vowel deletes shows that MAX-‐IO must outrank LINEARITY. Another constraint
which plays a direct role in discussing metathesized forms is HEAD-‐MATCH (52),
which demands that the underlying prosodic head of the word must also be the
surface prosodic head of the word (McCarthy & Prince 1995).
(52) HM (Head-‐Match): the prosodic head of the word.
HEAD-‐MATCH prevents Rotuman from employing strategies to adhere to the
incomplete phase by changing the foot structure. McCarthy (1995) identifies the
main stress foot of Rotuman to be a moraic trochee; consequently an input such
as pure is footed {púre} (the first syllable is the head). In the incomplete phase
adjustment is required and various alternatives are available; consider the forms
below, which list candidates satisfying HM (53a) and some that violate HM
(53b):
(53a) {puer} (b) *pu{er} {pur} *pu{ree} {pür} *pu{ret} *{per}
54
The words in the left column share the property that they have the same
segment in head position as the input pure, thus satisfying HM. It needs to be
noted that the umlauted form pür is a coalesced form and thus corresponds to
both vowels of the complete form pure; since the umlauted form contains the
dependent as well as the head element it still satisfies HM.58 The following
tableau depicts the interaction of the discussed constraints (the umlauted
candidate pür is not included; in the current analysis it would share winner
status with the actual winner puer but see below why puer is favored over pür):
(54)
The first candidate violates the undominated incomplete phase requirement.
(54de) satisfy RM but do not maintain the prosodic head of the input in the
output. (54b) survives RM and HM but at the cost of deleting a segment
(violating MAX-‐IO), while the winner only violates the lower-‐ranked LINEARITY
constraint.
However, metathesis is restricted by the constraint LIGHT-‐DIPHTHONG,
which limits light diphthongs to sequences that rise in sonority.
(55) LD (Light-‐Diphthong): light (monomoraic) diphthongs can only rise in sonority
This restriction on monomoraic diphthongs is widely recognized (Kaye 1983;
Kaye & Lowenstamm 1984; Rosenthall 1994). The preference for rising sonority
might be connected to the fact that the second element in the diphthong is the
58 It should also be noted that HM is an output-‐output constraint; that is, instead of a correspondence relationship between the input and output there is a relationship between one output form and another output form. Such an extension to relationships other than input-‐output is available in a multiple layered model of phonology such as Correspondence Theory but increases the power of the model greatly and should be regarded critically.
55
head of the nucleus and thus requires the most prominent material: the most
sonorous material in the case of nuclei (see also §4.1).
Since Rotuman resorts to umlauting rather than deletion when metathesis is
ruled out due to LD, LD must outrank MAX-‐IO. Umlauting involves a violation of
UNIFORMITY, a constraint which, as LINEARITY, is specifically developed to describe
coalescence and thus is effective yet relatively uninsightful.
(56) Uni (Uniformity): No element at the surface has multiple correspondents underlyingly
UNIFORMITY and LINEARITY cannot be ranked with respect to each other in
Rotuman, because in this language every violation of UNIFORMITY also involves a
violation of LINEARITY; in other words, umlauting only occurs in the incomplete
phase when they are realized on the first vowel with concomitant deletion of the
second vowel.
(57)
The metathesized candidate is ruled out due to LD and umlaut is favored over
deletion due to MAX-‐IO outranking both UNIFORMITY and LINEARITY.59 It is worth
mentioning that in the case of the incomplete phase of pure the metathesized
form puer wins from umlauted pür because the latter involves an unmotivated
additional violation of uniformity when compared to puer.
A final matter that needs to be addressed concerning the umlauting cases
is the form of the coalesced form; why does it adopt only the frontness of the
second vowel (taf) rather than for instance also its place features (tüf)? McCarthy
IDENT[ITY] constraints, which select 59 An apparently obvious candidate is left out of the tableau: tiaf, in which the final vowel of the complete form not only is transposed with the consonant but also with the first vowel, thus satisfying the restriction on monomoraic diphthongs. However, as mentioned in the introduction, metathesis is assumed to be restricted to local movement.
