Alec Marantz: Lecture notes - Fall morphology class

[Martha McGinnis]

Dear DM-List-ers,

With the hope of generating more productive discussion about morphological
theory, I have decided to post the lecture notes I wrote up for the fall 99
morphology course at MIT.  Anticipating some responses to the questions
raised in these notes, I delayed posting until after the semester was over,
since I was reluctant to both push forward with the topics I promised to
discuss in the course and engage thoughtfully in the discussion over the
notes on previous topics.  I still can't promise to reply to your comments
or to clarify everything that is left unclear in the notes.

Some formatting may be lost in the pasting of Word documents into these
e-mails.  I will monitor the results when I receive the postings myself.
There should be twelve sets of notes in all; I hope to finish posting them
by the end of January.  I'm re-reading the notes and making small
corrections, particularly in response to class discussions.  There will be
still be errors remaining, along with much opaqueness and unclarity.

Please think twice before quoting these notes in print.  In addition to
flat out errors, there is sometimes a dialog going on in the text, with
positions other than my own being presented for discussion.

Yours with some trepidation,

Alec Marantz



Lecture Notes, Morphology
9/10/99

Jackendoff, R.  1997.  "Idioms and Other Fixed Expressions."  In The
Architecture of the Language Faculty, LI Monograph 28, Cambridge, MIT
Press, pp. 153-177.

Marantz, A.  1997.  "No Escape from Syntax," in A. Dimitriadis et al.,
eds., U Penn Working Papers in Linguistics, 4.2, pp. 201-225.

Fodor, J.A., & E. Lepore.  1998.  "The Emptiness of the Lexicon."  LI 29.2,
269-288.

Pustejovsky, J. 1998.  "Generativity and Explanation in Semantics."  LI
29.2, 289-311.

1.	Lieber, Deconstructing Morphology

"The starting point for this theory of word formation is a somewhat odd one
- the fringes of morphology, so to speak, where the syntax of words and
that of phrases seem to converge."

a floor of a birdcage taste
over the fence gossip
off the rack dress
God is dead theology


2.	This betrays the hidden assumption that the syntax deals with the
arrangement of words.

3.	Jackendoff, The Architecture of the Language Faculty

"What is in the lexicon and how does it get into sentences?  So far, we
have dealt only with the lexical insertion/licensing of morphologically
trivial words like cat (chapter 4).  Now we turn to morphologically complex
elements of various sorts."

There's an illusion that we know what a word is --  "cat" is a word - and
an assumption that we do syntax with words.  Of course every syntactician
violates this assumption from square one.

D'ya wanna debate this point?

Latin:  fe:les NOM singular  fe:lem ACC  fe:lis GEN

4.	Suppose we begin with the features or categories of language.

	In phonology, phonological features are constituitive of sounds -
that is, if you strip the features and structure from a sound, there is no
remnant; there's nothing left.

	Are there non-constituitive (say, classificatory) features to the
terminal (combining) nodes of the syntax (or the Lexicon)?

5.	cat has at least a root and a category feature ("noun").  If
syntactic categories can be classificatory features of roots, then the word
cat could be a syntactic and Lexical atom - an atom with a syntactic
category.

However, if syntactic features are constituitive, then there must be a
remnant to cat when its nounness is removed (just like there's a remnant to
a /t/ when its [-voice] is removed).

6.	Word and Paradigm theories, like A-Morphous Morphology, assume that
words have properties required by the syntax.  Suppose Word and Paradigm
theories existed without the notion of a "paradigm."  The words of a
language, an arbitrary set, would have whatever features (properties) they
have.  Syntactic combination, however, requires words with particular
properties.  So Latin direct objects must have the ACC property.  Only if
there is a word meaning 'cat' with the property ACC in the language will
Latin allow 'cat' as a direct object.

7.	Paradigm theory says that "words" have different "forms."  The set
of features for, say, a Noun that the syntax might require (all the
different cases, for example) form a paradigm space (crossing the features,
e.g., case and number, yields an n-dimensional space, e.g., 2 dimensional
case by number space).  Words come in families identified as forms of the
same word.  The forms of a word fill out paradigm space, so a particular
form may fill several "cells."

8.	Anderson:  Forms of a word are created from a base form via
morphophonological rules.  On this logic, the base form is the (universal)
default form, since no morphophonological rules need to apply to a base
with a full set of syntactic features (assigned to the base by the syntax).

9.	Anderson:  The features realized by inflectional morphology - that
is, the features that determine paradigm space - differ from language to
language.  For example, in some languages number on nouns is inflectional
and in some it is derivational.  Since the syntax itself must treat
inflectional features specially, this means that the basic operation of the
syntax is different from language to language.

10.	Why isn't a whole sentence simply a "form of a word" in some
paradigm space defined by the syntax?

Everyone assumes that one reaches bedrock at the "content" words.  So, "The
cat is on the mat," can't be a form of the word "cat" since "mat" must be
an independent word.  However, "He's always singing," could be a form of
the word "sing," and might be realized as a single phonological word in
some language.

