Adger/Core syntax/ho 3/AS

Constituency and Theta Roles Usually, the label that is used is some subset of the features of the object,
most commonly the major category feature, but other features may also be
Syntactic structures are hierarchical in nature, with smaller structures coming relevant.
together to make larger structures, which, themselves, can be parts of yet
larger structures, and so on. We then join both the merged objects to a point placed higher up on the page,
to form a new object, the label of which is written on the higher point.
We shall motivate a syntactic operation, which we will call Merge, whose
function is to build up larger structures out of smaller ones, with the smallest We are merging are X and Y. e merge them to form a new object whose label
elements being lexical items. is Z.

Syntax starts off with features, as its basic atoms; these features are then
combined into lexical items, which are essentially bundles of features, which
we have assumed so far are unstructured.

These lexical items then combine into larger and larger structures, in a way It’s important to note that the new object is the whole structure, and not just
which we will motivate and make explicit. This phenomenon is known as the letter Z itself. Z is just the label of this whole structure. The lines joining
headedness. the labels are called branches, and the whole structure is called a tree, just as
we saw above.
Constituents
The new object created by Merge is said to immediately contain the original
Read Carnie for constituents and constituency tests
objects.
Fundamental notions: develop a theory which will give rise to these
We will adopt the idea that Merge joins two elements together as a working
hierarchical structures.
hypothesis. This will mean that all of the syntactic structures that we propose
we want our syntactic theory to do is to provide some operation that will will be binary branching.
build structures out of lexical items, and larger structures out of smaller ones.
Illegitimate structures Legitimate ones
Let us, then, posit an operation, which we call Merge, which has the property
that it joins two syntactic objects together.

This operation is motivated by the facts that we saw in the previous section,
which showed that words in sentences group into constituents. Merge is
essentially a constituent building operation.

This operation is usually notated is by firstly giving the objects that are being
joined labels.
Adger/Core syntax/ho 3/AS

One is practical: how do we determine the head of a constituent?

Merge does not specify the order of the elements that it merges. The other is more theoretical: what is it that makes a particular syntactic
object a head?
However the Order is crucial to natural languages. It is the order of the words
that makes (42) ok, while (43) is bad: the head of a constituent is the most important element in it. The vague term
“most important” here covers a multitude of sins, but essentially it picks out
(42) Harry collapsed. the element that is most characteristic of the constituent as a whole.
(43) *Collapsed Harry. It’s very easy to capture the idea that the head determines the distribution and
The usual term for this is that we need to linearize the structures. agreement properties of a constituent in the system we have so far.

Since Merge forms syntactic objects out of syntactic objects, it is a recursive We adopt a new notion projection, where features from a daughter node
operation. This means that we can build ever larger constituents using just project onto the mother node in a syntactic object.
Merge: Projection can be thought of as a copying operation: select some features on
a node and copy these onto another node.

The idea that the distribution of a constituent is determined by its head, we

simply say that it is the features of the head that project in any merged
structure.
Properties of merge
Syntax is sensitive to these features, it will be able to treat a head with a
Merge combines whole constituents feature, and a projection of that head, in the same way. So the fact that kiss
Peter behaves in the same way as drink Absinthe can be captured by the
claim that the heads of both of these constituents are the verbs that they
contain.

We will assume that Merge only applies to root nodes.

Determining the head

Adger/Core syntax/ho 3/AS

In each of these cases the new constituent which has been formed by Merge Let’s summarize this discussion:
is labelled by the features of its head.
1. A Predicate has θ-roles to assign
In each of these cases the new constituent which has been formed by Merge
is labelled by the features of its head. 2. These θ-roles are assigned to constituents in the sentence

Predicting the head - θ-roles and selectional features 3. The constituents to which the θ-roles are assigned are called arguments

θ Roles and the θ-Criterion 4. Every θ-role must be assigned to a constituent.

A proposition (like disappear), then it’s said to be a 1-place predicate. When Unassigned θ-roles
a predicate needs two other constituents (demonise), it’s a 2-place predicate,. The situation is a little more complicated than this. There are some predicates
3-place predicates require three. that seem to allow their θ-roles to remain unassigned. Examples include the
Verbs like rain, snow etc. are usually called 0-place predicates and need no verb donate:
semantic arguments to express whole propositions. (97) The landlord donated a helicopter
Agent and theme Donate is a 3-place predicate, but only two of its θ-roles have been assigned
One-place predicates which combine with an Agent are called Unergative in (97).
predicates, while one-place predicates which combine with a Theme are (98) The students demonstrated (the technique) this morning
called Unaccusative predicates.
(99) I have eaten (my hat) already
various other roles that we will come across (Goal, Source, Experiencer).
Selectional Features and Lexical Representation
how many thematic roles a predicate assigns, we refer to theta-roles (where
theta is short for thematic, and itself is often abbreviated to θ). So instead of Problem??
saying that a predicate needs two expressions with semantic content to make
up a proposition, we can say that the predicate has two θ-roles to assign. 100 Julie felt hot

The Unique Θ Generalization: Each θ-role is assigned to exactly one (101) Julie felt he was there
constituent in the sentence.
(102) Julie felt a twinge in her arm
Only some of the constituents of a sentence are assigned θ-roles, and these
we need to define a mechanism whereby certain syntactic properties are
are known as arguments. So an argument is defined as a constituent in a
associated with particular θ-roles; the θ-assigning properties of lexical items
sentence which is assigned a θ-role by a predicate.
Adger/Core syntax/ho 3/AS

don’t, on their own, allow us to determine the category of a syntactic We will assume that the syntactic structure to which the semantic interface
argument. rules apply should consist only of interpretable features. If this structure
contains uninterpretable features, then the semantic rules will not be able to
The standard way to express statements about the category of the constituent assign a complete interpretation to everything in the structure.
that a particular θ-role is assigned to is by means of categorial selectional
features (usually abbreviated to c-selectional features), and it is to these that We will call this general constraint Full Interpretation:
we now turn.
(104) Full Interpretation: The structure to which the semantic interface rules
A c-selectional feature is a categorial feature on a lexical item, which does apply contains no uninterpretable features.
not determine the distribution of the lexical item itself, rather it determines
the category of the elements which will be able to Merge with that lexical
item.

This sounds complicated, but an example should make it clear. Take a word
like kissed:

this clearly has a V-feature since it is the past tense of a verb, but it also has F on Y is uninterpretable by prefixing it with u. Now the uF on Y must be
(at least one) c-selectional N-feature. checked and it gets to be checked by being in a syntactic relation with
another F feature somewhere else.
This N-feature signifies that something that merges with kiss must itself have
a categorial N-feature. So we can Merge a noun like pigs or like Peter with Since Z is a sister to Y, the syntactic relation of sisterhood allows feature
kiss, but we cannot merge another verb, or a preposition: matching to take place, and uF to be checked.

(103) kissed Peter; kissed pigs; *kissed eat; *kissed by We will notate this by marking uF with a strikethrough. So the tree in (107)
will transform into the tree in (108):
Interpretable and uninterpretable features

Interpretable features are those features which have an effect on the semantic
interpretation of a category. Uninterpretable features are features which seem
to make no difference to the semantics of a sentence, but which are somehow
required if we are to explain the (un)grammaticality of certain sentences.

Case features, for example, simply regulate the position of certain nouns in This tree is generated from the lexical item kiss:
certain structures.
(110) kiss [V, uN]

Next topic S-selectional feature