Week 1

Lesson 1
24.11
Compositionality principle
Interpretation for complex natural language’s expressions is based on the syntactic structure.
In particular we are going to adopt a specific view that combines the syntactic component and the
semantic component, based on the principle of compositionality.
The compositionality principle concerns the idea that in general, the interpretation of complex
expressions is based on the interpretation of its constituent parts, and the way they are syntactically
combined.
The version of compositionality principle we will rely on is a very strict version of compositionality
which says that the interpretation for any given node in the tree is given by the interpretation of its
daughter nodes (which are called “immediate constituent”) 🡪 interpretation is always strictly local1.
The type of analysis we are going to develop also borrows from set theoretic interpretation, Tarski
and notion of models.
In the logic course we saw as a 1-place predicate is equivalent to a property; a binary relation is a
set of ordered pairs; a 3-place relation corresponds to a ternary relation and so on.

Models
A model is an abstract object that we use to clarify the interpretation of natural language’s
expressions. Compositional semantics is a development of artificial logic.
A model for interpretation is composed from three basic components:
- Domain of model (set of objects we want to talk about): we can allow in the domain of our model
special entities as “words” or “travels” that are ontologically unclear, or abstract.
But the model can contain them because natural languages allow us to refer to such entities.
- Interpretation function: something that associates any expression of natural language to an
interpretation (which we technically call a denotation) which is something constructed out of the
domain of entities.
- A set of possible words: it will allow us to make reference to alternative possible states of the
world, which is a basic cognitive ability that human beings have. We will maintain the uncommitting
attitude with respect to the philosophical issue of the words that can be contained. Intuitively, we
assume words just as possible ways that the world might be, or possible alternative states of the
world.

1
It applies combining two sister nodes (daughter nodes) that return their mother node’s connotation. Of
course, the process is recursive.

1
In a full-fletched analysis we will also introduce the set of moments in time, which represents the
temporal dimension, which is another important inherent aspect of semantics of natural languages.
Therefore, for a complete analysis we will need both the set of possible words and the set of moments
in time in order to represent the expressive richness of natural language. But these components
belong to the intensional level of semantic analysis.

Intensional and extensional semantics

Rome is Rome 🡪 not really informative.
Rome is the capital of Italy 🡪 adds information.
When we interpret sentences, we not only interpret them relatively to the world, but we consider a
variety of possible states of the world:
Rome is Rome is a tautology, a proper name like “Rome” will always denote exactly the same entity
in all possible worlds; while “Rome is the capital of Italy” is not a tautology, the definite description
“the capital of Italy” does not refer to the same city in all the possible worlds: in some
interpretations, the capital of Italy can be another city. This is why it is informative.
Semantics must make reference to the world (intentional, informative).
Anyway, in the traditional study of semantics, the first approach starts with a superficial,
extensional (uninformative) level of analysis, relative to a single possible world.
This is why we will start with extensional analysis of the world, to, then, proceed to an intensional
level only when the first step is acquired.
When we say that possible worlds are a theoretical construct which is used to capture a precise
cognitive ability, we refer to the fact that the we can imagine how a situation could be even if we do
not directly see or experience it: we can formulate hypothetical sentences and we have the ability to
understand what they mean even without having a practical example (when the covid-19 will be
over, we will be able to have classes in presence again 🡪 this sentence refers to a non-actual situation,
but everyone can imagine how it would be like, anyway).
The cognitive ability of interpreting states of the world is also proper of very small children and
some animals2. The uniqueness of humans being is the possibility to talk about it (while animals
cannot).

Semantic meaning
The basic notion of semantic meaning we will adopt is known as truth-conditional meaning: the
meaning of a sentence in natural language is mainly expressing truth-conditions.
Truth-conditions are not equivalent to truth-values.
Truth-values are true or false, if you are in a bivalent system; or true, false and undefined if you are
in a trivalent system. These are intuitive notions.
Truth-conditions is a view by Wittgenstein that discusses the idea that when you have a sentence
like “Valentina Smokes”, a person knows the meaning of such sentence if she’s able to discriminate
possible states of the world: states of the world in which is true that Valentina smokes, and those

2
Through some experiments has been proved that animals that get used to a certain routine, show surprise if
something changes respect to what they were expecting (Barbara Partee).

2
in which is not true that Valentina smokes.
Essentially, the meaning of a sentence is a way to discriminate between possible states of the world,
through the truth-conditions.
Truth-conditions, are, then, the conditions that must be satisfied in a sentence in order to
satisfy that particular sentence in a given interpretation.
The task of our semantic analysis is to make explicit the truth-conditions for sentences of a given
natural language (object language), and this is done by using a metalanguage (technical language
we use to specify truth condition, a set theoretic metalanguage).
However, the ultimate version of metalanguage we are going to use is actually functional.
Now let’s define what we mean by “function”.

Functions
Specifying the truth-condition of a sentence will take the form of a Tarski biconditional. It is a
sentence in the metalanguage that has the following form:
the object language: “Valentina Smokes” is true in an interpretation if and only if 🡪 metalanguage to
describe what the truth-condition are in order to make the sentence true.
To define a function, we first have to define a relation (function is a special type of relation). A
relation is defined as a set of ordered pairs.

Tarski biconditional
“…” is true in w iff (↔) …

Inventory:
< = indicates an ordered sequence of elements (ordered pairs)
D = domain of model
⟦⟧ = Interpretation function
W = set of possible words
R = relation
{} = brackets to define a set

Model:
< D, ⟦object language⟧, W >
D: {1, 2, 3, …} 🡪 domain coincides with natural numbers.
R : {<1, 2>, <2, 3>} 🡪 relation defined in the set of positive natural numbers, and corresponds to the
relation between a number and its successor.

A relation R is a function if and only if for every x there is

R is a function iff ∀x ∃!y s. t. <x, y> = exactly one element y such that the pair x, y belongs to the
∈R relation R.

3
 We cannot have a situation in which two pairs <x, y>
<x, y> ∈ R and <x, z> both belong to R.
<x, z> ∈ R If this happens, then,
or we do not have a function,
or y = z.

It is easily understandable if we consider the relation “being sisters”, in the domain of entities
consisting of human persons.
If Valentina and Annalisa are sisters, then <Valentina, Annalisa> is a pair, but Valentina can have two
sisters, so <Valentina, Ludovica> can also be a pair.
Therefore, the sisterhood relation is a binary relation but it is not a function, because you can have
two different pairs with the same element in the first position, and two different elements in the
second position.

● Rsister : {<v, a>, <v, l>} 🡪 not a function

If we consider a different kind of relation, “being biological father”, then we will see a different
outcome, precisely because each person can have maximum one biological father. Let’s suppose
Valentina’s father’s name is Tiziano. Then we can have the pair <Valentina, Tiziano>, and since
Valentina and Annalisa are sisters, also <Annalisa, Tiziano>, and, since Ludovica is another sister,
<Ludovica, Tiziano>. Here the second element is always the same, so we have a function.

● Rfather : {<v, t>, <a, t>, <l, t>} 🡪 function

The set of elements in the first position inside the pair is called domain (Valentina, Annalisa,
Ludovica); while the set of elements in the second position is called codomain (in this case only
Tiziano, but in general it can contain many elements).
In this case the domain’s and the codomain’s elements are part of the same set (individuals, or
human beings) but it can also be the case that the domain and the codomain pertain to two different
sets (for example, cakes and receipts).
⟦⟧ f :
𝑑𝑜𝑚𝑎𝑖𝑛 = 𝑠𝑒𝑡 𝑜𝑓 𝑜𝑏𝑗𝑒𝑐𝑡 𝑙𝑎𝑛𝑔𝑢𝑎𝑔𝑒 𝑒𝑥𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛𝑠 𝑐𝑜𝑑𝑜𝑚𝑎𝑖𝑛 = 𝑠𝑒𝑡 𝑜𝑓 𝑑𝑒𝑛𝑜𝑡𝑎𝑡𝑖𝑜𝑛𝑠 𝑑𝑒𝑓𝑖𝑛𝑒𝑑 𝑜𝑛 𝑡ℎ𝑒 𝑑𝑜𝑚𝑎

Rfather : {<v, t>, <a, t>, <l, t>} can also be written as
ffather : (v) = t 🡪 the function father maps the entity Valentina to the entity Tiziano.

● I (Valentina) = v 🡪 If I apply the function “I” to the entity “Valentina”, it will return the
element named Valentina (in metalanguage: v).
● ⟦Valentina⟧M 3 = v 🡪 In compositional semantics we use the object language in fat square
brackets
These two previous denotations are equivalent.

3
In the object language, outside the interpretation function, you normally indicate the Model as a small v
outside brackets. For the moment we omit this indication.

4
Frege’s Program
Frege pointed out that when we consider a simple sentence like “Valentina smokes” we have an
intuition that they don’t have exactly the same nature:
● The name “Valentina” it itself has a definite meaning: it refers to a particular entity of the
world (saturated).
● The predicate “smokes” seems to not have a complete meaning: there is something missing
in this description of the world. The meaning seems somehow incomplete (unsaturated). It
needs to be linked to an expression like “Valentina” in order to become what Frege defined
“a complete thought”.
Frege has this idea that some meanings are saturated and some others are not. He found in the
function’s application the way to saturate the unsaturated meanings.
To him, an incomplete (unsaturated) expression is a function which needs to take a some other
complete (saturated) element as an input in order to give as outcome a complete thought.

Sets
Suppose we define a set A. We can list all of its elements. In this case, the set contains all the
numbers from 1 to 5:
● A : {1, 2, 3, 4, 5}
Instead of listing all the elements in A (which is quite complicated if the set contains a lot of
elements, or is infinite), we can also define a set by specifying a membership condition. In this case,
the membership condition is the set of numbers n such that n is a positive number smaller or equal
to 5:
● A : {n | n ≤ 5}
In this case we have introduced a variable (n), and specified a condition that must be satisfied by the
variable in order to be a member of the set.
There is also an empty set (∅) which contains no elements.
It is also possible to define a set of sets.
● {{m | n ≤ 5}, {m | n > 5}}
This set contains two sets: the numbers smaller or equal to 5 ({n | n ≤ 5}), and the set of all numbers
bigger than 5 ({n | n > 5}).
Remember: the difference between a set (denoted by {}) and an ordered couple (denoted by <>) is
that in the ordered couple the order of elements matters, while in the set, the elements can appear
in a casual order.
Ordered pair: <2, 3> ≠ <2, 1>
Set: {2, 3} = {3, 2}

5
But let’s return to functions,
● f : ∣N → ∣N s. t. ∀x ∈ ∣N, f (x) = x + 1

This is the function that goes from the domain of natural numbers to the codomain of
natural numbers such that for every element belonging to the domain of natural numbers,
the function that applies to that element is x + 1.

Example of Tarski biconditional

⟦Valentina smokes⟧ = True (in W) iff … (1, 0)
⟦vP⟧ = True (in W) iff… ⟦S⟧

⟦smokes⟧ = {x ∈ D ∣ x smokes}

⟦VP⟧ = ⟦V⟧ = ⟦smokes⟧

⟦v’⟧ = ⟦VP⟧

⟦NP⟧ = ⟦N⟧ = ⟦Valentina⟧ = v ∈ D

⟦vP⟧ = 1 (True) iff ⟦NP⟧ ∈ ⟦v’⟧

iff v ∈ {x ∈ D ∣ x smokes}
iff v smokes

R. Montague

Interpretation
Valentina smokes is true if and only if… and then we specify the truth-conditions in the object
metalanguage.
When interpret such a structure, we are interpreting a syntactic structure. Therefore, we draw the
syntactic tree. We have the NP in the specifier of vP.

1. ⟦vP⟧ = True (in W) iff…

In the first passage, ⟦vP⟧ = True (in W) iff…, we use the denotation ⟦vP⟧ to indicate the
whole tree (because the elements we will deal with are all under the vP). ⟦vP⟧ is short form
to indicate the whole tree. 4

2. ⟦smokes⟧ = {x ∈ D ∣ x smokes}

We must start from the bottom of the tree and go upwards. Let’s start from the

4
Let’s note that for this first (extensional) analysis we will not take in account the TP domain, because it
would require an intensional treatment. In the same way, we will not consider v. We are pretending that TP
and CP are not involved just to explain the process in an easy way.

6
 interpretation of the single lexical items. In this case we start from the predicate (smokes).
This is a 1-place predicate (therefore, it refers to a property, or, a set of entities which share
the feature of smoking).
We put into fat squared brackets the object language’s element (⟦ smokes⟧), and then
rewrite in in our metalanguage: (x belongs to the domain iff, and then we specify the
condition that must be satisfied by x in order for x to be part of the set of predicate “smoke”
🡪 {x ∈ D ∣ x smokes}).
When we write “x smokes” in our metalanguage, what we are doing is just retrieving a
representation of the concept of smoking from our mental lexicon.
What compositional semantics sees of the lexical semantics of the verb “smokes” is that it
can be applied to an entity to distinguish between entities that have the property of smoking
and entities that do not have the property of smoking. So, it defines the meaning of “smoke”
as the set of entities that have the property of smoking.

3. ⟦VP⟧ = ⟦V⟧ = ⟦smokes⟧

R. Montague formulate the analysis of nodes as follows: the content of a node is denoted by its
daughter node.
In this perspective, VP’s content is identical to V’s content, which is denoted by the verb
“smokes”.

4. ⟦v’⟧ = ⟦VP⟧

We apply the same reasoning, therefore, v’ has the same content of its daughter node VP.

5. ⟦NP⟧ = ⟦N⟧ = ⟦Valentina⟧ = v ∈ D

The content of NP node is exhausted by the content of N head, which is the proper name
Valentina, which is a particular entity belonging to the domain D.

6. ⟦vP⟧ = 1 (True) iff ⟦NP⟧ ∈ ⟦v’⟧

iff v ∈ {x ∈ D ∣ x smokes}
iff v smokes

This passage reassumes the whole procedure:

On the one hand we have the NP that denotes an entity (Valentina);
On the other hand we have the v’ that denotes a set of entities (entities that smoke).

● Therefore, the overall structure (vP) is true if and only if (and here we specify our truth
conditions) the entity denoted by NP (Valentina) belongs to the set of entities v’ (smoke),
(⟦vP⟧ = 1 (True) iff ⟦NP⟧ ∈ ⟦v’⟧).

7
 ● It is true, if and only if Valentina (v) belongs to the set of x belonging to the domain of the
model such that x has the property of smoking (iff v ∈ {x ∈ D ∣ x smokes}).

● And Valentina (v) will satisfy the membership condition if and only if she smokes (iff v
smokes).

This is the compositional analysis for a simple vP, that just consists of a proper name and a simple
1-place predicate.
We said that “smoke” is denoting a set.
Now, we want to define the meaning of “smokes” as a function f.

The sentence as a function

⟦smokes⟧ = fsmokes : D → {1, 0} s.t.
∀x ∈ D : fsmokes (x) = 1 iff x smokes
{x ∈ D ∣ fsmokes (x) = 1}
= {x ∈ D ∣ x smokes}
⟦v’⟧ = ⟦VP⟧ = ⟦V⟧ = ⟦smokes⟧
⟦NP⟧ = ⟦N⟧ = ⟦Valentina⟧ = v

⟦vP⟧ = fsmokes (v) 1 𝑖𝑓𝑓 𝑣 𝑠𝑚𝑜𝑘𝑒𝑠 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

(= ⟦v’⟧ (⟦NP⟧))

Explanation:
1. ⟦smokes⟧ = fsmokes : D → {1, 0} s.t.
To understand what kind of function is this, we must look at the syntactic tree of the sentence.
Looking at it, we will figure out that the v’ node (smokes) is sister of the NP node (Valentina), which
denotes an entity. So, intuitively, this function maps the saturated entity to the unsaturated element.
Therefore, we imagine that the domain of our function is the set of entities.
What do we want this function to return? We want it to return a truth value.

∀x ∈ D : fsmokes (x) = 1 iff x smokes

For every x that belongs to the domain of the function, the function applies to x and returns a
truth-value true, if and only if x smokes.
A function from entities to truth-values is called a characteristic function, because it is a function
that characterizes a set.

{x ∈ D ∣ fsmokes (x) = 1}
In fact, we can define a set: the set of entities x that belong to the domain, such that the function
“smokes” applies to x and returns a value true.

8
In order to return the truth-value true, x must have the property of smoking:

= {x ∈ D ∣ x smokes}
We can always translate a set into a characteristic function, from the given domain to the set of
truth-values.

2. ⟦v’⟧ = ⟦VP⟧ = ⟦V⟧ = ⟦smokes⟧

3. ⟦NP⟧ = ⟦N⟧ = ⟦Valentina⟧ = v
We saw that the v’ node corresponds to the predicate smokes, and that the NP corresponds to the
entity Valentina.

4. ⟦vP⟧ = fsmokes (v) 1 𝑖𝑓𝑓 𝑣 𝑠𝑚𝑜𝑘𝑒𝑠 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Then we can say that vP will be the result of applying the function denoted by “smokes” to
“Valentina”, and it will result as true (1) if and only if Valentina smokes, and false (0) if and only if
Valentina does not.

(= ⟦v’⟧ (⟦NP⟧))
So, this is just the function denoted by v’ that takes as an input the entity denoted by NP, in the
specifier of vP.
In this way we have that the denotation of v’ to vP takes as input the denotation of the subject NP
and directly returns a truth-value as the denotation of the level of vP; and the truth-value will be
related to the particular circumstance whether in that circumstance Valentina has the property of
smoking or not.
Let’s now see the semantics for an example like: Valentina drinks and smokes.

Coordinate predicates example

Valentina drinks and smokes Nobody drinks & smokes ↔
● Valentina drinks Nobody drinks & nobody smokes
● Valentina smokes

⟦drinks⟧ = {x ∈ D ∣ x drinks}

⟦smokes⟧ = {y ∈ D ∣ y smokes}

⟦and⟧ = ⋂

⟦v’*⟧ = ⟦v’1⟧ ⋂ ⟦v’2⟧

{x ∈ D ∣ x drinks} ⋂ {y ∈ D ∣ y smokes}

⟦vP⟧ = ⟦NP⟧ ∈ ⟦v’*⟧

= v ∈ {x ∣ x drinks} ⋂ {y ∣ y smokes}
● v ∈ {x ∣ x drinks}

9
 ● v ∈ {y ∣ y smokes}

Explanation:
Remember that when we analyze natural language expressions we are interested in specifying the
truth-conditions with respect to the logical relations which naturally hold between natural language
sentences.
So, for instance, when we have a sentence like “Valentina drinks and smokes”, it implies that
Valentina drinks, and Valentina smokes.
This implication is intuitive in the sense that every competent speaker of English will catch this
relation.
Note that if we change the subject with “Nobody” it won’t be the same: saying that “Nobody drinks
and smokes” does not mean that nobody drinks and that nobody smokes, but that nobody both
drinks and smokes.5
The tree we gave is a simplified, non-standard analysis of the sentence.

1. ⟦drinks⟧ = {x ∈ D ∣ x drinks}
2. ⟦smokes⟧ = {y ∈ D ∣ y smokes}

When we have a 1-place predicate we will specify a set, or the corresponding characteristic
function. So, drinks specifies the set of entities x such that x has the property of drinking.
In the same way, with a different variable, we specify the set of entities y such that y has the
property of smoking.

3. ⟦and⟧ = ⋂

If we assume that each of the 1-place predicates (properties) denote a set, we can interpret
the conjunction “and” as simply introducing a set intersection.

4. ⟦v’*⟧ = ⟦v’1⟧ ⋂ ⟦v’2⟧

{x ∈ D ∣ x drinks} ⋂ {y ∈ D ∣ y smokes}

Now we can say that the denotation of v’* (the whole conjunction) will result from the
intersection of the set denoted by v’1 and the set denoted by v’2. That are, the set of
elements x belonging to the domain such that x drinks; and the set of the elements y
belonging to the domain such that y smokes.

5. ⟦vP⟧ = ⟦NP⟧ ∈ ⟦v’*⟧

Now, we can have the truth-conditions of the whole vP, that will be just that the individual
denoted by the subject NP belongs to the set denoted by the sister node v’*.

5
This gives us a proof of the fact that conjunction must take place at the level of predicate, and not at the level
of the whole phrase. “Nobody” is a quantifier.

10
 = v ∈ {x ∣ x drinks} ⋂ {y ∣ y smokes}

Namely, if and only if Valentina belongs to the intersection between the set of entities x who
drink and the set of entities y who smoke.

● v ∈ {x ∣ x drinks}
● v ∈ {y ∣ y smokes}

Of course, if it is true that Valentina belongs to this intersection, this implies that Valentina also
belongs just to the set of drinking entities, and to the set of smoking entities.

This is how we obtain the two entailments that we underlined at the beginning. In other terms, with this
analysis of 1-place predicates as sets we can account for the fact that conjunction is a polymorphic operator: it
can apply to the level of predicate and just give set intersection.

11
Lesson 2
25.11
A transitive structure
Let’s propose a new sentence, a more elaborate structure with a transitive verb.
⟦v⟧ = ⟦likes⟧ = {<v, t>, <v, s>, <v, a>, <v, s>}
= {<x, y> ∣ x likes y}
⟦NP2⟧ = ⟦N⟧ = ⟦Steven⟧ = s
{x ∈ D ∣ …}

⟦VP⟧ = {z ∈ D ∣ <z, ⟦NP2⟧> ∈ ⟦V⟧} =

= {z ∈ D ∣ <z, s> ∈ {<x, y> ∣ x likes y}}
= {z ∈ D ∣ z likes s}

⟦NP1⟧ = … = v

⟦vP⟧ = 1 (in w) iff ⟦NP1⟧ ∈ ⟦v’⟧ (in w)

iff v ∈ { z ∈ D ∣ z likes s}
iff v likes s

Explanation:
Having a transitive verb, we have a verb and an object (for the sake of simplicity, a simple proper
name). We already know that the NP containing a proper name denotes an entity. This is why we
distinguished NP1 and NP2.
At the level of the v’ we are having a set of entities x that belong to the domain of the model such
that … ({x ∈ D ∣ …}). Therefore, we can proceed to interpret the vP as the unione of the NP1 and
its sister node v’, as we did with the previous intransitive example.
Therefore:
NP1 and NP2 are entities;
V is a transitive verb;
v’ is a set of entities.

1. ⟦v⟧ = ⟦likes⟧ = {<v, t>, <v, s>, <v, a>, <v, s>}
= {<x, y> ∣ x likes y}
Notice that, in terms of sets, we will take a transitive verb (2-place predicate constant) as a 2-place
(binary) relation.
Then, it can correspond to “Valentina like Tiziano (<v, t>); Valentina likes Silvia (<v, s>), Valentina
likes Annalisa (<v, a>); Valentina likes Sara (v, s)…”.
Therefore, we have:
A set of ordered pairs at the level of the V node;
An entity at the level of the NP2 node;
and what we want to obtain for v’ is not a set of ordered pairs but a set of entities.

12
We cannot proceed in a direct way: if the structure were a flat structure such as:
Then, of course, we could interpret the whole VP by a simple rule as:
we could directly use the three denotations (the entity NP1; the set of ordered pairs
V; the entity NP2) saying that the structure is true if and only if the ordered pair
consisting of the entity denoted by the subject NP1 and the entity denoted by the object NP2
belongs to the denotation of the transitive verb.
In this case we could have written the truth-conditions in the following way:
⟦VP⟧ = 1 (in w) iff <⟦NP1⟧, ⟦NP2⟧> ∈ ⟦V⟧ 6.
BUT
we do not have a flat structure: in our syntactic tree, NP1 is in a higher position with respect to the
chunk of v+NP2. Therefore, we cannot proceed with this kind of analysis.
In our analysis we must first try to obtain a set of entities in the v’ node (combining the set of binary
relations V and the entity NP2); and then analyze NP1 and v’ as sister nodes.
Or, it is first necessary to combine the transitive verb (likes) with the object (Steven), and then
proceed to associate the subject (Valentina), that is the saturated part, with the obtained
combination (…likes Steven), which is unsaturated.

1. ⟦v⟧ = ⟦likes⟧ = {<v, t>, <v, s>, <v, a>, <v, s>}
= {<x, y> ∣ x likes y}
Going back to the first passage, we listed the possible ordered pairs inside the set ⟦v⟧. But we know
that, instead of listing all the elements, we can also establish a membership condition, which in this
case is {<x, y> ∣ x likes y} (the set of ordered pairs x and y, with x and y belonging to the domain of
the model, such that x likes y).

2. ⟦NP2⟧ = ⟦N⟧ = ⟦Steven⟧ = s

We already saw that a node takes its denotation from its daughter node.

3. ⟦VP⟧ = {z ∈ D ∣ <z, ⟦NP2⟧> ∈ ⟦V⟧} =

The denotation of VP is the set of entities x that belong to the domain of the model such that the
ordered pair with z in the first position and NP2 in the second position belongs to the denotation of
the verb.

= {z ∈ D ∣ <z, s> ∈ {<x, y> ∣ x likes y}}

This can be rewritten as above: the set of entities z belonging to the domain such that the pair with z
in the first position and Steven in the second position belongs to the set of ordered pairs which has
x in the first position and y in the second position, such that x likes y.
In practical terms, this means that we apply the ordered pair <z, s> to the set resulting from the
binary predicate, and then we obtain:

= {z ∈ D ∣ z likes s}

6
It corresponds to the previous description.

13
 4. ⟦NP1⟧ = … = v
As we did for NP2, we denote the node NP1 as Valentina.

5. ⟦vP⟧ = 1 (in w) iff ⟦NP1⟧ ∈ ⟦v’⟧ (in w)

iff v ∈ { z ∈ D ∣ z likes s}
iff v likes s
The denotation at the level of vP is true if and only if the entity denoted by the subject (Valentina)
belongs to the set denoted by v’ (the set of entities that like Steven).
Therefore, it is true if and only if Valentina belongs to the set of entities z belonging to the domain
such that z likes s.
Namely, if and only if Valentina likes Steven.

This is how we obtain the analysis of a transitive verb.

Note: because the subject is more prominent than the rest of the clause, we need an intermediate
step (3.) in which we define the set of entities which are I specific relation with the entity denoted
by the direct object.
This is imposed to us by the strict version of the compositionality principle that we have adopted,
but we can already recognize that this analysis makes one immediate empirical prediction:
It can be conjoined to an intransitive vP.
Let’s give an example: Valentina smokes and likes Marlboro
Where Marlboro is an abstract object, a proper name; and while smokes is intransitive (1-place
relation), likes is transitive (2-place relation).
Valentina [smokes] ⋂ [likes Marlboro].

The sentence as function

We have an entity at the level of NP1 and an entity
at the level of NP2.

f : D → {0, 1} What denotation can we give to V (likes) such that it

can yield a function of the set of entities at the level
of the v’ node?
Remember that a function is defined by the type of
possible inputs and the type of possible outputs:

The function denoted by the V (likes) will take an

input (the denotation of its sister node NP2) to
return an output at the level of VP node. And the
output will not be a truth-value, but another
function.
This is the exact definition of a function nest.

