Spanning

Peter Svenonius
CASTL, University of Troms

April 25, 2012

1 Introduction
I argue that the operation of Spell-out recognizes spans, dened in terms of a complement se-
quence of heads, normally in a single extended projection.
1
An extended projection (Grimshaw
2005) consists of a lexical head such as N or V and its associated functional projections (such as
D, for N, and T, for V). A functional sequence is a partial (or total) ordering of a set of syntactic
categories, for example if the nominal functional sequence is PDNumN, then PP, DP,
and NumP are possible extended projections of N.
2
In an extended projection, each element in
a functional sequence takes some lower element in the same sequence as a complement. A span
is a subpart of the complement line in an extended projection.
For example, in the following tree (for a PP containing a noun phrase DP
1
with a prenominal
possessor DP
2
), PD
1
is a span, and so is D
1
NumN.
(1)
PP
q
q
q
q
q
q
q

P DP
1
q
q
q
q
q
q
q

DP
2

D
1
q
q
q
q
q
q
q

D
1
NumP
q
q
q
q
q
q
q

Num NP
To be more precise, the exhaustive set of nontrivial spans in the main projection line in (1) is
as follows (there will be a distinct set of spans for DP
2
, nontrivial if D
2
has a complement):
(2) (i) PD
1
(ii) PD
1
Num (iv) D
1
Num
(iii) PD
1
NumN (v) D
1
NumN (vi) NumN

Thanks to Tarald Taraldsen, Marina Pantcheva, Michal Starke, Patrik Bye, Heidi Harley, and two anonymous
reviewers for discussion and comments on a previous draft.
1
The term span is sometimes used for portmanteaux (in Hocketts 1947 sense, meaning a single morpheme
which corresponds to two or more morphological slots) in the context of templatic descriptions of morphology,
for example by Melnar (2004, 17). It is also used by Williams (2003, 214) for head-complement sequences within
an extended projection, and is specically used in the context of spell-out.
2
For the purposes of this paper it does not matter whether the order is total or partial, or whether a given
extended projection can be missing elements in the functional sequence.
1
This is more transparent if the tree is represented mirror-theoretically, eliminating redundant
phrase labels and representing complementation with right-sloping lines (Brody 2000a,b).
(3)
P

D
1
q
q
q
q
q
q
q

D
2

Num

N
A head is a trivial span. I suggest that morphological exponents are always associated with
spans, trivial or nontrivial (see also Bye and Svenonius in press, where this assumption is explicit
but not specically argued for). A single morphological exponent (morpheme, for short) cannot
spell out two heads (cannot span two heads) unless those heads are in a complement relation
with each other. Thus, a single morpheme cannot spell out a head in an extended projection
together with all or part of a specier, nor can a single morpheme spell out a head in an
extended projection together with all or part of an adjunct.
3
The span is recognized indirectly in the theory of head-movement, since a head can normally
only move to the head which selects it (Travis 1984; Baker 1988). Thus, spans can be thought
of, roughly, as potential complex heads. However, the identity is not perfect; clitics can join in
head-clusters even when they are not part of spans (see Roberts 2010 for a recent proposal and
references). Furthermore, spans cannot be derived from head movement, as spans can be shown
to be relevant to Spell-out even when no head-movement takes place (I discuss this distinction
between spanning and head movement in section 4). in Brodys Mirror Theory, head-movement
is a matter of where in a span a word linearizes; Brody (2000a,b) argues for example that a
V-T span spells out parametrically in V (English) or T (French) (see also Adger et al. 2009 for
an application of this to head-nal structures).
In this paper I show how spanning accounts for the distribution of the French portmanteau
preposition-determiner forms such as du and au. First, I lay out some background assumptions
regarding the two-step procedure of Spell-out, returning to French preposition-determiners in
section 3.
2 Spell-out in two steps
Spell-out maps syntactic structures onto representations which are the input to the phonolog-
ical component (Chomsky 1993). The strict separation of phonology and syntax (Zwicky and
Pullum 1986; Pullum and Zwicky 1988) motivates a two-step approach to Spell-out, where the
rst step (L-Match) associates syntactic structures with the syntactic features in lexical entries,
and the second step (Insert) arranges the phonological content of those lexical entries into a
form which is legible to phonology. This can be modeled in a version of Distributed Morphology
3
For the purposes of this paper a span can be assumed to be conned to a single extended projection, without
embedding of extended projections. However, there are cases in which a single morpheme appears to spell out a
head such as V along with a head or heads in the extended projection of its complement (e.g. Gruber 1965; Hale
and Keyser 2002; Son and Svenonius 2008). I will not discuss those cases in this paper but will assume for the
time being that c-selection essentially turns a selected complement into part of the extended projection, at least
for the purposes of lexical insertion (see Baltin 1989; Svenonius 1994 on the relationship of c-selection and the
complement conguration).
2
(DM), and I will adopt the specic implementation detailed in Bye and Svenonius (2010, in
press), including the OT model of the phonological component.
The model assumes late insertion, meaning that the list of exponents in the lexicon is
accessed only after syntactic operations are complete in the relevant domain. The model also
assumes cyclic Spell-out, meaning that the domain relevant for lexical access is the phase, a
syntactic subconstituent of the utterance (Chomsky 2001).
The French determiner system distinguishes masculine from feminine (which I will represent
using f), denite from indenite (def), and singular from plural (pl). However, not all
feature combinations are distinguished at once. Applying the algorithm for feature assignment
developed in Adger (2006), we arrive at the following lexical entries.
(4) a. un D
[def,f,pl]
/ /
b. une D
[def,+f,pl]
/yn/
c. le D
[+def,pl]
/l@/
d. la D
[+def,+f,pl]
/la/
e. les D
[+def,+pl]
/le(z)/
f. des D
[def,+pl]
/de(z)/
These entries reect the assumption (following e.g. Tranel 1996) that liaison with les and des
(e.g. les enfants /lez Af A/ the children) reects the presence of a structurally decient /z/,
written /(z)/ in the lexical entry, which is deleted if it cannot be parsed as an onset (as in
les gar cons /legarsO/ the boys).
4
Similarly, le and the prevocalic alternate l are assumed to
correspond to a single lexical entry with a decient vowel /@/, one lacking a root node (Zoll
1993) or a mora (Tranel 1996). Before another vowel, it elides, but before a consonant, it
surfaces as schwa. No similarly regular phonological alternation could reduce feminine la to l.
Since l appears with both masculine and feminine nouns, the entry for le is underspecied for
gender, just like les and des. Its interaction with phonology is detailed below.
2.1 L-Match
The rst step in each cycle of Spell-out is called L-Match (L for lexical). In each phase, the
syntactic structure must be lexicalized, that is, linked to lexical entries. Lexical entries pair
syntactic and phonological information, but L-Match only operates on features on the syntactic
side of the pairing.
In (5), the operation L-Match is illustrated for a form like une ecole a school (f) with
squiggly lines representing lexical association to the syntactic tree, assuming the lexical entries
in (5). The additional lexical entries contain features which do not match the tree, hence remain
unassociated. The noun is assumed to have spelled out on a previous cycle (following Marantz
2007, who argues that all nouns contain a phase head n; see Marvin 2002; Newell 2008 on cyclic
spell-out of words).
4
The phonological analysis works the same whether or not les and des are bimorphemic and /(z)/ is a plural
morpheme, a detail which I will set aside here.
3
(5)
D
[def,+f,pl]

