To appear in: Cuadernos de Linguistica XV, 2008

Item-and-Arrangement or Item-and-Process?1
EULÀLIA BONET (UNIVERSITAT AUTÒNOMA DE BARCELONA)

Abstract. The choice of gender marker (or class marker) in Catalan masculine
nominals is determined partially by a phonological constraint. In this paper this
phenomenon is used to evaluate some differences between an Optimality Theory-
based Item-and-Arrangement (IA) model, in which each morphosyntactic feature has
a corresponding exponent (or more than one), and Item-and-Process (IP) models, in
which words are built through processes, not by the addition of affixes. While an
explanation of the phenomenon is fairly straightforward in the IA model, it runs into
several problems under an IP model.

1. Item-and-Arrangement and Item-and Process morphology

Hockett (1954) coined the terms Item-and-Arrangement (IA) and Item-and-Process
(IP) to refer to two different views of the mapping between phonological form and
morphosyntactic (and semantic) information. In the IA model a word like dogs is
viewed as a set of morphemes which include a root with its corresponding exponent
or morph, dog /d/, and a plural morpheme with its corresponding morph, s /z/. In
this model, then, each piece of morphosyntactic information is paired with some
morph or exponent. This relation appears schematized in (1).

(1) root [+plural]

/d/ /z/

In the IP model a word like dogs is viewed as the result of an operation, called
by some authors a Word Formation Rule, that applies to a root paired with a set of
morphosyntactic features, and modifies its phonological form. A Word Formation
Rule for plural appears in (2a); its application to the word dogs is shown in (2b). An
important point to bear in mind is that in this model the resulting phonological
sequence, here /dz/, is a single piece, not a composite of two morphs.

(2) a. Word Formation Rule for Plural

 +N 
 +Pl 
/X/ → /Xz/

b.  +N 
 +Pl 
/d/ → /dz/

In recent times the IA view of morphology has been argued for and developed
in theories like Distributed Morphology (Halle and Marantz 1993). The IP model has
been defended recently in works like Anderson (1992) and Aronoff (1994).

1
The content of this paper is part of a talk (“In defense of a morphous morphology”)
and subsequent discussion in a meeting of the Network Core Mechanisms of
Exponence, coordinated by Jochen Trommer, held in Leipzig, January 11-12, 2008.
All possible errors of interpretation of what was said are my own.

1
 One of the main problems for the IA model is that often the mapping between
morphosyntactic information and phonological information is not a one-to-one
relationship. In some cases, for a given word there are more morphosyntactic features
than morphs. For instance, in a Spanish verbal form like miro ‘I watch’ the ending -o
(all there is leaving the root aside) could potentially be associated with features
related to person, number, tense or, mood, to mention the more obvious options.
Models like Distributed Morphology make available different types of mechanisms to
account for such mismatches: fusion of morphosyntactic features, impoverishment
(loss) of one or more morphosyntactic features, realization as Ø for certain features,
etc. The question is then which are the more adequate operations in each case, the
usual critique being that the model offers too many solutions to a given problem.2 An
even worse problem for the IA model is the existence of phenomena in which
morphological information is not associated with a specific phonological segment or a
sequence of segments but with a phonological feature, a discontinous set of segments
(as in root-and-pattern morphology), or a loss of segmental information (subtractive
morphology), to name a few of them. In none of these cases is morphological
information encoded in a phonological piece affixed to a stem. This is the type of
evidence that led researchers like Anderson to argue for an IP model of morphology,
as illustrated by the quotation in (3).

(3) “If we accept the evidence that the range of morphological possibilities in natural
languages includes some processes that cannot properly be represented as the
addition of an affix, we must conclude that a general morphological theory
should admit both affixational and non-affixational rules. Since a process-based
approach naturally accommodates affixation, but not vice versa, the alternative
we should prefer is to explore a theory of morphological processes.”
Anderson (1992: 68)

The goal of this paper is to show that, in spite of the existence of morphological
phenomena that can be better viewed as processes, it is not true for all cases that “a
process-based approach naturally accomodates affixation, but not vice versa.” The
case that will be reviewed here is the choice of gender / class exponent in masculine
nominals in Catalan. The paper is organized as follows: in section 2 the crucial facts
of Catalan are presented. In section 3 it is shown how these facts can be accounted
without any problem in an IA model of Optimality Theory. Section 4 is devoted to
showing the types of problems an IP model has to face in dealing with the same facts.
Finally, section 5 contains some concluding remarks.