14
 1. ⟦V⟧ = ⟦likes⟧: g : D → (D → {0, 1}) s. t. ∀x ∈ D : g (x) = f x : ∀ y ∈ D : fx (y) = 1 iff y
likes x.
The transitive verb likes has a function g from the domain of entities taking an input to the domain
of functions from entities to truth-values.
The function g is such that for every entity x that belongs to the original domain, g taking an input x
will return a function that we will call f x.
f x is a function such that for every entity y that causes an input to the nested function f, fx applies to
y and returns the value true if and only if the entity y (which is the entity that causes the input to
the nested function) is in a like relation with x (or, y likes x.)
Function nesting 🡪 the function that we call g is the function denoted by the transitive verb which
takes as input an entity x (that corresponds to the object NP2), and it returns as output another
function f (x) which takes another entity y (that corresponds to the subject NP1) as input and
returns the value true, if and only if the second entity (y) is in a like relation with the entity x.
Let’s empirically explain how this works:

2. ⟦likes⟧ = {<v, t>, <v, s>} D = {v, t, s}

Let’s suppose that the binary relation “like” consists in two ordered pairs: Valentina likes Tiziano;
Valentina likes Steven.
The domain is {Valentina, Tiziano, Steven}.
Now, we can draw the function in the following way:

3. ⟦likes⟧ = ⟦likes⟧ =

The pair Valentina, Tiziano and the pair Valentina, Steven are true; all the other pairs are false.
We make up all the possible pairs and we assign to each one its correspondent truth-value.
In the second column, we define function nesting:
Likes takes as input an entity, which may be Valentina, Tiziano or Steven; and for each of these
entities we verify whether the other entities are in a “like” relation with them.
In this case, we have a true outcome only in the case:
t → [v → 1] 🡪 Valentina likes Tiziano.
and in the case

15
s → [v → 1] 🡪 Valentina likes Steven.
All the other outcomes are false.
As is possible to note, the three selected entities are the entities that we are testing to be liked, not
the entities that actively like: the three selected entities are the entities that we would find in the
second position of the binary relation <v, t>; <v, s>.
In this way we are obtaining one advantange:
We can assume that:

4. ⟦VP⟧ = ⟦V⟧ (⟦NP2⟧)

That means, the denotation at the level of VP is the denotation of the transitive verb applied to
an input, indicated with round parenthesis, that is: the denotation of the object NP2.
The way 1. to write the function:
⟦V⟧ = ⟦likes⟧: g : D → (D → {0, 1}) s. t. ∀x ∈ D : g (x) = f x : ∀ y ∈ D : fx (y) = 1 iff y likes x.
is a cumbersome method, therefore, we will need a more compact way to denominate this
function.
This is why we are introducing the notion of semantic types.

Semantic types
The idea is that every node in the syntactic structure is associated with a denotation belonging to
one specific semantic type.
Semantic types are subsets of possible denotations defined on the basis of the basic ingredients of
our model which can correspond to specific actual denotations of specific nodes in the tree.
In the previous model we already specified two (simple) semantic types7: NP1 and NP2.
What we still need is to specify the semantic type of the nodes that denote a function, and in doing
so we introduce the recursive definition of semantic types.

Semantic types (extensional)

1. e is a type 🡪 De : = D
2. t is a type 🡪 {1, 0}
3. For every types α and β, <α, β> is a type 🡪 D<α, β> : f : Dα → Dβ
4. Nothing else is a type

1. The semantic type e corresponds to the type of entities.

2. The semantic type t corresponds to the type of truth-values.
3. Then we have a recursive definition which says that “For every types α and β, the ordered
pair <α, β>8 is also defined as a type.
4. Finally, nothing else is a type.

7
The so-called “saturated types” in Frege’s analysis.
8
Where α is a well-defined type and β is another well-defined type (for instance, <e, t>, or <<e, t>, e>)

16
These definitions hold just for extensional semantics.
Pairs like <α, β> will correspond to a function that maps the first to the second element, with α
as domain and β as codomain of the function.
On the basis of this definition, we can define a type <e, t>, which is a function from entities to
truth-values. By applying this recursive definition, we could also create an ordered pair which
has the type e as first member and <e, t> as second member (<e, <e, t>>), that is what happens
in nested functions.
● <e, <e, t>> is a function that takes as input an entity and returns as output the characteristic
function of the set. Therefore, it straightly corresponds to the function g we have previously
defined.

Let’s draw again our syntactic tree and match each of these types with a specific node.
vP 🡪 t type

NP1 🡪 e type

v’ 🡪 <e, t> type

VP 🡪 <e, t> type

V9 🡪 <e, <e, t>> type

NP2 🡪 e type

From the tree is clear how function application corresponds to type simplification:
We have a transitive verb (V) of type <e, <e, t>>: it takes as input the entity (e) of NP2 and returns,
at the level of the mother node (VP) an output of the type <e, t>; And, at the level of vP we have two
daughter nodes: NP1 of type e, and v’ of type <e, t>, so that the v’ can take as input the NP1
denotation e and return a truth-value at the level of vP.
Therefore, zooming on the highest part of the tree, the first element of the v’ ordered pair <e, t> is
saturated by the e of NP1; and the second element of the pair <e, t>, the truth-value, is what returns
at the level of the mother node vP.
RECAP:
I. V: <e, <e, t>>. The first element e is saturated by the entity denoted by the object NP2 (e)
🡪 returns <e, t> at the level of their mother node VP.
II. v’ inherits the denotation of VP, which is <e, t>.
III. v’: <e, t>. The first element e is saturated by the entity denoted by the subject NP1 (e) 🡪
returns <t> at the level of their mother node vP.
From this analysis we see how saturating is a type simplification.

9
Remember: V in this case corresponds to a transitive verb, this is why it corresponds to the type <e <e, t>>.

17
The recursive character of this process allows us to have very complicated types. For instance, we
could find a type corresponding to <α, β> like:
<<e, <e, t>>, <e, t>>, where:
<e, <e, t>> = α,
<e, t> = β.
Going back to our definition of this sentence:
⟦V⟧ = ⟦likes⟧: g : D → (D → {0, 1}) s. t. ∀x ∈ D : g (x) = f x : ∀ y ∈ D : fx (y) = 1 iff y likes x.
All this definition denotes a function of the type <e, <e, t>> that we must compact through what is
called λ notation10.

λ notation
We use the Greek symbol λ as a piece of metalanguage used to define functions.
Whenever we find a λ, it denotes the definition of a function. It is often accompanied by squared
brackets []. Therefore, [λ opens the definition of a function.

Simple (intransitive) sentence

Let’s suppose we want to define a function of the type <e, t>, a simple intransitive verb like “smokes”,
which takes as input an entity (subject) and returns as output a truth-value.
λ means the function takes an input, therefore we put a variable which stands for any thing of the
appropriate semantic type that goes as input to the function. This is why after λ we always put a
variable11 and this variable must be defined as belonging to the particular denotational domain.
In the case of an <e, t>12 function, the variable will belong to the domain of type e (because it is an
entity: the first element, or, the input, of the function <e, t>):
At this point we put a full stop, and then we put the description of the value that must return for
each x that goes in input. What we want to describe is a truth-value. Therefore, we write the
truth-conditions that must be satisfied in order for the function to return a value true.
For instance, in this simple case, we will write that, in order to return a truth-value true, x must have
the property of smoking:

<e, t>
[λx ∈ De · x smokes]13

10
Lamda notation.
11
Corresponding to the input of the function.
12
Simple, intransitive.
13
Where “λx ∈ De” stands for e (entity) and “x smokes” stands for t (truth-value).

18
This denotation corresponds to:

f : De → {0, 1} s. t. ∀x ∈ De f (x) = 1 iff x smokes

Read as: Function from the domain of entities to the truth-values T or F, such that for every x
belonging to the domain of entities, the function of x is true if and only if x smokes.

All of these information are packed into the simple formula [λx ∈ De · x smokes], where the
information of x belonging to type e is expressed from the first part, and the information about the
truth-values is expressed by the second part.

Simple (transitive) sentence

Let’s now consider a transitive sentence of the type <e, <e, t>>, with the function nesting
corresponding to the transitive verb “likes”.
In this case, in the pair <α, β>, e = α; <e, t> = β 🡪 <e, <e, t>>.
Then we will write:

[λx ∈ De · [λy ∈ De · y likes x]], or, simplifying:

[λxe · [λye · y likes x].

<<e, <e, t>>

<α, β>

Here I introduced another variable y because we have two entities to represent (the object and the
subject).
The order is the following:
we first apply the most external function (which corresponds to the level of
the V + NP2).
As is possible to see from the branch of the tree we showed previously, the
V corresponds to <e, <e, t>>; and the first e is saturated by the object,
which in our λ denotation is x.
Then, the VP is denoted by the type <e, t>

Then, proceeding to a higher level,

in the type <e, t> inherited by the v’ from the VP (which corresponds to
the second element of our overall formula, β) the e is saturated by the
subject, therefore it returns t to the vP node.
In this sense we can VERY INFORMALLY, translate our formula as
follows (just to understand, read note):

<e, <e, t>> = <obj, <subj, tr.v>> = [λxe · [λye · y likes x]14

14
NOT STATED BY THE PROFESSOR. SPECIFICATION ADDED BY ME. BE CAREFUL IN USING IT.

19
Function application
Let’s explain now how to notate in λ notation the operation of functional application.

● ⟦[V likes]⟧ (⟦NP2⟧) = [λxe · [λye · y likes x] (s)

The verb likes takes as input the denotation of the object (NP2), and in the λ notation we write it as
we did just before: “the variable x belonging to the domain of entities (object)” corresponds to α;
and “the variable y belonging to the domain of entities (subject) + y transitive verb x” corresponds
to β. The final parenthesis indicates Steven (because the original sentence was Valentina likes
Steven).
How to simplify this formula in order to obtain the nested function we need?

● [λxe · [λye · y likes x] (s) In the λ notation we gave, Steven, who is the object,
saturates the first λ function (λxe).

● [λxe · [λye · y likes x] (s) s is, then, going to replace all the occurrences of the
variable x. So, what we will obtain is:

= [λye · y likes s] 🡪 <e, t> namely, λy belonging to the domain e such that y is in a
like relation with s (Steven), where s is now a constant.
It corresponds now to the notation of <e, t>

Of course, we can also have a double-function application. So, when we are working bottom up, we
will obtain <e, t> at the level of VP.
Proceeding at a higher level, we know that the VP transfers its denotation to the v’ node. Then, to
return the final output vP, we will have:

● ⟦v’⟧ (⟦NP1⟧) = [λye · y likes s] (v) Where I substitute all the occurrences of the
= v likes s variable y with NP1, Valentina (v), the subject.

The final result corresponds to the truth-value true of the sentence, if the truth-condition v likes s is
satisfied.
This strategy could also be expanded further. We can, for example, imagine a function such as:
<e, <e, <e, t>>
This is a function that takes as input an entity and returns a function <e, <e, t>> from entity to
characteristic function (which corresponds to a transitive verb).
This kind of function corresponds to a ditransitive verb, as in the sentence:

20
Valentina introduced [Steven] [to Tiziano]

[λxe · [λye ·[ λze · z introduce x to y]

Where:
x is the object (Steven), or, the person who is introduced;
y is the goal (Tiziano), or, the person who receives the action; z is
the subject (Valentina), who acts.
The ultimate analysis will have the verb at the little v level, but
for the moment we explain the relation like this.
Further, we will see the ultimate analysis.

21
Lesson 3
26.11
Two fundamental rules
The two rules that we use to characterize what we have been doing up to now are essentially two:

Non-branching node rule Branching node rule

α ⟦α⟧ = ⟦β⟧ α If ⟦β⟧ is a function s. t

⟦γ⟧ ∈ domain of ⟦β⟧,
then ⟦α⟧ = ⟦β⟧ (⟦γ⟧)
β β γ (function application)
α = {β, γ}

The non-branching node rule:

Suppose we have a node α whose daughter node is β. The rule says that in this case, the denotation
of α corresponds to the denotation of β (the denotation of a node in inherited by the denotation of
its daughter node).
The branching node rule:
A node α dominates two daughters (β and γ). If the denotation of β is a function, and its sister node γ
is a possible input for the function of β, then the denotation of their mother node α is equivalent to
applying the function of β to the denotation of γ (function β takes as input γ).

The structures that we interpreted all terminate with root nodes that have denotations of type t. but
what does it happen when we have a sentence which has an argument position that is not
syntactically saturated. For example, a sentence like “*Greeted Ann”, which is ungrammatical
because the subject is missing. But, in terms of syntactic analysis, we would assume that a transitive
verb like “greeted” has a so-called “thematic grid”15, commonly called “θ-grid”, which corresponds to
<Agent, Theme>, since the verb is transitive.
*Greeted Ann.

θ-grid <Agent, Theme>

Here we do not reach

θ-criterion: every thematic a “t” at the vP node.
role must be assigned an
argument, and every
argument in the structure
must receive one and only
one thematic role.

15
Which is abbreviated with the Greek symbol θ.

22
*Ann laughed Jan We already
reached a “t”
θ-grid <Agent>

Unergative Unaccusative

If we take this other example, we see that what happens in the unergative case is that the
truth-value is reached at the level of the VP, so, if we merge this structure with something else, we
will have an entity NP1 and a truth-value VP. Therefore, we are not able to combine them and assign
a denotation to their mother node vP.
Concretely, the structure violates the general principle that every node of the syntactic tree is in the
domain of the interpretation function: in fact, the interpretation function must be able to assign a
denotation to every node of the tree.

Difference unergative-unaccusative
The subject of an unaccusative verb is generated as the sister of the verb, whereas the subject of an
a is generated in a higher position.
The system fails in making a distinction between accusatives and unergatives:
in fact, if we rely on the denotational types of the nodes, we see in both the unergative and the
unaccusative tree, a 1-place predicate (laughed/fell) of the type <e, t>, and there is no explicit
matter that would lead to the projection of the argument in different positions.
Therefore, two possibilities can be involved in order to avoid this problem and be able to make such
distinction:
● One logical possibility is to assume that the thematic-theta theory (or, theory of argument
structure) is required: in this case, we need to have a representation of theta-structure
because we want certain types of arguments (themes and patients) to always be projected
according to the uniformity of theta assignment hypothesis in the complement position of
the big V, in order for agents to be merged in the higher position in the specifier of vP, or
anyway in a different syntactic position.
● Another possibility is to make reference to the notion of event.
There is in fact an interpretation, in literature, that refers to events as entities in the domain
of the model. This theory allows us to apply a different analysis of the thematic-roles.

Event semantics
The notion of event is introduced in semantic analysis via a root that is in fact independent from the
issue of thematic roles. It is proposed in a work by Davidson (1967).

23
For Davidson, the verb comes to denote a relation between two individuals and an event, rather than a
simple relation between two individuals (relation object-event; relation subject-event).
Davidson was concerned in the semantic implications (intuitively calculated by competent speakers
of the language) in sentences like:
[Brutus] stabbed [Caesar] [PPwith a knife] [PP(in) the senate]16.
● B. stabbed C. with a knife.
● B. stabbed C. in the senate.
● B stabbed C.
Here we have two PP (with a knife and in the senate).
This is a pattern of implication that we saw in “Valentina drinks and smokes” with conjunction.
We can, in fact, individuate four parts of the structure (1. Brutus; 2. Caesar; 3. With a knife; 4. In the
senate), therefore we consider the predicate “to stab”, in the complete sentence, a 4-place predicate,
since it could support four constituents.
[Brutus] stabbed [Caesar] [PPwith a knife] [PP(in) the senate]. 🡪 4-place predicate

B. stabbed C. with a knife. 🡪 3-place predicate

B. stabbed C. in the senate. 🡪 3-place predicate

B. stabbed C. 🡪 2-place predicate

Davidson’s idea is that we can actually have a simple analysis of the modifiers (with a knife and in
the senate), related by an intersective behavior, by assuming that a verb like “stabbing” implies a
reference to an event. According to Davidson, then, a verb like “stabbing” does not simply refer to a
relation between two entities, but also to a relation between two entities and an event.
Then, basically, we have a function from an entity to another entity, to a property of events (or a
1-place predicate of events), which is a characteristic function:
o <e, <e, <ε, t>>>

Here we mark the event with the Greek symbol ε17.

In this case, the nested function will return a property of events.

What are events?

We are going to say that the domain of the model contains such things as events: their ultimate
nature is not concern of linguistic semantics. Whether events exist or not in the actual world is not
our concern.

16
Remember: these are NOT arguments: these are adjuncts.
17
Epsilon

24
What concerns us is that natural languages work as events exist, and could even be counted (for
example I can say “my last travel to Siena”, or “three travels to Siena”). Therefore, events could be
counted as if they were actual physical objects.18
The first analysis Davidson gives us is the intersective treatment for the modifiers we have
introduced with the sentence “Brutus stabbed Caesar with a knife in the senate”.
How do we map these semantic types to a function? We are following the usual system.
o <e, <e, <ε, t>>>

λxe · λye · λeε · y stabbed x in e (or, e is an event of stabbing y by x).

Caesar Brutus Event = B. stabbed C. in Ev. (knife, senate)

Obj. Subj. Event. (function of event to truth-values).

Syntactic analysis

⟦VP⟧ = ⟦V⟧ (⟦NP⟧) = λye · λeε · y stabbed C. in e.

⟦v’⟧ = ⟦VP⟧

⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧) = λeε · B. stabbed C. in e.

⟦with a knife⟧ = λe’ inst. (e’…)

⟦vP*⟧ = ⟦vP⟧ ⋂ ⟦PP⟧

What happens here is that after we have applied the predicate “stabbed” to the object NP2 “Caesar”,
the entity denoted by NP2, “Caesar” is going to saturate the λxe function, so that we leave out the λxe
and the entity denoted by NP2 is substituting the variable x in our function:

λxe · λye · λeε · y stabbed x in e 🡪 λxe · λye · λeε · y stabbed Caesar in e.

18
Of course, an event coincides with a specific temporal or spatial location.

25
 o ⟦VP⟧ = ⟦V⟧ (⟦NP⟧) = λye · λeε · y stabbed C. in e.
This denotation is inherited by the v’ node, which saturates the variable y with the denotation of its
sister note NP1 Brutus:

o ⟦v’⟧ = ⟦VP⟧
o ⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧) = λeε · B. stabbed C. in e.
Now, the idea is that the modifier “with a knife” will also be a property of events (λe’) such that the
instrument of e’ is a knife.

o ⟦with a knife⟧ = λe’ inst. (e’…19)

Now, assuming that an adjunct denotes a property of events, and that the other PP (in the senate) is
λe’’, such that e’’ takes place in the senate; this type of analysis allows us to reduce this
combination to a simple intersective treatment.
So, we will just have characterized the set of events which belongs to the set characterized by vP
(Brutus stabbed Caesar) intersected with the set of events PP (of the instrument knife), and we will
obtain the intersection between the lower two (vP + PP) at the level of vP*.
In this way, we obtain another 1-place property of events which can be intersected with another PP
e’’ (in the senate); and the process can be iterated.

o ⟦vP*⟧ = ⟦vP⟧ ⋂ ⟦PP⟧20

When we have two sister nodes that are both denoting properties of events, we cannot make a
function application, because each node wants as input an event, and the sister node cannot take it
as an input since it is the same semantic type.
On the top of the structure, at a certain point, we will need to have an operator (existential
quantifier):

λf<ε, t> · (∃e’) f(e’) = 1 🡪 existential quantifier used to bind the event variable
There is an event e’, such that function f applies to e’ and returns the value true. Or, the set
characterized by f is not empty. So, when we apply the ∃ what we obtain is that there is an
event which belongs to the set of events of B. stabbing C. and also to the set of events that
take place with a knife.
In some cases we describe a unique event, as the murder of Caesar, but still, the event is not
just one event, but it is a property of events, so, a 1-place predicate, or, a set.
This is why we must maintain a difference between the single event and the property of
events, which is an object that can combine through intersection with other sets.

19
e’ stands for “a knife”
20
The symbol ⋂ is just to explain the idea of conjunction.

26
Pasons analysis (1990)
● Brutus stabbed Caesar
Parsons gave a new understanding of thematic-roles: they can be conceived as particular relations
between an individual and an event.
So, we can reanalyze our previous example as a set of relation between entities and events. In
other terms, we can describe the truth-conditions of an event such that: e is an event of stabbing,
and patient of e is Caesar, and the agent of e is Brutus:
o (∃e) stabbing(e) & patient(e, C.) & agent (e, B.)
This analysis allows us to reinterpret thematic roles as related to different components of an event.
In particular we can finally give a compositional account of the combination between vP and VP.
agent (e, B.) 🡪 v = introduces the agent, which is saturated by NP1 (spec of vP).

⟦v⟧ = λxe · λeε · e is a stabbing of x

⟦NP0⟧ = ⟦Caesar⟧ = c

⟦VP⟧ = ⟦V⟧ (⟦NP0⟧) = [λxe ·[ λeε · e is a stabbing of x]] (c)

= [λeε · e is a stabbing of c]

This analysis is proposed by Kratzer (1996).

The proposal is that we should analyze a verb lilke “stab” as consisting of a core predicate which is
just a relation between an entity and an event <ε, t> (VP), whose patient is Caesar.
As we did in the previous analysis, we substitute all the occurrences of x in our lambda notation
with Caesar.
Now we want to combine VP with v, such that we obtain v’.
v’, as we mentioned, must introduce the value of agent, and we will relate this entity (Brutus),
through the agent relation, to the same property of events denoted by VP (stabbed Caesar).
To do so, Kratzer, instead of applying a characteristic function (which is also possible), introduces a
new rule: the event identification rule.

27
Event identification rule:

s. t. ⟦γ⟧ ∈ D <ε, t> and ⟦β⟧ ∈ D <e, <ε, t>>

then ⟦α⟧ = [λye · [λe’ε · ⟦γ⟧(e’) = 1 and ⟦β⟧(y)(e’) = 1]]

This rule assigns to v a denotational type which is a relation between an entity and an event, so it is
of the type <e, <ε, t>>. And the rule of event identification is the following:

Given a tree of the form shown above, such that the denotation γ belongs to the domain of the type
<ε, t>, and the denotation of β belongs to the domain of the type <e, <ε, t>>,
the denotation of α is going to be given by the function λy of the type e and λe’ of the type ε
and then γ applied to e’ (remember that gamma is a function from event to truth-values) with a
truth-value true; and the function β applied to y and e’ also returns the value true.
In other words, v introduces the agent, and this will allow the fact that the event 1 (stabbed
Caesar) and the event 2 (Brutus stabbed) are both relating to the same event.
It is essentially a conjunctive rule that puts together VP and v.
Let’s see how this rule applies to the previous structure:
⟦v⟧ = λxe · λe’’ε · x is the agent of e’’21

⟦v’⟧ = [λye · λe’ε · [λeε · e is a stabbing of C.] (e’) = 1 &

[λxe · λe’’ε · x is the agent of e’’] (y) (e’)

= [λye · λe’ε · e’ stabbing of C. & y agent of e’]

⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧)
= [λye · λe’ · e’ stabbing of C. & y agent of e’] (B.)
= [λe’ · e’ stabbing of C. & B. is agent of e’]

o ⟦v⟧ = λxe · λe’’ε · x is the agent of e’’

Now, we put together the value of VP and v through the event identification rule:
⟦VP⟧ = ⟦V⟧ (⟦NP0⟧) = [λeε · e is a stabbing of c]
⟦v⟧ = λxe · λe’’ε · x is the agent of e’’

o ⟦v’⟧ = [λye · λe’ε · [λeε · e is a stabbing of C.] (e’) = 1 & [λxe · λe’’ε · x is the agent of e’’] (y) (e’)
At this point we have the denotation of v’ obtain through the EI rule: we insert a function of
the entity agent (y) and the final event (e’ 22), and we copy the formula of VP, applying the
event e’ to the single event of stabbing Caesar (e). And the VP formula must be true, as the v

21
(x = C.)
22
Which is the synthesis of the event of Caesar stabbed and Brutus stabbing.

28
 formula, that we copy, applying to it a variable that denotes the agent (y) and the denotation
of the event (e’) which is going to substitute the single event of Brutus stabbing (e’’).
In this way, we introduce the agent y and we state that event (e) and event (e’’) must be
identified in a single event (e’).

= [λye · λe’ε · e’ stabbing of C. & y agent of e’]

What I have obtained is an entity and an event, such that the event is an event of stabbing
Caesar, and the entity y is the agent of this event.

o ⟦vP⟧ = ⟦v’⟧ (⟦Brutus⟧)

= [λye · λe’ · e’ stabbing of C. & y agent of e’] (B.)
= [λe’ · e’ stabbing of C. & B. is agent of e’]
What we obtain is e’ (stabbing Caesar) and y (Brutus).

Exercise:
Compositional analysis of the sentence:
Valentina helped Sylvia
with:
I. ⟦helped⟧ = [λx · [λy · y helped x]]
II. ⟦helped⟧ = [λx · λe · x helped in e]]
⟦v⟧ = [λy · [λe · y agent of e’]]
Event Identification [λz · [λe’’ · ⟦helped⟧ (e’’) = ⟦v⟧ (z) (e’’) = 1]].
NB: with II you obtain a property of events as the vP denotation.

29
Week 2
Lesson 4
1.12
Correction of exercises:
I.
Valentina helped Sylvia

⟦NP0⟧ = ⟦N⟧ = ⟦Sylvia⟧

⟦V⟧ = ⟦helped⟧ = [λx · [λy · y helped x]]

⟦VP⟧ = ⟦V⟧ (⟦NP0⟧) 🡪 [λx · [λy · y helped x]] (s.) =

= [λy · y helped s.]

⟦v’⟧ = ⟦VP⟧ = [λy · y helped s.]

⟦NP1⟧ = ⟦N⟧ = ⟦Valentina⟧

⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧) 🡪 [λy · y helped s.] (v.) =

= [v. helped s.]

30
II.
Valentina helped Sylvia in event
⟦NP0⟧ = ⟦N⟧ = ⟦Sylvia⟧
⟦V⟧ = ⟦helped⟧ = [λxe · [λeε · x is helped in e]]
x relates to event, such that x is helped in the event.

⟦VP⟧ = ⟦V⟧ (⟦NP0⟧) = [λxe · [λeε · x is helped in e]] (s.)

= [λeε · s. is helped in e]
an event such that obj. is helped in the event.

⟦v⟧ = [λye · [λe’ε · y agent of e’]]

= y relates to event1 such that y is the agent of event1.

EI23 = ⟦α⟧ = [λye · [λe’ε · ⟦γ⟧(e’) = 1 and ⟦β⟧(y)(e’) = 1]]

⟦v’⟧ = [λze · [λe’’ε · ⟦VP⟧(e’’) & ⟦v⟧(z) (e’’) = 1]] =
= entity Z relates to event2 such that VP happens in event2 and v relates to entity Z in event2 = T.

= [λze · [λe’’ε · [λeε · s. is helped in e](e’’) = [λye · [λe’ε · y agent of e’]](z) (e’’) = 1]] =
= entity Z relates to event2 such that (event of VP: obj. is helped in event) happens in event 2;
and (y relates to event1 such that y is the agent of event1) is related to entity Z and event2 = T.

= [λze · [λe’’ε · [λeε · s. is helped in e](e’’) = [λye · [λe’ε · y agent of e’]](z) (e’’) = 1]] =
= [λze · [λe’’ε · [e’’ · s. is helped in e’’] & [z · [e’’ · z agent of e’’]] = 1]] =
= entity Z relates to event2 such that (event2: obj. is helped in event2) &
(z: in event2, z is agent of event2) = T.

= [λze · [λe’’ε · s. is helped in e’’ & z is agent of e’’]] = 1

= entity Z relates to event2 such that object is helped in event2 and z is agent of event2 = T.

⟦NP1⟧ = ⟦N⟧ = ⟦Valentina⟧

⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧) = [λze · [λe’’ε · s. is helped in e’’ & z is agent of e’’]] (V.)
= entity z relates to event2 such that (obj. helped in event2 and z is agent of event2) apply to Subj.
= [λze · [λe’’ε · s. is helped in e’’ & z is agent of e’’]] (V.)
= [λe’’ε · s. is helped in e’’ & v. is agent of e’’]
= event2 such that obj. is helped in event2 and subj. is agent in event2.

