1
DM, Spell-out, Phonological Representations, and the
PF-Interface
Masaryk University
March 4-April 15
Heather Newell
newell.heather@uqam.ca
§1 Introduction to the issues, Introduction to DM (March 4)
In this lecture we will look briefly at cyclic phonological effects
and discuss what the goals of a theory of the morphosyntaxphonology
interface should be, and how they might best be
achieved. We will then focus on an overview of Distributed
Morphology and its core principles, as a lead-up to a discussion of
cyclicity in the second lecture. The main source of information for
this overview will be Embick (2015)’s The Morpheme: A
theoretical introduction.
§1.1 Cyclic Phonology and the Structure of the Grammar
• The focus of this lecture series is on cyclic phonological
computation, and on how different proposed solution to this
computation impact our theories of grammar generally. By
grammar here I mean the morphosyntax, the phonology, and
the semantics, although we will mainly ignore the latter save
for some discussion of compositional vs. idiomatic meaning.
The main point of all of this is that our theories of
morphosyntax have an impact on our theories of phonology
and vice versa.
• So, what kind of data are we looking at?
(1) a. párent
b. paréntal
c. párenting
d. párenthood
• The question triggered by data like those in (1) is what
influences the shift in stress in (1b), but blocks a shift in stress
in (1c,d)?
There are three possible options for an explanation here.
(A) The distinction is lexical. The position of stress in (1a-d) is
memorized. This implies that the phonology of structurally
complex objects is stored, or that the examples here are not
morphologically complex.
(B) The distinction is morphosyntactic. There is a crucial
difference in morphosyntactic structure between (1b) and (1c,d)
that explains the different stress patterns.
(C) The distinction is phonological. There is a phonological
property of the affix in (1b) that distinguishes it in a relevant way
from the affixes in (1c,d) and explains the different positions in
stress.
Another possibility is that it is a combination of (B) and (C) that
leads to the distinction ((A) could also combine with the other
proposals if some forms are lexicalized and others are not).
• The analysis of the above will assume Full Decomposition. Full
Decomposition is defined as follows in Embick (2015:17):
2
(2) Full Decomposition: No complex objects are stored in memory;
i.e., every complex object must be derived by the grammar.
"The particular sense of complex that is at issue in (9) is
important. In principle, complex could mean either “consisting of
more than one primitive”, or it could mean something like “a single
primitive with some internal structure”. It is the former sense that
is crucial here: primitives—which in this theory are morphemes in
memory—may be internally complex, in that they may be
composed of multiple features. What is ruled out by (9) is storage
in memory of a representation that consists of more than one
morpheme."
• Now, in this particular room full of Nanosyntacticians, may
disagree that Full Decomposition should be restricted in this
way, allowing for feature bundles to be single morphosyntactic
pieces. And they might disagree that complex objects cannot
be stored as multi-morphemic lexical items. But, I think we
can all agree on the principle of Full Decomposition and
assume lexicalization as a last resort solution.
*N.B. that lexicalization has been proposed to coincide with Full
Decomposition and be subject to restrictions on processing. See
Bermúdez-Otero's (2013) discussion of transportation vs.
importation.
• The question of what exactly is going on in (1) and derivations
like (1) has been the subject of much work since the 1950's. It
has received new attention in the last decade in light of the
predictions of the frameworks of Distributed Morphology and
Phases.
• In order to understand why, we must first lay out our
theoretical assumptions. Today we will look at DM. Next week
we will look at cyclicity/Phases.
• After we have defined some morphosyntactic assumptions, we
will look at how these relate to the phonological patterns
important to analysing cyclic phonological domains. And will
complicate (1) with data like those in (3):
(3) a. paréntalize
b. parèntalizátion
• A note on dumping one's garbage in other people's yards.
Assuming that we have yards implies a commitment to
modularity. We will take this as the default position here.
o Syntacticians will sometimes dump problems that they
cannot solve in the phonologist's yard.
§ ex. *long vowel at right edge of NP
o Phonologists will sometimes dump problems that they
cannot solve in the syntactician's yard.