11.	Should we take anything for granted?  If we don't actually do
syntax with phonological words, is there any reason to suppose that we have
a notion of "word" suitable for the atoms of syntax?  Should we assume that
cat is atomic, that it is an atomic Noun, that it is a syntactic atom, that
we even know what Jackendoff is referring to when he writes,
"morphologically trivial words like cat"?

12.	The basic questions any theory of grammar must ask include, what
are the atomic, combinatory units of the syntax, what are the atomic,
combinatory units of the (morpho-) phonology, and how do these atomic units
connect in the derivation of a phonological form?

13.	One could imagine that, as far as the grammar is concerned, there
are only "constituitive" features, each feature is an atomic, combinatory
unit in the syntax, and each feature corresponds to an atomic
morphophonological unit, call it a "vocabulary item."  To the extent that
roots (the "remnants" of "content words" when all constituitive features
(save the root) are stripped away) have classificatory features, they may
play a role in semantic interpretation (at the LF interface with conceptual
systems) but would not enter into any grammatical principle (rule,
constraint, what have you=8A).

14.	Let's suppose that this idealized view of grammar is too
simplistic.  On the syntax side, for example, sets of features, as opposed
to individual features, may operate as atomic units in the syntax (which
means that the features aren't combined by "merge" and thus that their
combination need not be interpreted at PF or LF).  On the phonology side,
sets of syntactic features may correspond to single vocabulary items.  We
call the theory of pre-syntactic "bundling" of features into single atomic
units for the syntax the theory of the Lexicon, where the bundles are the
Lexical items.  The strongest claim we can make (other than the claim
implied in 13. - that there is no Lexicon in this sense and that therefore
every feature is a syntactic atom subject to "merge") is that there is no
"bundling" (Fusion) outside the Lexicon and that therefore every syntactic
atom is the locus for vocabulary insertion in the phonology.

15.	Jackendoff explicitly assumes that "mismatches" between "conceptual
structure" and "phonological structure" can include arbitrary many-to-one
correspondences wherein whole chunks of hierarchically arranged
syntactico-semantic atoms correspond to atomic phonological pieces.  So
simplex "went" can correspond to [GO + Past] (and simplex "kill" could be
CAUSE(X, DIE(Y))).

16.	Fodor, on the other hand, supposes (in some sense, to be made
clear) a one-to-one correspondence between words and concepts.  Note that
=46odor acknowledges that "cats" has "cat" in it, and that "book shelf"
probably has "book" in it as well.  He doesn't dwell on the issue of
whether "redden" has "red" in it, or "raise" includes "rise."  The issue
he's after is whether logical inference is run off constituitive features
of (in our terms) roots.  Is the conclusion that "John is unmarried" from
"John is a bachelor" made on the basis of a decomposition of "bachelor"
into features like "unmarried" (with our without remnants) or is this
conclusion drawn on the basis of classificatory features (or properties) of
"bachelor"?  Fodor never considers the possibility that "bachelor"
decomposes into the root "bachel" and the suffix "or," where the
nominalizing suffix -or here is associated with the meaning of "occupation"
or "socially identifying classification of a human" (cf. butcher,
pensioner, widower, spinster=8A).

17.	For Jackendoff, in the best of all possible worlds, phonological
structure transparently reflects conceptual structure and logical inference
can be run formally off conceptual structure.  What's interesting is that
both Jackendoff and Fodor argue essentially from a "what you see is what
you get" perspective.  Fodor see words and grapples with what he takes as a
transparent truth that linguistic structure doesn't support logical
inference.  Jackendoff sees just slightly messed up conceptual structures
in phonological structures, once the proper form of lexical representations
is revealed.

Both think they know a word and a noun when they see one (cat).

18.	The failure to question the obvious has led morphology into
darkness and despair.

A.	No one takes the serious morphologists seriously.  I don't know of
a single phonologist or syntactician that has really come to grips with
Anderson or Lieber (adopted their theories - I mean really adopted, rather
than waved at what they thought the theories were - or rejected the
theories on reasoned grounds), for example.  Most syntacticians and
phonologists operate outside any defensible or well-supported theory of
morphology (and I'm willing to name names here).

B.	The serious morphologists, on the other hand, who generally tend to
be phonologists at heart, end up making up their own theories of syntax, by
themselves.  Anderson and Lieber think they're adopting standard syntactic
theories, but they were several years out of date in syntax when they
created their theories and since syntacticians didn't take them seriously
anyway, they were never subjected to the type of critique of their
understanding of syntax that could have shown them the error of their ways.
[Wunderlich and Beard just make the syntax up - as a serious enterprise for
Wunderlich, although out of the mainstream.].

C.	Phonologists generally punt the morphology entirely these days.
Standard Optimality Theory is simply [i.e., literally] incoherent from the
standpoint of morphology.  [It is impossible to tell how standard OT deals
with the usual issues of morphological theory.]