23
Event identification.

31
Comments:
o The variable e’’ has been introduced in order to indicate that e and e’ are the same event.
o EI applies when we have <ε, t> in one branch and <e, <ε, t>> in the other.
o While in the first structure (I.) (v) is irrelevant; while in the second structure (II.) it assumes
the role of introducing the subject
o At the level of vP node we obtain a 1-place property of events.
o When we apply the EI rule, the first part (VP) returns the same truth-value as the second
part (v), this is why we put = between the two. In the last passage we drop = and substitute
it with & (because both conditions hold in e’’)
o We proceed by steps: the first one (VP) is to combine the object NP2 with the event (e) of
being helped. Then, at (v’) we introduce the fact that there is another event (e’) in which a
subject operates actively, and we establish that the event in which the object is getting
helped (e) and the event in which the subject is agentive (e’) are the same event (e’’). The
last step is (vP), in which we specify that the subject is NP1.
At this point we obtain a property of events at vP, and if this is a single event, we will have higher up
in the functional structure an aspectual head that binds this event (∃ 🡪 there is an event of helping
Sylvia where the agent is Valentina); but we can also have different kinds of aspect.
For example, we can have a habitual aspect (which corresponds to ∀ 🡪 whenever Sylvia is in
trouble Valentina helps Sylvia).

32
It is also possible to add a propositional phrase (PP) adjunct to the vP, like “in Paris”.
Let’s see how would we be analyzing the adjunct of such sentence:

Valentina helped Sylvia in Paris

(Former analysis until vP node +)
⟦NP⟧ = ⟦N⟧ = ⟦Paris⟧ = P.
⟦P⟧ = ⟦in⟧ = [λxe · [λe1ε · e1ε takes place in x]]
⟦PP⟧ = ⟦P⟧ (⟦NP⟧) = [λxe · [λe1ε · e1ε takes place in x]] (P.)
= [λxe · [λe1ε · e1ε takes place in x]] (P.)
= [λe1ε · e1ε takes place in P.]
⟦vP*⟧ = λe2ε · ⟦vP⟧ (e2) = ⟦PP⟧ (e2) = 1
= λe2ε · [λe’’ε · s. is helped in e’’ & v. is agent of e’’] (e2)
= [λe1ε · e1ε takes place in P.] (e2) = 1
= λe2ε · [s. is helped in e2 & v. is agent of e2
& e2 takes place in P.] = 1
The last rule we applied is a conjunctive rule. This is why we do not introduce a new variable: we
only have a PP (event), and we do not have a new agentive or passive entity.

This kind of analysis brings important empirical advantages: it makes possible to understand why
when we have verb nouns we can typically avoid to specify the agent argument.
Transitive verbs such as “destroy” need a specification of the agent and of the patient, while, if we
use the nominal form “destruction” we do not need to specify these entities anymore (gerund).
This happens because this kind of nominative structure derives corresponds to the lexical verb (V),
and in this case (v) is not needed (difference between the first and the second tree).

Classification of verbal predicates: eventuality

Vendler provided a specification of different subtypes of verbal predicates:

33
Let’s see in detail what these denominations mean.

Stative vs non-stative
Verbal predicates (eventuality) can be differentiated into two classes: stative and non-stative.

Stative Non-stative
This verbal category describes a state, something This verbal category is compatible with a
that is not in evolution, or, a type of event with no progressive event.
dynamic development implied.
For instance, “John knows her” is a stative form: it is It can be further divided into two classes:
not an event that undergoes a progression. Atelic and Telic verbal predicates.
If we turn the sentence in “John is knowing her”, then
the form is not stative anymore: in fact, in this case Progressive and imperative forms are
we have a progressive event. compatible with non-stative verbs.
Progressive form is not compatible with stative, nor
is imperative form.
“John is stupid” 🡪 stative.
“John is being stupid”24 🡪 non-stative.

Atelic vs Telic (non-stative)

Non stative verbs can be disambiguated into: atelic and telic.25

Atelic Telic

24
Means “John is acting in a stupid way”, refers to an event that is happening, not to a state of being.
25
It corresponds to the distinction between perfective and imperfective aspects. If we use the aspectual
denomination we have to specify that we are dealing with a lexical aspect (independent from grammatical
aspect).

34
This denotation is given to events that involve This kind of label is given to events that involve
no clear ending point. a process that reaches a goal.26
A sentence like “John walked in the park” A sentence like “John built a house” specifies an
specifies an action that John has done for a action that involves a building process starting
certain amount of time, but it is not specified from the basement of the house and
whether the action of walking lasted five culminating with the last element that
minutes or one day; nor is specified if John completed the house.
reached a specific place in the park, or if he just The reach of the goal (the moment in which the
walked around a little. house has been entirely built and now is finally
The action of walking did not end with a complete) coincides with the Telos.
completion of the progression. The action of building ended with the
completion of the progressive action.

Achievement vs Accomplishment (telic)

Among the telic events, we can distinguish between: achievement and accomplishment.

Achievement Accomplishment
This kind of verbs refer to a progressive This kind of verbs refer to a progressive
process (non-stative) which involves a process process (non-stative) which involves a process
and a goal (telic); but they do not really focus and a goal (telic); and they do focus both on the
on the process part. process and on the goal part. Or, in other
Saying “John is born” indicates the moment in words, the process is effectively accounted as
which John has seen this corrupt world for the part of the event, as in the case of “John built a
first time. The process of “being born” involves house”.
a pregnancy (process) and a goal (the very
moment in which the baby gets out from his
mother’s womb), but the sentence above
intuitively focuses solely on the second part
(goal).

Affection of telic verbs on the IA

Telic verbs are linked to the lexical semantics of the sentence, but they also affect the internal
argument.
Let’s take as example the sentence “John repaired the computer”.
In this case, the computer undergoes the process of being fixed, therefore, it experiences a number
of changes until the verb is accomplished.
The maximum degree of the property applied by the verb to the computer (the moment in which
the computer, after all the process, reaches the state of being fixed) corresponds to the telos.
From this analysis we understand that the verb inevitably affects the IA (object), which is modified
by it. John’s computer, in the sentence, is broken at the beginning, it gets fixed and it ultimately
reaches the final status of fixed.

26
Telic literally means “goal” or “aim”.

35
In this sense, the idea, studied by Verkuyl and Tenny is that the argument (DP/NP) measures the
development of the event.
Take as example the sentence “John ate the pancake”.
Here we have an event in which the IA (pancakes) is gradually affected until the telos is reached and
the pancake no longer exists.

Unaccusative verbs
Whenever we have an unaccusative sentence, as “the boat sank”, the boat is the only argument, and
it originates in the complement position (sank (__the boat__)), and then moves to a higher position
in the structure (the boat sunk (__)). In fact, also in this case we see that is the boat that measures
the event of sinking: the sinking event starts with the boat that is still on the surface of water, keeps
happening with the boat taking on water, and ends with the boat that is gone completely under
water.
Another instance is “the letter arrived”, that was originally “arrived __the letter__”, but then “the
letter” moved to a higher position.
In this case too, the letter, which is the argument, changes its position (starts its journey in a place
and ends up being in another place), which corresponds to the verb “arriving”. The moment in
which the letter ends up being in its final location coincides with the telos.
Then, we can say that the argument that measures the event is always the IA. Never the EA.

Stative events
In a stative event, IA does not measure the event:
“John knows Mary” 🡪 there is no process that the IA Mary undergoes that measures the event.
The distinction between unaccusatives and unergatives is related to this fact: the syntactical
symmetry in the position where the IA is externally merged as opposed to the EA corresponds to
this property of measuring the event.
Unergative verbs do not allow the IA to measure the event.

Telic & Atelic switch

Verbs like eating allow certain property of the IA affects the telicity of a predicate: let’s take the
former example “John ate a pancake” and compare it with the form “John ate pancakes”.
In the first case we have an action that starts with the first bite of the pancakes and ends with the
last bite of the pancake. We consider it a telic event.
In the second case we do not know how many pancakes John ate: maybe 2, maybe 45. The IA
(pancakes) is a bear plural. Therefore, we consider it an atelic event.
We can express the difference between telic and atelic actions considering two different ways to
indicate time:
Telic Atelic

36
 In x time For x time

John ate a pancake in 5 minutes John ate pancakes for 5 minutes

*John ate a pancake for 5 minutes *John ate pancakes in 5 minutes

The sentences marked with * are wrong.

Intuitively, we can say that an even happens in 5 minutes, or that an event lasts for 5 minutes; but,
“in” has to be anchored, measures, the interval between the beginning of the event and the reaching
of the telos; while “for” measures the duration of the event without including a telos.
If we combine “for” with telic predicates, it has the property to turn the predicate into an atelic one:
“John ate the pancake for 5 minutes” means that the John did in fact eat the pancake for the duration
of 5 minutes, but maybe he did not finish it, we do not know at which point of the pancake he did
stop (the telos is not reached, therefore, the sentence turns into atelic).
Same with “a letter arrived 🡪 in two days” (telic); “letters arrived 🡪 for two days” (atelic).
All of these properties can be analyzed assuming the composition projecting the IA.

Psych verbs
Another distinction has to do with psych verbs (Belletti, Rizzi, 1988)
● This worries John
When we have this kind of verbs, we have a causer (this) which is the EA, and an entity that is
affected by the verb (John) is the IA. 🡪 non-stative.
● Lucy hates John
In this other case, is the EA that is affected by the verb (Lucy), the theme; while the IA (John). So, it
expresses exactly the opposite order 🡪 stative.

Verkuyl and Tenny’s analysis

According to them, we can distinguish two parts in an eventuality: the v head regards the subject
(John eats) and the VP (with V head and NP/DP) involving the object, and therefore, implicitly, the
change of state of the IA (eats the pancakes).
Is like if there were two subcomponents of the event “eating”, such that one is the causing event
(involving subj. EA, NP1) and the other is the changing event (involving obj., IA, NP2).
E3 = E1 (causing) + E2 (changing)

Example: John washed Lucy.

⟦VP⟧ = λe2 · e2 washing of L.

37
⟦v⟧ = [λp<ε, t>27 · λze · λe3 · (∃e2) P (e2) = 1 & (∃e4) z agent (e4) & e4 causes e2 & e3 = e2 + e4]*28
= takes as input a property of events and defines an event e3 which is the sum between event e2 VP (changing) and
e4 v (causing), and the event e4 is caused by agent z which is the subject of the sentence NP1.

R. Ramchand (2008) was one of the promoters of this analysis.

Lesson 5
2.12
In this lesson we will see some properties of predicates and nouns.
In detail, we will discuss the notions of:
● Cumulativity
● Divisivity
● Quantization

Predicates
Atelic (non-stative) predicates and stative predicates can be associated with cumulativity and
divisivity;
Telic (non-stative) predicates can be associated with quantization.

Atelic predicate:
If is say “John ran from A to C”, I am implicitly saying:
E.

But, in order to reach C, from A, there is certainly a middle place B that John reaches before
continuing in the direction of C.
It is implicit, that, in order to reach C, there are two middle steps (I) and (II) in which John goes
from A to B, and then from B to C.

27
Where P denotes property of events
28
Contains an error. Correct structure will be given in the next lesson. For this lesson the important thing is to
get the general idea.

38
What happens in I. is equivalent to what happens in II. : the two intervals are equal, and if we sum
them we will obtain the whole path that John made from A to C.
John ran from A to B (I.) and John ran from B to C (II.) are both fractions of the whole activity that
John made running from A to C (E). Therefore, they are two different intervals (or, subevents) of the
whole event that happens from A to C.
If we overlap (I.) and (II.) we will obtain two equivalent events:
III.

The property of cumulativity applies to atelic and stative predicates and refers to the fact that
summing up the two events (I. 🡪 A to B) and (II. 🡪 B to C) we obtain the whole event E.

Telic predicate:
With telic predicates this does not happen. Remember, in fact, that telic predicates relate to events
that have a start and an ending (coinciding with the telos). Therefore, if we try to do the same thing,
we will get to the conclusion that it is impossible to reach a unique, whole, entity by adding parts of
a telic event: a telic event always refers to an event that reached a telos, a final point. Therefore, if
we consider the sentence with the telic verb:
“Between 12 and 14, Mary grew up” we will not have a uniform event that can be equally divided into
equivalent, summable parts.
E.

39
If we try to overlap the two subparts of the event, we will see that (I. 🡪 12 to 13) and (II. 🡪 13 to 14)
are not equivalent, not uniform: the path that Mary made between her 12 and her 13 years is not
equivalent to what she did between her 13 and 14 years:

III.

The idea is that we represent part-hood relations (parthood is expressed by the symbol ≤), and if
these parts are uniform, equivalent one to each other we use <. In telic predicates they are not
equivalent, therefore, there is no cumulativity property.

Parthood 🡪 x is a proper part of y if x is a part of y and x ≠ y

“Proper part” (binary relation) must necessarily be:29

● Reflexive: ∀x ∈ De , x ≤ x
For every x belonging to the domain, x is a part of itself

● Transitive: ∀x ∀y ∀z [x ≤ y, y ≤ z → x ≤ z]
For every x, every y and every z, if x is a part of y and y is a part of z,
then x is a part of z

● Antisymmetric: ∀x ∀y [x ≤ y, y ≤ x → x = y]
For every x and every y, if x is a part of y and y is a part of x, then x = y

Overlapping:
Overlapping is expressed with the symbol ∘:
x ∘ y : = ∃z [z ≤ x, z ≤ y]
x overlaps with y if = there is an entity z such that z is a part of x and z is a part of y

sum (x, P) ↔ ∀y [y ∘ x → ∃z [P(z) & z ∘ y]]

x is the sum of P30 if and only if everything that overlaps with x overlaps also with z in P.

29
Pag 371 of Aloni, Dekker.
30
P denotes a set of entities.

40
What follows from these assumptions is that:
∀P ∃! z [sum relation (z, P)]
for every set of entities P there is a z such that the sum relation of P returns z.

⊕ = sum operator which takes the parts present in set P and returns the whole P with the three of
them as one entity. Therefore:
∀P ∃! z [z = ⊕ P]
for every set P only one entity z corresponds, such that z is equal to the sum of the elements of P.

In less technical terms, the sum (⊕) is a function that applies to the set P and returns a single
entity (= single event) from the union of all the elements of P.
P = set = (<e, t>);
⊕P = entity = (e).
P = {a}{b}{c} 🡪 atomic entities
⊕P = {a}∪{b}∪{c} = z.
(where a, b, c are atomic entities, and their sum is a unique entity).

Join operation in semi-lattice structure

This schema shows three atomic entities a, b, c
that undergo what is called by Aloni, Dekker the
join operation, which is
Associative: (a+(b+c)) = (a+ b(+c))31,
Commutative: a⊕ b = b ⊕ a,
Idempotent: a ⊕ a = a,

In lattice structure what we have an operation that is defined as a primitive but it gives us the same
result as the sum relation we have explained: summing a and b, a and c, a and b, we will obtain a+b,
a+c and b+c, that, summed again between them, will return a unique entity a+b+c (join).
In order to have the join operation to be entirely equivalent to the sum operation, we must add the
property of the so called unique separation:
Unique separation: ∀x ∀y [x < y → ∃! z [¬z ∘ x & y = x ⊕ z]]
for every x and y, if x is a proper part of y then this entails that there is an element z such that z is
not overlapping with x and y is the sum of x and z.

This means that whenever x is a proper part of y there is one entity z which, summed up with x, will
give y as a result.

31
That is: join of (a & b) + c = join of a + (b & c). General schema: Aloni, Dekker, pg. 374.

41
Definition of cumulativity
P is cumulative if and only if, for every entity x and y, if x has the property P and y has the
property P, then the sum of x and y also belongs to P.
P is cumulative ↔ ∀x ∀y [P(x) & P(y) → P(x ⊕ y).
Returning, as instance, to an event as “running”, we can say that:
Run 🡪 run(e1) & run(e2) = run(e1 ⊕ e2)32
Stative predicates work in the same way as atelic predicates, while telic predicates do not.

Divisivity
Divisivity is related to cumulativity, and in fact, it applies to statives and atelic predicates as
cumulativity.
A predicate is divisive if, for every y and every x, if x belongs to P and y is a proper part of x,
then P also applies to y.
P is divisive ↔ ∀x ∀y [P(x) & y → P(y).
In other words, every proper part of an x which is in P, is also in P.
While the cumulative property applies from the low upwards in the lattice structure, the divisive
property applies from the high downwards:
Cumulativity Divisivity

P(a) & P(b) → P(a ⊕ b) P(a ⊕ b) → P(a) & P(b)

Divisivity deals with the problem that, going downward dividing the whole event is subparts, there
is a moment, the very initial part of the event, that cannot be further divided in parts (in the case of
John running, let’s say, the very moment in which John lifts his foot to start his walk). This problem,
of course, occurs at the lowest level of the structure (at a certain point we are forced to stop dividing
the parts of the event in other subparts).

32
Any two events described by run can be summed up if they relate.

42
Quantization
Quantization is the opposite of divisivity.
P is quantized if for every x and every y, if x has the property P and y is a proper part of x,
then y does not have the property P.
∀x ∀y [P(x) & y < x → ¬P(y)]
A predicate P is quantized if and only if whenever it holds of something, it does not hold of any of its
proper parts.
Quantization applies to telic predicates: any part of an event that reaches a telos is not equivalent to
another part, because there is no uniformity between parts.
If we consider a telic event such as “drawing a circle”, this distinction will be clearer.
ATELIC: run: run1 + run2 = run3 = (run1+run2);33
TELIC: drawing a circle: drawing a circle1 + drawing a circle2 = drawing two different circles.
This happens because a telic event presupposes a telos, a final part which is reached. We cannot
imagine a telic event without a telos: in order to be telic, the predicate must indicate the whole
action of drawing from the first to the last point of the circle.

Nominal predicates
All this discussion will become clearer to this point: we started from predicates distinguishing
among telic (quantization), atelic and stative (cumulativity and divisivity); but actually, this analysis
also applies to nouns.
We have quantization in case of singular countable nouns; and we have cumulativity and divisivity
in mass nouns and plural count nouns.

Singular countable nouns

If I have a singular countable noun as “table”, I will have a set of atomic sentences P:
⟦table⟧ 🡪 {table1, table2, table3} = 3 tables (Quantification)
Here we see as there are no proper parts that can be described as equal components of a table;
instead, we have, as sum of the entities of the set, three different entities (no singular in join
semilattice structure).

33
Look at page 38 where John is happily running.

43
Mass nouns
If I consider a mass noun as “water”, I will have a different situation:
⟦water⟧ 🡪 {water1, water2, water3} = water (Cumulativity and Divisibility)
Here, join semilattice structure without bottom (atomic) layer (in natural language water is
considered as infinitely divisible; there is no problem of divisibility).

The cumulativity or non-cumulativity of predicates affect the type of IA: whenever we have a
predicate that corresponds to a mass noun, we always have an atelic predicate.

● Mass: “John drinks water” 🡪 atelic

When we have a count noun, we have a telic predicate.

● Sing. Count. “John drinks a glass of water” 🡪 telic34

Plural countable nouns

With plural nouns we see that something similar to mass nouns happens. In this sense, despite a
morphological appearance, plural count nouns are more similar to mass nouns than singular count
nouns.
⟦tables⟧ 🡪 {table1, table2, table3} = tables (Cumulativity and Divisibility)

34
“Glass” denotes a measure, not really a proper glass 🡪 measure function.

44
Closure undersum
Plural is part of inflectional morphology. Therefore, we need an operator that maps a set of atoms
into a join semi-lattice structure: the closure undersum.
*P = {x |∃P ⊆ P [P’ ≠ ∅ & x = ⊕ P’]}
Set of entities x such that there is a subset P1 of P such that P1 is non-empty, and x is identical the sum of P1.

It means that we start to take all the subsets of our original P and to sum up the elements in the
subsets.
*{a, b, c} = {a, b, c, a ⊕ b, a ⊕ c, b ⊕ c, a ⊕ b ⊕ c}.
In generating the closure undersum, we will obtain the entire structure including the original
atoms.

The idea is that the plural is an operator that applies to a set of atoms and returns the entire
semi-lattice 🡪 [+Pl[table]] = [tables]. In this way, the plural acquires the cumulative and the
divisivity properties.

Comparing: Singular & plural countable, and mass nouns

Article Determiner (s.) Determiner (pl./m.) Determiner (general)

SING. COUNT *ate candy *Lot of candy Every candy No candy
PLUR. COUNT ate candies Lot of candies *Every candies No candies
MASS drank water Lot of rice *Every rice35 No rice

As is possible to see in the table, some elements can be applied to all kinds of nouns36 (as the
determiner “no”), but, if some are not appliable to all of them, the opposition is between singular
nouns on one hand and plural/mass nouns37 on the other.

35
Unless singular.
36
As always, forms marked with * are considered to be wrong, ungrammatical.
37
Of course, mass nouns do not have plural unless it is a fixed expression (the waters of the ocean).

45
Recap
Cumulativity & Quantization
Divisivity
Predicates Atelic Telic
Stative
Nouns Mass Singular countable.
Plural countable

Lesson 6
3.12
In the last lesson we spoke about how certain predicates can become telic when combined with a
quantized IA, and atelic when combined with a cumulative IA.
The, we used the semi-lattice structure to properly characterize cumulativity (and divisivity).
Cumulativity is the property to sum up the single atoms of which a predicate (event or noun) is
composed and to be able to return a unique, whole entity (from atoms it goes upwards until the top
layer). Divisivity is the same property seen from the opposite point of view: we can take a unique
predicate (event or noun) and subdivide it into different, equivalent parts until reaching its atoms
(starting from the top layer to the bottom layer composed by atoms) ; these two properties apply to
atelic events (divisible because they do not reach a telos), mass nouns (because uncountables) and
countable plural nouns (because they can divided in singular nouns).
Quantization is when all of this is not possible because we cannot divide a single event or noun in
different parts, and, therefore, we do not have equivalent atoms to sum up to reach a whole entity. It
happens with telic verb (because, reaching a telos, they imply a “path” from the beginning to the end
of the action, and therefore a telic event is not divisible, otherwise, if we cut the telos, it would have
been an atelic event) and with singular countable nouns (because they cannot be divided into
equivalent atoms).
Closure undersum 🡪 with plural countable nouns we need an operator that maps the atoms of the
bottom layer to the entity revealed in the top layer of the semi-lattice structure.
[table1][table2][table3] (atoms, bottom layer) 🡪 [tables] (final entity, top layer).
This operator is the closure undersum, and its job is to sum up all the possible subsets of the atoms,
in order to return the plural entity:
table1,
table2,
table3,
table1+ table2,
table1+ table3,
table2+ table3,
table1 + table2 + table3.

46
The closure undersum does not apply to mass nouns because it is not necessary: the atoms are
identical to the final entity (look at example “water” pg. 44), therefore, the pluralization operator is
vacuous here.

Scalar implicature
In pragmatic, scalar implicature is an implicature that attributes an implicit meaning beyond the
explicit or literal meaning of an utterance.
Let’s give an example:
1. “I have got candies”
2. “I don’t have any candies”
3. “Do you have candies?”
If we consider sentence 1., it literally means “I have one or more candies”. In fact, if someone asks us
question 3. (Do you have candies?), and we only have one candy, we will answer “Yes! I have one.”, we
would not answer “No! I have one”. This, because we automatically consider the question as “Do you
have one or more candies?”. In other words, the plural “candies” entails the singular “candy”.
Therefore, the assumption “I have got candies” entails that I have one candy (and possibly other
candies). In fact, “I have got candies” is true if I have only one candy (this is what is meant by “scalar
implicature”).
But no speaker would say “I have got candies” to indicate that he only has one candy. This happens
because, in the actual world, saying such sentence in itself implies the negation of not having
candies, and the negation of having just one candy. In other words, it is faster to say with respect to
“I have got candies but I do not have got just one candy”.
For the same reason, if we have just one candy, we would say “I have got one candy”, and not “I have
got candies but I only have one”.
In fact, the sentence “I have got one candy” is more informative, and stronger, than the sentence “I
have got candies” used to indicate that I have one or more candies.

1 candy 5 candies
I have got (1 or more) candies T T
I have got one candy T F
I have got candies (pl.) F T

Since it is more informative, the sentence “I have got candies”, even though in principle entails the
possibility to have one or more candies, is commonly used with the meaning of excluding that the
person has one candy, and it is perceived as denoting just the possibility that the person has got
more than one candy (because it is a stronger specification).
Another example can be done to clarify the situation:
● “I have three kids”.
If I am a parent and I have three kids, for sure I have one kid, and for sure I have two kids. And
maybe I also have a fourth and a fifth kid. But this sentence is intended in a sense that excludes that

47
I have more than three: it implicitly denies that I have more than three, otherwise I would have said
“I have four kids”, or, “I have five kids”.
This phenomenon has to do with pragmatics. Intuitively, even though a sentence like “I have three
kids” could imply that I have more, in pragmatics, we perceive it as we have only those three kids.
Same, even though “I have got candies” could, in principle, imply that I only have one, it is excluded
in a pragmatic interpretation, and the sentence is necessarily interpreted as if I have a plural
number of candies.

Downward monotone
Downward monotone is an entailment from a superset to a subset. The presence of negation
reverses the degree of implication.
Let’s give an example:
Superset: Italy
Subset: Rome
“I am from Rome” entails that I am from Italy. Subset 🡪 Superset
“I am not from Italy” entails that I am not from ¬Superset 🡪 ¬Subset
Rome.

“I am not from Rome” does not necessarily entails that I am not from Italy (maybe I am from Siena).
“I am from Italy” does not necessarily entails that I am from Rome (maybe I am from Siena).

Maximality effect and Uniqueness requirement

Think about the sentence “The x is in the middle of the room”, where “x” can be substituted by any of
the subsequent elements:

● Plur. count noun: ⟦tables⟧ 🡪 maximality effect

● Mass noun: ⟦furniture⟧🡪 maximality effect
● Sing. count noun: ⟦table⟧ 🡪 uniqueness requirement

Then we will obtain:

● The tables are in the middle of the room
● The furniture is in the middle of the room
● The table is in the middle of the room
Intuitively, the first two sentences (PlN tables and MN furniture) will indicate that all the elements
(atoms) are in the middle of the room.

48
Intuitively, the third sentence implies that just one of the tables is in the middle of the room, while
the others are not.
Therefore, if I can represent the sentences as:

● ⟦tables⟧ 🡪 table1+ table2 + table3.

● ⟦furniture⟧🡪 furniture1 + furniture2 + furniture3.
● ⟦table⟧ 🡪#

In fact, we will need something that allows us to identify which table is in the middle of the room, or,
something that allows us to distinguish the table that is in the middle of the room from the tables

49
that are not. This is called “the uniqueness requirement”. In fact, we cannot just say that one casual
table is in the middle of the room, since it is not the blue-dotted table, nor is the orange-dotted
table. It is the green-dotted table that is in the middle of the room, and in order to identify it we
must use the uniqueness requirement.
When we use a definite description, there must be only one relevant entity (that we must be able to
distinguish from the others) that can be referred to.
In practical terms, we can apply a function between sister nodes from the specifier to the head in
order to derive from the atoms (bottom layer of the semi-lattice structure) the final entity (top layer,
or join of the semi-lattice structure).
Let’s see how it works:

Mass nouns

The determiner D, the, takes directly as input the NP furniture.

Remember: in mass nouns, “furniture” represents the join semi-lattice with no
bottom layer (no atoms).
Maximality effect: furniture (as join semi-lattice) refers to the maximum amount
of furniture that we obtain summing up the atoms (furniture1 + furniture2 +
furniture3).

Plural nouns
In plural nouns, the determiner D the, takes as input the plural projection Pl.P
table+s.
Remember: in plural nouns, the combination of N table with “Pl –” coincides
with the bottom layer (atoms), while the combination of N table with “Pl +”
coincides with the join semi-lattice (top layer).
Maximality effect: tables (join semi-lattice) refers to the maximum quantity
of tables that we obtain summing up the atoms (table1+ table2 + table3).

Singular noun
When we see the join semi-lattice of a singular noun (table) we cannot say that it refers to the
maximum quantity of tables that we obtain summing up the atoms. In fact, as we said before, in
such nouns we obtain the uniqueness requirement instead of the maximality effect.
In order to obtain the uniqueness requirement, that allows us to establish that the join semi-lattice
coincides with a single particular entity selected among others, we refer to the notion of Supremum
operator.