un D
[def,f,pl]
NP

une D
[def,+f,pl]
ecole
le D
[+def,pl]

la D
[+def,+f,pl]

Still following Adger (2006), nothing guarantees that there is a one-to-one function from syn-
tactic trees to strings of exponents. L-Match can associate more than one exponent with a
node in a tree. For example, as noted above, the denite article le is not marked for gender. In
a denite singular noun phrase with a feminine noun, the set of associations illustrated in (6)
arises.
(6)
D
[+def,+f,pl]

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

un D
[def,f,pl]
NP

une D
[def,+f,pl]
ecole
le D
[+def,pl]

la D
[+def,+f,pl]

At this point I depart from a standard assumption in DM. In DM, conicts like the one
in (6) are normally assumed to be resolved by reference to properties of the syntactic features
involved. In most such cases, the Elsewhere Principle systematically chooses the more fully
specied alternative, blocking the other option from occurring (Halle and Marantz 1993). This
kind of blocking can be straightforwardly computed only when the features on one lexical entry
are a subset of the features on another. In the case in (6), the Elsewhere Principle would
force the choice of la, and the association link to le would not be retained. If this model is to
be reconciled with the assumption that the Elsewhere Principle operates in L-Match, then I
would have to recongure the lexical entries here so that neither was a subset of the other, for
example by positing an additional feature on le. However, I will instead assume that L-Match
tolerates indeterminacy, and that the Elsewhere Principle is not invoked at this level (essentially
following Adger 2006).
5
5
Adger (2006) does not discuss phonologically conditioned allomorphy, but uses underspecication of this
kind for sociolinguistic variability. What I am suggesting here is that as far as the syntax is concerned, the
alternation between la and l with feminine nouns can be treated the same way, letting phonology rather than
sociolinguistic factors sort out the distribution.
4
2.2 Insert
Once the exponents have been matched systematically in this way, Spell-out proceeds to a
second step, called Insert in Bye and Svenonius (2010, in press), in which the syntactic features
are no longer available and only the phonological qualities of the matched exponents are relevant.
At this point, even the syntactic nodes themselves become irrelevant, as do their dependent
syntactic features, as they are mapped onto phonological constituents (here, for phonological
word and for phonological phrase, cf. e.g. Selkirk (2011); the details of whether is recursive
or n-ary branching and so on are irrelevant for present purposes).
(7)

.
.
.
.
.
.
.
.