§ ex. There are extra heads in the syntax that
simplify the phonological account, but do
nothing for the syntax.
• The goal of a fully modular theory of the grammar is to define
the boundaries of each yard, and to stay within it. Hopefully
that will lead to further insightful analyses.
• Realizational theories, such as DM and Nanosyntax, further
the modular viewpoint in some ways (and perhaps not others)
in that they derive the behaviour of (morpho)phonological
forms from the properties of the morphosyntactic
computational system.
3
§1.2 Distributed Morphology
§1.2.1 The architecture and the 'distribution' of lexica
(4) The Architecture of Distributed Morphology
(Embick 2015: 4)
• Feature Types: Sound and Meaning
a. Phonological Features: Features from the universal
inventory of phonological features; e.g. [±voice],
[±labial], etc.
b. Synsem Features: Features from the universal
inventory of syntacticosemantic features; e.g. [past]
(‘past’), [def] (‘definite’), [pl] (‘plural’), etc.
(Embick 2015: 6)
(more later on morphological features)
• Two Types of Syntactic Terminals
a. Functional Morphemes: These are, by definition,
composed of synsem features such as [±past], or
[±pl], or [±def]. A further hypothesis is that they
do not possess phonological features as part of their
basic representation (see below).
b. Roots: These make up the open class or “lexical”
vocabulary. They include items such as √CAT,
√OX, or √SIT. Roots do not contain or possess
synsem features; a working hypothesis is that in the
default case, they have an underlying phonological
representation.
(Embick 2015: 7)
• Why make this distinction between whether morphemes are
specified with phonological features or not?
o Allomorphy
o Syncretism
• Allomorphy (Non-Roots)
(5)
(Embick 2015: 9)
(6) a.
4
b.
(Embick 2015: 10)
• This is not agreed upon by all researchers who work within
DM. On roots and argument structure/suppletion see (among
others) Bobaljik and Harley (2013) on Hiaki suppletion.
(7)
• Embick believes that roots may be underspecified or not, and
if so, underspecification gets us root allomorphy. But, he says
that underspecification/VI is mostly there to account for
syncretism, and roots don’t display syncretism, so it’s not
really important. I think some people in this room might
disagree with that, but Embick leaves the door open for
contextual allomorphy/syncretism of roots.
• Allomorphy of 'functional' heads that appear to be roots
• Roots are a special kind of acategorial morpheme, according
to Embick (they need structure to be interpreted) but words
don’t have to have a root. Light verbs, and other function
words (e.g. pronouns) do not need to be built around a root.
• The classic view of DM is that these 'light' functional items
are not roots and their mis-analysis as roots is what has led to
the conclusion that some roots display allomorphy.
(8)
P.56
• We will come back to this, as some people have also proposed
that derivational morphemes (normally considered to have
categorial/functional features) are roots.
• What is important here is that the syntactic environment
determines the phonological form of certain morphemes, and
therefore allomorphy gives evidence that some (if not all)
morphemes are bereft of phonological form in the syntax.
• Not often discussed in DM is the evidence from the opposite
direction: Phonological features never impact the syntactic
derivation (Scheer 2010 and references therein).
5
• Syncretism
• Illustration of Syncretism (P.25)
(9) Latin laudāre ‘praise’; Spanish hablar ‘speak’):
p/n Latin Latin American Spanish
1s laud-ō habl-o
2s laudā-s habla-s
3s lauda-t habla-Ø
1p laudā-mus habla-mos
2p laudā-tis habla-n
3p lauda-nt habla-n
Vocabulary Insertion provides a systematic (that is to say, nonaccidental)
analysis of syncretism.
Underspecification:
[+1,-2,+pl] à -mos
[+1,-2,-pl] à -o
[-1,+2,-pl] à -s
[-1,-2,-pl] à -Ø
[-1,+pl] à -n
(27)
"If there were no Vocabulary Insertion operation, and functional
morphemes were represented with their phonological form
inherently, the Spanish pattern could not be analyzed as a
systematic effect. Rather, the grammar of Spanish would contain
two distinct functional morphemes [-1,+2,+pl,/n/] and [-1,-
2,+pl,/n/] whose phonological form just happens to be the same.