50
The supremum operator
Let’s see how it works.
This is our semi-lattice structure.
Supremum: is the unique element of which all
the other elements in the structure are part,
namely, the top layer (join semi-lattice). It is
often denoted by P or by σ.38 It indicates the
semi-lattice structure.
We can define it as:
σ(P) = Jx : P(x)& ∀y [y ∈ P → y ≤ x]
sigma (semi-lattice structure P) is equal to = the unique
entity39 x such that x belongs to the semi-lattice structure
and all the other entities y belonging to P, are proper or
improper parts of x.

In other words, the supremum is a unique element x, and it belongs to itself. All the other entities of
the structure are proper or improper parts of x.
The proposal for singular nouns is that the denotation of the definite determiner is in fact a
supremum operator.
In other words, we build a function that goes from a set of entities (that coincides with the
structure) to a unique entity. This function takes as input the whole semi-lattice structure and
returns the entity of which all the other entities of the structure are parts.

In terms of semantic types, the supremum is: <<e, t>, e> (function from a set of entities <e, t> to an
entity <e>).
When we apply the supremum operator to a mass noun, or to a plural noun, we will obtain the
maximality effect (the unique entity that is the sum of all the lower entities of the semi-lattice
structure).
When we apply the supremum operator to a singular noun we will obtain:
● σ {a, b, c} = undefined,
because the function looks for one element of the set of which everything is a part, but in singular
nouns, atomic entities are no subparts of something else, just of themselves (a is only a proper part
of itself, b is only a proper part of itself, c is only a proper part of itself).
For singular nouns we must apply the supremum operator to a single entity, in order to obtain a
unique entity that is a proper subpart of itself
● σ {a} = a

38
Sigma (σ), denotes the semi-lattice structure.
39
Jota (J) is a symbol meaning “the unique entity”.

51
Therefore, in dealing with single nouns, we build a function that takes as input a singleton set
(exactly one element, that is, in the case of tables, the green-dotted table) and returns the
uniqueness effect (that coincides with the selection of that particular element).
In other words, the set of atoms of the semi-lattice structure of a singular noun consists just in a
single atom (the green-dotted table), which is part of itself (because the parthood relation is
reflexive), and this is why it can return the unique entity corresponding to itself.

Recap
o Mass nouns/Plural nouns = applying the supremum operator, the function goes from the single
atoms a unique entity of which all the other entities are part.
σ {a, b, c} = maximality effect

o Single nouns = applying the supremum operator, the function goes from the only atom and returns
the unique entity that coincides with the only atom, which is a proper part of itself.
σ {a} = uniqueness effect

Compositional analysis (plural noun)

⟦NP⟧ = ⟦table⟧ = λxe · x table

= entity x such that x is a table (<e, t>). Denotes the set of entities x, which is true if x is a table.
⟦+Pl⟧ = λf<e, t> · λye · ∃ X ⊆ {ze | f(z) = 1} y = ⊕X
= the plural is the closure undersum operator (it is an inflectional feature of the noun): it has the function to add
the singular elements and to return the unique entity at the top of the semi-lattice structure. It will take as input
the set characterized by the function denoted by the NP (<e, t>), which is the set of atoms denoted by the NP
“table”, and it will return, at the level of its mother node, another set which is larger, because composed by the
all the atomic entities and their summed up subsets. For this reason, +Pl is an operator from an <e, t> type to
another <e, t> type. Since Pl+ belongs to the semantic type <<e, t>, <e, t>>, the most external lambda prefix (λf)
will bound the variable corresponding to the input of the function +Pl. In this case, <e, t>.
Therefore, we write “λf<e, t>”, which is the valence that must be saturated by the NP (<e, t>).
So, the function takes as input “λf<e, t>”, and it will return an <e, t> valence. This is why we put another lambda
prefix binding a variable of type e (λye) and then we specify the truth-conditions t. 🡪 such that there is a set X

52
 which is a subset of the set characterized by f (f of λf<e, t>), that is [the set of entities z such that f applies to z and
returns the truth-value true], such that y is identical to the sum of the entities in X, or, in other words, y is the
result of applying the sum operator to the elements of the set X.
To summarize all the reasoning, we have an element f, saturated by NP (λf<e, t>) that corresponds to the input of
the function <e, t>, then we have another <e, t> as output. Therefore, we assign a lambda to the entity of the
output e (λye) and the truth-value of the output consists in the fact that there is a set X (which is part of the NP,
because it contains all the atoms) such that y (the entity of the output) corresponds to the sum of all the
elements of such set X (that is part of NP).

⟦Pl.P⟧ = ⟦Pl+⟧ (⟦NP⟧) = [λf<e, t> · [λye · ∃ X ⊆ {ze | f(z) = 1} y = ⊕X]] ([λx
e · x table])
= in this passage we copy the ⟦Pl+⟧ and the ⟦NP⟧, and we saturate the most external
lambda prefix with the NP.

= [λf<e, t> · [λye · ∃ X ⊆ {ze | f(z) = 1} y = ⊕X]] ([λxe · x table])

= here we drop the lambda f and substitute all the occurrences of f with the NP ([λxe · x table]).

= [λye · ∃ X ⊆ {ze | [λxe · x table](z) = 1} y = ⊕X]

= let’s substitute x of NP with the variable z.

= [λye · ∃ X ⊆ {ze | [λxe · x table](z) = 1} y = ⊕X]

= at this point we know that, in order to return 1 (true) z, that saturates x, is a table.
So, I rewrite the formula dropping “=1” because it is entailed by the substitution of x with z.

= [λye · ∃ X ⊆ {ze | z table} y = ⊕X]

= what we have obtained now is that Pl.P that has the semantic type <e, t>, where the entity e
is “λye” and the truth-value t is “there is a set X which is a subset of the entity “table”, such
that the entity y is equal to the sum of all the subsets of X.”. In other words, we have obtained
the characteristic function that corresponds to the entire semilattice structure (taking an
atom y as input and returning the sum of all the possible subsets of the set X composed by the
atoms). Now let’s proceed with the denotation of the definite determiner “the”.

⟦D⟧ = ⟦the⟧ = [λf<e, t> · Jy : f(y) = 1 & ∀ue : f(u) = 1 → u ≤ y]

= the determiner has the semantic type <<e, t>, e>: in fact, the first part <e, t> is going to be saturated by
the Pl.P, and the output, at the level of the mother node DP must be a unique entity e. Therefore, the
semantic type of the determiner must be <<e, t>, e>.
So, again, we will have: a lambda prefix binding a variable f <e, t> (input), and returning as entity the
unique entity y such that f (<e, t>) applies to y (e) and returns the truth-value true (1); and all the
entities u that belong to the set f are parts of y (y must be the unique supremum).
In other words, we have an input that corresponds to a set f, and the output is an entity y, of which the
set f is part, and every element (u) of the set (f) is part of the entity (y).

⟦DP⟧ = ⟦D⟧ (⟦Pl.P⟧) = [λf<e, t> · Jy : f(y) = 1 & ∀ue : f(u) = 1 → u ≤ y] ([λye · ∃ X ⊆ {ze | z table} y =
⊕X])
= here there has been a technical problem: we wrongly used the variable y for both the D and
the Pl.P. Let’s simply substitute y with another letter in the Pl.P just not to confuse it with the D
variable.

= [λf<e, t> · Jy : f(y) = 1 & ∀ue : f(u) = 1 → u ≤ y] ([λxe · ∃ X ⊆ {ze | z table} x = ⊕X])

53
 = giving Pl.P as input of D, as always, we drop the most external lambda prefix and we substitute
all the occurrences of the corresponding variable (f) with the Pl.P.

= [λf<e, t> · Jy : f(y) = 1 & ∀ue : f(u) = 1 → u ≤ y] ([λxe · ∃ X ⊆ {ze | z table} x =

⊕X])
= at this point we drop the first lambda and actually substitute the circled f with the Pl.P in bold.

= Jy : [λxe · ∃ X ⊆ {ze | z table} x = ⊕X](y) = 1 & ∀ue : [λxe · ∃ X ⊆ {ze | z table} x = ⊕X](u) = 1 → u ≤
y
= now I search if there is any variable that can be substituted, and I see that I have two of them (“y” and “u”).

= Jy : [λxe · ∃ X ⊆ {ze | z table} x = ⊕X](y) = 1 & ∀ue : [λxe · ∃ X ⊆ {ze | z table} x = ⊕X](u) = 1 → u ≤
y
= here I simply copied the same formula highlighting in bold the variables (y and u) that will substitute the circled
variables (x). In the next formula, as always, I will also drop the most external lambdas.

= Jy : ∃ X ⊆ {ze | z table} y = ⊕X = 1 & ∀ue : [∃ X ⊆ {ze | z table} u = ⊕X = 1 → u ≤ y]

= now, since I must verify that both the first part and the second part are true (=1), I rewrite the formula simply dropping
the =1 and stating an equivalence between the first and the second part.

= Jy : ∃ X ⊆ {ze | z table} y = ⊕X & ∀ue : [∃ X ⊆ {ze | z table} u = ⊕X → u ≤ y]

= this formula describes the unique entity y (Jy) such that y is the sum of the members of the some subsets of X set (⊕X),
which atoms are z (table), and the subsets of all the other members of the set X are parts of y. Namely, it is the unique
supremum. In this case, we obtain the maximality effect (Plural: tables).

⟦DP⟧ = Jy : ∃ X ⊆ {z e | z table} y = ⊕X & ∀ue : [∃ X ⊆ {ze | z table} u = ⊕X → u ≤ y].

54
Week 3
Lesson 7
9.12
Abbreviations
In this lesson we will introduce some forms of abbreviation to denote some of the passages we went
through in the previous lesson’s exercise.
= λf<e, t> · λye · ∃ X ⊆ {ze | f(z) = 1} y = ⊕X
⟦+Pl⟧ = this was the denotation of +Pl that we used in the previous
exercise.
In fact, it is a semantic type <<e, t>, <e, t>>. So, the first lambda f
represents the input <e, t>, the second lambda y represents the
entity of the output and then we have the truth-value, that is
“there is a set X which is a subset of f (composed by elements z
that are part of f), such that y is equal to the closure undersum of
the elements of set X.”
Therefore, we can also write it as:

= λf<e, t> · λye · x ∈ *P

= here, the star operator abbreviates the truth-values formula.

⟦D⟧= = [λf<e, t> · Jy : f(y) = 1 & ∀ue : f(u) = 1 → u ≤ y]

⟦the⟧ =this part of the formula denotes the definite description ”the”,
and we said that it means that we have an input that corresponds
to a set f, and the output is the unique entity y, of which the set f is
part, and every element (u) of the set (f) is part of the unique
entity (y).”
Therefore, it corresponds to:
= [λf<e, t> · Jy [y = σ {ze | f(z) = 1}]40
= that is, λf<e, t> that returns as output the unique entity y such that
y is the supremum of the set f of entities z.

This can be abbreviated as:

= λf<e, t> · Jy [σ (y, f)]41, or

= λf<e, t> · Jy [y = σ(f)]
= The part into squared parenthesis means that y is the
supremum of the set f. The whole formula can also be written as:

= λf<e, t> · σ(f).

= it literally means that the input is the characteristic function
from entities to truth-values represented by λf<e, t>, and the
output is the supremum of the set f.

40
Here sigma is used as function from the set f and returns an entity y (supremum).
41
Here sigma is used as relation between an entity y and the set f.

55
Correction of exercises
1. Which of these nouns have cumulative reference?
silver, moon, clouds, branch, foliage, leaves, tree

2. Take the denotation of "tree" to be the set {t1, t2, t3}.

2.1 What is the denotation of the plural noun "trees"?
⟦trees⟧ = ⟦Pl.P⟧ ({t1, t2, t3}) = λx · x ∈ * 42 {t1, t2, t3}
= {t1, t2, t3, t1⊕t2, t2⊕t3, t1⊕t3, t1⊕t2⊕t3}
= function of x such that x belongs to the closure undersum of the set.

2.2 What is the denotation of the definite description "the trees"?

⟦the trees⟧ = ⟦D⟧ (⟦Pl.P⟧) = ⟦the⟧ (⟦trees⟧)
= σ {t1, t2, t3, t1⊕t2, t2⊕t3, t1⊕t3, t1⊕t2⊕t3} = t1⊕t2⊕t3
= operator that returns the supremum of the set (the entity that contains all the atoms)

2.3 And what about the definite description "the tree" (singular)?
⟦the tree⟧ = # undefined (uniqueness requirement)
in this case the result is undefined because if we apply the supremum operator σ to “the tree”, this will look
for the element of the set of which all the others elements are part. But, since it is a singular count noun,
there is no element of which all the other elements are part. Therefore, we say that there is no supremum.
This is why the definite determiner is not able to select the unique entity that is the sum of all the other
entities. This is why this remains undefined.
⟦the tree⟧ = ⟦the⟧ (⟦tree⟧)= σ {t1, t2, t3} = #
When a definite description has no denotation, any sentenece that contains this definite description will not
express truth-conditions. Because we cannot denote the element that the supremum is looking for.
In these cases we can consider the supremum operator as a partial function, because it takes as function
only sets that have a supremum, namely: plural count nouns and mass nouns.

The only case in which is possible to have a supremum for a singular count noun, is if the set contains only
one element. In fact, in that case, all the elements of the set (the only element) will be the entity containing
all the other entities (always itself).
σ {t1} = t1 is the supremum operator of itself = t1 ≤ t1.

Formalization by Heim, Kratzer

“rich” here denotes a characteristic function of a set

(discrimination between rich and non-rich people). However,
being a grade adjective, an adjective like “rich” entails a certain
level of vagueness: I can be richer than someone and poorer
than someone else, therefore, my richness can be debatable, or,
it depends on the context whether I am considered. We can,

42
* stands for the closure undersum

56
anyway, consider “rich” as a characteristic function from entity to truth-value (Rich: person 🡪 T or
F)

⟦AP⟧ = ⟦A⟧ = ⟦rich⟧ = λx · x’s wealth exceeds t43

The next node we are going to analyze is Aux “is”. In this sentence, the relevant nodes are NP and AP:
in fact, if I say “John rich” (even though it is ungrammatical) I am already conveying the information
that John is rich. The Aux node (is) does not really bring new semantic information in the sentence.
In other, more specific, words, if rich is an <e, t> semantic type, and the NP is an <e> semantic type,
combining the two of them I will already obtain the truth-value at the level of AuxP. For this reason,
Aux “is” is considered to be semantically empty.
But remember that it is not possible to combine two nodes that are not sister nodes!
Therefore, how do we treat the (semantically empty) Aux node in order to obtain the right
denotation of its mother node?
There are two ways to treat a semantically empty node:

Using the T-model44:

The idea behind the T-model is that the structure that we interpret is a structure that starts
from the lexicon and at a certain point we have a split in two branches: on one side we have
the building of what is spelled out in the sentence, or, the perceptible representation of the
sentence (phonological form); and, on the other branch, we find the logical form, or, a
transformation of the tree that yields to a structure that can be interpreted.
We start from the lexicon with the syntactic operation. At a certain point the structure
shows its phonological interface (the sentence is translated into a sequence of phonemes);
on the other hand, we continue modifying the syntactic tree until we get to a point in which
is possible to interpret elements by semantic compositional rules. The final syntactic
representation contains some elements that are different from the ones of the phonological
representation of the structure. In the last stage of the logical form, certain phonological
elements, as the auxiliary, simply got deleted.

Using Identity function:

We assume that Aux is simply semantically empty, and that, for this reason, it takes as input a
predicate P of type <e, t> and returns as output exactly the same semantic type <e, t>, at the level of
its mother node.

🡺 ⟦Aux⟧ = ⟦is⟧ = λP<e, t> · P (identity function)

43
“t” here is a contextual parameter denoting a scale.
44
Pg. 185 Heim, Kratzer.

57
Analyses:
John is fond of Mary
The AP will have an <e, t> denotation, which will
be inherited by the previous node Aux’ because
its sister node Aux is semantically empty. Then,
the Aux’ will take as input the denotation of its
sister node NP1, which is an entity, and it will
return a truth-value at the level of AuxP.
How are we going to analyze the lower part of
the structure AP (fond of Mary) in such a way
that it denotes an <e, t> semantic type?
We know that the proper noun Mary, NP2,
denotes an entity. P is semantically empty,
therefore, it will return <e> at the level of PP.
Then, we must combine P with its sister node A.
A is analogue of a transitive verb, therefore, it
has a function nest of the semantic type <e, <e,
t>>, which will be saturated by the entity of the
PP, returning <e, t> at the level of its mother
node AP.

Therefore, the compositional analysis of this sentence is:

⟦NP2⟧ = ⟦N⟧ = ⟦Mary⟧ = (m)
⟦P⟧ = ⟦of⟧ = λx · x45
⟦PP⟧ = ⟦NP2⟧ = (m)
⟦A⟧ = ⟦fond⟧ = [λye · [λze · z is fond of y]]46
⟦AP⟧ = ⟦A⟧ (⟦PP⟧) = [λye · [λze · z is fond of y]] (m)
= [λye · [λze · z is fond of y]] (m)
= [λze · z is fond of m]
⟦Aux’⟧ = ⟦AP⟧47
⟦NP1⟧ = ⟦N⟧ = ⟦John⟧ = (j)
⟦AuxP⟧ = ⟦Aux’⟧ (⟦NP1⟧) = [λze · z is fond of m] (j)
= [λze · z is fond of m] (j)
= [j is fond of m]
= 1 iff j is fond of m48
45
It takes as input an entity and it returns an entity, because it is semantically empty (Identity function).
46
Remember: fond is treated as a transitive verb. Therefore, it implies a nested function.
47
Because the Aux node is semantically empty, therefore, the mother node of Aux and AP inherits the
denotation of AP.
48
“= 1” is because AuxP has the denotation of a truth-value.

58
An adjective can be analyzed as a transitive verb (nested function of the type <e <e, t>>).
Therefore, we can simply extend the treatment we use for transitive verbs49 to prepositions of the
non-functional type if they do not convey meaning (functional prepositions, distinct from lexical
prepositions that express a lexical content, like “above”, “below”).

John is in Texas (non-verbal predicate)

Here, NP2 is an entity, and it saturates the input of its sister node P
(in), which is not semantically empty, and in fact it expresses a
2-place relation (combines NP1, John, and NP2, Texas), and for this
reason it represents a nested function (<e, <e, t>>).
Then, combining NP2 with P we obtain, at their mother node PP, the
denotation of <e, t>, which gets inherited by Aux’ (because Aux node
is semantically empty).
The last passage is to combine Aux’ node with its sister node NP1,
which denotes an entity, that saturates the function of its sister node
and for this reason, the AuxP node has the semantic type <t>.
Let’s see the compositional analysis of the sentence:

⟦NP2⟧ = ⟦N⟧ = ⟦Texas⟧ = (t)

⟦P⟧ = ⟦in⟧ = [λxe · [λye · y is in x]]
⟦PP⟧ = ⟦P⟧ (⟦NP2⟧) = [λxe · [λye · y is in x]] (t)
= [λxe · [λye · y is in x]] (t)
= [λye · y is in t]
⟦Aux’⟧ = ⟦PP⟧
⟦NP1⟧ = ⟦N⟧ = ⟦John⟧ = (j)
⟦AuxP⟧ = ⟦Aux’⟧ (⟦NP1⟧) = [λye · y is in t] (j)
= [λye · y is in t] (j)
= [j is in t]
= 1 iff j is in t
It is never possible to combine two nodes that are not sister nodes.

49
Without event semantics, of course.

59
Rome is a city
Here we start from NP2, which is a function from an
entity to a truth-value. In fact, city is a function from the
entity city to the truth-value 🡪 x is a city T/x is not a
city F).
Then, since both D and Aux are semantically empty, the
denotation of NP2 is inherited first by DP and then by
Aux’ node.
The next passage is the combination of Aux’ node with
NP1, which is an entity. This combination yields to the
denotation of <t> at the mother level AuxP.

⟦NP2⟧ = ⟦N⟧ = ⟦city⟧ = [λxe · x is a city]

⟦DP⟧ = ⟦NP2⟧
⟦Aux’⟧ = ⟦DP⟧
⟦NP1⟧ = ⟦N⟧ = ⟦Rome⟧ (r)
⟦AuxP⟧ = ⟦Aux’⟧ (⟦NP1⟧) = [λxe · x is a city] (r)
= [λxe · x is a city] (r)
= [r is a city]
= 1 iff r is a city.

Rome is a city in Italy (restrictive modification)

Here, the analysis of the higher part is the same
as the previous one. So, let’s focus on what
happens below the NP2 level.
The part of the sentence below the NP2 level is
called “restrictive modification”: it gives a
more restrictive definition of the subject, or, it
specifies a subset (cities in Italy) of the original
set (cities). In other words, this specification is
optional: the sentence could have ended with
the word “city”50. The further description that is
made is a way to restrict the domain we are
talking about (cities in Italy ⊆ cities).

50
As in the previous analysis, pg. 60.

60
 Starting from the lowest node, we have an NP3
that denotes an entity, and the P node denotes a
nested function from an entity to a function
from an entity to a truth-value. In fact, it
combines city (N of NP2) with Italy (NP3)51.
The entity of NP3 saturates the input of P,
returning the semantic type <e, t> at the level of
their mother node PP.
Therefore, PP has the semantic type <e, t>; but
its sister node (N of NP2) too, city, has the same
semantic type.

This situation, that sees two sister nodes having the same semantic type is called “type mismatch”,
because the two denotations do not match. So, how can we combine them such that we obtain
another <e, t> at the level of their mother node52?
We have two possible solutions to do so:

1. Assume new compositional rule (Predicate Modification Rule).

2. Modify the semantic type of one of the two sister nodes when there is a restrictive
modification (Type Shifting Solution).
Remember that we want to obtain an <e, t> semantic type at the level of NP2 to proceed
with the further analysis of the sentence.
To obtain it, we create an intersection between the set denoted by “city”, and the set denoted
by “in Italy” 🡪 ⟦N⟧ ⋂53 ⟦PP⟧
🡺 Rome is a city in Italy = truth-conditions: 1 iff R ∈ ⟦city⟧ ⋂ ⟦in Italy⟧
🡪 R ∈ ⟦N⟧
🡪 R ∈ ⟦PP⟧

Solution 1: Assume new compositional rules

Predicate modification rule (Heim&Kratzer):

We essentially introduce a conjunction rule.
⟦γ⟧ = λxe · ⟦α⟧(x) = ⟦β⟧(x) = 1
Introduce λ prefix (set of entities x, for which both the set α and the set β applied to x return the
value true).

Let’s apply it to our structure.

This is the interested piece of structure.
⟦NNP2⟧ = ⟦α⟧ = λye · y is a city

51
As in the example of pg. 59 (John is in Texas)
52
In fact, we want <e, t> of NP2 to be inherited by Aux’ and to be saturated by NP1, such that AuxP is a t type.
53
This symbol is used here as intersective modifier.

61
⟦PP⟧ = ⟦β⟧ = λze · z is in Italy
⟦NP2⟧ = λxe · ⟦N⟧(x) = ⟦PP⟧(x) = 1
⟦NP2⟧ = λxe · ⟦λye · y is a city⟧(x) = ⟦λze · z is in Italy⟧(x) = 1
⟦NP2⟧ = λxe · ⟦λye · y is a city⟧(x) = ⟦λze · z is in Italy⟧(x) = 1
⟦NP2⟧ = λxe · x is a city = x is in Italy = 1
⟦NP2⟧ = λxe · x is a city ⋂ x is in Italy = 1
From this analysis we obtain that x is both a city and located in Italy (<e, t> type at the NP2 node). In
the next step, proceeding through the full structure (shown in the previous page) the denotation of
NP2 is inherited by Aux’, and NP1 Rome is going to saturate the x of NP2, returning the final formula
(R. is a city ⋂ R. is in Italy = 1) at the level of the final mother node AuxP.
This solution, anyway, is not general enough to analyze all the possible modification structures. In
the next lesson we will see why.

Lesson 8
10.12
Solution 2: modify the semantic type of one of the two sister nodes (PP)

We know that, in order to return a semantic type <e, t> at the mother
node, its daughter nodes should have the semantic type <e, t> and <<e,
t>, <e, t>>, because one (<e, t>) is going to saturate the input of the
other one (<<e, t>, <e, t>>), and the result will be <e, t>.
Since N is necessarily a semantic type <e, t>, we try to treat the PP as if
it has the semantic type <<e, t>, <e, t>> instead of <e, t>, and to do so,
we essentially use a recursive rule.

The type-shifting rule was introduced by Barbara Partee in 198754, and proposed by Partee and
Portner in 2002.
In case that we have a noun and an adjectival modifier, there are morphological reasons, according
to Partee, to assume that is the adjective that undergoes the type-shifting rule, as in certain
languages we find a different morphological aspect when the adjective is in adnominal position,
with respect to when the adjective is in a predicative position (in the deeper analysis of the PP, “in”
would not have an <e, <e, t>> type, according to this interpretation, but a different, more complex
semantic type).
Otherwise, we can just assume that the type-shifting rule is appliable when we have two nodes that
are not matching, as an automatic adjustment (in this case, the rule is applied at this level without
considering what happens below).
Type-flexibility, the assumption that a given element can have denotations of different semantic
types, is a departure from the strict version of compositionality proposed by Montague. He

54
“Noun phrase interpretation and type-shifting principle”

62
published several works between ’60s and ‘70s of the last century, defining compositional
interpretation as isomorphic to syntactic structure, and part of this isomorphism was concerning
the fact that every syntactic category corresponds to one and only one semantic type. Partee
proposed a deviation from this strict view, allowing different syntactic categories to possibly belong
to different semantic types.
⟦PP⟧ = [λf<e,t> · [λxe · f(x) = 1 and x is located in Italy]
= lambda function of type <e,t>, there is an entity x that belongs to the function f (being a city) and is in Italy.

⟦N⟧ = ⟦city⟧ = λye · y is a city

⟦NP2⟧ = ⟦PP⟧ (⟦N⟧) = [λf<e,t> · [λxe · f(x) = 1 and x is located in Italy] (λye · y is a city)
= [λf<e,t> · [λxe · f(x) = 1 and x is located in Italy] (λye · y is a city)
= [λxe · (λye · y is a city)(x) = 1 and x is located in Italy]
= [λxe · (λye · y is a city)(x) = 1 and x is located in Italy]
= [λxe · x is a city = 1 and x is located in Italy]
= [λxe · x is a city and x is located in Italy]
In this case, as well as in the previous solution, we see an intersective interpretation of the sentence:
x is both part of set 1 and set 2. And this fact entails that x is part of set 1; and x is part of set 2.
● x ∈ ⟦PP⟧ ⋂ ⟦N⟧ (x is in Italy, and x is a city)
🡺 x ∈ ⟦PP⟧ (entails that x is in Italy)
🡺 x ∈ ⟦N⟧ (entails that x is a city)

This second solution is appliable in all cases in which modifiers give rise to an intersective
interpretation. But it is not always the case: in fact, not all of them have this behavior. Now we will
see some cases of non-intersective behavior, and we will explain why these two analysis cannot hold
with the kind of modifiers that do not have an intersective interpretation.
Let’s say that the modifiers showing a complete intersective behavior are sort of a standard, or
default case. But what about modifiers that are not intersective?