56
among the confli Identity
constraints refer to the degree of correspondence between input and output; for
instance, IDENT(PLACE) demands identical values for the feature [place] in both
the input and output form. In the previous example shown in the tableau in (56),
tafi becomes taf in the incomplete phase and not *tüf because the latter does not
satisfy IDENT(PLACE).
In the case of the third strategy that Rotuman employs to resolve the incomplete
phase, i.e. deletion, McCarthy (ibid.) again resorts to the same set of ranked
Identity constraints responsible for the featural make-‐up of the umlauted form;
will block coalescence when conflicting featural allegiances are
onstraints are required for a
comprehensive analysis of Rotuman, an abstract IDENT-‐X will be used in the
set of Identity constraints, which
determines the format of the coalesced form and which blocks umlauting in
favor of deletion in the right contexts.
(58)
is only opted for after metathesis and umlaut are ruled out due to higher-‐ranked
constraints.60
In Rotuman, the template for the incomplete phase demands a change in the
underlying structure in the case of words ending in VCV sequences. The
following constraint ranking is responsible for the strategies of Rotuman;
60 Note that MAX-‐IO(V) must be low-‐ranked, since the vowel is deleted, while the consonant is retained (high ranking of MAX-‐IO(C)).
57
deletion is only resorted to if it cannot be resolved by metathesis (violating LD)
or umlaut (violating IDENT-‐X):
(59) RM, LD, HM, IDENT-‐X » MAX-‐IO » UNIFORMITY, LINEARITY
5.2.2 Coloured Containment
some features. Based on the
phonemic inventory, Rotuman vowels have the following feature geometry:61
(60)
It is assumed that these nodes can move around more or less independently;
however, a head node always takes its dependent nodes with it. Hence, if a
vocalic node is transposed, its place features as well as its front or round features
are automatically taken along; however, in the case the front node is transposed
only its roundness feature would be carried along. In this view, metathesis and
umlauting share the principle of preservation of material, but in the case of
metathesis more material is preserved since the higher place node spreads
towards a new position, while in the case of umlauting only the feature for
frontness is retained.
Again the incomplete phase is regarded as a morphemic template, with the
concomitant constraint to realize the morpheme (RM). Focusing on metathesis,
the transposition of the final segments is represented by two constraints: PARSE-‐
(VOC) and PARSE-‐µ(VOC) .
61 The features are privative and [front] and [round] are not necessarily specified; for instance, is specified for [round] but not for frontness, while contains the [front] feature but no
roundness feature. Moreover, segments which are both [front] and [round] are marked structures and only derive from umlauting in the incomplete phase (see also below).
° voc ° place ° front ° round
58
(61) Parse-‐ e incorporated in the phonology
(62) Parse-‐µ(voc): the phonological feature (voc) must be incorporated in the morphology
Since these constraints are violated there must be a higher-‐ranked constraint,
which drives these violations. In the case of metathesis it is the preservation of
the morphological [place] feature:
(63) Parse-‐ ed in the phonology
The incomplete phase and RM forces the final vowel to remain unrealized;
however, since its [place] features need to be retained by virtue of PARSE-‐
(PLACE) this forces the phonological insertion of a vocalic node (a violation of
PARSE-‐µ(VOC)) onto which the [place] node can dock. In (64) this process is
schematically represented:62
(64) The final vocalic node remains unparsed at the phonological level; however, the
place features of the final vowel have to be realized phonologically due to PARSE-‐
(PLACE). Consequently, they spread to the {voc} node that is only linked to the
phonological level. The final aspect of metathesized forms includes a constraint
that rules out differences between the foot structure in the morphological
structure and the phonological structure (cf. HM above).63
62 Note that this situation is different from consonantal metathesis discussed above (§5.1.2);
association lines. 63 Note that Coloured Containment cannot refer to output-‐output relationships (see also footnote 58); instead, mirroring constraints are used to reflect the structures between the morpho-‐
p u {e} r _ ° ° ° ° ° {voc} voc place
59
(65) phonology
The tableau below shows the interaction of the constraints involved in choosing
the metathesized candidate as the winner:
(66)
The final candidates (66de) do not conform to the constraint requiring
consistency between the prosodic structure at the morphological level and the
phonological level, that is, PARSE-‐ (FTS). Metathesis is favored over deletion
since it preserve the original place features, even though they are maintained at
the cost of violating PARSE-‐µ(VOC) and PARSE-‐ (VOC).