2. Gender/class assignment in Catalan

In Catalan, most masculine nominals end in a consonant or a stressed vowel. In IA
terms they are said to have a Ø morph for masculine. We can call this set class 1. The
plural is formed by adding a final s, as in Spanish and other languages. Some
examples of class 1 nominals appear in (4).

(4) Catalan class 1 nominals

singular plural

2
Oltra-Massuet and Arregi (2005), for instance, claim that the ending -o is the
exponent of a terminal node that corresponds to the fusion of Tense and Agr.

2
 llit llits ‘bed(s)’
cor cors ‘heart(s)’
mussol mussols ‘owl(s)’
amic amics ‘friend(s)’

A much smaller set of masculine nominals end in o, realized as /u/ in many

dialects; let us call them class 2.
(5) Catalan class 2 nominals
singular plural
mico micos ‘monkey(s)’
toro toros ‘bull(s)’
lavabo lavabos ‘bathroom’

So far, the basic system of gender-related vowels (or class markers or theme
vowels, as are often called in the literature) is fairly similar to that of Spanish, the
main difference being that in Spanish the more common exponent is o, not Ø.3
However, in Catalan there is a set of words which belong to class 1 in the singular but
to class 2 in the plural. Some examples appear in (6).4

(6) singular (class 1) plural (class 2)

gos gossos ‘dog(s)’
cas casos ‘case(s)’
embús embussos ‘obstruction(s)’
peix peixos ‘fish(es)’
despatx despatxos ‘desk(s)’

The set of words that show the pattern illustrated in (6) is not random; all the
words share the property that they end in a sibilant. The presence of the o marker in
the plural avoids having a sequence of sibilants (*[oss], *[pes], *[mbuss], etc.,
which would be the outcome if the class 1, Ø, marker were chosen. Sequences of
sibilants are categorically prohibited in Catalan, and the language resorts to different
strategies to avoid this contact. The strategies used are fusion (between words),
epenthesis (in cliticization), and the choice of a different exponent, as in the present
case.5

3. An account in an Item-and-Arrangement model of Optimality Theory

3
Throughout this paper I will ignore the debate concerning the nature of endings in
nominals, and I will be using, for convenience, either terms related directly to gender
(like ‘masculine’) or class-related terms (like ‘class 1’). This question is tangential to
the main issue under discussion.
4
In Catalan spelling, a ss sequence represents a single voiceless alveolar fricative; the
sequence ix represents a postalveolar voiceless fricative.
5
For a description and an analysis of the different strategies used to avoid sequences
of sibilants see Bonet and Lloret (2002).

3
Within Optimality Theory (OT), sequences of identical elements are penalized by the
constraint OCP.6 And this constraint, here referring to sequences of sibilants, is
ranked very high in Catalan. Other constraints have to determine what strategies are
chosen to avoid the conflict. Some facts and aspects of the analysis I outline here for
the choice of class marker in Catalan are ignored for the sake of argumentation; for a
full analysis see Bonet, Lloret and Mascaró (2007).
As mentioned in section 2, most nominals in Catalan have a Ø exponent, while
a much smaller set have the o /u/ exponent. This preference relation can be captured
using Distributed Morphology notation as is done in (7).

(7) Vocabulary Item for masculine

[–feminine] ⇔ {Ø > u}

What (7) says is that the feature [–feminine], a terminal node, has two exponents, with
a preference relation, and both exponents constitute the input to the phonology (they
are inserted at the same time). Many cases of phonologically conditioned allomorphy
have been studied in Optimality Theory, and many of the approaches assume multiple
inputs (see, for instance, Mascaró 1996a,b, Tranel 1996, Perlmutter 1998 or Lapointe
1999). One crucial aspect of multiple inputs is that candidates with either allomorph
do not violate faithfulness constraints; therefore the choice of allomorph is left to
makedness constraints. The novelty in the present case is the preference relation
between the allomorphs, a preference relation that has been argued for in Mascaró
(2007), and Bonet, Lloret and Mascaró (2007). The faithfulness constraint that
ensures that the less preferred allomorph is penalized is PRIORITY, informally defined
in (8).