une
/yn/
ecole
/ekOl/
A linearization algorithm determines the order in which the phonological units are concatenated,
on the basis of the nature of the syntactic dependencies (in the diagrams, squiggly lines are
lexical dependencies, and straight lines are structural dependency relations right-sloping
for complements, and left-sloping for speciers). For unincorporated complements, dominance
maps to precedence: if lexicalizes A (i.e. is a lexical dependent of A), then precedes any
material lexicalizing any complement of A. Speciers do not gure into the examples in this
paper, but I assume that the linearization algorithm has the eect that speciers precede heads
(as in Kayne 1994). I also follow Brody (2000a,b) in assuming that morphologically incorporated
complements precede heads, as a principle of linearization. The temporal precedence relation is
represented in the tree in (7) by left-right order of the phonological elements. This is the input
to the phonological component (which may realign prosodic constituency, for example making
the /n/ of /yn/ into the onset of the following syllable, hence realigning the phonological word
boundary).
For L-Match, all features manipulated by syntax are relevant (including gender, which is
copied by Agree), but L-Match is blind to phonological features. L-Match cannot resolve cases
of contextual allomorphy, where the phonological environment determines which allomorph is
selected. A well-known example is the English indenite article, where a /@/ is used before
consonants and an /@n/ before vowels. In such cases, the output of L-Match can be assumed
to include two options, resolved later in the phonology. Thus, the two options {a, an} must
be indistinguishable to the syntax, and so the single lexical entry for the indenite article can
simply include both options.
(8) a, an D
[def]
{/@/, /@n/}
Either the marked form has a phonological condition associated with it (Hayes 1990), or else
both forms are entered into Gen and evaluated together by the phrasal phonology (Mascaro
1996).
Recall from the above that the alternation in masculine nouns between le and l (as in
le gar con the boy, lenfant the child) is not a case of contextual allomorphy, but of pure
phonology; the defective segment /@/ is deleted unless required to support syllable structure.
But the choice between la and l with feminine nouns (as in la profession the profession,
lusine the factory) is a case of contextual allomorphy just like the choice between a and an
5
in English.
The L-Match tree from (6) is repeated here in (9). It is converted in Insert to the structure
on the right.
(9)
D
[def,f]