However, the restrictions on the nucleus in metathesized items remain the same
in a CC as well as in a CT analysis; the prohibition against non-‐rising monomoraic
diphthongs is captured in LD. In the case that undominated LD rules out
metathesis, the final vowel, which is forced to remain unrealized due to RM, is
not able to maintain its [place] features. However, umlauting at least allows its
[front] feature to be maintained, which is expressed in the constraint in (67).
(67) Parse-‐ into the phonological structure
syntactic and phonological structures; however, it is not immediately apparent why a morpho-‐syntactic structure would already be parsed prosodically. A possible solution lies in assuming different strata through which a form proceeds; the derivation of incomplete phase stems then would be on a lower stratum than the derivation of a complete phase stem. Any item including an incomplete phase stem first has to go through the higher stratum, which in this case involves the derivation of the complete stems, to be able to reach the lower stratum. In effect, this means that an incomplete phase stem goes through a prosodic parsing as a complete phase stem before entering the derivation as an incomplete phase stem; the prosodic parse in the complete phase stem stratum would serve as a reference in the constraint in (65) then.
60
Note, though, that it is only the [front] feature, which is preserved; final vowels
with a back specification do not bestow their backness on front vowels (ti u
becomes ti in the incomplete phase). This situation is predicted in the current
account since [front] is a privative feature and back vowels are simply not
specified for it. Rather, back vowels identify themselves as [round], a feature that
does not spread when its source vowel is deleted (see also below). The tableau in
(68) shows that the umlauted candidate is optimal:
(68)
The metathesizing candidate (68c) does not survive due to high-‐ranking LD;
umlauted (68d) wins from the candidate which opts for deletion because in the
former the frontness feature of the final vowel is preserved. The fact that the
umlauted candidate retains the ([round]) specifications of the first vowel with
only the [front] feature of the second vowel added follows naturally from the
featural hierarchy displayed in (60), since only the [front] feature requires
retention and it can simply dock onto the empty [front] node of the first vowel.
The preservation of the frontness of the final vowel comes at a cost though; the
spreading of the features onto the previous vowel results in a segment specified
for both [front] and [round]. Such complex segments are not observed anywhere
in the language except when they result from umlauting in the incomplete phase;
consequently umlauted forms violate a markedness constraint against complex
segments:
(69) *Complex: Segments cannot be specified both [front] and [round]
Crucially, PARSE-‐ (FRONT) outranks *COMPLEX: the preservation of the feature
[front] is more important than keeping segments simple.64
64 Presumably, in forms in which both vowels are front (e.g. tokiri) the [front] feature of the final vowel also spreads but since they are both front this spreading is not registered phonetically; technically, though, these cases should not be regarded as examples of deletion, though.
61
Complete phonetic loss of the final vowel is a direct result of LD preventing non-‐
rising monomoraic diphthongs combined with the fact that
vowel does not contain a [front] feature, which is protected by PARSE-‐ (FRONT).