(8) PRIORITY: Respect lexical priority (ordering) of allomorphs

In cases where there is no phonological conflict, PRIORITY favors the choice of

the preferred allomorph, Ø, as illustrated in (9). The candidates (9a) and (9c) both
have the Ø allomorph, but (9c) is ruled out because it has an epenthetic vowel (it
appears underlined for clarity).

(9) gos [os] ‘dog’

/os + {Ø > u}/ OCP DEP PRIORITY
a. ☞ os
b. osu *!
c. os *!

When the nominal masculine element is plural, the exponent of the [+plural]
terminal element, /s/, also appears in the input. In this environment, OCP is violated
when the Ø allomorph is chosen and nothing is done to prevent the contact between
the sibilants, as shown in (10a). This fatal contact can be avoided if the less preferred
allomorph /u/ is chosen, (10b), or if the preferred allomorph is chosen and an
epenthetic vowel is inserted, as in (10c). The ranking DEP >> PRIORITY ensures that

6
OCP stands for Obligatory Contour Principle, the term coined by Leben (1973) to
refer to sequences of identical tones.

4
the optimal candidate is the one that violates P RIORITY, the candidate with the /u/
allomorph, (10b).

(10) gossos [osus] ‘dogs’

/os + {Ø > u} + s/ OCP DEP PRIORITY
a. oss *!
b. ☞ osus *
c. oss *!

When OCP is not violated in plural nominals, PRIORITY determines that the Ø
allomorph is chosen consistently. This is shown in (11).

(11) cors [ks] ‘hearts’

/k + {Ø > u} + s/ OCP DEP PRIORITY
a. ☞ ks
b. kus *!

In a fairly small set of nominals the marked allomorph /u/ is chosen over Ø, as
the examples in (5) illustrate. This choice has to be marked lexically somehow
because it is not predictable. This lexical marking is assumed to be a
subcategorization frame indicated by a subscript u on the root; a word like toro ‘bull’
has the root /t  u/. The constraint that tries to ensure that subcategorization
requirements are obeyed is RESPECT, informally defined in (12):

(12) RESPECT: Respect idiosyncratic lexical specifications

The tableau in (13) illustrates how words like toro are obtained. Notice that, in
spite of the lexical specification, all masculine allomorphs appear in the input.

(13) toro [tu] ‘bull’

/tu + {Ø > u}/ RESPECT PRIORITY
a. tr *!
b. ☞ tu *

As it has ben shown, then, an IA approach, an approach based on affixes,

combined with Optimality Theory can account in a fairly easy way for the distribution
of masculine-related endings.

4. A possible account within an Item-and-Process model?

As mentioned in section 1, in IP models like Anderson (1992) or Aronoff (1994)
inflected words are obtained through Word Formation Rules. In Anderson (1992) it is
explicitly stated that the morphosyntactic representation of a given word consists of
“an unordered list of its morphologically relevant features” (Anderson 1992: 90). The
example in (14) illustrates this point. At the output of syntax the Georgian adjective

5
axalebi ‘new-PL- N O M ’ that appears in (14a) consists of the root /axal/ plus an
unordered set of morphosyntactic features, as in (14b).

(14) a. (Anderson 1992, 139 (1g))

es axalebi
this new-PL-NOM
“these new ones (nom. pl.)”

b.  +N 
 +Pl 
 +Nom 
/axal/

In order to obtain the right phonological expression of this set of features, two
Word Formation Rules have to apply, one targeting the feature [+Pl] and the other one
targeting [+Nom]. These word formation rules appear in (15).