Non intersective, subsective modifiers

Some modifiers are subsective instead of intersective.
● x ∈ ⟦AP⟧ ⋂ ⟦NP⟧
🗶 x ∈ ⟦AP⟧
✔ x ∈ ⟦NP⟧

In other words, if an entity x belongs to the combination of the adjective phrase and the modified
noun phrase, this does not entail that x belongs to a set characterized by the adjective itself, but we
have the entailment that x belongs to the set of the modified noun phrase.
It is not a fully intersective behavior, but what we obtain is a subset of the set characterized by the
noun phrase.
For instance, consider the following sentences:

63
 1. John is a skillful surgeon
2. John is a pianist
In this case, the modifier “skillful” refers to John as surgeon, not as pianist. Maybe John is an
excellent surgeon but a mediocre pianist. This means that the entailment relation does not hold in
the pianist case.
These two sentences entail that:

✔ John is a surgeon
✔ John is a skillful surgeon
✔ John is a pianist
🗶 John is a skillful pianist (no)

John belongs to the set of surgeons, and also to the subset of skillful surgeons; and he belongs to the
set of pianists, but not to the subset of skillful pianists. If he had belonged to the set of skillful
people, then he would have been both a skillful surgeon and a skillful pianist. But he only belongs to
the set of skillful surgeons, therefore this does not automatically entail that he is also a skillful
pianist. In fact, “skillful surgeon” is a subset of “surgeon”, therefore, it can only refer to that domain.
We call this behavior subsective.

Implicit relativization
Anyway, we have some cases in which this adjective carries an intersective interpretation.
Saying “John is skillful” is a general statement, that means “he is skillful as a person”. But, according to
context, this sentence could be denoting different meanings: if we have “skillful” applied to some
specific characteristics, it can be that he is not skillful in all aspect of his life, as seen before.
In this sense, the denotation “skillful” is relative to the context we have (implicit relativization).
This kind of behavior cannot be captured by the purely intersective rule of Predicate Modification;
therefore, we need a more specific analysis of this kind of situation.
Let’s take as example:
“John dances beautifully”, or “John is a beautiful dancer”
The modifier is subsective, because the sentence entails that John is a member of the set of dancers,
and in particular a member of the subset of beautiful dancers, but he is not a member of the set of
beautiful people in general. Therefore, in this case too, we cannot apply the intersective analysis
proposed in the two possible solutions that we analyzed before.
✔ John is a beautiful dancer
✔ John is a dancer
🗶 John is beautiful
“beautiful” modifies, somehow, the event, retuning as outcome a non-purely intersective
interpretation, but a subsective one, instead.

64
Non intersective, non subsective
Some cases show that modifiers can be neither intersective nor subsective.
Think about:
John is a former55 student of linguistics.
🗶 John is a former (non-sense)
🗶 John is a student of linguistics (not entailed)
John is an alleged murderer.
🗶 John is an alleged (non-sense)
🗶 John is a murderer (not entailed)
In these previous examples, John does not really belong to the set of students of
linguistics/murderers (not in the present world, at least); nor to the set of former/alleged (these
are words that do not have a sense if not referring to a nominal element).
These adjectives operate in an intentional level (they make reference to a state of the world that is
not actual, in a temporal level: past, future; or a hypothetical level: possible non-actual world…).
In certain languages, modifiers with a non-intersective behavior, can precede the noun phrase but
there is no context in which they can be placed in a post-nominal position, or in a predicative
position.
For instance, “presunto”56 can be only placed before the noun it refers to (presunto colpevole, not
colpevole presunto), and it cannot be used in a predicative position (John è un presunto is a senseless
sentence without a noun).

Non intersective, gradable adjectives

Gradable adjectives as “rich”57, because of their vagueness, do not allow an intersective meaning.
Let’s take as example the sentence:
Jumbo is a small elephant
✔ Jumbo is an elephant
🗶 Jumbo is small
Being a small elephant is a subset of being elephants, therefore, it shows a subsective behavior. In
fact, the adjective “small” is relative to the fact that he is small with respect to the average size of an
elephant, but probably he is not small if compared to a person.
In the same way, saying “I saw a big butterfly” means that the butterfly is big with respect to the
average size of butterflies.
We can consider these adjectives as a scale going from one extreme to the other.

55
“Former” is collocated in past, but also “future” belongs to the same category.
56
That is, alleged.
57
See pg. 57

65
Analysis 1 of grade adjectives
Therefore, an analysis that has been proposed is to attribute to this kind of adjectives the semantic
type <<e, t>, <e, t>>, where the first <e, t> is the input “x such that x is an elephant”, and the second
<e, t> is the output corresponding to the standard of the scale (relative to the noun: small/big to be
an elephant).

Analysis 2 of grade adjectives

In her analysis, Partee arrived to the conclusion that standards for comparisons are not relevant:
she imagined Jumbo to be an elephant that lives in an imaginary world with gigantic elephants, so,
even though Jumbo is considered to be a big elephant in our world, it remains the smallest elephant
in a world of giant elephants.
The point of this anecdote is that the standard is provided by context, not necessarily from the
average size of the entity taken into consideration.
If I refer to a big snowman, it could be a giant, 6-meters snowman if I am talking about “a big
snowman” in a context of a world competition; but it could also be a 1-meter snowman if I am
talking about “a big snowman” in a context in which my three-years old son made it.
This discourse becomes important in underlying that context determines part of truth-conditions.
1. If we assume the hypothesis that the standard is provided by a function from the set of
entities to a degreed adjective, then, of course we assume an <<e, t>, <e, t>> analysis.
2. If, on the other hand, we assume Partee and Heim&Kratzer’s analysis, that the standard is
provided by context, then we could assume that even in these cases in which the gradable
adjective does not allow an intersective interpretation, in fact, the adjective is intersective
anyway, and therefore it acquires the semantic type <e, t>. The reason why we do not have
an inference is that the contextual standard shifted from the interpretation of “Jumbo is a
small elephant” to “Jumbo is small”; therefore, the second sentence becomes an inference of
the first one (Jumbo is in fact small, relatively to the context).

Analysis 3 of grade adjectives

In a scale, degrees must be considered.
This is what C. Kennedy and L. McNally stated in a study conducted in 2005 “Language”. They
provided an analysis of gradable adjectives as expressing a relation degree-individual: function
from degree to individuals <d, <e, t>>.
So, an adjective like “small” takes as input a particular degree and returns the characteristic function
of the set of entities which possesses the relevant property of that degree.

Positive morpheme
To saturate the degree (d), in English we use the positive or negative grade -er, or -less:
🡺 Jumbo is [+/-] small

66
 🡺 Jumbo is [+] small
In languages like English, this [+] is not pronounced, but according to this analysis, we would say
that Jumbo belongs to the set of entities which possess smallness to the degree specified by the
covert morpheme [+].

Comparative morpheme
The positive morpheme is in complementary distribution with the comparative morpheme
🡺 Jumbo is small[er]
🡺 Jumbo is [less] small
These degrees are used in comparison between two entities, not related to a general contextual
interpretation.
✔ Jumbo is smaller than his mother
🗶 Jumbo is small
The first sentence does not entail the second one.

Degree scales
Our analysis in enriched by the introduction of these degree scales, which are essentially abstract
objects that establish an order relation with some abstract mathematical property, which allows to
capture some relevant properties of these entities.
● An interesting discussion has rose by Kennedy and McNally’s analysis, which is related to
this argument: some adjectives have a scale that goes in one direction only, as, for example,
“noisy”: it has a minimal degree, but no maximal degree: it is open in just one direction.58 For
example, it is possible to say “this baby is completely quiet”, but not “this baby is completely
noisy”, precisely because this noise scale is closed in its minimal degree, but not in its
maximal degree.
● Other adjectives have a scale that is closed in both directions: they have a minimal and a
maximal degree. The discussion focused on the fact that adverbial and adjective modifiers
are sensitive to the degree scale.
● Another interesting case is about adjectives like “possible” that have antonyms “impossible”.
This kind of adjectives, by definition, should not be gradable (something can be either
possible or impossible), but they actually are (think of: “this is almost impossible”).
This general discussion tells us that in some cases it may be better to adopt a solution in terms of
Type Shifting instead of Predicate Modification, because some special cases cannot be treated
with intersection (mainly adjectives).
For this reason, the Predicate Modification solution (solution 1) just covers part of these
phenomena (the intersective ones), but not all of them.

58
In fact, a noise can be progressively reduced until it is no more a noise (because it becomes silence), but
there is no point in which, increasing the noise, it assumes another connotation. It would only become a
louder (and louder) noise.

67
Syntactic structures-compositional interpretation
⟦the green book⟧ = book2
σ ⟦green book⟧
“the” (σ) should impose the uniqueness requirement in this case. Since it is a
singular count noun, the supremum is undefined unless “green book”
characterizes a set of just one element.
This case can be treated as intersection between the set of books and the set
of green elements:
{book1, book2, book3…} ⋂ {book2, tree, frog} = {book2}.
Analyzing this case, we realize that if we want to attach the determiner “the”, it should be attached
on the top of the structure:

Let’s consider another example to explain in detail why is so.

The book on the table

This is a singular count noun, therefore, as our previous analysis shows, its supremum must be the
uniqueness requirement instead of the maximality effect.
⟦NP⟧ = ⟦book⟧ = {book1, book2, book3…} (set of books)
⟦PP⟧ = ⟦on the table⟧ {book1, pen1, pencil1} (set of things on the table)
⟦NP⟧ ⋂ ⟦PP⟧ = {book1}
This is the procedure we want to apply to our sentence in order to obtain the uniqueness
requirement (supremum), that corresponds to book1.

🗶 If we consider this following structure:

Is not possible to intersect the NP and the PP before applying the DP to

the resulting node, because they are not sister nodes.

68
In this structure, we would combine NP with D (the book) and then we can combine this resulting
DP with the PP. This would not allow the definite determiner “the” to look for the unique element
that is the only element of the set resulting from the intersection of NP and PP.

✔ Therefore, the structure we want to work with is the following:

In fact, this structure allows us to combine N and PP resulting in an

intersective set of books that are on the table (NP). Then, at this point,
the definite determiner D “the” can search for the only element present
in the set NP (books N that are on the table PP), returning the
uniqueness requirement.

Historical overview
This is a point stressed by Heim & Kratzer in chapter 4.
But let’s talk about the historical aspect of these analyses.
This argument was proposed by Barbara Partee, who was the most important scholar in
compositional semantics in the phase between the beginning of the ‘70s of the last centuries, when
Montague prematurely died and so he could not develop his approaches. It was Barbara Partee that
took his projects and tried to establish a connection between Montague’s approach to compositional
semantics and Chomsky’s syntactic analysis.
Actually, at the beginning of the ‘70s, Chomsky’s and Montague’s approaches were incompatible, for
a number of reasons concerning the general character of the theories they were carrying on
(Chomsky was a cognitivist, Montague was a logician). Barbara Partee tried to establish a
connection between the two lines of research so that today we do a compositional analysis of
syntactic structures in the perspective of generative grammar. In ’75 she proposed this argument
about modifier structures. Chomsky’s reaction was extremely negative, based on the presupposition
that syntax is autonomous, and blind to any consideration of external component.
Partee’s position was strongly opposed by orthodox of Generativism.
Only after many years, after 2000, Chomsky re-evaluated Partee’s approaches collocating them in
the framework of an interface between the phonological form and the conceptual structure of
language.
Somehow compositionality is one of these interface conditions posed by Chomsky.

69
Week 3
Consolidation exercise n #3/restrictive modifiers
Analyze the following determiner phrase by using
a) the Predicate Modification rule (Heim & Kratzer 1998, p.65)
b) the type shifting solution (Heim & Kratzer 1998, p. 66)
with [[ the ]] = λP. σ(P)
"the Italian boy"

a. Predicate Modification rule

RULE: ⟦γ⟧ = λxe · ⟦α⟧(x) = ⟦β⟧(x) = 1

⟦NP⟧ = ⟦boy⟧ = λxe · x is a boy

⟦A⟧ = ⟦italian⟧ = λye · y is italian

⟦AP⟧ = λze · ⟦A⟧(z) ⋂ ⟦NP⟧(z) = 1

= λze · ⟦λye · y is italian⟧(z) ⋂ ⟦λxe · x is a boy⟧(z) = 1
= λze · z is Italian and z is a boy = 1

⟦D⟧ = ⟦the⟧ = ⟦λP · σ(P)⟧

⟦DP⟧ = ⟦D⟧ (⟦AP⟧) =

= ⟦λP<e, t> · σ(P)⟧ (λze · z is Italian and z is a boy = 1) =
= σ (λze · z is Italian and z is a boy = 1) =
= Jz · z is Italian and z is a boy59

b. Type shifting solution

⟦NP⟧ = ⟦boy⟧ = λxe · x is a boy

⟦A⟧ = ⟦italian⟧ = [λf<e, t> · [λye · f(y) = 1 and y is italian]]

⟦AP⟧ = ⟦A⟧ (⟦NP⟧) =

= [λf<e, t> · [λye · f(y) = 1 and y is italian]] (λxe · x is a boy)=
= [λye · [λxe · (x is a boy) (y) = 1 and y is Italian]] =
= λye · y is a boy and y is Italian

⟦D⟧ = ⟦the⟧ = ⟦λP · σ(P)⟧

⟦DP⟧ = ⟦D⟧ (⟦AP⟧) = [λP · σ(P)] (λye · y is a boy and y is Italian)

=
= σ (λye · y is a boy and y is Italian) =
= = Jz · z is Italian and z is a boy

59
“The unique entity z such that z is Italian and z is a boy”

70
71
Week 4
Lesson 9
15.12
Recap
To recapitulate the difference between the two solutions to analyze modifiers:
● Predicate Modification can only be applied when we have an intersective inference,
because it results from the intersection of two sets. Adjectives that do not show an
intersective behavior do not allow this kind of solution.
● Type Shifting can be applied in any case because it is more general. Adjectives with a
subsective behavior can be analyzed through this method.

Definite determiner: Heim & Kratzer analysis60

Condition in quantized nouns (singular)
We have seen how the definite determiner, in single nouns, can be interpreted as the supremum
operator of type <<e, t>, e> (in fact it is a function that takes as input a set and returns an entity)
that looks for the uniqueness requirement.
Since Heim & Kratzer do not analyze plural and mass nouns, but they only focus on singular count
nouns, they do not use the supremum operator, because it is useful for words whose semilattice
structure presents a cumulative property. Instead, they adopt a choice of function going from a
(sub) set to the unique entity that is a member of that set.
In this sense we consider the function to be partial: the domain is not the whole denotational
domain of type <e, t>, but just a subset of it consisting in the singleton (containing just the selected
atom) of the set of all atoms.

60
Chapter 4, section 4

72
This condition can be specified in the lambda prefix, before the value description (before the dot).
Let’s see how it is possible to label the definite determiner of a single count noun, according to Heim
& Kratzer.
⟦the⟧ = ⟦D⟧ = λf<e, t> -here, before the dot, we specify the condition that limits the domain to a subset of type <e,
t>. 🡪 : ∃!x f(x) = 1 -now, I specify the value description

● = λf<e, t> : ∃!x f(x) = 1 · jy f(y) = 1

input is a function from entity to truth-values : condition : there is exactly one element x
such that x is part of the set f61 · value description · output is the unique entity y such that
y is part of f. Therefore, of course, x must coincide with y.

Of course, this analysis cannot be applied to nouns with a cumulative reference (mass nouns or
plural nouns, because it is relative to the unique entity of the set that will result in the uniqueness
requirement).
The point here is that we can define a condition that we apply to f in order for it to go as input to the
function.
What is relevant to underline is what happens when the definite determiner is syntactically
combined with a NP whose denotation does not characterize a singleton set: in this case, as we saw,
the function will not be able to take as input the denotation of this NP, and no denotation will be
returned. This is the situation that rises when we have a singular definite description in a context in
which more than one entity satisfies the definite description. In this case, the use of the definite
description is infelicitous.

Semantic presupposition
Definition: A sentence S carries a semantic presupposition P whenever P must be true in order for S to
have a truth-value.
Whenever P is not satisfied, the sentence will not express complete truth-conditions and therefore it
will not be endowed with a truth-value in any particular circumstance of evaluation. In this sense
we consider the domain condition that we have posed to our input function, to be inherited by the
whole sentence that contains a singular definite description. In other words, the condition we
established must be satisfied in order for the sentence to have a true/false truth-value. If the
condition is not satisfied (if there is more than one element or no elements) the sentence cannot
receive any truth-value.
In Heim & Kratzer there is also another interesting discussion about the distinction between a
sentence that is void in truth-value (the presupposition is not satisfied; therefore the sentence has
no truth-value62), and a false sentence (its presupposition is satisfied but the sentence results to be
false). These two categories are often grouped together in contraposition to true sentences, but
actually there are some ways to disentangle the two cases.
Russell tried to analyze the definite description in quantificational terms, such that no truth-value
gap can rise. This topic is a big issue in Philosophy of Language.

61
This means that the set is a singleton.
62
This is also called “presupposition failure”.

73
Condition in cumulative nouns (mass and plural)
In this case, there is one entity that is selected to be the supremum of the structure, but that entity is
not the only entity present in the set that constitutes the domain of the function: instead, there are
different entities in the set of domain, and all of them are part of the unique entity that results from
their closure undersum. The result (the unique entity) is the supremum.

● ⟦the⟧ = λf<e, t> : (∃!x) f(x) = 1 & ∀z [f(z) = 1 → z ≤ x] · σ(f)

there is exactly one x such that x belongs to f, and all the entities z that belong to the same set are parts of x.

The condition here (between : and ·) includes the fact that only one entity must be selected from the
function, but it also specifies the maximality effect underlining that all the entities that are part of
the set are also parts of the final entity.63

Passives and smuggling

In terms of semantics, the head that we have called in our analysis “little v” is called “voice head” by
Kratzer. Moreover, there are some specific analyses that distinguish the role of little v head and the
role of voice head.
The Kratzer approach with event identification is only useful if we want to give an analysis of
passives in which there is no syntactically realized agent; whereas we are aware that, since Chris
Collins’ paper in 2005, the main view now, in syntactic literature, is that there is actually a
syntactically realized external argument in passive clauses: what happens is that a chunk of the
sentence moves towards the syntactic spine in such a way that it reaches the subject position.
For this reason, the manipulation of the theta-role for the External Argument that could be done
using event semantics is not so relevant, because in terms of compositional analysis, if we adopt the
smuggling derivation, there is an argument that is simply phonologically unrealized.
At this point, the most interesting part of the analysis of passives would concern the question of
what the effects of this smuggling are under an interpretive point of view.
The view that Collins and Belletti are assuming is that the smuggling operation (movement of this

63
This formula for plural and mass nouns is not explicitly treated in Heim & Kratzer.

74
verbal chunk across the External Argument) is only motivated by purely syntactic reasons, namely
by the need to make it possible for the Internal Argument to move to the specifier of TP, or, however,
to the subject position.
There is another proposal in the literature: that this smuggling operation could also have a semantic
effect (but this proposal did not have much success).
The point, however, would be to consider what the consequences are of promoting the Internal
Argument to the subject position.
Intuitively, the Internal Argument, in this way, becomes more prominent, and this also has the effect
of assimilating this IA to something like a “topic of the clause”, or, the entity that the rest of the
sentence is about. But this is an inaccurate characterization, in the sense that clearly, preverbal
subjects, are not completely identical to an actual topic.
At this point, the question becomes: what are the semantic consequences of this syntactic
promotion to the subject position? This is a very interesting question. Bianchi has been working a
little bit on this, but, to be frank, she thinks that the solution only comes in an intensional system,
and in particular in an intensional system with situation semantics instead of possible world
semantics.

Quantifiers
With this argument we start the second module of this course.64
In first order logic, we translate a sentence like: “Every man is a mortal” as: “For every x, if x is a man
then x is mortal”. In other words, we use quantifiers and implications to derive the meaning that we
find in natural language.
This treatment is not compositionally very satisfying: in fact, “every man is mortal” has two
constituents: “every man” (subject) and “is mortal” (predicate), and these pieces are distributed over
in different parts of the formula in a non-straightforward way.
Technically, “every man” has somehow its own denotation by itself. Therefore, compositionally, it is
not entirely clear how we should treat this sentence.
If we have to translate this syntactic tree in logic language, it would
be something like:
∀xe [x man → x mortal]
every entity x such that if x is a man, x is mortal

Or, (if we have “some” instead of “every”)

∃xe [x man ⋂ x mortal]
there is an entity x such that x is a man and x is mortal

First of all, we have two different pieces of the formula (quantifier

and implication/conjunction), that somehow seem to be
pertaining to the D part of the three.
But, the most important point is that it is not clear what denotation will we have at the level of the
DP: it gives the truth-conditions for the entire sentence, but how do we label it? We know that the

64
Chapter 6/7 Heim & Kratzer.

75
predicate denotes the property of being mortal (<e, t> type), but what about the DP?
In this frame, two things are not completely clear:
1. Which denotation does the DP have?
2. When we consider NPs that have a quantifier, they are not denoting just a specific entity, as
“John” or “city”. In an intensional level we have a difference, but on an extensional level it
seems that these quantifiers denote an entity, just as a simple proper name.
The expressive power of these NPs goes beyond what can be expressed simply using the
existential or the universal quantifier.
For example, when we say “most students are brilliant” we do not refer to a specific singular
or plural entity, but to a subset (most) of a set (students), and we cannot explicitly define the
meaning of “most” just by using quantifiers.
Numerals, modified numerals and quantifiers act this way (“three of them”, “almost three of them”,
“none of them”) carry a specific meaning that involves a limited expressivity with respect to natural
languages.
The only cases that can be captures in terms of first order logic are the cases of the simple universal
quantifier (∀ every), the simple existential quantifier (∃ some) and their negation (not every
student; no student).
The domain of quantification is empirically extremely rich, and its expressive power dramatically
increases the possible truth-conditions: the quantifier “every” is necessary when we talk, for
example, of infinite sets. If we didn’t have this quantifier, the only way to say something like “every
even number is divisible x2”, would be through an infinite chain of coordinate sentences “2 is divisible
x2 and 4 is divisible x2 and 6 is divisible x2…”. In the same way, the quantifier “some” has a wider
expressivity range than what it could seem: saying “an even number is divisible x2” can assume the
meaning of a general statement that is possible to refer not just to one element, but to every even
number.

Quantifiers as sets
How should we analyze these NPs that do not refer to specific entities? Heim and Kratzer
suggest using quantifiers to denote sets.
● Every man is mortal
⟦DP⟧ = ⟦every man⟧ = {x | x man}
(⟦DP⟧ =⟦Ann⟧ = {a} (singleton)) 🡪 substituting “every man”
⟦T’⟧ = ⟦is mortal⟧ = {y | y mortal}
⟦TP⟧ = 1 iff ⟦DP⟧ ⊆ ⟦T’⟧
● {x | x man} ⊆ {y | y mortal}
● ({a} ⊆ {y | y mortal})
“Every man is mortal” = true if and only if the set of men is a subset of
the set of mortals.
“Ann is mortal” = true if and only if the singleton Ann is a subset of the
set of mortals.
Apparently, this analysis in terms of sets returns the correct outcome of truth-conditions: intuitively,
if every man is mortal, this means that men constitute a subset of the set of mortal entities.

76
The problem comes when we consider the denotation of a quantifier like “no man” (negation of
some).

● No man is immortal
The only possible guess is to assign to this quantifier the
denotation of the empty set (∅), because the sentence literally
means that there is no man in the set of immortal entities.
⟦DP⟧ = ⟦no man⟧ = ∅
According to this analysis, the empty set (denoting “no man”)
is a subset of the set of immortal people.
∅ ⊆ {y | y immortals}
But here a problem arises: the empty set is a subset of any
set.
This means that it would always be true that no man is
immortal, but also that no man is mortal (because, being
necessarily a subset of any set, the empty set is also a subset
of the set of mortals), and so on.
For this reason, the lexical content would become vacuously true according to this analysis: it
would become impossible to discriminate the differences in content if “no man is immortal” is
equivalent to “no man is mortal”.
The symmetric problem is that if “no man” denotes the empty set, we would be able to replace the
set with any other existing set and we will always obtain the value true.

77
Solution
Which is the solution?
Heim & Kratzer try to explain it step by step, first considering simpler quantifiers and then
analyzing more complex cases.
Let’s start from simpler cases:

● Everything vanished
We know that the intransitive verb is a function from entity to
truth-values (<e, t>), and we want to obtain a truth-value (t) at the
level of the VP node. For this reason we construe a function,
for the DP, that takes as input the denotation of its sister node and
returns the denotation of its mother node (<<e, t>, t>)

⟦DP⟧ = ⟦everything⟧ = λf<e, t> · ∀xe f(x) = 1

= for all the entities x, f applied to x returns the value true.

This treatment can also be generalized to other cases:

● Nothing vanished

⟦DP⟧ = ⟦nothing⟧ = λf<e, t> · ¬∃xe f(x) = 1

= there is no entity x such that f applied to x returns the value true.

= λf<e, t> · ∀xe f(x) = 0

= for all entities x, f applied to x returns the value false.

(both denotations are correct because they are equivalent).

● Something vanished

⟦DP⟧ = ⟦something⟧= λf<e, t> · ∃xe f(x) = 1

= there is an entity x such that f applied to x returns the value true.

Now, what about “some man”, “no man” and “every man”?

78
 Let’s see how we treat complex cases:

● Every man vanished

Following the previous reasoning, since we have

established that we want to obtain a <<e, t>, t> type at the
DP level, and we know that NP is a function of type <e, t>,
we deduce that D should be an <<e, t>, <<e, t>, t>> type.
It is a rather complex semantic type, but we can find a
parallelism between this semantic type of the quantifier
and the denotation of a transitive verb:

So, while the transitive verb expresses a relation between entities ([λxe [λye · y loves x]), quantifiers
express a relation between sets ([λf<e, t> [λg<e, t> · {x | f(x) = 1}65 is in relation with {y | g(y) = 1}66]).
The idea that natural language quantifiers can be expressed through sets was proposed by Barwise
and Cooper (1981), and reported by Porter and Partee in their studies.
Barwise and Cooper captured that this treatment can express all the quantifiers of natural
languages.
Quantification is also an interesting topic regarding language acquisition.
Let’s see the analysis of “every”, “no” and “some” (man).

Every: ⟦D⟧ = ⟦every⟧ [λf<e, t> · [λg<e, t> · {x | f(x) = 1} ⊆ {y | g(y) = 1} = 1]

=
Set f (applied to entities x) is a subset of set g (applied to entities y) returns the value true

No: ⟦D⟧ = ⟦no⟧ = [λf<e, t> · [λg<e, t> · {x | f(x) = 1} ⋂ {y | g(y) = 1}] = ∅
Set f (applied to x) intersected with set g (applied to entities y) returns the empty set

Some: ⟦D⟧ = ⟦some⟧ = [λf<e, t> · [λg<e, t> · {x | f(x) = 1} ⋂ {y | g(y) = 1}] ≠ ∅
Set f (applied to x) intersected with set g (applied to entities y) does not return the empty set

In fact, if we empirically imagine what these formulas mean, we will see that:

65
Set f.
66
Set g.

79
● Every Christmas gift is a box

The set of Christmas gifts is a subset of the set of boxes =

true
(Inclusion)

● No Christmas gift is blue

The intersection
between the set of
blue things and the
set of Christmas gifts
returns the empty set
(no entities in
common).
(Disjunction)

● Some Christmas gifts are round

The intersection between the set of round
things and the set of Christmas gifts does
not return the empty set (there are
entities in common)
(Overlap)

80
Lesson 10
16.12
Recap
We have seen how quantifiers express relations between sets of entities.
The set characterized by the NP (normally called “restriction set”) combines immediately with the
quantification determiner, the second set (usually called “scope of the quantifier”) denoted by the
predicate part (VP).
We saw that the relations between these sets can be expressed in terms of diagrams:
● In the case of every, we have an inclusion relation (restriction set is a subset of the scope
set)
● In the case of no, we have that the intersection of the restriction and the scope set returns
the empty set (disjunction).
● In the case of some, we have that the intersection of the restriction and the scope set does
NOT return the empty set (overlap).
We also went through reasoning in terms of semantic types of the denotations of a quantifier. In a
compositional analysis, in fact, we want to have a specific denotation for the DP node.
The semantic type of DP should be a function that takes as input its sister node containing the
predicate, and returns the truth-value at the level of its mother node. Therefore, the semantic type
we have selected for the DP is <<e, t>, t>. Going downwards through the syntactic spine, we see how,
to obtain the type <<e, t>, t> at the DP level, we should combine D with NP67. Therefore, if NP is an
<e, t> type and it should saturate the input of its sister node to return <<e, t>, t>, this means that the
D node has the denotation <<e, t>, <<e, t>, t>>; where the first part is saturated by the NP
denotation, and the second part is the value that returns to the DP node.
It is possible to individuate a parallelism between this semantic type that we have obtained at the D
level <<e, t>, <<e, t>, t>> and the semantic type of transitive verbs <e, <e, t>> (nested function): in
the same way that the transitive verb’s type represents a two place relation between entities, the D
semantic type represents a two place relation between sets.
Notice that the term quantifier is sometimes used in an ambiguous way.
To be clear, “quantifier” refers to the denotation of the whole DP (<<e, t>, t>). Sometimes we sloppily
call “quantifier” only the D part (every/no/some), but actually, in a compositional analysis, the
quantifier indicates the DP node.
In principle, when we have to establish the semantic type of a node, and we already have the
denotation of its sister node, we have two possibilities:
● Consider our node to be the input of its sister node;
● Consider the sister node to be the input of our node.
In this case we could not apply the first possibility: if we try to consider D as the input of NP
function, since NP is <e, t>, the D node should have had the denotation <e>, but we know that
“every” does not just denote an entity. Therefore, this explanation would have been too restrictive.