Rather, the fact that deletion is favored over umlaut is the fact that the constraint
that demands the realization of roundedness, PARSE-‐ (ROUND), is crucially
outranked by *COMPLEX.65
(70) Parse-‐ morphological feature [round] must be incorporated into the phonological structure
The full set of constraints is depicted in the tableau below, which shows that if
metathesis and umlaut are not warranted, deletion of the final vowel is
observed:
(71)
The undominated RM constraint in Rotuman forces the prosodic structure of any
incomplete phase stem to conform to the morpheme of the incomplete phase, i.e.
a monosyllabic foot. In the set of underlying VCV words this necessarily leads to
a restructuring. The constraint ranking in (71) explains the data; if preservation
of the place features by metathesis is possible it is the optimal resolution;
realization of the front feature is the next solution, even at the cost of creating a
complex segment; finally, maintaining roundness is lower ranked than the ban
against complex segments and thus deletion is observed.
(72) RM, LD, PARSE-‐ FTS) » PARSE-‐ FRONT) » *COMPLEX » PARSE-‐ ROUND) » PARSE-‐ PLACE) » PARSE-‐µ(VOC), PARSE-‐ VOC)
65 As in the CT account, it should be noted that deletion of the consonant is favored over deletion of the vowel (the incomplete form of ti u is ti rather than *tiu).
62
5.3 Correspondence Theory versus Coloured Containment
Comparing CT and CC there are some constraints which are clearly equivalent:
RM and LD are identical in both models, HM and PARSE-‐µ(FTS) are comparable,
and MAX-‐IO and PARSE-‐ VOC) correspond to each other straightforwardly as
well. The remaining constraints in a correspondence based account are: IDENT-‐X,
UNIFORMITY and LINEARITY. The two latter constraints are specifically designed to
account for the processes discussed here and thus are naturally descriptively
adequate but not very insightful. The last remaining constraint, IDENT-‐X, is an ill-‐
defined set containing an unclear number of constraints. It is claimed that
metathesis is the most faithful strategy, followed by umlaut and that deletion is
the least faithful resolution; however, this does not follow directly from the
nature of the constraints involved. In particular, the set of Identity constraints
responsible for both the formation of the umlauting forms and the decision
whether forms are subject to deletion is of the same nature, which groups these
two strategies together as opposed to metathesis, which only involves a
straightforward violation of LINEARITY. Furthermore, umlauting and metathesis
are intuitively a similar process of preservation of material that cannot be
pronounced at the place of origin; although both are united in violations of
LINEARITY it is an accidental aspect of the Rotuman grammar that all umlauting
cases also involve metathesis LINEARITY and UNIFORMITY do not have anything in
common as such.
Turning to Coloured Containment, this analysis leaves us with the
following constraints (aside from those similar to the ones employed in CT):
*COMPLEX, PARSE-‐ PLACE), PARSE-‐ FRONT) and PARSE-‐µ(VOC). Constraints against
complex segments are commonplace and one against front rounded vowels is
particularly unsurprising since these segments are universally rare (Lass 1989).
The three other constraints follow directly from the CC model; PARSE-‐ PLACE)
and PARSE-‐ FRONT) unite the strategies of metathesis and umlaut by virtue of
identifying the same motivation for both: the pressure to realize morphological
material. Furthermore, deletion is identified as the failure of the
preservation of roundness, which finds a natural explanation in PARSE-‐µ(ROUND)
being lower-‐ranked than *COMPLEX.
63
Clearly, it is unnecessary to posit metathesis as a separate phonological process:
CC can easily accommodate for surface metathesis. Moreover, an analysis based
on Coloured Containment explains the lack of dorsal-‐labial metatheses, since
consonantal metathesis is restricted to the crossing of unspecified segments (i.e.,
coronals). Moreover, a detailed analysis of Rotuman CV metathesis reveals that
regarding metathesis as a by-‐product of other phonological processes is more
insightful than positing phenomenon-‐specific constraints, since a more
comprehensive relation between metathesis and umlaut is established.
64
3 CONCLUSION
65
6 Conclusion
The current chapter will recapitulate the arguments in favor of placing the
majority of metatheses in the phonological module, although a few well-‐defined
cases are argued to be the result of perceptual confusion. The implementation of
the view that metathesis is phonological in nature will be summarized in §6.2
and the superiority of a view maintaining metathesis as a by-‐product will be
reiterated.