(15) Number and Case assignment in Georgian

a. (Anderson 1992, 139 (2))
 +N 
 +Pl 
/X(a)/ → /Xeb/
b. (Anderson 1992, 140 (4a))
 +N 
 +Nom 
/X/ = /Y[–Syllabic]/ → /Xi/

Crucially these two word formation rules have to apply in a fixed order to
ensure that the output /axalebi/ and not */axalieb/ is obtained. This is illustrated in
(16).

(15a) (15b)
(16) a. /axal/ /axaleb/ /axalebi/

(15b) (15a)
b. /axal/ /axali/ */axalieb/

Going back to Catalan, and following this type of approach, two Word
Formation Rules can introduce class 1 and class 2; these rules appear in (17);7 an
additional rule, (18), targets the [+plural] feature.

(17) Possible word formation rules for classes

a.  +N 
 class 1 
/X/ → /X/

7
Anderson (1992) does not address the issue of noun classes, but Aronoff (1994)
does. Here I adapt the approach in Aronoff (1994) for Spanish to the notation used in
Anderson (1992).

6
 b.  +N 
 class 2 
/X/ → /Xu/

(18) Word formation rule for Number

 +N 
 +Pl 
/X/ → /Xs/

These rules can apply with no problems to nominals that consistently belong to
class 1 or class 2, as shown in (19). First the rules referring to the realization of
classes apply, (17a) and (17b), and afterwards, through rule (18), the plural is realized.

(19) a. cors [ks] ‘hearts’ (cf. cor [k] ‘heart’):

(17a) (18)
/k/ /k/ /ks/

b. toros [tus] ‘bulls’ (cf. toro [tu] ‘bull’):

(17b) (18)
/t/ /tu/ /tus/

However, a problem arises with nominals that belong to class 1 in the singular
but to class 2 in the plural. This problem is illustrated in (19). First, rule (17a) applies
to the stem gos, since it belongs to class 1 (at least in the singular); afterwards the
word formation rule for plural adds the final s. At that point a sequence of sibilants is
created, and the question is what to do next.

(19) gossos [osus] ‘dogs’ (cf. gos [os] ‘dog’):

(17a) (18)
/os/ /os/ /oss/ ??

Since the rule realizing the class has already applied it cannot apply again; and,
if it did, it would attach the vowel at the edge of the stem, not inside it (*/ossu/), and
that would not solve the OCP problem. The solution could be left to the phonology,
since the problem is phonological. But then the question would concern the nature of
the inserted vowel, the epenthetic vowel in (Eastern) Catalan being a schwa, [], as
illustrated by the examples in (20).

(20) []stop pobl[] ‘village’

[]scannejar ‘to scan’ vendr[] ‘to sell’

While [] is a typical epenthetic vowel, a completely unmarked segment, [u] is

hardly ever chosen as an epenthetic vowel in any language, unless some copy
mechanism favors it.8 In this case, moreover, this epenthetic segment would be
restricted to masculine nominals. A purely lexical solution to the problem should also
be rejected. Saying that there is a set of items that are lexically specified as belonging

8
On vowel quality in epenthesis see, for instance, Uffman (2006).

7
to class 1 in the singular and to class 2 in the plural would be just a way of restating
the facts, plus it would miss the generalization that this difference in behavior is due
to a phonological problem; it is predictable.
The main problem for this IP model arises from the claim that rules have to
apply sequentially, while in the case at hand the assignment to a class in the plural
depends on information that is not yet available when realizing class information.
Within an IP model it is not easy, though, to give up this claim. Given that it is
assumed that inflection-related features are part of an unordered set, the only way of
getting the right surface order is to apply Word Formation Rules sequentially. This is
not a problem in an IA model as the one assumed in section 3, because in this type of
model it is the syntax (in some cases together with some morphological operations)
that determines the surface order of morphs. The basic differences between the two
models are illustrated in (21). In both models a set of features has to be ultimately
associated with a phonological sequence, here /tus/ ‘bulls’. In (21a) only the
sequential application of Word Formation Rules can give as a result /tus/ instead of
*/tsu/. In (21b), in which for concreteness the type of structure argued for in Oltra-
Massuet and Arregi (2005) has been assumed, this problem does not exist, because
there is a one-to-one mapping between the structurally and linearly organized
morphosyntactic information and the corresponding exponents.