19.	We know some things about morphology.

A.	First, there's s paradigmatic dimension to at least non-root morphemes.

	i.	"gaps" are the exception rather than the rule.  So we
expect every noun to be able to appear in every case position, every verb
to have a past tense, etc.
	ii.	there is "competition" among forms for the realization of
features, a competition that yields syncretism and (at least the appearance
of) undespecification.

This paradigmatic quality of morphology already disconfirms standard
theories of Lexical Morphology in which the feature structure of a word is
built up via percolation or other computation over the feature structure of
morphemes qua lexical atoms of sound/meaning correspondence.

You can't prevent, "He comb his hair" for "He combs his hair" unless you
have competition, and thus "separation" in Beard's sense.  You have to know
that you're shooting for third person singular and thus "combs" (really,
/z/) can beat out "comb" (really, /=F8/) for the realization of third person=
,
being more specific than (default) "comb"/zero.

Proponents of contemporary Lexical Morphology reintroduce paradigms in some
other way (see Lieber and Wunderlich), but these mechanisms have never
seriously been examined by anyone.

B,	Second, the phonological realizations of syntactic features are
pieces with properties (we call them vocabulary items); thus A-Morphous
Morphology as a general approach to morphophonology can't be right.

In fact, as shown by Halle and Marantz, A-Morphous Morphology as a theory
(as opposed to a descriptive framework in which anything could be
described) had no properties independent from its principles of
disjunction, which are empirically disconfirmed with data that no one has
ever disputed.

And Anderson's new version of A-Morphous Morphology has pieces - his theory
of clitics (which must be treated as the output of morphophonological rules
by the general principles of A-Morphous Morphology) requires constraints
ordering the clitics as pieces with properties.

20.	The paradigmatic and piece properties of morphology require
something like Distributed Morphology - that is, they require a theory in
which phonological pieces are organized into hierarchical tree structures
and in which these pieces are underspecified with respect to syntactic
features and compete with each other for the realization of syntactic
features.  All the burning issues in morphology can be stated as disputes
within this general framework.  There are no alternatives to Distributed
Morphology (i.e., a theory with late insertion of piece-like vocabulary
items into non-root terminal nodes, where the vocabulary items are
underspecified with respect to the syntactic features they realize and
where vocabulary insertion is governed by "elsewhere" competition).  A real
alternative would need some explicit account of the paradigmatic and piece
properties of morphology.

21.	Nevertheless, we will find some recent proposals for alternative
theories in the literature.  In particular, we will review some proposals
for notions of "lexical relatedness" other than "shares the same pieces,"
where connections between "stored" derived and/or inflected forms play a
role in the grammar.  An extreme here is the Bybee/Burzio position that
speakers store all tokens of all words (for Bybee, probably all sentences)
that they hear and draw generalizations over stored forms.  Steriade makes
related claims.  We have to ask how we can evaluate these claims, which are
literally incoherent - i.e., do not cohere with any general theory of
grammar.

22.	Learning roots like "cat" probably requires a prior conceptual
space, such that you know the meaning of "cat" before you learn the word
("oh, that's what we call those things!).

Still, an innate conceptual space may or may not imply a decomposition of
meaning into constituitive features (is "fuzzy" a necessary/definitional
property of "cat").

In any case, the essential issue for linguistics is whether the feature
space that creates the meaning of "cat" plays any grammatical role, similar
to that played by features like [+past].  If the features of roots play a
role in the grammar, then we might expect syncretism (from
underspecification) and/or suppletion (contextual allomorphic vocabulary
items) for roots.

23.	Suppose true modularity for root semantics.  The features of roots,
be they consituitive or classificatory, may be irrelevant/invisible to the
grammar.  This assumption predicts no syncretism/suppletion for roots.

24.	General assumptions:  The Distributive Morphology (Sept. 99 Marantz
version) Framework.

* = inevitable assumption (conceptual necessity)
! = arguable assumption/assumption with testable consequences

There exists a Universal set
of grammatical features, U 				*

A language chooses a subset of U for its grammar	!

The language bundles some subsets of its subset
into Lexical Items					!

The Lexical Items include only a generic
root node and no specific roots				!

Root nodes aren't bundled with any features
into Lexical Items					!

Syntactic Merger is the only process that
combines Lexical Items					*

The phonological word is not a cyclic
domain of the syntax					!

Re-Merger (Morphological Merger) occurs
at PF (i.e., without LF consequences)			!

Vocabulary Insertion and phonological realization
occurs at the "phase" level				*
As McCawley emphasized, the cyclic nature of
grammatical derivation may be the most
important and least questionable discovery of
generative grammar.

Vocabulary items are connections between
features of U and phonological forms			!/*

Vocabulary insertion works bottom up			!

Features from U in Lexical Items may be
deleted prior to VI (Impoverishment)			!

There may be multiple VI in a single
Lexical Item (fission)					!

There are Vocabulary Item specific ordering
constraints at a single hierarchical level		!

marantz at mit.edu