67
See tree pg. 79.

81
For this reason, we considered the second possibility: that the sister node NP becomes the input of
our D node.

Compositional exercise
Every boy smokes
⟦V’⟧ = ⟦V⟧ = ⟦smokes⟧ = [λxe · x smokes] (scope)

⟦NP⟧ = ⟦N⟧ = ⟦boy⟧ = [λye · y boy] (restriction)

⟦D⟧ = ⟦every⟧ = [λf<e, t> · [λg<e, t> · {ue | f(u) = 1} ⊆ {ve | g(v) = 1}]

⟦DP⟧ = ⟦D⟧ (⟦NP⟧) = [λf<e, t> · [λg<e, t> · {ue | f(u) = 1} ⊆ {ve | g(v) = 1}]
([λye · y boy])

= [λf<e, t> · [λg<e, t> · {ue | f(u) = 1} ⊆ {ve | g(v) = 1}] ([λye · y boy]) =

= [λg<e, t> · {ue | ([λye · y boy])(u) = 1} ⊆ {ve | g(v) = 1}] =

= [λg<e, t> · {ue | ([λye · y boy])(u) = 1} ⊆ {ve | g(v) = 1}] =

= [λg<e, t> · {ue | u boy = 1} ⊆ {ve | g(v) = 1}].

⟦VP⟧ = ⟦DP⟧ (⟦V’⟧) = [λg<e, t> · {ue | u boy = 1} ⊆ {ve | g(v) = 1}] ([λxe ·
x smokes])

= [λg<e, t> · {ue | u boy = 1} ⊆ {ve | g(v) = 1}] ([λxe · x smokes]) =

= {ue | u boy = 1} ⊆ {ve | ([λxe · x smokes])(v) = 1} =

= {ue | u boy = 1} ⊆ {ve | ([λxe · x smokes])(v) = 1} =

= {ue | u boy = 1} ⊆ {ve | v smokes = 1} =

= {ue | u boy} ⊆ {ve | v smokes} = 1.

Generalized quantifiers properties

Once we assume that all quantifier determiners express relations between sets, we can try to
describe their properties and whether this classification yields interesting empirical consequences.
In Barwise & Cooper paper it is already clear that in fact, we have some empirical outcomes that are

82
worth mentioning: the relevant properties of generalized quantifiers allow us to formally capture
the asymmetries and contrasts between quantificational determiners.

Conservativity property
Barwise & Cooper, 1981; E. Keeman, 1995
All natural-language quantifiers are conservative:

RQ is conservative iff RQ (A, B) ↔ RQ (A, A ⋂ B)

This means that a relation is considered to be conservative whenever the relation that holds
between the restriction set (A) and the scope set (B) entails the relation between the restriction set
and the intersection between the restriction and the scope set.
In empirical terms, this means that:

● Every (inclusion)🡪 A “man”; B “mortal”

“every man(A) is mortal(B)” entails that every man(A) is a mortal man(B ⋂ A).
Every is conservative because this logical equivalence holds:
A⊆B↔A⊆A⋂B

● No (disjunction)🡪 A “man”; B “immortal”

“no man(A) is immortal(B)” entails that no man(A) is an immortal man(B ⋂ A).
No is conservative because this logical equivalence holds:
A ⋂ B = ∅ ↔ A ⋂ (A ⋂ B) = ∅

● Some (conjunction) 🡪 A “man”; B “happy”

“some men(A) are happy(B)” entails that some men(A) are happy men(B ⋂ A).
Some is conservative because this logical equivalence holds:
A ⋂ B ≠ ∅↔ A ⋂ (A ⋂ B) ≠ ∅

Numerals are another example of conservative quantifiers

● Two 🡪 A “man”; B “happy”

“two men(A) are happy(B)” entails that two men(A) are happy men(B ⋂ A).
Two is conservative because this logical equivalence holds:
|A ⋂ B| ≥ 2 ↔ |A ⋂ (A ⋂ B)| ≥ 2

Many quantifiers satisfy this behavior of conservativity.

What conservativity tells us is that in a relation between the restriction and the scope set, we are
only interested in part of the scope set: that part that is in common with the first set. The elements
that are in B but not in A are irrelevant for the relation.

Non conservative relations

Some quantifiers show a non-conservative relation.
Take, for instance, equality.
Suppose we have a quantificational determiner that indicates an equality relation.

83
 ● A = B ↔ A = (A ⋂ B)
it does not hold in general that if A is identical to B, A is also identical to the
intersection of A and B.
In fact, if we imagine an inclusion relation, A= (A ⋂ B) is true, but A = B does not hold.
Therefore, equality does not satisfy the property of conservativity.
The Barwise & Cooper’s proposal is not that natural languages do not have the expressive power of
articulating relations of equality between two sets; but that no simple quantificational determiner
expresses this kind of relation.
Therefore, non-basic quantificational determiners must be used.
For example, in English, complex expressions as “all and only” are used:
“All and only the students are present”. In this case, the set of students and the set of present people
are equivalent, but this relation cannot be expressed through simple quantificational determiners.

Symmetry and asymmetry

Another property that is not universal but describes relations between the restriction and the scope
set is symmetry.
A relation is symmetric if it also holds in the opposite relation:

RQ is symmetric iff RQ (A, B) ↔ RQ (B, A)

This means that if there is a relation between the restriction and the scope set, the same relation
holds between the scope and the restriction set. If the relation cannot be reversed, it is considered
to be asymmetric.
Some of the quantification relations we have investigated before are symmetric.

● Every 🡪 “every student is brilliant” ≠ “every brilliant person is a student”

The symmetric relation does not hold here. 🡪 ASYMMETRY
● No 🡪 “no student is brilliant” ≡ “no brilliant person is a student”
The symmetric relation holds in this case. 🡪 SYMMETRY
● Some 🡪 “some students are brilliant” ≡ “some brilliant people are students”
The symmetric relation holds here. 🡪 SYMMETRY
● Most 🡪 “most students are brilliant” ≠ “most brilliant people are students”
The symmetric relation does not hold in this case 🡪 ASYMMETRY

The property of symmetry turned out to be particularly relevant in the domain of a particular
structure which have been studied in ‘70s and ‘80s: the “Existential There Structure” of English.

Existential There Structure

This structure refers to when we have a post-verbal NP, and the subject position is filled by the
simil-adverb “there”, commonly analyzed as an expletive element.
There has been a long discussion in the history of syntactic literature; but in the semantic literature,

84
the analysis that has been drawn, carried by Milsark (1977), refers to “weak and strong
quantifiers”.68
The observation was that in this “Existential There Structure”, some quantifiers are admitted in the
position following the copula and some others are not.
If we look at their distribution in terms of the denotational type of the quantifiers we will see that
the discriminating factor is precisely symmetry:
o The quantificational determiners that denote a symmetric quantification relation can
introduce the post-copula NP in the Existential Structure (weak quantifier);
o The quantificational determiners that denote an asymmetric quantification relation
cannot introduce the post-copula NP in the Existential Structure (strong quantifier).
For instance,

✔ “There are some students”, “There is no student”, “There are two students” are possible
realizations (symmetric quantifiers can be put after the copula in the There Structure).
🗶 “There are every student”, “There are most students” are not possible realization
(asymmetric quantifiers cannot be put after the copula in the There Structure).
This demonstrates a discrimination between two subclasses of quantificational determiners which
clearly have some empirical consequences.
Notice that the Italian translation of this structure, which is expressed through a locative clitic
“c’è/ci sono gli studenti” actually has a slightly different property: it does not impose the same rule
on the post-copular NP. Certain sentences that sound perfectly grammatical in Italian, in English are
impossible.
There is a limited number of cases in which we can actually find in the Existential There Structure a
strong determiner as “every” but these are exceptions.69
This restriction on the quantifiers that can be found in the post-copular position was originally
called “the definiteness restriction”, because of the basic asymmetry that holds between the
indefinite determiner and the definite determiner:
✔ There is a student in the classroom
🗶 There is the student in the classroom

Why is the definite determiner impossible to use here? Because it carries an asymmetrical feature.
And we will see why.
Zamparelli, 1995, published in 2000 “Layers in the Determiner Phrase”
He stated that the symmetric weak quantifiers correspond to a functional projection which is lower
than the projection that strong quantifiers host.

68
Weak stands for symmetrical; strong stands for asymmetrical.
69
Louise McNally (Pompeu Fabra, Madrid).

85
Monotonicity70
Monotonicity refers to the ability that certain determiners have to give rise to inferences which can
be described through two directions:
● Upward monotonicity (↑)
inference from a set to a superset
● Downward monotonicity (↓)
inference from a set to a subset

RQ (A, B)
The monotonicity property can be referring to the first or to the second argument (restriction set
and scope set).
Barwise & Cooper, ’81,
o use the term (upward/downward) monotonicity with respect to the second argument
(scope set) 🡪 rightward monotonicity
o use the term persistence with respect to the first argument (restriction set) 🡪 leftward
monotonicity
The idea is that a given quantificational relation can give rise either from a set B to a superset of B
(if it is rightward upward monotone); or from a set B to a subset of B (if it is rightward downward
monotone); similarly, it is possible to give rise to a superset A from the set A (if it is leftward upward
monotone); or to a subset of A from a set A (if it is leftward downward monotone).
Let’s give some simple examples for these cases:

Some students are from Italy

If we consider the set of people from Italy to be set A, and the set of people from Europe to be set B,
we will see that A ⊆ B (people from Italy is a subset of people from Europe, which is a superset of
it), therefore, people from Italy will necessarily be from Europe; while the reverse does not hold: it
is not true that people from Europe are necessarily from Italy.
In the same way, if we consider set B to be a subset of A and not a superset of it, A does not
necessarily entail that people from Italy are from Rome (which is a subset of set A).
Set A: people from Italy.
Superset B1: people from Europe.
Subset B2: people from Rome.
● A entails B1
● B1 does not entail A
● A does not entail B2
● B2 entails A

70
There is a small introduction of this topic at pg. 47.

86
Exercise
Consider the case of “every”;
Consider pairs of sets and subsets:
{students, female students}/{from Italy, from Europe}
1. Every student is from Italy → every student is from Europe
Every student is from Europe ↛ every student is from Italy
2. Every female student is from Italy ↛ every student is from Italy
Every student is from Italy → every female student is from Italy
3. Every female student is from Italy → every female student is from Europe
Every female student is from Europe ↛ every female student is from Italy
Note that the negation reverses the entailment
4. Not every student is from Italy ↛ not every student is from Europe
Not every student is from Europe → not every student is from Italy
5. Not every female student is from Italy → not every student is from Italy
Not every student is from Italy ↛ not every female student is from Italy
6. Not every female student is from Italy ↛ not every female student is from Europe
Not every female student is from Europe → not every female student is from Italy

Lesson 11
17.12
Recap
Generalized quantifiers
Conservativity
We defined the notion of conservativity: we saw that there is a proposed universal, that was, in
particular, discussed by Barwise & Cooper in 1981, Keeman & Stavi 1886.
The conservativity tells us that all quantification relations expressed by basic quantificational
determiners satisfy the equivalence that if a relation holds between A and B, it also holds between A
and the intersection of A and B. This means that when we are interpreting a quantifier, what we take
into account are only the entities present in the restriction set and the entities that the restriction
set has in common with the scope set (entities in the scope set that overlap with the restriction set).
In other words, when we interpret a sentence like “every student smokes”, we do not take into
account the members of the scope set (smokers) that are not in the restriction set of students.
This holds in general for all natural language determiners and this is the universal proposal.
We saw some kinds of exceptions (as only, all and only), a series of problematic cases for this
conservativity universals.

87
Symmetry
Intuitively, those quantificational relation which are based on the intersection between the
restriction and the scope are symmetric in the sense that since intersection is a symmetric
operation, reversing the roles of the restriction and the scope set will give equivalent
truth-conditions; whereas, when we have asymmetric quantificational relations as inclusion, then of
course this reversal will give different truth-conditions.
This property was identified as the relevant one to characterize the DP that can appear in the
“Existential There Construction” of English (that was called “The weak noun phrase” by Milsark,
1977); whereas non-symmetric quantifiers and definite determiners are not allowed in this context,
and these others are the NPs that Laudsaw called “strong”, in contrast with the weak, symmetric
ones.
Whenever we see in literature differences between weak and strong determiners, this is the
distinction that is taken into consideration. Weak determiners are sometimes called “indefinites”,
even though in general, the indefinite determiners are a specific category of determiners. They are
called indefinites because in certain kinds of analysis they show many common properties with
indefinite determiners.
This very simple property that we have defined actually yielded to rich discussions in various
directions. Some of them will be analyzed in detail in further lessons.
Zamparelli 1995 tried to provide a sort of cartography of determiner phrase, such that the weak
determiners are located in a lower functional position than the higher ones. An immediate
argument for this is the fact that certain weak quantificational determiners (as numerals) can be
preceded by strong quantifiers (ex: every two weeks, or the two students).

Monotonicity
It can be upward (when it gives an entailment from a set to its superset), or downward (if the
entailment holds between a set and its subset).
Then we noticed that a quantificational relation can yield a monotonicity either relative to the
second argument (scope set), and in this case it is called rightward monotonicity; or it can hold
relatively to the first argument (restriction set), and in this case it is called leftward monotonicity71.
We saw how in the case of “Every”, the monotonicity can be either upward and rightward; or
downward and leftward 🡪 downward monotonicity pertains the restriction set, upward
monotonicity pertains the scope set.
In the case of “Some”, it is upward monotone with respect to both arguments (rightward and
leftward). This is because while “every” represents an inclusion relation, “some” represents an
intersective relation. Therefore, the overlap of the restriction and the scope set yields to an upwards
monotonicity.
Our last case, “No” is, instead, downward monotone with respect to both arguments (restriction and
scope set), because, being a disjunctive relation, any subset of the first set is disjunct from the
second set.
With respect to monotonicity, we also observed how negation reverses the monotonicity.

71
Or persistency, in Barwise & Cooper.

88
EVERY If “every student is from Italy”, it means
that the set of students is a subset of the
set “from Italy”. Since “from Italy” is a
subset of “from Europe”, if the set of
students is a subset of “from Italy”, it is
also a subset of “from Europe”.
Moreover, any subset of “students” will
also be a subset of “from Italy” and
“from Europe”.
This explains why here we have a
downward leftward monotonicity
(every student is from Italy implies that
every female student is from Italy), and
an upward rightward monotonicity
(every student is from Italy implies that
every student is from Europe).

SOME
If “some students are from Italy”. This
means that we have some people that
belong both to the set of students and to
the set of people from Italy.
It is not necessarily entailed that a
subset of the restriction is part of the
scope: it can be that only some male
students are from Italy, and no female
students are from Italy.
This shows why here we have both an
upward rightward (some students are
from Italy entails that some students
are from Europe) and an upward
leftward monotonicity (some female
students are from Italy entails that
some students are from Italy).

NO
Finally, in the case of “no students are
from Italy” we have no overlap of the
restriction and the scope. For this
reason, we have a downward leftward
(no students are from Italy entails that
also no female students are from Italy)
and a downward rightward
monotonicity (no students are from
Europe entails that no students are
from Italy): there is no possibility to
entail that a subset of the restriction set
Is also a set of the scope set. The
entailment can only occur from a set to
its own subsets.

89
Monotonicity’s empirical consequences
Monotonicity property of quantifiers is relevant for the identification of what we call negative
polarity items (Ladusaw & Fauconnier, 1975/1979)

Negative polarity items

This phenomenon refers to some elements that only survive in presence of a particular negative
operator in the sentence, as “any, ever72”.

✔ I did not read any paper yesterday

✔ Nobody read any paper yesterday
🗶 I read any paper yesterday

Ladusaw observed that this happens not only with negative operators, but also with downward
monotone quantifiers, with respect to the part of the clause that contains the polarity item itself:
? At most three students read any paper
🗶 At least three students read any paper

This contrast between the two quantifiers “at most” whose outcome is not totally unacceptable, and
“at least” which is totally ungrammatical in this frame, lead to establish the difference between the
two:
● At least is upward monotone
● At most is downward monotone
Interestingly, it was noticed that even a quantifier like “every” can license a negative polarity item,
but only within the restriction and not within the scope. The reason is that, as we saw before,
“every” is downward monotone only with respect to its restriction set (downward leftward), while
it is upward monotone with respect to the scope (upward rightward).
For this reason, it is possible to find sentences as:

✔ [Every student who read any paper] got an A73

Whereas it is not grammatically correct to say:

🗶 Every [student] [read any paper]74

It also works in Italian:

✔ Nessuno studente ha letto alcun articolo 🡪 “nessuno” is the negative operator.

🗶 Almeno tre studenti hanno letto alcun articolo

72
Or, in italian, “alcuno”
73
The relative clause is part of the restriction of the quantifier.
74
Here the restriction set is just “student” and the scope is the predicate part.

90
There is a rich literature about polarity items. In particular, there are some recent analyses by A.
Giannakidou about this topic; but historically, the distinction between upwards and downwards
monotonicity can be captured in semantic terms more than in syntactic terms (formally, there is no
way to discriminate why at least and at most are different in their distribution).
According to Chierchia, scalar implicatures are sensitive to downward monotonicity.

Italian polarity items

In Italian we have a category of polarity items that are mostly idiomatic expressions.
For example, consider:
“Non ho capito un tubo”
This is clearly an expression that shows polarity behavior:
✔ Nessuno ha capito un tubo
🗶 Qualcuno ha capito un tubo
Presumably, these kinds of expressions are also at the origin of certain elements that developed
their meaning in negative markers, as, for instance, French negative particle “pas”, which literally
means “step”, and originated in an expression like “I did not walk a single step”. Then, the lexical
meaning of “step” got lost and the word assumed a new meaning which is the negative marker.
In contemporary colloquial French it is also possible to say “je comprend pas”, instead of “je ne
comprend pas”75, avoiding the negation because “pas” already conveys the negative feature.
The same can be seen in some Italian dialects: in Bergamo’s dialect it is usual to say “so mia”76 (non
so mica), where “mia” derives from “mica” than means “crumb”. Originally, mica was used in
sentences to express reinforcement of negation. Then, it underwent a process of grammaticalization
and it ended up to have a negative polarity denotation. In fact, nowadays, in Bergamo’s dialect it is
also possible to omit the negation “non” because “mia” already carries the negative meaning.
This is a rather complex discourse. And it becomes even more complex if we consider Italian N
words such as “Nessuno” that seem to work in certain context as negative quantifiers and in other
contexts as polarity items:

● Nessuno ha capito 🡪 (preverbal position, acts as negative quantifier)

● Non ho visto nessuno 🡪 (postverbal position, acts as polarity item)

There is a wide literature about these topics.

Issue of presuppositionality
This is another property of generalized quantifiers: some quantifiers introduce a presupposition
about the cardinality of the restriction.
Let’s take as example the quantifiers “both” and “neither”.

75
Translated: I do not understand.
76
Translated: I don’t know.

91
When we say a sentence like “both students are brilliant” we are implying that the cardinality of the
student set is 2.
|⟦student⟧| = 2
We can associate the semantic type of “both” and “neither” to the uniqueness requirement: we
assume that the quantifier determiner is a partial function that takes as input exactly a function of a
set composed by two elements.
“Both” is also equivalent to the universal quantifier (because we use it when we denote all of the
two entities taken into consideration). For these reasons, the semantic type we can associate to
words like “both” and “neither” is <<e, t>, <e, t>, t>.
⟦both⟧ = [λf<e, t> : |{xe | f(x)} = 2|77 · [λg<e, t> · {xe | f(x)} ⊆ {ze | g(z) = 1}
= input is a function with the condition that the cardinality of the restriction set is 2 (or, condition: there are
exactly 2 x) and the output is composed by a function g such that f is a subset of g.

Both is exactly a universal quantifier that is appliable only in the case that the restriction set
contains just two elements. If we do not specify the condition that there are exactly two elements in
the restriction set, it is not possible to return a truth-value.

Italian’s quantifiers
Surprisingly, quantifiers turned out to be very hard to learn for children: even though they have
acquired the meaning of a quantifier like “every” (in Italian “tutti”), they still struggle to understand
the meaning of expressions like “both”, in Italian “entrambi”/”tutti e due”.
The colloquial form is “tutti e due” which is semantically transparent because it literally means “all
and two”, with a sort of conjunction (tutti e due gli studenti = tutti gli studenti e due studenti / all and
two the students = all the students and two students) that we can logically define as follows:
{x | s ⊆ x} ⋂ {x | x ⋂ s} ≥ 2 🡪 conjunctive meaning
x such that student is included in x AND x such that x and s are equal or major than 2
(all the students) AND (two students)

If we intersect these two sets, what we obtain is that the set of students, that has exactly two
members, is included in x.
About the acquisition of quantifiers an interesting research has been conducted by N. Katsos

Empty restriction
Let’s now consider the last issue of presuppositionality constraint.78
Now we will see how the treatment of quantifiers that we have proposed in terms of relations
between sets gives rise to an intuitively correct entailment pattern, except in the case that the
restriction is empty.

Every
Let’s start from the assumption that if every entity x has the property P, of course there is at least
one entity x that has the property P, that is:

77
It is also possible to write “∃!2x”
78
Last part of 6th chapter of Heim & Kratzer.

92
 ● ∀xP(x) → ∃xP(x)
Consider the entailment in terms of reinterpretation of the universal quantifier and the existential
quantifier in sets.
● Every student is brilliant → Some students are brilliant
S⊆B→S⋂B≠∅
Note that this entailment pattern fails in the case that the set of students is empty: in that case, the
first term will remain true (the empty set is a subset of any set, therefore, ∅ ⊆ B = 1), but the second
term will no longer be true (the intersection between the empty set and another set returns the
empty set, ∅ ⋂ B ≠ ∅ = 0).
● If S = ∅,
(∅ ⊆ B) → (∅ ⋂ B ≠ ∅)
T → F = F
The entailment (∀ → ∃) does not hold if the restriction set is empty.
It directly follows that not only the entailment pattern with the existential quantifier, but any
sentence with the universal quantifier “every” applied to the empty set is vacuously true. As
explained a few lines above, the empty set is a subset of any set, therefore, ∅ ⊆ B ALWAYS = 1.
Barwise and Cooper define this kind of situation (universal quantifier applied to empty set)
“degenerate”, because it would be always vacuously true that the empty set is a subset of any of
the possible sets.
Something similar happens also in the case of “some”

Some
If S = ∅,
● (∀X ⊆ D) ∅ ⋂ X = ∅
For every x belonging to the domain, the empty set conjunct with any set will always be the empty set.

Therefore, the quantificational relation expressed by “some” will necessarily be false, no

matter what it applies to (it is the reverse effect of what happens with “every”).

No
If S = ∅,
● It will be again trivially true that the empty set is a logical consequence of the disjunction
of any set, because no set can be the product of a disjunction between other two sets if not ∅.

Hypothesis
The hypothesis carried by Strawson is that quantifiers should impose on quantificational
determiners general presuppositionality constraint: they cannot combine with a NP that denotes a
function characterizing an empty set.
But then a problem arises: what happens in cases like:
● There are some students
● There are no students

93
Because in these cases, intuitively, we are a priori asserting that S ≠ ∅. If we assume that the
quantifier “some” already carries the presupposition that the set is not empty, this assumption is
totally uninformative; while, in the case of “no” it clearly indicates that the set is empty.

● There are some students 🡪 S ≠ ∅

● There are no students 🡪 S = ∅

We can assume that these quantifiers carry a presupposition in all contexts (that the set is not
empty in the case of some, and that the set is empty in the case of no). This approach would lead to a
contradiction in certain contexts.
To avoid this contradiction, we must assume that there are two versions of weak determiners79: a
weak version (that does not carry the presupposition) and a strong version (which carries the
presupposition).
Diesing, 1992 discussed this topic in “Definites” and she showed that there are syntactic correlations:

🡺 Weak (non-presuppositional) 🡪 inside the predicate nucleus of the sentence.

🡺 Strong (presuppositional) 🡪 outside the predicate nucleus of the sentence.

Anyway, in some languages, the surface form of the sentence can be misleading: we can have a
quantifier that seems to be outside the predicate nucleus, but it allows a weak interpretation. In
these cases, Diesing assumes a so-called reconstruction process, namely a syntactic process that
happens in mapping two logical forms, putting back the quantifiers within the predicate nucleus.
So, if we have a sentence like:
● Some students showed up
This allows for
1. a presuppositional interpretation (the set of students is non empty and some of them
showed up = “among the students, some of them showed up”);
2. a non-presuppositional interpretation (the set of students might be empty, but someone
showed up and they had the property of being students = “some people showed up and they
were students”).
In Heim & Kratzer the example is:

● Some American kings live in New York 🡪 infelicitous (presupposition failure)

● There are some American kings in New York 🡪 false

The difference is that when we find the quantifier in the preverbal position it has a strong
presuppositional meaning; while if it is in a postverbal position it carries a weak
non-presuppositional meaning.80
Moreover, we can semantically consider the strong quantifiers with a presuppositional meaning
(preverbal position) as functions that only accept as input non-empty sets.

79
Weak determiners are basically intersective in their interpretation (symmetric). See pg. 85
80
The topic was studied also by Ladusaw.

94
Status of definites
We said that we have this treatment of definite descriptions like “the students” as having a
denotation of type <e>
But now we saw how it is possible to conjoin proper or common names:
● John and many other students
● The professor and most of the students are tired
What happens here is that we are conjoining an entity with a family of sets (quantifiers of type <<e,
t>, t>). So, how is it possible for a conjunction to put together such different types? In this case we
again need a type shifting to assign also to “John” and “the professor” a more complex semantic type
<<e, t>, t>.
⟦John⟧ = ⟦j⟧ = {x ⊆ D e | g ∊ X}
= λf<e, t> · f(j) = 1 (function from entity to truth-value such that j belongs to f)
We can define a general type-shifting rule that lifts the denotation of a definite description or a
proper name to the quantifier type whenever this is needed by content (as in this case, where we
have a conjoint).
Montague assumed that all NP have the denotational type of a quantifier. He generalized it because
whenever we use the quantificational type, this will also give equivalent truth-conditions to having
just a denotation of an entity type; while in some contexts, the truth-conditions can only be
expressed through the quantificational type.
His assumption that all NP denote a quantifier was a strategy of generalization to the worst case
based on the assumption that, according to him, there is an homeomorphism between the syntax
and the semantics, in the sense that for every syntactic category there is just one possible
denotational semantic type. Since we considered that “John” and “most of the students” are both NPs,
in order to have a uniform type, we must assign the same denotational type to both of them.
With Partee’s introduction to type-shifting rules it became possible to assume that the basic
denotational types of the lexical nouns in itself is just an entity, and then, when needed it can be
lifted and become more complex.