.
.
.
.
.
.
.
.
le D
[def]
NP

la D
[def,f]
ecole

/l@/

/la/ /ekOl/
Lexicalizations of are linearized before material under , as expected, but the multiple ex-
ponents associated with are not linearized with respect to each other. This provides Insert
with two options for lexicalizing the node, la and le; Gen applies to both of these options
to generate multiple candidates for pronunciation. Candidates in which le appears without its
defective vowel violate a low-ranked constraint against deleting defective segments, which I call
Parse-V here, following van Oostendorp (2000) (deleting the /a/ in la, on the other hand,
would additionally violate a higher-ranked Parse constraint against deleting fully specied
segments, following Tranel (1996), not shown here).
(10)
/{l@, la} ekOl/ Ons Parse-V *@
a. le.kOl *
b. [email protected] *! *
c. la.e.kOl *!
I have also included a relatively low-ranked constraint against realizing the defective vowel as
a schwa (Tranel 1996 posits a constraint AIF avoid integrating oaters, and van Oostendorp
2000 uses *[cons] to punish vocalic roots; I simply call whatever constraint is responsible
*@, and place it lower than Parse-V on the basis of forms like prenez-le /pr@nel@/, take it).
This set of rankings causes la to be preferred over le (the Onset violations in (11) are for the
syllables consisting only of /l/).
(11)
/{l@, la} tabl/ Ons Parse-V *@
a. l.tab.l **! *
b. [email protected] * *!
c. la.tab.l *
There are several cases like this in French where adnominal elements have two contextual
allomorphs, one prevocalic and one preconsonantal. Tranel (1996) and Mascaro (1996) discuss
what the latter calls belle-allomorphy, based on alternations like the following.
(12) Masculine nouns Feminine nouns
C-initial V-initial C-initial V-initial
beautiful beau cure bel abbe belle table belle arme
parson abbot table weapon
6
A closed class of adjectives in French show this pattern (noveaunouvelle new, vieuxvieille
old, foufolle crazy, etc.). As Mascaro notes, bel /bEl/ cannot be derived from the same base
as beau /bo/ by any rules motivated in French phonology, so the alternatives must be lexically
listed suppletive forms. However, feminine belle and masculine prevocalic bel are phonologically
identical, /bEl/, and so can be treated as orthographic variants (which I will represent bel(le)).
The learner, then, is faced with an alternation between /bo/ and /bEl/ with masculine nouns, in
syntactically identical contexts. Thus, according to the algorithm adopted from Adger (2006),
the form /bEl/ is underspecied for gender, while beau /bo/ is specied as [f].
This ensures that beau is prevented in L-Match from appearing with feminine nouns, and
so only bel(le) can surface there. But with masculine nouns, both alternants are associated
by L-Match, just as happened with feminine nouns and the denite article, with the version
being chosen which gives the best syllable structure. This is formally represented by Mascaro
as follows for bel ami beautiful friend (m) and beau mari beautiful husband (m), omitting
several constraints which do not come into play here.
(13)
/{bEl, bo} ami/ Ons No-Coda
a. bE.la.mi
b. bo.a.mi *!
(14)
/{bEl, bo} mari/ Ons No-Coda
a. bEl.ma.ri *!
b. bo.ma.ri
Zwicky (1987) analyzes this same data set, without using underspecication or a two-step lexical
insertion procedure. Instead, he posits a Rule of referral which stipulates that the feminine
form is substituted for the masculine in prevocalic cases (his (53), p. 223, reads, [minor,
+def, masc, sg] in morphosyntactic structure is referred to the corresponding [fem] when a
V-initial word follows, where minor is a diacritic to mark the special function words to which
the rule applies). The rule invokes both syntactic and phonological information in a single
statement, something which I am proposing to do without. Furthermore, it cannot be extended
to the denite article, or to mon my which also shows a masculine form being substituted
for the feminine (e.g. mon arme my weapon (f), in Tranels example).
(15) Masculine nouns Feminine nouns
C-initial V-initial C-initial V-initial
my mon cure mon abbe ma table mon arme
parson abbot table weapon
Tranel (1996) proposes an OT analysis in which agreement in gender can be ranked among
phonological constraints, which again violates the separation of phonology and syntax. The
account of Perlmutter (1998) similarly mixes phonological and morphological constraints. Fur-
thermore, these proposals overgenerate unless the constraints are arbitrarily associated with
individual lexical items, as the substitution only occurs for a xed set of adnominal elements.
Mascaro (1996) points out that it does not apply, for example to favori favorite m and favorite
favorite f, e.g. it is favori ami favorite friend (m), not the phonologically superior *favorite
7
ami.
6
The account proposed here does entirely without indexed constraints (following Bye and
Svenonius 2010, in press, and references there).
The analysis proposed here holds that a learner posits underspecied entries for forms like
le, les, mon, bel(le), and so on, because they are observed to appear with both genders of noun.
Forms like la, ma, beau, and so on are specied for gender because they consistently appear
with nouns of a particular gender. In cases where L-Match nds two suitable exponents, Insert
may distinguish them on the basis of phonology (otherwise, they will presumably appear as
variants, as in the cases discussed in Adger 2006).
This analysis is therefore not consistent with common practice in DM, where underspec-
ication is used extensively and the Elsewhere Principle is assumed to adjudicate at what
corresponds to the L-Match level here (Halle and Marantz 1993). I am thus assuming a more
restricted role for underspecication and the Elsewhere Principle (it may apply to certain kinds
of features, for example categorial features, or interpretable features, if the -features on D
can be assumed to be uninterpretable manifestations of features interpretable elsewhere; space
prevents me from pursuing this matter further here).
3 Preposition-determiner fusion
Having established how the two-step model of Spell-out works, I turn to portmanteaux.
It is relatively common for prepositions and determiners to interact morphologically (cf. e.g.
Napoli and Nevis 1987 on Italian, Waldm uller 2008 on German). In a cyclic model of spell-out,
the interactions suggest that they are in the same phase and are located in the same spell-out
domain.
7
A well-known example is the French fusion of the masculine denite article with the
prepositions de and `a. This happens only if the determiner is not contracted with a vowel-initial
noun phrase, as seen in the table below (from Zwicky 1987, 212).
(16) Feminine nouns Masculine nouns
V-initial C-initial V-initial C-initial
lecole la maison lh opital le parc
/lekOl/ /lamEzO/ /lOpital/ /l@park/
`a `a lecole `a la maison `a lh opital au parc
/alekOl/ /alamEzO/ /alOpital/ /opark/
de de lecole de la maison de lh opital du parc
/d@lekOl/ /d@lamEzO/ /d@lOpital/ /dypark/
The pattern shows that the phonology of the noun phrase is relevant to the allomorphy of
the preposition: if the noun phrase is vowel-initial, then the determiner is procliticized and
the preposition is regular; if the noun phrase is not vowel-initial, then in the right context of
morphological features (here, masculine singular; there is a parallel interaction in the plural,
6
Mascaro (1996) proposes to treat belle-allomorphy in French much like the English {a, an}, simply listing
{beau, bel } as alternate allomorphs of the masculine adjective, and leaving the phonological identity of belle
and bel (and other such cases) to factors outside the grammar proper (e.g. historical factors). Adopting such a
proposal would leave unaected the main point of this squib concerning spans. See also Mascaro (2007); Bonet
et al. (2007) for a development in which allomorphs are partially ordered by a preference relation, and adherence
to this preference can be interranked with phonological constraints.
7
Law (1998) suggests languages with fused P-D elements never allow P-stranding. If that is correct, then it
suggests a syntactic prerequisite for P-stranding which is incompatible with P-D fusion. In my account, such a
prerequisite could be DP having phase status independent of PP (e.g. in English, which allows P-stranding, but
not in German, which has fused P-D elements). However, Laws generalization is based on a very small set of
European languages, and so the correlation could be accidental.
8
not illustrated here), a special portmanteau allomorph is chosen. Thus, allomorph selection
for the adposition must happen after the phonological content of the noun phrase has become
accessible.
The two prepositions involved can be given the following entries, letting [rel] and [loc]
stand in for whatever features distinguish the two.
(17) a. de P
[rel]
/d@/
b. `a P
[loc]
/a/
For a form like `a lecole, L-Match associates the entry in (17b) to a tree like that in (18).
(18) L-Match
P
[loc]