(21) a. IP model b. IA model

 +N  √ n Th Nº
 +Pl  [class 2] [+pl]
 class 2 
/tus/ /t/ Ø /u/ /s/

Unfortunately, assuming, for the sake of argumentation, that in the IP model

Word Formation Rules can apply freely does not readily solve the OCP-related
problem.9 The Word Formation Rules in (17) can apply only if the right class (class 1
or class 2) is mentioned in the morphosyntactic feature matrix. However, this type of
information should not be present in examples like gossos ‘dogs’, because the choice
of marker both in the singular and in the plural is predictable: the Ø that appears in the
singular is predictable because this is the unmarked class marker for masculine
nominals; the -o /u/ that appears in the plural is the marker that prevents a contact of
sibilants.
Stephen Anderson (p.c.) suggests that in fact the o in words like gossos could
have an epenthetic nature, following the analysis he outlines for the extension // in
the verbal system of Surmiran, a form of Swiss Rumantsch (Anderson 2008). If this

9
One line of approach that could be pursued in the IP model, without having to
abandon the idea of unordered feature matrices, might be to say that word formation
rules apply freely and that phonological constraints determine which one is the
optimal output. For the particular example used in (21), if we compare /tus/ and
*/tsu/, we can see that */tsu/ violates FINAL-C (a prosodic word has to end in a
consonant), while this constraint is not violated by /tus/. Of course, one should
check whether this line of approach would work for other cases, in Catalan (nominal
but also verbal inflection) and in other languages with a richer inflectional system, a
mission that lies far beyond the scope of this paper.

8
were the case, this vowel would violate the faithfulness constraint D EP, a violation
that could only be acceptable if the higher ranked constraint OCP were also violated.
This is schematically shown in (22), for the singular, and in (23), for the plural.

(22) gos [os] ‘dog’

OCP DEP
a. ☞ os
b. osu *!

(23) gossos [osus] ‘dogs’

OCP DEP
a. oss *!
b. ☞ osus *

The question is now where this type of epenthetic, non-input, segment comes
from. As mentioned above it cannot be a regular epenthetic vowel; it has to be related
somehow to masculine nominals, the only ones that can surface with this vowel.
Following the parallelism with Surmiran, we could say that a rule like the one in (24),
actually a constraint, would introduce the /u/.10

(24) /X/ → /Xu(Y)  _____ 

 +Masculine 

The rule / constraint in (24) would evaluate positively candidates like *[ossu] and
[osus], but the candidate *[ossu] would be ruled out by OCP (like [oss] in (23b)).
Although this solution would work mechanically, it presents two problems:
first, it does not relate the “epenthetic” [u] with the [u] that appears in words like toro
(the two identical vowels have a totally different origin); second, (24) is a very
peculiar and totally ad hoc constraint within Optimality Theory, and it would be

10
Surmiran has a set of vowels that can appear only in stressed syllables and another
set of vowels that can appear only in unstressed syllables. Unlike other vowel
reduction patterns, here the alternation between vowels in stressed position and
vowels in unstressed position is not predictable (e.g. a stressed [a] alternates in some
cases with [], in some cases with [], in some cases with []). So, allomorphs have to
be posited for each verb, one with a vowel possible in the stressed system, and
another one with a vowel possible in the unstressed system. The extension // is
inserted only in those verbs that lack an allomorph with a stressable vowel, and it is
inserted only in those cases in which the stress would fall on the unstressable vowel.
To account for the presence of the extension Anderson (2008) proposes the following
rule, which would have the same properties as the rule suggested in (24) for Catalan:

(i) /X/ → /X/  _____ 

 +Verb 

9
desirable to have a theory as restrictive as possible, not a theory in which anything
goes.
There could be a different line of approach that might escape part of the
problems just mentioned, although other potential problems would need to be worked
out. Some authors (Hammond 2000, Russell 1995 or Yip 1998, among others) have
proposed that exponents of morphemes are introduced by morphemic constraints; they
are not present in the input. This line of research could be adapted to an IP model by
having all the Word Formation Rules in (17) as ranked constraints, interspersed with
phonological constraints; that would have the same effect as apply the Word
Formation Rules freely. We could try to capture the difference between gos ‘dog’ and
gossos ‘dogs’ as is shown in (25) and (26). Notice that the features related to class
have been ignored both in the input and in the constraints; the order between the
constraint realizing /X/ → /X/ (what has been called class 1) and the one realizing /X/
→ /Xu/ (what has been called class 2) reflects the markedness relation between the
two classes, which in section 3 was expressed through the ordering of morphs in the
input. This move voids the need for specifying class membership in predictable cases.
Under this view accounting for the singular gos is straightforward.