6.1 The Locus of Metathesis
The majority view regards metathesis as a fully-‐fledged phonological process,
but proponents of diachronic phonetics have challenged this claim. In chapter
three it is shown, though, that a phonetic account cannot be maintained. Notably,
synchronic metathesis raises difficulties concerning the core idea of how
grammar should be perceived; a diachronic phonetic account necessarily has to
assume a usage-‐based grammar, while the status of such grammars is
controversial (see also chapter 3). Furthermore, diachronic phonetics fails to
capture generalizations arising from especially morphophonologically driven
metathesis and has to regard them as accidental coincidences. Finally,
phonotactic considerations have shown to be relevant in both CC and CV
metathesis; in particular the sonority pattern identified for syllables plays a role
in metatheses, since languages tend to reduce the markedness of their
structures. However, abstract structures such as the syllable are not available in
a view based on diachronic phonotactics.
Nevertheless, there seem to be instances in which phonetics is relevant;
for some cases of metathesis there are strong arguments to regard them as
motivated by perceptual considerations. In particular, variable metathesis is best
analyzed as phonotactically based; although it can be modeled phonologically an
analysis in which listeners have difficulty identifying the exact location of
segments accounts naturally for the observed alternations.
66
Consequently, although most metathesis is assumed to be firmly
grounded in the phonology, the existence of perceptual confusion leading to a
variable ordering of segments is acknowledged. Note, though, that this is not
metathesis in its strict sense of segment reversal.
6.2 The Reality of Metathesis
Whether metathesis is regarded as a phonological process or a by-‐product is
intrinsically related to which framework is chosen to represent the grammar
(see chapter 2). Although descriptive adequacy is naturally a prerequisite for any
theory, it cannot be a measure in the investigation which of the two approaches
is superior, since an account that incorporates the process of metathesis in its
very grammar will of course be fully accurate. Instead, the challenge lies in
whether such an elaborate account is necessary or whether it is also possible to
maintain a more restrictive model in which metathesis is constrained by other
processes. It turns out that the more restrictive theory in which metathesis as a
process is ruled out is not only descriptively satisfactory, but it makes the
particular prediction that dorsal-‐labial metathesis is impossible, which indeed is
borne out. Furthermore, in a detailed study of CV metathesis an account based
on Coloured Containment proved to be explanatorily more satisfying as
metathesis and umlaut were clearly related to each other, functioning as
conservation strategies involving different degrees of preservation. Note also
that a view of metathesis as a by-‐product predicts that metathesis should be
infrequent since a particular blend of restrictions is responsible rather than a
single constraint; thus the relative scarcity of metathesis cross-‐linguistically is
accounted for.
The marginal cases of perceptual metathesis identified in §3.1 and the
interpretation of metathesis as a phonological by-‐product lead to the conclusion
that metathesis is only an illusionary effect and that it does not exist in its
traditional meaning of segment reversal; at the level of phonetic reality the
segments seem to have switched place but they do not transpose grammatically.
67
grammatical constraints some of the segments cannot be parsed into the
phonetics; however, if the pressures to express their material are strong enough
the subsegmental features spread to a different position to ensure their
realization, creating the illusion of metathesis.
6.3 Beyond Metathesis
Up to this point, the case of metathesis has been discussed in isolation; however,
the current analysis naturally has implications on other aspects of phonological
theory.
Diachronic phonetics has been shown not to be able to account for
role of phonetics should not be underestimated; although it might play a
marginal role (such as in the case of metathesis), implementing it in a
phonological theory skews the model in the sense that it might widen its scope to
a degree that it explains more than desired; such overgeneralization can be
avoided if phonetic processes are acknowledged and excluded from the
phonology.