This latter analysis, which treats the identity in form as accidental
homophony, is not capable of stating the important generalization
that (featurally) similar meanings are realized with the same
morphology in language after language." (27-28)
(10) The distribution of lexica/"Notions of Lexical
a. The idea that the lexicon is a generative system in which
words (as opposed to syntactic objects) are derived.
b. The idea that basic elements (morphemes) must be
listed, because they are underived.
c. The idea that the unpredictable behavior of complex
objects must be listed."
(14)
(11) The three lists of DM: The 'Distributed' factor
a. the list of Syntactic Terminals
b. the Vocabulary.
c. the Encyclopedia.
(17)
6
§1.2.2 More on Functional morphemes and Roots
§1.2.2.1 Functional morphemes
• Bundling
• Functional morphemes are bundles of features. The feature set
is universal, but the inventory applied within each language is
a subset (like in inventories of phonological features).
• Languages differ in whether certain features are bundled
together or not, but this is constrained (not all combinations
of features are possible, e.g. {past, inalienable}).
o Note that this falls out of a theory of general syntactic
structure/selectional restrictions (as in Nanosyntax)
but requires an extra theory of feature
hierarchies/relationships in DM.
o Buuuut, an extra theory of morpheme internal
structure has a parallel in phonology too, at the
segmental level (although work like Pöchtrager’s makes
the phonological hierarchy look more like the syntax).
• Examples of different kinds of bundling
(12) English we vs.
(34)
• On the subject of bundling:
o Are these differences due to bundling?
o Are they both bundled and then Mandarin split?
o Are they both split and then English bundled?
• Or is it all just VI? Can VI span nodes in a syntactic structure?
In classic DM (Embick) spanning is disallowed. VI only targets
terminal nodes/bundles).
• The inventory of features
• Do languages employ different (subsets of universal) features?
(13) SG/PL languages vs. SG/PL/DU languages
English I/we vs. Greek
• English only needs one feature to account for the surface
morphemes, but Greek needs two.
• Unless of course English has the same morpho-syntactic
distinction, but just has no overt Vocabulary Items to
distinguish it.
7
§1.2.2.1 Roots
• Categorization of Roots
Roots are category free. They are categorized by category-defining
heads. These are derivational heads, that are often overt in other
languages (languages with theme vowels, root-and-pattern
langauges). In English they can be null or pronounced, but the
assumption is that the syntax is cross-linguistically stable.
Important to what we will discuss is that these category-defining
heads have been proposed to define cyclic interpretation.
Consider (1) again, in (12).
(12) a. párent
b. paréntal
c. párenting
d. párenthood
(13) a. n(or v)
4
√PARENT Ø
b. a
4
√PARENT al
c. n(or Asp)
4
v																			ing
4
√PARENT Ø
c. n
4
v																			hood
4
√PARENT Ø
• Spanning/non-terminal spell-out vs zero heads? Does it make
a difference? Maybe? Let's discuss.
• The current view of phases is that it is only their complement
that is sent to (full) PF interpretation (the 'full' will be
discussed next week). The null head is proposed to not be fully
interpreted, and so can be influenced allomorphically (and
phonologically) by an adjacent, higher phase head. A spanning
account would, without further modification, imply that in
(13c,d) the outer n heads should be inside the same cycle as
the root. This is not the case.
• Samuels (2009): [Level 2 morphemes]…”undergo vocabulary
insertion on different cycles and thus cannot undergo fusion.
This explains an observation attributed to David Perlmutter
to the effect that there are no portmanteaux derivational
affixes; see Embick (2010) for discussion.” (P. 91 of PDF)
• But, there are proposed to be portmanteau that cross cycles
and include two phase heads. A classic example of this is the
French P+D patterns. If this is allomorphy, then there should
be no reason that category-defining heads should not be
portmanteau.