`a P
[loc]
D
[+def,+f,pl]

de P
[rel]
NP

la D
[+def,+f,pl]
ecole
le D
[+def,pl]

les D
[+def,+pl]

...
Given the lexical items listed here, the only match for P is `a, but D has two matches,
the more specied la and the less specied le. The plural form les is not a match, on these
assumptions, since it bears a plural feature. Nor is the preposition de, because it also has a
feature, [rel], which is not matched in the syntactic tree.
Once the exponents have been matched syntactically in this way, spell-out proceeds to
Insert, in which the phonological qualities of the matched exponents are relevant, but not the
syntactic features.
(19) Insert

`a
/a/

.
.
.
.
.
.
.
.

la
/la/

le
/l@/
ecole
/ekOl/
As before, the linearization algorithm, which puts unincorporated higher heads before lower
structure, linearizes /a/ to the left of both /la/ and /l@/, and all of those to the left of /ekOl/,
9
but does not linearize /la/ with respect to /l@/. They depend on a single node, and only one
exponent is needed to lexicalize that node, so they compete. Thus, we now compare /alaekOl/
to /al@ekOl/ (along with the prosodic structure represented in (19)). This is easier to see if we
redraw (19) as a set of trees in which each node is represented only by one exponent.
(20) Insert

`a
/a/
la
/la/
ecole
/ekOl/

`a
/a/
le
/l@/
ecole
/ekOl/
Alternatively, we can represent the same information in the form of a tableau. Here, for
simplicity of formatting, I omit the prosodic structure represented in the trees from the input
line of the tableau.
(21)
/a {l@, la} ekOl/ Dep Ons Parse-V *@
a. a.le.kOl * *
b. ta.le.kOl *! *
c. [email protected] **! *
d. a.la.e.kOl **!
Essentially the same factors adjudicate here as in the form for lecole seen in (10). I have
added a Dep constraint against epenthesis in order to show a candidate, ((21)b), which does
not violate Onset.
In addition to the regular Ds and Ps, there are the portmanteau forms, which associate
to the span of PD. This is specied in their lexical entries (compare the proposals of Zwicky
1987 and of Mascaro 1996).
(22) a. au P
[loc]
,D
[+def,f,pl]
/o/
b. du P
[rel]
,D
[+def,f,pl]
/dy/
First, note that the new entry au does not falsely predict *au ecole: because au is specied as
masculine, it will never be associated to the tree with a feminine feature on D.
Now consider the spell-out of du parc (setting aside the pl features, for simplicity). Here,
because the entry for du species both P and D features (it is a portmanteau), L-Match asso-
ciates it to two nodes. At the same time, de is also a match, and so is le.
(23) L-Match
10
P
[rel]

`a P
[loc]
D
[+def,f]

de P
[rel]
NP

du P
[rel]
,D
[+def,f]
la
D
[+def,+f]

le
D
[+def]

parc
Moving to Insert, we discard the unassociated morphemes and consider only the phonological
information in the remaining lexical entries. Although de can be linearized before le, and le
before parc, and du before parc, there is no relative linearization of de le and du. They lexicalize
the same nodes, as shown in the left-hand tree below, and thus they compete.
(24) Insert

,
,
,
,
,
,

du
/dy/

de
/d@/
le
/l@/
parc
/park/

.
.
.
.
.
.
.
.

du
/dy/
parc
/park/
The correct result is of course du parc, as indicated in the right-hand tree above. The question
is why this is preferred over *de le parc. There are many cases in which a portmanteau morph
is observed to block an apparently equally fully specied sequence of two separate morphemes,
and it has been proposed that there is a general principle favoring fewer exponents over more
exponents, all else being equal (e.g. Muriungi 2009; Caha 2009; Siddiqi 2009; Taraldsen 2010).
However, in the case at hand it can be seen (in (16)) that the principle is overridden by
concerns of syllabic well-formedness, since vowel-initial masculine nouns like h opital take the
bimorphemic sequences de l, `a l rather than du, au.
8
(25) Insert
8
As usual, the Insert trees represent intermediate stages in the phonological derivation. Syllable well-
formedness constraints will ensure that the denite article l is contained in the same phonological word as the
vowel-initial noun hopital in the nal output of phonology, although in the righthand tree here it is represented
as being outside the word.
11

,
,
,
,
,
,

du
/dy/

de
/d@/
le
/l@/
h opital
/Opital/

de
/d@/
l
/l@/
h opital
/Opital/
Given the strict separation of syntax and phonology adopted here, this means that somehow, the
phonological component must have a way of preferring fewer exponents over multiple exponents,
if onset requirements are satised. I will assume that this is because phonology must concatenate
sequences of exponents, and each instance of concatenation creates prosodic structure. In other
words, the two nodes in (24) and (25) are distinct when associated to distinct exponents, but
they are one and the same when only one exponent (du) is associated. A general constraint
*Struct punishes candidates with more structure compared to those with less.
In the tableaux below, this is represented by assigning one violation mark under *Struct
for each instance of concatenation. The constraint *Struct is crucially ranked below Onset;
I place it above Parse-Vand *@ here for concreteness.
(26)
/{d@ l@, dy} park/ Ons No-Coda *Struct Parse-V *@
a. dy.park ** *
b. [email protected]@.park ** **! **
c. [email protected] ***! ** * *
(27)
/{d@ l@, dy} Opital/ Ons No-Coda *Struct Parse-V *@
a. dy.O.pi.tal *! * **
b. [email protected]@.O.pi.tal *! * *** **
c. [email protected] *** *** * *
4 Comparison with an alternative
Embick (2007, 2010) develops an alternative account within a more standard version of Dis-
tributed Morphology. In that framework, spanning is not possible, and lexical insertion is
restricted to terminal nodes. However, certain operations may apply to fuse adjacent terminal
nodes together, creating new, composite terminal nodes. One such operation is local dislocation,
and Embick states two rules of local dislocation, specic to French, in (28).
(28) Embick (2010, 88)
a. Article Cliticization
D[def]