The fact that metathesis is better analyzed in a Coloured Containment
model naturally has repercussions for phonological theory in general. The fact
that a mono-‐stratal phonological module can account for a phenomenon
originally regarded as a process raises the question whether more phenomena
traditionally analyzed as processes might be better accounted for in a different
framework. Notably, Coloured Containment is not impaired with the
shortcomings of traditional Containment Theory. From Containment Theory it
retains one phonological representation, which carries the burden of explaining
differences between lexical forms and their phonetic implementation. However,
different from the original conception of Containment Theory, it incorporates the
notion of faithfulness as mirroring constraints between the morpho-‐syntactic
and phonological structure. The obvious advantage of Coloured Containment is
the restrictiveness of the model due to its single level of representation;
68
however, it remains to be seen whether it can stand firm in light of other
phonological phenomena.
6.4 Future Research
Some of the predictions that the current analysis of metathesis makes should be
put to further scrutiny, since they are essential to the claim that Coloured
Containment is the superior model. In particular, the cases listing dorsal-‐labial
metathesis, even if these are diachronic in nature, should be investigated in more
detail, especially the evidence for metathesis in Aklanon. Additionally, the theory
of contrastive specification should be verified by carefully investigating
languages displaying coronal-‐coronal metathesis; they should include one
segment that is not grammatically specified for coronality and one segment that
has to be specified to ensure that PARSE-‐µ(F) can see the coronal feature.
Finally, the conclusions drawn here serve future research into formalizing
phonology in general and the framework of Coloured Containment in particular.
The theory should be assessed against more phonological phenomena, but as has
been preliminarily shown here for umlauting, assimilations seem to pattern
naturally in Coloured Containment. Further advantages of Coloured Containment
might stem from investigating opacity; for instance, in Faroese, the underlying
form ; aside from metathesis the nasal
clearly is affected by the velar stop, which in Coloured Containment is still
present to give the nasal its place of articulation, while in Correspondence
Theory it has been deleted in the output and is not able to leave a trace. The
future for Coloured Containment seems to be quite promising but careful
research will have to show whether it can be maintained.
69
References
Avery, P. and K. Rice (1989). Segment structure and coronal underspecification. Phonology 6. Barker, M. (1964). Klamath grammar. Berkeley: University of California Press. Bermúdez-‐Otero, R. (2007). Diachronic Phonology. In P. de Lacy (ed.) The Cambridge Handbook
of Phonology. Cambridge: Cambridge University Press. Besnier, N. (1987). An autosegmental approach to metathesis in Rotuman. Lingua 73. 201-‐223. Blevins, J. (1995). The syllable in phonological theory. In J. Goldsmith (ed.) The Handbook of
Phonological Theory. Oxford: Blackwell. Blevins, J. (2004). Evolutionary Phonology. Cambridge: Cambridge University Press. Blevins, J. and A. Garrett (1998). The origins of consonant-‐vowel metathesis. Language 74. 508-‐
556. Blevins, J. and A. Garrett (2004). The evolution of metathesis. In B. Hayes, R. Kirchner and D.
Steriade (eds.) Phonetically based phonology. Cambridge: CUP. Blevins, J. and A. Garrett (2009). Analogical morphophonology. In S. Inkelas and K. Hanson (eds)
The Nature of the Word: Essays in Honor of Paul Kiparsky. Cambridge, MA: MIT Press. Boersma, P. (1998). Functional Phonology: Formalizing the interactions between articulatory and
perceptual drives. The Hague: Holland Academic Graphics. Bonthuis, F. (2001). Metathesis in Leti. In E. Hume, N. Smith and J. van de Weijer (eds) Surface
Syllable Structure and Segment Sequencing. Leiden: HIL. Botma, B. and C. Ewen (2008). Against rhymal adjuncts: the syllabic affiliation of English. To
appear in P. Backley and K. Nasukawa (eds) Strength Relations in Phonology. Berlin: Mouton de Gruyter.
