morph: [cn w]
(pl: -a (= pl(-um)))
sem: PL(〚 addend 〛)
implicature: something has gone before
- Reprise on Features and Values
Features are functions from linguistic signs to elements
in various value spaces. For example, the feature Case is defined for signs
in the nominal categories CN, DP,... There is room for variation here, for
example, Case determines or checks case only for pronouns in English and
Haisla, for DP's and CN's in Latin, DP's in Japanese. The value space for
Case is a set of cases: 6 in Latin, perhaps 17 in Finnish, and so on.
There are a number of variations in how morphosyntactic features may be and
have been viewed and used.
- Strict lexicalist:
Linguistic signs come fully inflected from the
lexicon -- viewed as a repository of all lexemes (see below). An inflected
form has everything specified, including values for all the features defined
for its category. The feature mechanisms of the grammar are used just to
check agreement and government. For example, trying to combine a Preposition
governing a certain Case value with an argument showing an incompatible case
value will fail.
Linguistic signs come in their base forms from the
lexicon. As part of the construction of complex forms, these base forms are
subject to morphological operations which contribute to the phonological,
syntactic, morphological, semantic representations or interpretations. For
example, a Nominative(Case) operation, along with whatever it does to the
other elements of a complex sign, specifies the value Nominative for the feature
- Traditional paradigms
display the forms and values for various morphosyntactic
- Regular and irregular:
We need some way to deal with regular forms, such as the plural form for the
CN, the past and participial forms for the regular (weak) verb ” ring (2).
One way to do this is to incorporate something like an "elsewhere" condition:
if an item is not prespecified for a value for a given feature, then it gets
the "regular" treatment.
is dealt with by assigning unions of values (or "OR" values). So in
German, for example, the value for Case of an inflected word like • Herrn
is (LISPily speaking) a list (GEN DAT ACC).
means this: a functor category checks or assigns values on an
argument for a government feature such as Case.
means this: a functor category checks or assigns values for an
agreement feature on itself according to its argument.
- Percolation of features
is dealt with according to general principles:
features are defined for the various syntactic categories.
Given now a functor X/Y:
if X = Y, all feature values are passed from the argument to the
if X and Y share no features, no feature values are passed to the
if there is overlap in the feature spaces between X and Y, the
resultant expression gets the union of the features and values for the functor
and the argument.
These principles are almost automatic from the architecture of the system. An
implementation or "execution" does require spelling out procedures, for
example, for a parser or constructor.
(See Bach, 1983, for one (pre-unification) working out of all this formally. I
have implemented some of this in LISP (actually SCHEME) for a simple parser
and morphological constructors for a number of languages: English (most
detail), Haisla, morphological fragments for Czech, Classical Arabic, Yup'ik,
- Features and the core relations of CG.
There are two primary possibilities in pure categorial systems. Given a
functor of category X/Y (or Y\X ), if X = Y: we have
X/X ( X\X henceforth I'll ignore the directionality) we have a pure
X/X X ==> X.
Otherwise, that is when X and Y are different, we have an
X/Y Y ==> X.
Given a feature system like that outlined above, there are two further
X and Y share no features: This is a pure exocentric
X and Y share some but not all features: This is a mixed
- Ungoverned (Choice) features;
Some features are not governed overtly by functor elements. This situation is
one where in some frameworks Functional Categories are posited, which may or
may not have any segmental realization. In categorial systems, we can use
unary categories (with the operation UP). (We expect to publish a paper on
this: Bach and Chao In preparation,s but it is still pretty much "in
- big houses: CN/CN CN so all features are shared
- see him: IV and DP share no features
- every fish: DP and CN share some features (language dependent?)
- G: der junge Herr `the young gentleman'
An interesting case: mixed agreement and government? Kase values (NOM) are
expressed all the way through, the functor der agrees with the noun in
gender but governs the form of the adjective.
(This section owes a great deal to David Dowty (p.c), who raised the initial
question in class and then sent me a detailed note about the issues. Much of
the following is paraphrased from his email.)