95
Week 4
Exercise
Compositional analysis of the sentence: "Three dogs barked"
assuming that "three" expresses a quantificational relation whereby the intersection between the
restriction and the scope has cardinality equal (or greater than) 3 (without plurality)

⟦V’⟧ = ⟦V⟧ = ⟦barked⟧ = [λxe · x barked] (scope)

⟦N⟧ = ⟦NP⟧ = ⟦dogs⟧ = [λye · y dogs] (restriction)

⟦D⟧ = [λf<e, t> · [λg<e, t> · |{ue | f(u) = 1} ⋂ { ve | g(v) = 1}| ≥ three]]

⟦DP⟧ = ⟦D⟧(⟦NP⟧) = [λf<e, t> · [λg<e, t> · |{ue | f(u) = 1} ⋂ {ve | g(v) =
1}|] ≥ three] ([λye · y dogs]) =

= [λg<e, t> · |{ue | (λye · y dogs)(u) = 1} ⋂ {ve | g(v) = 1}| ≥ three] =

= [λg<e, t> · |{ue | u dogs = 1} ⋂ {ve | g(v) = 1}| ≥ three]

⟦VP⟧ = ⟦DP⟧(⟦V’⟧) = [λg<e, t> · |{ue | u dogs) = 1} ⋂ {ve | g(v) = 1}| ≥

three] ([λxe · x barked]) =

= |{ue | u dogs = 1} ⋂ {ve | (λxe · x barked)(v) = 1}| ≥ three

= |{ue | u dogs = 1} ⋂ {ve | v barked = 1}| ≥ three

= |{ue | u dogs ⋂ ve | v barked}| ≥ three

96
Week 5
Lesson 12
22.12
Review
In this and the next lesson we will focus on the revision of the main topics we have discussed in
terms of functional metalanguage, going from the simplest to the more complex exercises.

Valentina smokes (intransitive verb)

⟦smokes⟧ = ⟦V⟧ = ⟦VP⟧ = ⟦V’⟧ [λxe · x smokes]

⟦Valentina⟧ = ⟦N⟧ = ⟦NP⟧ = Val
⟦VP*⟧ = ⟦V’⟧ (⟦NP⟧) = [λxe · x smokes] (Val) = [Val smokes] =
= 1 iff Val smokes
NOTE: in this case, little v head (v) does not give any semantic
contribution.

John likes Bill (transitive verb)

⟦Bill⟧ = ⟦N⟧ = ⟦NP2⟧ = B
⟦likes⟧ = ⟦V⟧ = [λxe · [λye · y likes x]
⟦VP⟧ = ⟦V⟧ (⟦NP2⟧) = [λxe · [λye · y likes x] (B) =
[λye · y likes B]
⟦v’⟧ = ⟦VP⟧
⟦John⟧ = ⟦N⟧ = ⟦NP1⟧ = J
⟦VP*⟧ = ⟦v’⟧ (⟦NP1⟧) = [λye · y likes B] (J) =
[J likes B]
NOTE: in this case, little v head (v) does not give any semantic
contribution.

97
John helped Bill (with event semantics)
To analyze this sentence, we will use the Davidsonian event position (Davidson ‘67).
We use e1, e2, e3… to denote variables over events (ε is the semantic type of events).
⟦Bill⟧ = ⟦N⟧ = ⟦NP2⟧ = B.
⟦helped⟧ = ⟦V⟧ = [λxe · [λe1 · x helped in e1]]
⟦VP⟧ = ⟦V⟧ (⟦NP2⟧) = [λxe · [λe1 · x helped in e1]] (B.) =
[λe1 · B. helped in e1]
NOTE: here, the little v head is not semantically empty: it
introduces the agent (John), correlating it with the VP.
Kratzer (1996) calls little v head “voice”, because it is the active
voice that introduces the agent.
⟦v⟧ = [λze · [λe2 · z agent of e2]]
At this point we have VP (<ε, t>) and v (<e, <ε, t>>). Since we must
combine them to obtain v’, we need a compositional rule to return
an <e, <ε, t>> semantic type at the v’ level.
This rule is called Event Identification (Heim&Kratzer).
Event Identification Rule correlates e1 and e2, merging them into the same event (e3) to make clear
that the event we are talking about is the same (event of helping = e1 is help received by Bill; e2 is
help given by John).
⟦v’⟧ = [λue · [λe3 · ⟦VP⟧(e3) and ⟦v⟧(u)(e3)]]
⟦v’⟧ = [λue · [λe3 · [λe1 · B. helped in e1](e3) and [λze · [λe2 · z agent of e2](u)(e3)]]]
⟦v’⟧ = [λue · [λe3 · [B. helped in e3] and [u agent of e3]]]
⟦John⟧ = ⟦N⟧ = ⟦NP1⟧ = J.
⟦vP⟧ = ⟦v’⟧ (⟦NP1⟧) = [λue · [λe3 · [B. helped in e3] and [u agent of e3]]] (J.) =
[λe3 · [B. helped in e3] and [J. agent of e3]]
The further compositional step to obtain a truth-value, is to bind the event (λe3). To do so, we must
add an aspectual head on the top of the structure, which tells us when does the event happen (with
a certain frequency, just once). In this case we imagine a non-iterative episode.
⟦Asp⟧ = [λf<ε, t> · ∃e4 [f(e4) = 1]]
Input: a function from events to truth-value;
Output: the truth-value such that there is an event e4, and f
applied to e4 returns the value true. (Asp, in this case,
represents the existential quantifier).

⟦Asp*⟧ =⟦Asp⟧ (⟦vP⟧) = [λf<ε, t> · ∃e4 [f(e4) = 1]] ([λe3 · [B. helped in e3] and [J. agent of e3]]) =
= ∃e4 [λe3 · [B. helped in e3] and [J. agent of e3]] (e4) = 1
= ∃e4 [B. helped in e4 and J. agent of e4] = 1

98
John helped Bill in Paris (with adjunct)
⟦Paris⟧ = ⟦N⟧ = ⟦NP⟧ = Par.
“in” is a nested function from an entity to a function
from event to truth-value (Paris, such that event is true
if it happened in Paris).
⟦in⟧ = ⟦P⟧ = [λze · [λe5 · e5 takes place in z]
⟦PP⟧ = ⟦P⟧ (⟦NP⟧) = [λze · [λe5 · e5 in z] (Par.) =
= [λe5 · e5 in Par.]
⟦J. Helped B⟧ = ⟦vP⟧ = [λe3 · B. helped in e3 and J. agent
of e3]
To analyze VP, since both vP and PP have the same
semantic type, we can apply the rule of Predicate
Modification, which is an intersective rule appliable to
two sister nodes that have the same semantic type. It
can be expressed as follows:
⟦γ⟧ = λxe · ⟦α⟧(x) = ⟦β⟧(x) = 1
In our actual structure, the incognita will refer to a variable of events. Therefore, we introduce a
lambda binding an event.
⟦VP⟧ = λe6 · ⟦vP⟧(e) = ⟦PP⟧(u) = 1
= λe6 · [λe3 · B. helped in e3 and J. agent of e3] (e6) = [λe5 · e5 in Par.] (e6) = 1
= [λe6 · B. helped in e6 and J. agent of e6] = [e6 in Par.] = 1
= [λe6 · B. helped in e6 and J. agent of e6 and e6 in Par.] = 1
Now, what we have obtained is the set of events that the aspect head will close through a quantifier.
In other words, the AspP will return the value true if and only if, in the circumstance of evaluation,
there is a non-empty set of events of the type described.
⟦Asp⟧ = [λf<ε, t> · ∃e7 [f(e7) = 1]]
⟦Asp*⟧ = ⟦Asp⟧ (⟦VP⟧) = [λf<ε, t> · ∃e7 [f(e7) = 1]] ([λe6 · B. helped in e6, J. agent of e6 & e6 in Par.] = 1)
= ∃e7 [[λe6 · B. helped in e6, J. agent of e6 & e6 in Par.] = 1 (e7) = 1]
= ∃e7 [B. helped in e7, J. agent of e7 & e7 in Par.] = 1

99
Lesson 13
23.12
Review
In this lesson we will consider the problem of plural nouns reference.

The students (determiner with plural noun)

We have seen how a singular count noun as “student” denotes a set of atomic entities; while the
plural “students” denotes all the possible pluralities derived by the sum of the atomic entities
represented by the singular.
⟦student⟧ = {a, b, c}
⟦students⟧ = {a, b, c, a⊕b, a⊕c, b⊕c, a⊕b⊕c}closure
( under sum) 🡨 join semilattice structure
We encode the closure under sum as a plural morpheme represented by the functional head Pl.P
added on the top of the NP.
For this reason, Pl.(+) is considered to be an operator that takes an input the set of atomic entities
{a, b, c} and returns as output their closure under sum {a, b, c, a⊕b, a⊕c, b⊕c, a⊕b⊕c}.
⟦NP⟧ = ⟦student⟧ = [λye · y is a student]
⟦Pl. +⟧ = [λf<e, t> · *{x | f (x) = 1}]81
⟦Pl.P⟧ = ⟦Pl. +⟧(⟦NP⟧) = [λf<e, t> · *{x | f (x) = 1}] ([λye · y is a
student])
= *{x | [λye · y is a student](x) = 1}
= *{x | x is a student = 1}
The idea is that the plural is an inflectional head and
therefore it belongs to the functional structure of the NP.
What we obtain through this reasoning is an entity that is
composed in its parts by all the possible sums of the atomic
entities, and we call it supremum.
To denote this supremum, we add, on the top of the structure, an additional layer (determiner
phrase, DP).
What we want to obtain at the highest level is just a unique entity (the students). For this reason, we
attribute the semantic type <<e, t>, <e, t>> to the determiner “the”.
Remember that, in plural nouns, D is a function that takes as input the Pl.P (students) and
returns to his mother node the unique entity y which is part of f, and for every part z that
belong to f, z is a subpart of y.
Or, in other words, it returns the unique entity y (the students) that derives from the closure
undersum f (students), and for every part z of this closure undersum (student1, student2, student3,
student1⊕student2, student1⊕student3, student2⊕student3, student1⊕student2⊕student3…), these
parts z are subparts of the final entity y that we obtain.

81
* is the operator that denotes the closure under sum.

100
⟦D⟧ = ⟦the⟧ = [λf<e, t> · ∃! y [f(y) = 1 & ∀z [f(z) = 1 → z ≤ y]]
Or,
⟦D⟧ = ⟦the⟧ = [λf<e, t> · J y [f(y) = 1 & ∀z [f(z) = 1 → z ≤ y]]
Of course, we can use an abbreviation to indicate this formula:
⟦D⟧ = ⟦the⟧ = [λf<e, t> · σ {y | f(y) = 1}]
= [λf<e, t> · σ(f)]
Remember: σ indicates the supremum of the set characterized by f.
All of the previous are equivalent formulas to indicate D, which is simply the operator that takes as
input the set characterized by the closure under sum of the entities and returns the unique entity y
represented by the closure under sum of all elements, and every element of the closure undersum is
also a part of y.
⟦DP⟧ = ⟦D⟧ (⟦Pl.P⟧) = [λf<e, t> · σ {y | f(y) = 1}] (*{x | x is a student = 1})
= σ {y | *{x | x is a student = 1} (y) = 1}
= σ *{y | y is a student = 1}
We obtained the unique entity (σ) that is the maximality effect (student1⊕student2⊕student3) of the
closure under sum (*) of the set of students.
With this last step we will obtain:
● The maximality effect [if plural +]
● The uniqueness requirement [if plural -]

Implicatures
o If we say a sentence like “there are no students”, this sentence can be falsified by the
presence of a single atomic student. But if we assume that the denotation “students” only
indicate a plurality of students, it is not explainable why the presence of a single student
would falsify the previous sentence. This means that, in a certain sense, students also involve
the possible meaning of only one student.

o In the case of “there are students” we can see a Gricean issue: if there was only one student
we would have said “there is a student”, because it would have been more informative. So, if
we use “there are students” the sentence is already implying a plurality of students and
excluding the possibility of a singular student being present. In an upward monotone
perspective, with this sentence, the possibility of a singular student is filtered out.

101
The Italian student (determiner with singular noun)
⟦NP⟧ = ⟦N⟧ = ⟦student⟧ = [λxe · x is a student]
⟦AP⟧ = ⟦A⟧ = ⟦italian⟧ = [λye · y is Italian]
Here, to combine these two sister nodes, we must introduce a
combination rule, since they have the same semantic type.
Therefore, we use the Predicate Modification rule82:
⟦γ⟧ = λxe · ⟦α⟧(x) = ⟦β⟧(x) = 1
⟦NP*⟧ = λze · ⟦NP⟧(z) = ⟦AP⟧(z) = 1
⟦NP*⟧ = λze · [λxe · x is a student](z) = [λye · y is Italian](z) = 1
= λze · z is a student & z is Italian = 1
[- plural] 🡪 if we had a positive plural feature, we would have had
another branch above the NP node and below the NP* node
indicating the pluralization. But since here we have a singular noun, the pluralization is absent, and
we can analyze the sentence as if it was not there.
In fact, note that there is a difference between:

● multi-value features: number 🡪 2 (singular, plural); person 🡪 3 (1st, 2nd, 3rd)

● binary features: plural 🡪 (+/-)
privative feature: plural 🡪 (+/unspecified)

Depending whether we assume a binary feature or a privative feature, we draw the

projection of plurality or not. In fact, it is possible to omit it if we have pl. – (singular), in the
case that we assume a privative feature (as in the previous example the Italian student). If
we assume a binary feature, the projection of plurality - must be present even though it does
not give any semantic contribution to the sentence.
In our context is not necessary to establish that we are dealing with a single atomic entity in the set
of students. The only important thing is that there is a single atomic entity in the intersection
between the set of students and the set of Italian people.
Definite determiner here is applying to a singular noun. Therefore, instead of using the
supremum operator, we use the jota operator, which indicates the unique entity. This
determiner will take as input the characteristic function of the set “Italian and student”, with
the uniqueness condition that there is exactly one element that belongs to the function (without
specifying the subparts, because it is singular); and it will return the unique entity that is part
of that set.
⟦D⟧ = ⟦the⟧ = λf<e, t> : ∃! u f(u) = 1 · Ju [f(u) = 1]83

82
PAY ATTENTION: Predicate Modification rule is only appliable when the adjective shows an intersective
behavior!! (pg. 72)
83
Determiner analysis by H&K.

102
No student smokes (without event semantics)
⟦v’⟧ = ⟦VP⟧ = ⟦V⟧ = ⟦smokes⟧ = [λxe · x smokes]
⟦NP⟧ = ⟦N⟧ = ⟦student⟧ = [λye · y student]
⟦D⟧ = ⟦no⟧ = [λf<e, t> · [λg<e, t> · {z | f(z) = 1} ⋂ {v | g(v) = 1} =
∅
⟦DP⟧ = ⟦D⟧ (⟦NP⟧) = [λf<e, t> · [λg<e, t> · {z | f(z) = 1} ⋂ {v |
g(v) = 1} = ∅ ([λy
e · y student]) =

= [λg<e, t> · {z | [λye · y student] (z) = 1} ⋂ {v | g(v) = 1} = ∅ =

= [λg<e, t> · {z | z student = 1} ⋂ {v | g(v) = 1} = ∅

⟦vP⟧ = ⟦DP⟧ (⟦v’⟧) = [λg<e, t> · {z | z student = 1} ⋂ {v | g(v) = 1} = ∅ (λx

e · x smokes) =

= {z | z student = 1} ⋂ {v | (λx
e · x smokes) (v) = 1} = ∅ =

= {z | z student = 1} ⋂ {v | v smokes = 1} = ∅ 🡨 you can write like this

= |{z | z student = 1} ⋂ {v | v smokes = 1}| = 0 🡨 or like this

In this case, the determiner “no” expresses the relation between the student set and the smokers set
(there are no elements in the set derived by the intersection of the student set and the smokers set).

Three dogs barked (without plurality, H&K analysis)

⟦NP⟧ = ⟦N⟧ = ⟦dogs⟧ = [λze · z is dog]
⟦v’⟧ = ⟦V⟧ = ⟦barked⟧ = [λve · v barked]
⟦D⟧ = ⟦three⟧ = [λf<e, t> · [λg<e, t> · |{x | f(x) = 1} ⋂ {y | g(y) = 1}| ≥ 3
⟦DP⟧ = ⟦D⟧ (⟦NP⟧) = [λf<e, t> · [λg<e, t> · |{x | f(x) = 1} ⋂ {y | g(y) =
1}| ≥ 3 ([λze · z is dog]) =
= [λg<e, t> · |{x | [λze · z is dog] (x) = 1} ⋂ {y | g(y) = 1}| ≥ 3 =
= [λg<e, t> · |{x | x is dog = 1} ⋂ {y | g(y) = 1}| ≥ 3
⟦VP⟧ = ⟦DP⟧ (⟦V’⟧) = [λg<e, t> · |{x | x is dog = 1} ⋂ {y | g(y) = 1}| ≥
3 ([λve · v barked]) =

e · v barked] (y) = 1}| ≥ 3 =

= |{x | x is dog = 1} ⋂ {y | [λv
= |{x | x is dog = 1} ⋂ {y | y barked = 1}| ≥ 3
The reason why we write “≥ 3” instead of “= 3” is because “3”, pragmatically means “at least three”.

103
Three dogs barked (with plurality)
⟦NP⟧ = ⟦N⟧ = ⟦dogs⟧ = [λxe · x dog]
⟦Pl+⟧ = [λf<e, t> · *{y | f(y) = 1}]
⟦Pl.P⟧ = ⟦Pl+⟧ (⟦NP⟧) = [λf<e, t> · *{y | f(y) = 1}] (λxe · x dog) =
= *{y | [λxe · x dog] (y) = 1} =
= *{y | y dog = 1}
At this point, we consider the numeral as a filter that takes as
input the whole semilattice structure and filters all the atoms
to return as output a set of plurality with exactly three
members (it filters the layer in which three atoms are summed
up inside the semilattice structure).
Therefore, we put on the top of the numeral phrase an
existential quantifier, meaning that there is a non-empty set (of
three entities) that is the intersection between the set of dogs and the set of entities that barked.
This is because “three” is a cardinality predicate.
The rest of the analysis is more complex than the ones we will investigate during this course (and
not needed for the exam).
If you want to know more, here are the references:
● Zamparelli, 2000, Layers in the DP (Routledge);
● Reuland & ter Meulen, 1987, The representation of (in)definiteness, (Heim, Higginbotham)
● Aloni & Dekker, chapter 9.
In the next lectures we will see how do we interpret structures in which the quantifier is not in a
subject position.

104
Week 6
Lesson 14
7.01
In this and the in next lectures we will finish to explore the discussion about quantifications.
Until now, we have seen cases in which the sister node of a quantifier
can go as input to the quantifier node, because their semantic types
matched. In all the examples that we have analyzed, the quantifier was
in a subject position.

But what if the quantifier is not in the subject position? What if its denotational semantic type
mismatches with its sister node’s type?
We have already outlined two possible solutions to solve a configuration of Type Mismatching. And
these are:
1. Introducing a new compositional rule to combine the two mismatching nodes;
2. Changing the semantic type of one of the two sister nodes (Type Flexibility84)
In principle, if we follow the Type Flexibility solution, we can decide which of the two sister nodes
to change in order to allow their combination; but there are some simplifications that cannot be
made. Let’s see an example:

Valentina interviewed every candidate

In this structure, we can combine NP (candidate)
with its sister node Q (every). And we will obtain the
semantic type <<e, t>, t> at the level of QP. At this
point, we find a type mismatching: QP and V are not
compatible, therefore, the function application is not
possible.
Let’s try to solve this problem with Type Flexibility:
we want to have <e, t> at the level of VP, to be
inherited by v’ that will, then, combine with NP
(Valentina) to obtain a truth-value at the vP level.
For this reason we must change the QP type in:
<<e, <e, t>>, <e, t>>, so that it can take as input V
(<e, <e, t>>) and return as output <e, t> at the VP
level.
It is not possible to simplify QP in <e>, such that V
can take it as input and return <e, t>.

84
Heim&Kratzer

105
 At this point, we have the mismatching problem
and we shift the QP semantic type to <<e, <e, t>>,
<e, t>>.
For this reason, we changed the semantic type of
QP in the syntactic tree.

⟦QP⟧ = λf<e, e. t> · [λxe · ∀ y : y candidate → f (y)

(x) = 1]]
The truth-value means that for every y that is a
candidate, the function f (the two place relation denoted
by the transitive verb interview) first applies to y
(candidate, the object, IA), and then to x (subject, EA).

This formula is equivalent to:

⟦QP⟧ = λf<e, e. t> · [λxe · {y | y candidate} ⊆ {ze | f

(z) (x) = 1}
The set of entities y (candidates) is a subset of the set of entities z (every) such that f (transitive verb:
interview) applies first to z (object) and then to x (subject). Here, the truth-value expresses the subset relation
between the set of candidates y, and another set of entities which stand in the relation f (interview) with the
entity x (EA).
We can informally translate the last formula with: There is a function f (interviewing) that expresses a
transitive verb (<e, <e, t>>), and it presupposes that there is an entity x (Valentina) such that the set of
candidates y is a subset of the set of entities (z) that get interviewed (f) by Valentina (x).

Now that we have the new denotation of QP, we can obtain VP simply by proceeding in the function
application as usual.
⟦V⟧ = ⟦interviewed⟧ = [λue · [λze · z interviewed u]]
⟦VP⟧ = ⟦QP⟧(⟦V⟧) = λf<e, e. t> · [λxe · ∀ y : y candidate → f (y) (x) = 1]] ([λue · [λze · z interviewed u]])
=
= [λxe · ∀ y : y candidate → [λue · [λze · z interviewed u]] (y) (x) = 1]] =
= [λxe · ∀ y : y candidate → [x interviewed y]] = 1]].
Here we have obtained a function of type <e, t>

Now we proceed as always. The denotation of VP gets inherited by v’.

⟦v’⟧ = ⟦VP⟧
⟦NP⟧ = ⟦N⟧ = ⟦Valentina⟧ = val.
⟦vP⟧ = ⟦v’⟧ (⟦NP⟧) = [λxe · ∀ y : y candidate → [x interviewed y]] = 1]] (val.) =
= [∀ y : y candidate → [val. interviewed y]] = 1]]
For every candidate y, Valentina interviewed y.
Let’s now see how we change the semantic type of V, instead of QP.

106
 Here we changed the semantic type of V from <e, <e, t>>
to <<<e, t>, t>, <e, t>>; although this solution has some
limitations.
⟦NP⟧ = [λxe · x candidate]
⟦Q⟧ = [λf<e, t> · [λg<e, t> · {y | y (f) = 1} ⊆ {z | g (z) = 1}]]
⟦QP⟧ = ⟦Q⟧ (⟦NP⟧) = [λf<e, t> · [λg<e, t> · {y | y (f) = 1} ⊆ {z
| g (z) = 1}]] ([λxe · x candidate]) =
= [λg<e, t> · {y | ([λxe · x candidate]) (y) = 1} ⊆ {z | g (z) = 1}]] =
= [λg<e, t> · {y | y candidate = 1} ⊆ {z | g (z) = 1}]]

107
Quantifier raising (QR)
R. May (1977/1985)
Let’s suppose that the quantifier always has the same denotational semantic type <<e, t>, t>.
In this scenario we know that the
restriction set, indicated by the QP, is the
set of candidates. But what about the
scope set? The problem here is that we
must somehow find a set in the rest of
the clause (Valentina interviewed), which
is the nuclear scope of the quantifier. But
how can we construct a set out of a piece
of sentence that is not even a
constituent?
The solution that was proposed by
Robert May is to assume a movement
operation to move the QP on the top
of the structure. When this movement
is done, it leaves a trace in the previous
QP position.
The trace is an empty category, but the
movement of QP creates a connection
between the two nodes (the node where the
QP originated and the node where QP ended
up). This relation is represented by an index,
that we collocate both on the side of the
QP new node, and on the side of the trace
left in its old node.
According to May, once we have removed QP
from its original position below the vP* node,
we can attribute to the vP* node the scope
set. This is why we label it with the semantic
type <e, t>, because it constitutes the scope
set that must be correlated to the restriction
set (candidates) represented by QP. vP*, that
now we consider as the scope set, intuitively
indicates the set of people that get
interviewed by Valentina.
This is why we use the vP* node to
compositionally build the scope set to
intersect with the quantifier.
Even though the trace has no phonological content (it is an empty category), we use it as syntactic
object, assuming its index (1) to be visible for the purpose of the compositional interpretation.

108
The important point is to remember that this rule called Quantifier Raising, is a so-called covert
movement rule, because it is not visible in the pronounced sentence: when we pronounce this
sentence, QP is in IA position, while when we interpret it in a semantic perspective, the QP lands on
the top of the structure.

QR: a covert movement

In this sense we find a divergence between the logical form and the phonological form of the
sentence.85
In the T model of grammar, we take elements
from the lexicon, we build a syntactic
structure, and at a certain point, which is
called s-structure (surface structure), we
have a divergence: one branch refers to the
phonological interface; the other branch
keeps having other applications that lead to
logical form (as the Quantifier Raising): the
second branch does not correspond to the
spelled-out sentence, but it allows to
compositionally interpret the structure.
If we did not have this divergence, the
problem of quantifier’s type mismatching
could not be solved.86
There have been some attempts to reduce
Quantifier Raising to an ordinary type of movement, but these solutions have not gained much
ground, mainly because if it was a regular movement it is not clear why it is covert in so many
languages. Normally, movement operations are explicitly visible.

Interpretive compositionality
This movement of the quantifier requires a weaker version of compositionality interpretation: the
interpretive compositionality, by which we mean something different from the rule-by-rule
compositionality.
Until now we have used the rule-by-rule compositionality, individuating the semantic type of each
node and combining it with its sister node, so that we could obtain their mother node; and we
proceeded like this until the whole structure was analyzed. This is why we started our process from
the bottom of the structure, and considered one node at a time.
BUT our actual analysis is not semantically interpretable in its phonological form. It means that
before analyzing the nodes, we must draw out the entire sentence in a syntactic tree, and then move
the constituents (in this case the Quantifier Phrase) until we reach a logically interpretable
sentence, which does not correspond to the phonological shape of the same sentence. This is why it
is not possible to start from the lowest node and proceed with its interpretation, to further combine

85
See pg. 57 for more clarifications about the logical/phonological distinction.
86
About this topic: Houstein, 2005; Kennedy, 1997.

109
it with its sister node exc. What we must do now, is to first establish the logical form of the sentence
(which has the QP on the top of it), and once it is possible to compositionally analyze it, we proceed.
In this sense, within the analysis of this kind of sentences, we must take in account that semantic is
applied to a sentence only after the syntactic level, and syntax is completely independent from the
semantic level.
The issue of the Quantifier Raising is still very debated, and there are many alternative approaches
to its resolution; and various rejections of this theory.
After we remove the quantifier from the problematic
position, how are we going to interpret the remaining
structure such that I can obtain a denotation of type <e,
t> as the sister node of QP?87
We know that the transitive verb has normally the
denotation <e, <e, t>>. For this reason, we consider the
trace to be an entity, to make it combine with the
transitive verb simply by function application.
We attribute the semantic type <e> to the trace, but we
know that it cannot simply denote an entity as NP.
Instead, it denotes a set of entities: the set of people
interviewed by Valentina 🡪 {x | v interviewed x}, which
corresponds to the scope set, or, in functional
metalanguage, [λxe · v. interviewed x] (this is what we
must obtain at the vP* level).
For this reason, we introduce a rule which allows us to interpret the trace as a variable over entities,
that fills the IA position, so that we can combine it with the transitive verb. Moreover, the variable
over entities that I am introducing at the t1 level must be bound. And vP* is the level that will bind
it. We can do it thanks to the index (1) that we have collocated both to the trace and to the QP new
position.88
Heim and Kratzer treat this issue in a slight different way, appending the index 1 both to the trace
and to a new node instead that appending it to the trace and to the QP node. Now we must
introduce the semantic of indices to see this treatment (5.3.4 H&K).