Bregman, A. (1990). Auditory scene analysis. Cambridge, MA: MIT Press. Browman, C. and L. Goldstein (1989). Articulatory gestures as phonological units. Phonology 6. Brugmann, K. (1902). Kurze vergleichende Grammatik der indogermanischen Sprachen. Berlin:
Walter de Gruyter. Bunye, M. and E. Yap (1971). Cebuano Grammar Notes. Honolulu: University of Hawaii Press. Butskhrikidze, M. and J. van de Weijer (2001). On v-‐metathesis in Modern Georgian. In E. Hume,
N. Smith and J. van de Weijer (eds) Surface Syllable Structure and Segment Sequencing. Leiden: HIL.
Campbell, L. (2004). Historical Linguistics: an Introduction (2nd edn). Cambridge, MA: MIT Press. Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge, Mass: MIT Press. Clements, G. (1988). Towards a substantive theory of feature specification. Papers from the
Annual Meeting of the North East Linguistic Society 18. Coleman, J. (1998). Phonological representations: their names, forms and powers. Cambridge: CUP. Connell, B. (1994). The structure of labial-‐velar stops. Journal of Phonetics 22. Derrick-‐Mescua, M. (1980). A phonology and morphology of Mikasuki. Ann Arbor, MI: University
Microfilms International. Ewen, C. H. van der Hulst (2001). The Phonological Structure of Words: An Introduction.
Cambridge: CUP. Friend-‐Pereira, J. (1909). A grammar of the Kui language. Calcutta: Bengal Secretariat Book
Depôt. Festschrift Jacob Wackernagel zur Vollendung des 70.
Lebensjahres am 11. Dezember 1923. Göttingen: Vandenhoeck & Ruprecht. Goddard, P. (1912). Elements of the Kato language. Berkeley: University of California Press. Goldsmith, J. (1976). Autosegmental Phonology. Ph.D. dissertation, MIT. Hale, M. and M. Kissock (1998). The phonology-‐syntax interface in Rotuman. In M. Pearson (ed.)
Proceedings of the Austronesian Formal Linguistics Association, UCLA, 1996 and 1997. Los Angeles: Department of Linguistics, UCLA.
70
Halle, M. and V. Zeps (1966). A survey of Latvian morphophonemics. Quarterly Progress Report of the Research Laboratory of Electronics, MIT.
Harris, J. (1994). English Sound Structure. Oxford: Blackwell. Hock, H. (1985). Regular metathesis. Linguistics 23. Howell, R. (1991). Old English Breaking and its Germanic Analogues. Tübingen: Max Niemeyer
Verlag. Hulst, H. van der and A. van Engelenhoven (1994). Metathesis effects in Tutukeian-‐Letinese. Ms.
Leiden University. Hume, E. (1991). Metathesis in Maltese: Implications for the Strong Morphemic Plane Hypothesis.
Proceedings of NELS 21. University of Massachusetts, Amherst. Hume, E. (2001). Metathesis: Data, Motivation and Phonological Theory. In E. Hume, N. Smith and
J. van de Weijer (eds) Surface Syllable Structure and Segment Sequencing. Leiden: HIL. Jakobi, A. (1990). A Fur grammar. Hamburg: Helmut Buske Verlag. Johnson, H. (2000). A Grammar of San Miguel Chimalapa Zoque. Ann Arbor, MI: Bell & Howell
Information and Learning Company. Johnson, K. (1997). Speech perception without speaker normalization. In K. Johnson and J.
Mullennix (eds) Talker variability in speech processing. San Diego: Academic Press. Jordan, R. (1974). Handbook of Middle English grammar: Phonology. Translated and revised by E.