X [D[def][X]], X V-initial
b. P-D Axation
P
+
D[def]
+
[P
+
[D
+
]]
where
+
is a diacritic for the particular terminals that are subject to this process
12
The rst rule, Article Cliticization, states that if a noun phrase (X) is vowel-initial, the denite
article (D[def]) will cliticize to it, bleeding the following rule. The second rule, P-D Axation,
states that certain specially-marked denite articles (namely, the masculine ones) form a con-
stituent with an adjacent specially-marked P (namely, the ones that would otherwise spell out
as `a or de).
The rst thing to note about these rules is that they are much more powerful than spanning,
since they operate under adjacency; any two heads could be combined in this way, after syntactic
movement, whereas spanning is restricted to spans (i.e. subsegments of a functional sequence).
Therefore, the burden of proof should be on this version of DM to show that such powerful
mechanisms are empirically necessary.
The second thing to note about these rules is that they mix syntactic and phonological
information. If the formation of a terminal node is syntactic, then the reference to vowel-
initial elements in (28a) is a violation of the well-motivated Principle of Phonology-Free Syntax
(Zwicky and Pullum 1986; Pullum and Zwicky 1988), which states that syntactic rules do
not make reference to phonology. If the formation of a constituent in (28a) is meant to be
phonological, then the rule violates the converse of the same principle, formalized for example
in Inkelas (1989) Indirect Reference principle, which states that phonological rules can only
refer to syntactic structure via prosodic constituency; (28a) refers to the syntactic features D
and [def]. If there is a module of morphology which mixes syntax and phonology in the way
that (28) does, then the empirical observations which motivate the Principle of Phonology-Free
Syntax and the Indirect Reference principle are left unexplained.
I claim rather that the simple cliticization of the denite article is due to phonological
properties of the exponent le, which has a defective vowel. Once the phonological properties
of the element lexicalizing D have been accessed, a syntactic constituent cannot be formed
out of P+D without opening up the possibility of phonology inuencing syntax; therefore, the
portmanteau status of P-D cannot be a syntactic fact, but is a fact about spell-out and the
exponents it associates with syntactic structures, one which is handled here through spanning.
The problem for standard implementations of DM such as Embicks is essentially that if
lexical insertion is to be applied to terminal nodes, then before lexical insertion occurs, there
must be some independent organization of the syntax into terminal nodes. Here, items which
are only visible through lexical insertion itself (namely the portmanteaux au and du) are the
only clue that P and D must be reorganized into terminal nodes. Hence the diacritics
+
in
Embicks rules, which have no other motivation. All theories have to list irregular portmanteaux
like au and du; on the spanning theory, that is all that need be done, no additional features
need be posited.
5 Extending the account to en
In this section I briey discuss a pattern concerning a class of proper place names which provides
a few clues about the nature of L-Match and Insert and the lexical entries that it deals with.
With certain proper place names (including many names of countries), the preposition en is
ordinarily used, in lieu of `a (examples from Zwicky 1987).
9
9
The exact distribution of this pattern is complex, with some place names showing dierent patterns. I will
assume for the time being that the dierences can be attributed to dierences in the featural make-up of D, and
restrict the analysis to the class discussed by Zwicky.
13
(29) Feminine place names Masculine place names
V-initial C-initial V-initial C-initial
lAmerique la France lIran le Canada
/lamerik/ /lafr A:s/ /lir A/ /l@kanada/
loc en Amerique en France en Iran au Canada
/ Anamerik/ / Afr A:s/ / Anir A/ /okanada/
The exception, as can be seen in the table, is masculine nouns beginning with consonants. This
pattern can be straightforwardly described in the spanning model on the assumption that en is
specied like au in being a PD portmanteau, and in being specied as [loc], but diering in
the features present on the determiner. First, en is not restricted to masculines, as au is, and
second, en is used in certain marked cases with defective determiners. I will model this by
specifying en with a distinct feature on D, [prpr] (for proper). The two entries are presented
side-by-side for comparison.
(30) a. P
[loc]
,D
[+def,f]
au /o/ (repeated from (22a))
b. P
[loc]
,D
[prpr]
en / A(n)/
The [prpr] feature is intended to ensure that en is never a contender in the cases already