(25) gos [os] ‘dog’

 +N  OCP  +N  [+N] [+N]
 –Pl   +Pl  /X/ → /X/ /X/ → /Xu/
/os/ /X/ → /Xs/
a. ☞ os *
b. osu *!

In the plural, candidates like *[oss], (26b), and *[ossu], (26c), are ruled out
by OCP, which is highly ranked. Candidate (26a), the singular, would constitute a
strategy to avoid the effects of OCP, but this candidate is ruled out because it does not
show the added [s] which is required by the rule realizing the plural, /X/ → /Xs/.11
The question now is how candidate (26d) must be evaluated with respect to the
morphemic constraint realizing the plural, /X/ → /Xs/. Literally, this constraint states
that the candidate should be identical to the input with the addition of an adjacent [s],
the problem being that the [s] is not adjacent. So, strictly speaking, candidate (26d)
should be ruled out by this constraint. The same problem arises for several candidates
with respect to other morphemic constraints (indicated with a question mark in the
relevant cells). Under this interpretation, candidate (26a) would wrongly be selected
as the optimal candidate (because they would fare even with respect to the second
constraint and the decision would be left to the third constraint. Giving up adjacency
(by proposing formulations of the type /X/ → /X(Y)s/) might be a dangerous move. I
leave aside other aspects that should be addressed in relation to the type of morphemic
constraints used here.12

11
In (10) (section 3) a singular candidate was not considered as an alternative to avoid
the effects of OCP. Such candidate would be ruled out by the constraint MAX,
because a segment from the input would not appear in the output.
12
The use of morphemic constraints presents additional difficulties related to the fact
that the phonological information introduced by them cannot be controlled by
faithfulness constraints. For a discussion of this issue, see Bonet (2004).

10
(26) gossos [osus] ‘dogs’
 +N  OCP  +N  [+N] [+N]
 +Pl   +Pl  /X/ → /X/ /X/ → /Xu/
/os/ /X/ → /Xs/
a. os *! *
b. oss *! ? *
c. ossu *! ? ?
d. ? ☞ osus ? ?

Additional questions would arise with words like toro(s) ‘bull(s)’, which have a
marked masculine ending both in the singular and in the plural; it is not clear how this
information should be encoded. A more general problem, not restricted to this
particular case, is how the ordering of “morphemes” would be determined in this
approach, if constraints / rules introducing this information are not applied
sequentially.

5. Concluding remarks
In Catalan the choice of masculine marker depends not only on the stem of the word
but also on the presence or absence of a plural marker; there is “look-ahead”. The
interaction between phonological and morphological factors is fairly easy to account
for if one assumes an account based on morphemes, an IA model, in which a
preference relation between morphs is combined with an evaluation of possible
candidates through ranked constraints, as in Optimality Theory. We have considered
different explanations within an IP model, but all of them presented very serious
problems, contrary to what might have been expected a priori, namely the idea,
mentioned in (3), that anything that can be accounted for in an IA model can also be
explained in a process-based approach, an IP model.
Many of the problems that an IP model has to face are related to the assumption
that inflected words contain all inflection-related features in an unordered feature
matrix, which means that some other mechanism must ensure the right surface
ordering of phonological information. So, an obvious question would be whether this
assumption is really needed in an IP model. The answer, in my view, is that, if one
wants to keep an IP model, it would not make sense to have different terminal nodes
with specific morphosyntactic features (e.g. one with gender related features, another
one with number related features) but without the possibility of hosting corresponding
exponents; this move would take us precisely to an IA model.