Crook. The Hague: Mouton. Kaplan, A. (2006). Vowel Length and Coda Cluster Interactions in Misantla Totonac. Proceedings
of the 29th Annual Penn Linguistics Colloquium. Penn Linguistics Club. Ladefoged, P. (1993). A course in phonetics. Forth Worth, TX: Harcourt Brace Jovanovich. Ladefoged, P. and I. Maddieson (1996). . Oxford: Blackwell. Ladefoged, P., I. Maddieson and M. Jackson (1988). Investigating phonation types in different
languages. In O. Fujimura (ed.) Vocal physiology: Voice production, mechanisms and functions. Raven: New York.
Lass, R. (1989). System-‐Shape and the Eternal Return: Front Rounded Vowels in English. Folia Linguistica Historica 10.
Lyche, C. (1995). Schwa metathesis in Cajun French. Folia Linguistica 29. 369-‐393. McCarthy, J. (1989). Linear order in phonological representation. LI 20. 71-‐99. McCarthy, J. (1995). Extensions of Faithfulness: Rotuman Revisited. ROA-‐110. McCarthy, J. and A. Prince (1995). Faithfulness and Reduplicative Identity. UMOP 18. 249-‐384. Mester, R. and J. Itô (1989). Feature predictability and underspecification: palatal prosody in
Japanese mimetics. Language 65. Mielke, J. and E. Hume (2001). Consequences of Word Recognition for Metathesis. In E. Hume, N.
Smith and J. van de Weijer (eds) Surface Syllable Structure and Segment Sequencing. Leiden: HIL.
Papers in Romance 3. Ohala, J. (1971). The role of physiological and acoustic models in explaining the direction of
sound change. Project on Linguistic Analysis Reports (Berkeley) 15. 25-‐40. Ohala, J. (1981). The listener as a source of sound change. In C. Masek, R. Hendrick and M. Miller
(eds) Papers from the Parasession on Language and Behavior. Chicago: Chicago Linguistic Society.
Ohala, J. (1993). The phonetics of sound change. In C. Jones (ed.) Historical linguistics: problems and perspectives. London: Longman.
Oostendorp, M. van (2007a). The Theory of Faithfulness. Ms. Meertens Instituut. Oostendorp, M. van (2007b). Derived Environment Effects and Consistency of Exponence. In S.
Blaho, P. Bye and M. Krämer (eds) Freedom of Analysis? Berlin: Mouton de Gruyter. Oxford English Dictionary Online (2009). Oxford: Oxford University Press. [http://www.oed.com]
(accessed on May 20th 2009) Phillips, B. (2006). Word Frequency and Lexical Diffusion. London: Palgrave.
71
Pierrehumbert, J. (2002). Word-‐specific phonetics. In C. Gussenhoven and N. Warner (eds) Laboratory Phonology 7. Berlin: Mouton de Gruyter.
Prince, A. and P. Smolensky (1993). Optimality Theory: Constraint interaction in generative grammar. ROA-‐537.
Rice, K. (1992). On deriving sonority: a structural account of sonority relationships. Phonology 9. 61-‐99.
Seo, M. and E. Hume (2001). A comparative OT account of metathesis in Faroese and Lithuanian. In E. Hume, N. Smith and J. van de Weijer (eds) Surface Syllable Structure and Segment Sequencing. Leiden: HIL.
Steriade, D. (2001). Directional asymmetries in assimilation: A directional account. In E. Hume and K. Johnson (eds) The Role of Speech Perception in Phonology. New York: Academic Press.
Ultan, R. (1978). A typological view of metathesis. In J. Greenberg (ed.) Universals of human language, vol. 2, Phonology. Stanford, CA: Stanford University Press.
Vennemann, T. (1988). Preference Laws for Syllable Structure and the Explanation of Sound Change: With Special Reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter.
West, P. (1999). Perception of distributed coarticulation properties of English /l/ and /r/. Journal of Phonetics 27.
West, P. (2000). Long-‐distance coarticulatory effects of English /l/ and /r/. Ph.D. thesis, University of Oxford.
Zorc, R. (1975). The Bisayan dialects of the Philippines: Subgrouping and Reconstruction. Ann Arbor, MI: University microfilms International.