discussed, since L-Match will not form associations with a lexical entry that is specied for
features not appearing in the syntactic tree. By the same convention, au will not be considered
for nonmasculine nouns, since it has a feature [f]. Phonological competition in Insert will
therefore only occur in the masculine, and only with the special D such as that occurring in
this class of proper place names.
10
In that narrow context, the empirical fact is that en beats both au and ` a le before vowels
(en Iran), and au beats both en and `a le before consonants (au Canada). This is consistent
with the phonological assumptions already made here. First, consider the situation before
vowels. The forms */oir A/ and */al@ir A/ lose out because of Onset; the form */alir A/ requires
the moraless /@/ to be deleted in order to avoid hiatus, violating Parse-V, and also requires
an extra application of *Struct compared to the form / Anir A/. The latter requires a rootless
consonant /(n)/ to receive a root node, violating a Dep-root constraint, which must be ranked
lower than one or the other or both of *Struct and some Parse constraint.
Turning to the example with a consonant (au Canada), it violates Onset once, but so do
its closest competitors, and it introduces no other phonological complications, whereas both
of the other forms require either deleting a (possibly rootless) consonant to avoid a coda (*en
Canada) or inserting a root or mora (*`a le Canada). Furthermore, the monomorphemic status
of au gives it an additional edge over `a le, because of there being one less *Struct violation.
6 Conclusion
I conclude that spans are recognized by the grammar: Lexical insertion targets spans, including
trivial spans, of functional sequences (possibly including complement sequences derived through
c-selection, not discussed here). This promises to shed new light on head movement, extended
projections, and the status of such objects as morphemes, words, and lexical items. The signif-
icance of spans for grammar also suggests that the complement relation must be distinguished
from the specier relation; thus complements, at least within extended projections, are not
10
Proper place names will also have to have the [+def] feature or else au would not be considered. The
featural specications used here are an oversimplication. For example, even the proper place names in (29) can
appear with `a in certain contexts (Zwicky mentions penser `a, think about, and toasts). These involve additional
features which can be expressed by la/le but not en, which only spells out certain featurally restricted D.
14
simply the rst-merged constituents, but have a distinct status. This has implications for the
theories of selection and projection.
References
Adger, David. 2006. Combinatorial variability. Journal of Linguistics 42 3: 503530.
Adger, David, Daniel Harbour, and Laurel J. Watkins. 2009. Mirrors and Microparameters:
Phrase Structure Beyond Free Word Order. Cambridge University Press, Cambridge.
Baker, Mark C. 1988. Incorporation: A Theory of Grammatical Function Changing. University
of Chicago Press, Chicago.
Baltin, Mark R. 1989. Heads and projections. In Alternative Conceptions of Phrase Structure,
edited by Mark R. Baltin and Anthony S. Kroch, pp. 116. University of Chicago Press,
Chicago.
Bonet, Eul`alia, Maria-Rosa Lloret, and Joan Mascar`o. 2007. Allomorph selection and lexical
preferences: Two case studies. Lingua 117: 903927.
Brody, Michael. 2000a. Mirror Theory: Syntactic representation in Perfect Syntax. Linguistic
Inquiry 31 1: 2956.
Brody, Michael. 2000b. Word order, restructuring, and Mirror Theory. In The Derivation of
VO and OV, edited by Peter Svenonius, pp. 2743. John Benjamins, Amsterdam.
Bye, Patrik and Peter Svenonius. 2010. Exponence, phonology, and non-concatenative mor-
phology. Ms. CASTL, University of Troms; available at ling.auf.net/lingbuzz/001099.
Bye, Patrik and Peter Svenonius. in press. Non-concatenative morphology as epiphenomenon.
In The Morphology and Phonology of Exponence: The State of the Art, edited by Jochen
Trommer, pp. 427495. Oxford University Press, Oxford.
Caha, Pavel. 2009. The Nanosyntax of Case. Ph.D. thesis, University of Troms.
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The View from Building
20: Essays in Linguistics in Honor of Sylvain Bromberger, edited by Kenneth Hale and
Samuel Jay Keyser, pp. 152. MIT Press, Cambridge, Ma.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in Language, edited by
Michael Kenstowicz, pp. 152. MIT Press, Cambridge, Ma.
Embick, David. 2007. Linearization and local dislocation: Derivational mechanics and interac-
tions. Linguistic Analysis 33 3-4: 235.
Embick, David. 2010. Localism versus Globalism in Morphology and Phonology. MIT Press,
Cambridge, Ma.
Grimshaw, Jane. 2005. Extended projection. In Words and Structure, edited by Jane Grimshaw.