References
ANDERSON, STEPHEN R. (1992): A-Morphous Morphology, Cambridge:
Cambridge University Press (Cambridge Studies in Linguistics, 62).
ANDERSON, STEPHEN R. (2008): “Phonologically conditioned allomorphy in the
morphology of Surmiran (Rumantsch), to appear in Word Structure.
ARONOFF, MARK (1994): Morphology by Itself. Stems and Inflectional Classes,
Cambridge (Mass.): The MIT Press (Linguistic Inquiry Monographs, 22).

11
BONET, EULÀLIA (2004): “Morph insertion and allomorphy in Optimality Theory”,
International Journal of English Studies 4: 73-104 (special issue: “Advances
in Optimality Theory”, edited by Paul Boersma and Juan Antonio Cutillas-
Espinosa) [ROA 734-0505].
BONET, EULÀLIA AND MARIA-ROSA LLORET (2002): “OCP Effects in Catalan
Cliticization”, Catalan Journal of Linguistics 1: 19-40.
BONET, EULÀLIA, MARIA-ROSA LLORET AND JOAN MASCARÓ (2007):
“Allomorph selection and lexical preferences: Two case studies”. Lingua 117:
903-927.
HALLE, MORRIS AND ALEC MARANTZ (1993): “Distributed Morphology and
the Pieces of Inflection”, in Kenneth Hale and Samuel Jay Keyser (eds.), The
View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger.
Cambridge (Mass.): The MIT Press, pp. 111-176.
HAMMOND, MICHAEL (2000): “There is no Lexicon!”, Coyote Papers 10: 55-77
[ROA 43-0195, 1995].
HOCKETT, CHARLES F. (1954): “Two models of grammatical description”. Word
10: 210-234.
LAPOINTE, STEVEN (1999): “Stem selection and OT”, in Geert Booij and Jaap van
Marle (eds.), Yearbook of Morphology 1999, Dordrecht: Kluwer, pp. 263-297.
LEBEN, William (1973): Suprasegmental Phonology, PhD Dissertation MIT.
MASCARÓ, JOAN (1996a): “External allomorphy and contractions in Romance”,
Probus 8: 181-205.
MASCARÓ, JOAN (1996b): “External allomorphy as emergence of the unmarked”,
in Jacques Durand and Bernard Laks (eds.), Current Trends in Phonology:
Models and Methods, Salford, Manchester: University of Salford, European
Studies Research Institute, vol. 2, pp. 473-483. Reprinted in John McCarthy
(ed.), Optimality Theory in Phonology: A Reader, Oxford: Blackwell, pp. 513-
522 (2004).
MASCARÓ, JOAN (2007): “External allomorphy and lexical representatiom”,
Linguistic Inquiry 38: 715-735.
OLTRA-MASSUET, ISABEL AND KARLOS ARREGI (2005): “Stress-by-Structure
in Spanish”, Linguistic Inquiry 36: 43-84.
PERLMUTTER, DAVID (1998): “Interfaces: Explanations of allomorphy and the
architecture of grammars”, in Steven Lapointe, Diane Brentari, Patrick M.
Farrell (eds.), Morphology and Its Relation to Phonology and Syntax,
Stanford: CSLI Publications, pp. 307-338.
RUSSELL, KEVIN (1995): “Morphemes and candidates in Optimality Theory”, ROA
44-0195.
TRANEL, BERNARD (1996): “French liaison and elision revisited: A unified
account within Optimality Theory”, in Claudia Parodi, Carlos Quicoli, Mario
Saltarelli and Maria Luisa Zubizarreta (eds.), Aspects of Romance Linguistics,
Washington, D.C.: Georgetown University Press, pp. 433-455.
UFFMAN, CHRISTIAN (2006): “Epenthetic vowel quality in loanwords: empirical
and formal issues”, Lingua 116: 1079-1111.
YIP, MOIRA (1998): “Identity Avoidance in Phonology and Morphology”, in Steven
G. Lapointe, Diane K. Brentari, and Patrick M. Farrell (eds.), Morphology and
Its Relation to Phonology and Syntax, Stanford California: CSLI Publications,
pp. 216-246.

12