There is a problem with the way that Verb was defined in class (Session 4) as
anything that ended up with the final category IV (IV$): the problem was that
according to this definition in a sentence like John slowly walked
slowly would be a Verb. The problem is also inherent in early
definitions of the operation RightWrap that was supposed to account for the
positioning of the object with complex transitive verb phrases like
persuade - to go (cf. Bach 1979, 1980) which appealed to structures
like this: [XP X Y] with the niche after the X. (This is sorta OK
if you think of X as a word.)
Chierchia (1984) and Hoeksema (1984) tangled with this problem, seeking to
define the concept of Head on a categorial basis: so that for both in an X/X
X structure (ignore directionality) X is the head, but you have to add the
notion "lexical head" (i.e. word in the morphological sense) (Dowty p.c.).
[Note: I haven't had a chance to check the exact details of Chierchia's
dissertation on this point.]
The upshot of this is that you need to keep track of both morphological
categories and syntactic ones. For some kinds of cliticization you need to
have information corresponding to descriptions like "the first word" and "the
first major [?] constituent" (cf. Wackernagel's position phenomena)
[illustrations on "black"-board].
References: on RightWrap and Heads: Chomsky early work (1957 et seq.) found - studying
in the library as a transitive verb phrase, Thomason, Partee on
persuade - to go [need exact references: see Bach 1979, 1980], Pollard
and Sag 1994 (and early HPSG work going back to Pollard's dissertation),
Chierchia 1984, Hoeksema 1984,
Corbett et al. 1993, Zwicky 1985, 1993, Nichols 1986, 1992.
- More on Number
- Word internal government?
Part B: Questions
- Lia's question
Is there any reason to want to compose morphemes together before applying them
to a root or stem?
So, in some theories, you have to have a word at every step in the derivation,
and even if that might not be the case, there's an explicit or implicit
assumption that (compounding aside) you build words from some core outwards.
Does that necessarily have to be the case, though?
For instance, I can imagine a system where you might put affixes together
first, to make uberaffixes, and then apply those as a unit to the starting
point, and I can imagine how to implement it in a categorial sort of system.
What I'm wondering, then, is would there ever be a reason to want such a
system? And are there compelling reasons to forbid such a system?
- Melvin's question
One topic I would like to explore in more detail is the morphosemantics of
Tense, Mood, and Aspect in some indigenous languages. For example, in Rukai
stativity is marked morphologically with the prefix –ma, e.g. ma-rav?rav?r?
(lit. to be happy). I was wondering if, for instance, Haisla has this same
phenomenon, ie. if Aktionsarten is marked morphologically in Haisla or other
indigenous languages you know.
- Pilar's question
Chung and Ladusaw (2004) claim that object incorporation in the Chamorro verbs
gäi- (have) and not have is a process of multiple linking in which the
expression composed with the incorporated argument must be composed via
Restrict. They claim that the incorporated object is semantically incomplete
(it denotes a property rather than an individual) and that the extra object is
not a syntactic complement of V, but an adjoined constituent. Do they make the
claim that the incorporated object is semantically incomplete because it is
reduplicated with an extra NP? What does ‘adjoined constituent’ mean?
Is this object incorporation of type IV: N +TV > TV Classificatory Noun
Incorporation, in which the valence of the verb does not change (from your
Is it assumed cross-linguistically that incorporated objects are semantically
incomplete (i.e., denote a property)?
- Chris's question
If I'm understanding correctly, you seem to conceptualize word formation
operations (both derivational and inflectional) in the same way, namely, not
strictly lexical. I'm curious how you conceptualize these formally -- do you
typically treat each morpheme as a function in a CG sense? What does this
approach buy us (other than obviously, the ability to make some
generalizations) as opposed to one which views morphology as ENTIRELY lexical?
- Rachel's question
Could you go over how to formalize a sentence in one of the languages we have
been talking about? I know you have gone over how you formalize things, but I
was wondering if you could apply that formalism to a specific sentence, and go
through it, almost like one would for a problem set in English in an
introductory formal semantics course.
Here's a partial answer from a transcript of a
warts and all.
- Emmon's question
Is Semantics universal?
will be found
at "http://www.people.umass.edu/ebach/courses/mrphrefs.htm". They will be
given in short form e.g.: Frege 1892. Please note that these references
have been updated regularly as we go along.