(14) a. de la à de la ‘of the(f)’
b. de le à du / _C, de l' _V ‘of the(m)’
c. le à le /_C ‘the(m)’
• Here, like I stated at the beginning of this handout, there are
multiple possible solutions to the above issue.
o Lexical/allomorphy: Either du is a portmanteau, or du
spells out the P or the D head and the other is null.
o Syntactic: There is some exceptional syntactic reason
that these two heads should see each other, where de is
permitted to overwrite le
o Phonological: de le à du is not allomorphy.
• Note that there are further ‘portmanteau’ patterns in
Quebecois French that could support or confirm a hypothesis
(Alexander-Renaud, Léveillé & Saucier, current
undergraduate project at UQAM)
8
(15) These alternations are optional)
a. sur la à [sa:] ‘on the(f)’
b. sur les à [se:] ‘on the(pl)’
c. sur un à [sœ̃ː] ‘on a(m)’
• These and other P+D alternations point to a larger
phenomenon.
• [l] in French (like [l] synchronically in Portuguese) historically
alternates with a vowel [o] in certain coda positions.
(16) cheval à chevals à chev[o] (chevaux) ‘horse~horses’
• This is exactly the kind of alternation that is of interest to us
when building theories of the interface. If de le à du is a
morphological operation, then we need to modify our theory
of the morphology; if it is phonological, then we do not.
• Why not? Phonology is also subject to cyclic effects. Are the
domains for morphological and phonological
accessibility/locality different?
o No and yes.
• Yes : Cyclic Containment (Arad 2003, Bermúdez-Otero 2018)
(17) a. áffix (n) ‘an affix’
b. affíx (v) ‘to attach’
c. áffix (v) ‘to attach an affix’
(18) accòmodátion (vs àbracadábra)
(19) Root-derived vs Word-derived items (Arad 2003)
a.
b.
• Note that outer affixes do not have access to the root (yld),
but only to the output of the first cycle.
• No: autosegmental representations (additive, copying),
infixation (Embick 2014, Newell 2017)
(20) a. fanfuckintástic
b. rigoddamndículous
c.f.
(21) a. That’s fucking fantastic!
b. You’re being goddamn ridiculous.
(Newell 2017:37)
(22)
(Embick 2014:7(pdf))
9
§1.3 Conclusions for today
• The morphosyntactic structure/derivation influences cycles of
phonological interpretation.
• The cycles of interpretation appear to be the same, but active
morphological and phonological operations (e.g. allomorphy vs
harmony) appear to be different.
• This has implications for our theories of morphosyntax and
phonology.
• The research I will present in the following weeks (mostly
mine, but others also for context) explores how an elaborated
theory of phonology and the phonology-syntax interface can
lead to a simplified, modular theory of grammar.
References
Arad, Maya. 2003. Locality constraints on the interpretation of
roots: The case of Hebrew denominal verbs. Natural Language &
Linguistic Theory, 21(4), pp. 737–779.
Bermúdez-Otero, Ricardo. 2018. Stratal phonology. In The
Routledge handbook of phonological theory (pp. 100-134).
Routledge.Bermúdez-Otero, Ricardo. (2013). The stem-level
syndrome. Speaker Series of the University of Pennsylvania
Linguistics Department, Philadelphia, 11.
Embick, David., 2015. The morpheme: A theoretical introduction
(Vol. 31). Walter de Gruyter GmbH & Co KG.
Embick, David. 2014. Phase cycles, φ-cycles, and phonological (in)
activity. In Bendjaballah, S., Faust, N., Lahrouchi, M. and
Lampitelli, N. eds., The form of structure, the structure of forms:
essays in honor of Jean Lowenstamm. John Benjamins Publishing
Company. pp.271-286.
Newell, Heather. 2017. Nested phase interpretation and the PIC.
In ed. Heather Newell, Máire Noonan, Glyne Piggott, & Lisa
Demena Travis (Eds.) The structure of words at the interfaces,
pp.20-40. Oxford University Press.
Scheer, T., 2010. A guide to morphosyntax-phonology interface
theories: how extra-phonological information is treated in
phonology since Trubetzkoy’s Grenzsignale. Walter de Gruyter.