CSLI, Stanford, Ca. Final version of 1991 ms.
Gruber, Jerey. 1965. Studies in Lexical Relations. Ph.D. thesis, Massachusetts Institute of
Technology, Department of Modern Languages.
Hale, Ken and Samuel Jay Keyser. 2002. Prolegomenon to a Theory of Argument Structure.
No. 39 in Linguistic Inquiry Monographs. MIT Press, Cambridge, Ma.
Halle, Morris and Alec Marantz. 1993. Distributed Morphology and the pieces of inection. In
The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, edited
by Kenneth Hale and Samuel Jay Keyser, pp. 111176. MIT Press, Cambridge, Ma.
Hayes, Bruce. 1990. Precompiled phrasal phonology. In The Phonology-Syntax Connection,
edited by Sharon Inkelas and Draga Zec, pp. 85108. University of Chicago Press, Chicago.
Hockett, Charles F. 1947. Problems of morphemic analysis. Language 23 4: 321343.
15
Inkelas, Sharon. 1989. Prosodic Constituency in the Lexicon. Ph.D. thesis, Stanford, Palo Alto,
Ca.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, Ma.
Law, Paul. 1998. A unied analysis of P-stranding in Romance and Germanic. In Proceedings of
the North East Linguistic Society, edited by Pius N. Tamanji and Kiyomi Kusumoto, vol. 28,
pp. 219234. GLSA, Amherst, Ma.
Marantz, Alec. 2007. Phases and words. In Phases in the Theory of Grammar, edited by
Sook-hee Choe, pp. 196226. Dong-in, Seoul.
Marvin, Tatjana. 2002. Topics in the Stress and Syntax of Words. Ph.D. thesis, MIT, Cam-
bridge, Ma.
Mascaro, Joan. 1996. External allomorphy and contractions in Romance. Probus 8 2: 181206.
Mascaro, Joan. 2007. External allomorphy and lexical representation. Linguistic Inquiry 38 4:
715735.
Melnar, Lynette R. 2004. Caddo Verb Morphology. University of Nebraska Press, Lincoln, Ne.
Muriungi, Peter Kinyua. 2009. The union spell-out mechanism. In Troms Working Papers on
Language and Linguistics: Nordlyd 36.1, Special issue on Nanosyntax, edited by Peter Sveno-
nius, Gillian Ramchand, Michal Starke, and Knut Tarald Taraldsen, pp. 191205. University
of Troms, Troms. Available at http://www.ub.uit.no/baser/nordlyd/.
Napoli, Donna Jo and Joel Nevis. 1987. Inected prepositions in Italian. Phonology 4 1:
195209.
Newell, Heather. 2008. Aspects of the Morphology and Phonology of Phases. Ph.D. thesis,
McGill University, Montreal.
van Oostendorp, Marc. 2000. Phonological Projection: A Theory of Feature Content and
Prosodic Structure. Mouton de Gruyter, Berlin.
Perlmutter, David M. 1998. Interfaces: Explanation of allomorphy and the architecture of
grammars. In Morphology and its Relation to Phonology and Syntax, edited by Steven G.
Lapointe, Diane K. Brentari, and Patrick M. Farrell, pp. 307338. CSLI, Stanford, Ca.
Pullum, Georey K. and Arnold Zwicky. 1988. The syntax-phonology interface. In Linguistics:
The Cambridge Survey, edited by Frederick J. Newmeyer, vol. 1, pp. 255280. Cambridge
University Press, Cambridge.
Roberts, Ian. 2010. Agreement and Head Movement: Clitics, Incorporation, and Defective
Goals. MIT Press, Cambridge, Ma.
Selkirk, Elisabeth. 2011. The syntax-phonology interface. In The Handbook of Phonological
Theory, edited by John Goldsmith, Jason Riggle, and Alan Yu. Blackwell, Oxford, 2nd edn.
Siddiqi, Daniel. 2009. Syntax within the Word: Economy, Allomorphy, and Argument Selection
in Distributed Morphology. John Benjamins.
Son, Minjeong and Peter Svenonius. 2008. Microparameters of cross-linguistic variation: Di-
rected motion and resultatives. In Proceedings of the 27th West Coast Conference on Formal
Linguistics, edited by Natasha Abner and Jason Bishop, pp. 388396. Cascadilla, Somerville,
Ma.
Svenonius, Peter. 1994. C-selection as feature-checking. Studia Linguistica 48 2: 133155.
Taraldsen, Knut Tarald. 2010. The nanosyntax of Nguni noun class prexes and concords.
Lingua 120: 15221548.
Tranel, Bernard. 1996. French liaison and elision revisited: A unied account within Optimality
Theory. In Aspects of Romance Linguistics, edited by C. Parodi, C. Quicoli, M. Saltarelli,
and M. L. Zubizarreta, pp. 433455. Georgetown University Press, Washington, D.C.
16
Travis, Lisa. 1984. Parameters and Eects of Word Order Variation. Ph.D. thesis, Mas-
sachusetts Institute of Technology, Cambridge, Ma.
Waldm uller, Estela. 2008. Contracted Preposition-Determiner Forms in German: Semantics
and Pragmatics. Ph.D. thesis, Universitat Pompeu Fabra, Barcelona.
Williams, Edwin. 2003. Representation Theory. MIT Press, Cambridge, Ma.
Zoll, Cheryl. 1993. Ghost segments and optimality. In The Proceedings of the West Coast
Conference on Formal Linguistics, edited by Erin Duncan, Donka Farkas, and Philip Spaelti,
vol. 12, pp. 183199. Cascadilla Press, Somerville, Ma.
Zwicky, Arnold. 1987. French prepositions: No peeking. Phonology 4: 211227.
Zwicky, Arnold M. and Georey K. Pullum. 1986. The Principle of Phonology-Free Syntax:
Introductory remarks. OSU Working Papers 32: 6391.
17