Semantics of indices
Heim and Kratzer assumed that a quantifier is
attached at the top of the structure, and its
sister node has an index that is appended as
independent daughter; and below all of this
we have the rest of our structure.
87
Remember: we want to obtain an <e, t> type because it corresponds to the semantic type of a set, and what
we want the sister node of QP to represent is the scope set.
88
Montague assumed the opposite QR movement for this kind of sentences: “Quantifying in”. He assumed that
the quantifier originates at the top of the structure and then it moves to the IA position. Exactly the opposite
of what May theorized.

110
In other words, they assume the structure to have another intermediate level between the vP of QP
and the vP of NP, an intermediate vP level that has the index 1 as head. Of course, the index must
match with the trace index.
This interpretation is not very standard, in the sense that an index is normally not appended is a
syntactic structure by itself.
Now, first of all, let’s give an interpretation for indices: they represent the fact that a syntactic
element that represents an entity has a non-fixed denotation (in fact, they are also applied to
pronouns)89.
The index specifies the syntactic element that it is attached to, as being dependent on an
assignment function.

Assignment function:
An assignment function is a function from indices to entities, or, in other words, it is a function that
assigns an entity to an index.
Let’s see an example of assignment function:

a:

● She1 likes him2

In this case what tells us the identity of she and he is the function that assigns to Valentina the index
1 and to Luca the index 2. The sentence “she likes him” cannot be interpreted unless we have an
assignment function that must necessarily be appropriate (that means, it must have, in its
domain, all the indices that are contained in the sentence).
Therefore, a sentence like “she likes him”, if analyzed through our function assignment, has the
meaning of “Valentina likes Luca” 🡪 ⟦she1 likes him2⟧a = Valentina likes Luca.

Of course, the entity attributed to the index entirely depends on the context of utterance.
For any element that carries an index, its interpretation relative to an assignment is going to be the
value that the assignment associates to the index. We translate this rule in technical terms as:

⟦α i ⟧a = a (i)

α = any syntactic node

i = index
a = assignment function
So, whenever we have an index attached to the structure, but it stands by itself, as in the structure
proposed by Heim & Kratzer, we introduce another interpretive rule that manipulates the

89
In the sentence “John says that he is happy” we can apply indexes to nouns and pronouns to clarify if the
word refers to the same entity or it does not: “John1 says that he1 is happy” = John says that John is happy;
“John1 says that he2 is happy” = John says that someone else is happy.

111
assignment function itself. And we set this rule at the vP* level (the layer that has the index as
head).

⟦vP*⟧a = [λxe · ⟦vP**⟧a x/1]

The interpretation of vP* node in the assignment

function depending by our context, is going to be a
function that introduces a lambda prefix x of type
entity; and the rule that we place is that the sister
node of the index (1) has to be interpreted as a
modified assignment a, in which the index is
replaced by the variable x at the vP* level.
This literally means that at the level of the vP* we
introduce the definition of a function binding the
variable x, and we specify the instructions:
analyzing our structure, whenever we encounter
an element bound by the same index (1), we
substitute it with variable x.

ax/1 = modified assignment function.

Our modified assignment function associates an entity to any other index different from (1),
and the variable x to any occurrence of 1. For this reason, we call it “modified”.
Assignment function “ax/1” is identical to “a”, except for index 1, which is modified into
variable x in our modified assignment function.
The variable itself is not a denotation or an object. “x” only makes sense because it is
introduced in the scope of the λ binder. So, it is going to specify a position that will be filled in by
any entity that will go as input to the function [λxe · ⟦vP**⟧a x/1].

Predicate Abstraction Rule

The rule ⟦vP*⟧a = [λxe · ⟦vP**⟧a x/1] is called Predicate Abstraction. This rule applies whenever we
have a configuration in a compositional triangle in which one of the daughter nodes is an index; and
it does two important things:
1. It introduces a lambda prefix that binds a
variable;

2. Is modifies the assignment function in

such a way that every lower position bound by
the same index is going to be interpreted as an
occurrence of the same variable x.

In fact, this rule allows us to attribute also to t1

the x variable. In fact, it would be non-sense to
simply write: ⟦t1⟧ = x, because x is not an entity,

112
it is just a variable, a piece of our metalanguage. To associate the variable x to t1 and to all the
occurrences of the same index 1, we must pass through this rule 🡪 ⟦vP*⟧a = [λxe · ⟦vP**⟧a x/1].

Directionality problem
Now, we have a directionality problem: if we start to interpret our syntactic structure from bottom
to top as we always did until now, we will face t1 before vP* node, which is the node that tells us in
which way we must modify our assignment function in order to obtain x as substitute of index 1.
In other words, we cannot compositionally interpret t1 until we specify our Predicate Abstraction
rule.
For this reason, in this particular configuration, we proceed analyzing the structure from the top to
the bottom, such that we encounter vP* node before, and when we approach to t1 we have already
introduce the rule that allows us to compositionally interpret t1.

FAST RECAP OF WHAAAT IS HAPPENING HERE

1. We are trying to denote the vP* node as the scope set of the sentence, which, combined with
the QP that has been moved in the top of the structure as its sister node, will give us the
relation between the restriction set QP (every candidate) and the scope set vP* (-people that-
Valentina interviewed).
2. To obtain a set <e, t> at the vP* level, we must introduce a rule, the Predicate Abstraction
rule, at the vP* node.
3. This rule does 2 things:
a. it introduces a λ prefix at the vP* node that binds a variable x;
b. it uses this variable to bind every occurrence of index 1, including the trace left by the
rising movement of QP.
4. We analyze the structure from the top to the bottom such that we first encounter the vP*, so
that we can introduce the rule that will be essential to interpret t1 node.

Top-down assignment function a (∅)

The assignment function could even be an empty
assignment function (an assignment function that is
not specifying an information). In this structure, for
example, is not the assignment function that
specifies the entity of index 1, but the compositional
rule that we introduce at the vP** level.
Let’s start our top-down compositional analysis.

⟦vP⟧a = ⟦QP⟧a (⟦vP*⟧a)

here we simply proceed by function application

= ⟦QP⟧a (λxe · ⟦vP**⟧a x/1)

vP* is composed by the numerical index 1 and vP**.
The index triggers the Predicate Abstraction rule.

= ⟦QP⟧a ([λxe · ⟦v’⟧a x/1 (⟦NP⟧a x/1)])

we proceed analyzing vP** as v’ that takes as input

113
NP. From this point downward, everything is interpreted through the modified assignment.

= ⟦QP⟧a ([λxe · ⟦VP⟧a x/1 (⟦NP⟧a x/1)])

we just substitute v’ with VP, since its denotational type has been inherited by its lower node.

= ⟦QP⟧a ([λxe · ⟦V⟧a x/1 (⟦t1⟧)a x/1 (⟦NP⟧a x/1)])

here, the underlined part corresponds to VP node: it is the compositional result of V + t1 node, both
interpreted through the modified assignment function ax/1.

= ⟦QP⟧a ([λxe · [λue · [λze · z interviewed u]] (x) (Val.)])

at this point of the analysis I have reached the t1 and the bottom of the structure. Now, I translate
each node in its equivalent denotational type. For this reason, I keep the lambda prefix that I have
introduced through my Predicate Abstraction rule (λxe), and write down the denotational type of
the V node (which is <e, <e, t>>). Being a transitive verb, it will be a nested function binding a first
variable λue corresponding to the object and a second variable λze corresponding to the subject.
Then, I substitute the trace with index 1 with the variable x, according to the Predicate Abstraction
rule. Last, I substitute the NP node with its denotational type e (Val.).
NOW, I proceed with the usual treatment of functions, dropping lambdas according to their
corresponding variables.

= ⟦QP⟧a ([λxe · Val. interviewed x])

Now I have obtained the set that I wanted to obtain (scope set) to combine with the QP (restriction),
and the truth-condition of the entire sentence is that the set of candidates (QP) is a subset of the set
of people that Valentina interviewed (vP).
In principle it is also possible to proceed with a bottom-up analysis, labeling t1 as ⟦t1⟧a, and then
modifying it into ⟦t1⟧a x/1 when we reach the vP** node.
This is not, anyway, ideal, for our compositional purposes. Bianchi prefers the top-bottom
analysis.

Week 6 exercise
John solved two problems
Remember:
● the QP must move to the top of the structure.
● We can analyze the QP at the beginning, but it is preferable to analyze it at the very
end of the compositional analysis just to save space: as in the previous example, it is
less confusing to analyze the scope set and then, when we have disambiguated it, we
switch to the QP analysis.
● ⟦two problems⟧ = λf<e, t> · |{u | u problem} ⋂ {y | f(y) = 1}| ≥ 2

Compositional analysis
⟦vP**⟧a = ⟦QP⟧a (⟦vP*⟧a)
= ⟦QP⟧a ([λxe · ⟦vP⟧a x/1])

114
= ⟦QP⟧a ([λxe · ⟦v’⟧a x/1 (⟦NP⟧a x/1)])
= ⟦QP⟧a ([λxe · ⟦VP⟧a x/1 (⟦NP⟧a x/1)])
= ⟦QP⟧a ([λxe · [⟦V⟧a x/1 (⟦t1⟧a x/1)] (⟦NP⟧a x/1)])
= ⟦QP⟧a ([λxe · [λye · [λze · z solved y]] (x) (John)])
= ⟦QP⟧a ([λxe · John solved x])
= λf<e, t> · |{u | u problem} ⋂ {y | f(y) = 1}| ≥ 2 ([λxe · John solved x])
= |{u | u problem} ⋂ {y | [λxe · John solved x] (y) = 1}| ≥ 2
= |{u | u problem} ⋂ {y | John solved y}| ≥ 2
Which is the denotational type of t1?
Being an index, t1 is semantically empty, because it does not have a specific denotation, by itself. It
just triggers the application of the Predicate Abstraction rule (and, consequently, the insertion of a λ
prefix). For this reason, is not really necessary to assign a denotational type to t1.90

90
Just remember, the lambda prefix introduced by PA rule binds an entity. In fact, in our tree, V gets somehow
saturated by t1 and VP from NP, such that we obtain a <t> denotation at the vP level. At this point, the PA rule
introduces a lambda binding an entity and it returns an <e, t> type at the vP* level, which gets saturated by QP
and all the tree ends up with a <t> type at the vP** level. For this reason, in a compositional perspective, we
can VERY VERY INFORMALLY PRETEND to assign the denotational type <e> to t1. BUT ACTUALLY, t1 IS
SEMANTICALLY EMPTY.

115
Week 7
Lesson 15
12.01
Top-down procedure: an example
Valentina saw Julius
⟦vP⟧a = ⟦v’⟧a (⟦NP1⟧a)
= ⟦VP⟧a (⟦NP1⟧a)
= ⟦V⟧a (⟦NP2⟧a)(⟦NP1⟧a)
= [λxe · [λye · y saw x]] (Jul.) (Val.)
= Val. Saw Jul.
This is a simple compositional top-down analysis.
“a” is always present because it is an arbitrary
assignment: it can even be empty, but we always must
imagine to interpret the nodes according to the value that
the assignment function gives.
The typical denotation of a node under an assignment
function is: ⟦α⟧M, g, w, where M stands for “module”, g is
the assignment function91, and w representing the
intention. H&K only use the g notation, omitting M and w.

91
We used “a” to denote it.

116
Empirical issues related to QR
In particular, we will see how do sentences with more than one quantifier behave.

A boy climbed every tree

This is an ambiguous sentence in English. It could mean:
● One boy climbed many trees
(i) ∃x · boy (x) & ∀ y | y tree → x climbed y (existential quantifier has scope over

universal)

● For every tree, a boy climbed it

(ii) ∀ y | y tree → ∃x · boy (x) & x climbed y (universal quantifier has scope over
existential)

117
In the interpretation (ii), we have a distributive situation, in which the universal quantifier takes
scope over the existential one.
Plus, there is an entailment relation between (i) and (ii): every situation that can be described by (i)
can be also described by (ii). In other words, whenever (i) is true, (ii) is also necessarily true, but
not vice versa 🡪 (i) ⊆ (ii).
In fact, if we think about it, whenever is true that one boy climbed some trees, it is also true that for
every tree, a boy climbed it: in (ii), in fact, is not specified how many boys are climbing (it could also
be one).

Scope reversal (inverse scope)

The problem of scope reversal relates to the fact that, in some sentences, interpreting the quantifier
in its visible position is not possibly yielding a disambiguation in the interpretation of the sentence:
the mechanism of Quantifier Raising, as other mechanisms that we have mentioned before the
Christmas break (as Montague’s Quantifying In), have been proposed in the literature to remove a
quantifier from its visible position and to assign it to a larger scope, and this was initially motivated
not so much by the problem of type mismatching, because type mismatch could also be solved
through type flexibility. The actual empirical motivation was precisely the mechanism of scope
reversal.
For this kind of ambiguous sentences, we need some kind of mechanism of scope reversal for
interpreting one of the alternative interpretations.
In other words, the most straightforward interpretation of our sentence “a boy climbs every tree” is
(i) ∃x · boy (x) & ∀ y | y tree → x climbed y, where the existential quantifier has scope over the
universal (∃ > ∀); but, if we want to interpret the sentence as (ii) ∀ y | y tree → ∃x · boy (x) & x
climbed y, which is a possible interpretation too, we must consider the universal quantifier as
having scope over the existential (∀ > ∃), and for this reason we need a mechanism (that we call
scope reversal) in order to move the universal quantifier in front of the existential one, to obtain
the description of (ii) in bold, such that we can interpret ∀ as having scope over ∃.
Let’s see how the Quantifier Raising is capable of expressing this kind of ambiguity.

118
Syntactic structures: a boy climbed every tree (ii) and (i)
∀ y | y tree → ∃x · boy (x) & x climbed y

The movement of QP2 allows us to assign wide

scope to the universal quantifier above the
existential quantifier through Quantifier Raising.
QP2 (∀) takes scope over QP1 (∃)

∀>∃

∃x · boy (x) & ∀ y | y tree → x climbed y

(QUANTIFIER IN SITU via type shifting)

This analysis involves the type shifting of QP2. Within
this analysis in situ we do not move QP2, such that QP1
(∃) takes scope over QP2 (∀)

∃>∀

119
 ∃x · boy (x) & ∀ y | y tree → x climbed y

(BOTH QUANTIFIERS RAISING)

In this analysis we move both the quantifiers,
because we must preserve the order that allows
us to intepret the existential quantifier (a boy)
as having scope over the universal (every tree).

∃>∀
Quantifier Raising is one of the possible
solutions to reverse the scope of the two
quantifiers.

Robert May originally formulated the QR rule (1977/85): QR is just a free rule that can apply
whenever we have one or more quantifiers that can be attached in any possible order.
But the point is: if we have no constraint, whenever we have two quantifiers interacting in a
sentence, we can have an ambiguity. But this prediction is not correct.

Scope ambiguities
If QR was completely unconstrained, we would have had an ambiguity in sentences like the previous
one, in all languages. This is not the case, instead.
mIn languages as Italian and Chinese, a sentence like “a boy climbed every tree” does not allow the
interpretation of the universal quantifier taking scope over the existential quantifier.
It seems that the scope reversal problem is particularly linked to the subject/object relation: in a
study conducted by Aoun & Li in 1993, the focus of the discourse relates exactly to the relation
between subject and object: when we have two quantifiers in the object position (as in “I introduce
every boy to a girl”, or “I introduced a boy to every girl”) the distributive interpretation is a possible
interpretation (also in Italian).

Scopally commutative quantifiers

Not all the pairs of quantifiers allow the scope reversal interpretation. Some pairs of quantifiers are
scopally commutative. It means that the interpretation of the sentence does not change if we move
one of the two quantifiers over the other:
● QP1 > QP2 ↔ QP2 > QP1
It does not matter which quantifier has scope over the other: the truth-conditions do not change.

120
Example: SCOPALLY COMMUTATIVE (QP1 > QP2 ↔ QP2 > QP1)

● Every student1 hates every professor2 = Every professor2 is hated by every student1
∀1 > ∀2 ↔ ∀2 > ∀1
But even when we have scope non-commutative pairs, some sentences do not allow the type of
interpretation that sentences like “a boy climbed every tree” has.

Example: SCOPALLY NON-COMMUTATIVE (QP1 > QP2 ↔ QP2 > QP1)

● Every boy1 solved most of the problems2

Here we notice that this pair is scopally non-commutative: reversing the scope of the quantifiers, we
obtain two possible interpretations, not the same one.
1. For every boy, there is a subset of the set of problems, that he solved92 (∀ > most)
2. There is a subset of problems that corresponds to most problems, and all the boys solved it93
(most > ∀)

● A boy1 solved every problem2

It is possible to have “every problem” (∀) taking scope over “a boy” (∃), giving a distributive
interpretation (for every problem there is a boy such that every problem is solved by a boy. In this
interpretation we have maybe more than one boy, and each of them solved a problem. At the end all
of the problems are solved).

● Most boys1 solved every problem2 🡪 NO POSSIBLE SCOPE REVERSAL

It is not possible, instead, to have “every problem” taking scope over “most boys”. In other terms, this
sentence does not allow a distributive interpretation (this sentence cannot mean that different
subsets of boys related to every problem: every problem is solved by only one set that correspond to
the set of most boys together) 🡪 *∀ > most

● Every student1 joined no committee2 🡪 NO POSSIBLE SCOPE REVERSAL

We cannot have an interpretation in which “no” takes scope over the universal quantifier
(impossible interpretation: there is no committee such that it was joint by every student). The only
possible interpretation is ∀ > no, or, “no student is present in any committee” 🡪 *no > ∀

From this analysis we draw the conclusion that the mechanism of inverse scope is sensitive to the
nature of the quantifier: it is possible when the quantifier is a weak quantifier, but not when it is a
strong quantifier.94

92
Distributive interpretation: 10 problems 🡪 boy1 solved problems 1-3, boy2 solved problems 4-6… and,
summing the small subsets of problems that have been solved, most of the problems are solved, each subset
by a different boy.
93
Non-distributive interpretation: 10 problems, the subset including 6 of them is solved by all the boys
together.
94
See pg. 85 for discussion on weak and strong quantifiers (symmetric 🡪 weak; asymmetric 🡪 strong).

121
Conclusion
These few examples give us the idea that the problem of scope interaction between two quantifiers
is much more complex than what can seem at first sight, and there is also an important role of the
context of interpretation which may point out a particular interpretation over the other or
completely delete one of the two interpretations.
Moreover, we must take into account that we are dealing with sentences that in principle allow for
two interpretations and then we try to see whether the scope reverse interpretation is really
available or not.
It turned out that for some non-commutative quantifiers, the scope reversal is not available, that
means that the QR mechanism must be somehow constrained.
Since the mechanism of scope reverse is analyzed in terms of Quantifier Raising, with one quantifier
being moved in a higher position with respect to the other one, a solution that comes to mind is
something similar to the minimality principle: for certain uses of quantifiers, the second one cannot
be moved beyond the first one.
Another solution, stated by Beghelli & Strowell, 1997, proposed that in the functional structure of
the sentence there are dedicated positions for different subclasses of quantifiers: for certain
quantifiers the landing site can never be reversed. Using our previous examples, an existential
quantifier as “a boy” can either stay inside the scope of the other quantifier, or be moved in a
position which is external to the scope of the universal quantifier; while a quantifier like “most
boys” will always necessarily land in a position that is higher of the universal quantifier.
In this perspective, the ambiguity that we find in “a boy solved every problem” is simply due to the
fact that the existential quantifier can cover two different positions in the structure, while we have
no ambiguity in “most boys solved every problem” because the quantifier “most” must necessarily be
analyzed in a position external with respect to the universal quantifier.

Inverse linking
Scope reversal is a phenomenon that has been invoked to justify a scope extension mechanism like
Quantifier Raising, but QR must me somehow constrained, to explain the fact that in some cases
scope reversal is not possible.
Another interesting phenomenon is called inverse linking95:

An apple in every basket is rotten

∀ y | y basket ∃ x | x apple & x is in y & x is rotten
This sentence means that there are as many apples as the number of baskets (distributive
interpretation). Here, the universal quantifier takes scope over the existential (basket over apple).
If we look at the syntactic structures, we will see that one of the two quantifiers is nested inside the
structure of the other (universal nested in existential).

95
Chapter 7 of H&K.

122
An apple in every basket is rotten
Here we see how QP2 (∀) is nested in QP1 (∃). The
scope extension mechanism works in such a way to
remove QP2 from its position and move it outside the
QP1, to finally make it reach a higher position in the
structure.
If we do not do so, and leave QP2 in its position, trying to
analyze it in situ, the only possible interpretation that we
will get is the implausible interpretation of the existential
quantifier having scope over the universal quantifier that
is nested in it, giving the meaning that we only have one
apple and this apple is collocated in many baskets (∃x |
x apple & ∀ y | y basket → x is in y & x is rotten not
possible).
Therefore, here, we necessarily need to apply some
mechanism of scope reversal.
This mechanism is called inverse linking in the sense that
the nested quantifier has to be removed and be assigned
scope above the originally containing quantifier.

Anaphor binding
In a sentence like:
Mary1 likes herself1
we consider Mary and herself to be referring to the same entity, and this is normally expressed with
an index shared both by the anaphor and the antecedent. This is called “anaphor binding”, that is
subject to locality condition in the sense that the anaphor must be locally close to the referent.
Since the referent is the same, our sentence has the same meaning of “Mary likes Mary”. So, we could
say that whenever an anaphor is close to an antecedent, it simply inherits the denotation of the
antecedent.
BUT NO. TOO EASY.
Let’s see where the complication is.
If we consider a sentence with a quantificational subject, we will see that we cannot apply this
intuitive deduction: we cannot generalize that in all cases the anaphor inherits the denotation
of its antecedent, as in the following instance:
Every girl1 likes herself1 ≠ Every girl likes every girl
In this example, we can see that the first sentence means that each of the girls loves herself, but if
we consider the anaphor “herself” being equivalent to the antecedent “every girl”, we will have the
meaning that each of the girls loves each of the girls, and this is not the case.

123
Solution:
Apply Quantifier Raising to the subject.
The idea is: the subject quantifier is removed and
attached on the top of the clause. It triggers the
application of the Predicate Abstraction rule.
⟦vP**⟧a = ⟦QP1⟧a (⟦vP*⟧a)
= ⟦QP1⟧a (λxe · ⟦vP⟧a x/1)
= ⟦QP1⟧a (λxe · ⟦v’⟧a x/1 (⟦t1⟧a x/1))
= ⟦QP1⟧a (λxe · ⟦v⟧a x/1 ⟦DP⟧a x/1 (⟦t1⟧a x/1))
= ⟦QP1⟧a (λxe · [λye · [λze · z likes y]] (x) (x))
= ⟦QP1⟧a ([λxe · x likes x])
As we see, in the structure we have two occurrences of the
variable x, one binding the trace of QP1, and the other one
binding the anaphor herself.
All cases of anaphor binding can be analyzed through the
insertion of the index that triggers the Predicate
Abstraction.96

96
With this lesson we concluded chapter 7 of H&K. The next lessons will be focused on chapter 10.

124
Lesson 16
13.01
Syntactic binding & semantic binding
In the last lesson we introduced the notion of semantic binding through the Predicate Abstraction
rule. Today, we will start recalling the distinction between syntactic binding and semantic binding.

Syntactic binding (every boy loves his mother)

Syntactic binding refers to the position in which we see elements in
the syntactic structure 🡪 co-indexing 2 positions.
QP syntactically binds the pronoun DP1 (his), because:
1. the two positions carry the same numerical index,
2. QP c-commands the pronoun from the surface position (the
position that it has when it is pronounced, without necessarily taking
into consideration the optional movement; the position where the
th-role is assigned).
3. We can add a locality constraint: there is no other element in
the thematic position that shares an index with this pronoun and that
syntactically intervenes (in terms of c-commanding) between the QP
and the DP. In other words, QP is the closest argument position that
c-commands the pronoun and shares an index with it.
In this case, QP syntactically binds the pronoun.

Semantic binding (every boy loves his mother)

Semantic binding was formulated not in terms of having an index
shared by the QP and the pronoun, but, instead, appending an
index to the sister node of the quantifier itself.
So, semantic binding is obtained by adjoining the quantifier on the
top of the structure, adding one layer below it and above the rest of
the vP, assuming this intermediate layer vP* to have an index as
daughter node (and sister of vP). This triggers the Predicate
Abstraction rule. What binds the pronoun here is the same index
that binds the trace of QP. So, we want to obtain both of them to be
two occurrences of the same bound variable.
⟦vP*⟧a = [λxe · ⟦vP⟧a x/1] = λxe · x loves x’s mother (x loves the unique
entity y such that y is mother of x).
As we can see, what we obtain is a double occurrence of x bound by
the lambda prefix (one corresponding to the t1 of QP; the second
corresponding to the pronoun D1).

125
H&K established a connection between the syntactic binding configuration and the semantic
binding configuration: QP semantically binds pronoun “his” iff “his” is bound by the Abstraction
Operator that binds the trace of QP.
In a derivative sense, QP semantically binds a pronoun but this means that the pronoun is captured
by the same lambda that binds the QP.97 It is another way to say that the lambda prefix necessarily
semantically binds both the QP and the pronoun.
On the basis of this connection, Heim & Kratzer proposed a binding principle.

Binding principle
● “α semantically binds β if and only if α syntactically binds β”
In other words, every semantic relation relies on a syntactic binding and vice versa, because there is
a double implication between them.
The empirical consequence of this is that a quantifier only binds a pronoun that it c-commands from
its visible position98.
The prediction is: if QP does not c-command the pronoun from its surface position, it cannot
bind it semantically.

His mother loves every boy

Quantifier Raising can move the QP1 outside the vP
structure, in a higher position, and attach an index 1
to the sister node of vP, triggering the Predicate
Abstraction rule.
As we can see from the structure, the binding index 1
has in its scope the pronoun “his”. However, the D1
“his” is not c-commanded by QP1 in its visible
position, but only from the derived position that it
assumes in the logical form.
In fact, according to Reinhart 1983, we see that a
sentence like this cannot mean that for every boy
there is a mother such that this mother loves her son
(*∀ x | x boy → x’s mother loves x):
native speakers will notice that this interpretation is
not possible, because the only possible interpretation
is that someone’s mother loves every boy, where
“every boy” is not sharing the index with the pronoun
“his”.
So, even if we have Quantifier Raising, the pronoun is not related to the same index in this sentence.
This phenomenon is called weak crossover (wco).

97
Pg. 263 H&K
98
With “visible position”, or “surface position”, we mean the position that the QP occupies before undergoing
QR.

126
Weak crossover (T. Wasow 1968)
It is a piece of evidence that the pronoun must be c-commanded from QP by its visible position (=
syntactically bound) to be semantically bound.
The binding principle (that links syntactical and semantic binding) captures the requirement that
the pronoun must be c-commanded by the argument position that the QP originally fits.
In this structure movement of QP is allowed, but still, it is impossible for it to bind the pronoun
because it does not c-command it: if QP does not c-command the pronoun, it cannot syntactically
bind it; and if it cannot syntactically bind it, it cannot semantically bind it.

Conclusion
In order for the pronoun to be semantically bound, two conditions must be present:
● The pronoun must be in the semantic scope of the quantifier;
● It must be syntactically bound (pronoun must be c-commanded by quantifier).

Counterexamples to the c-command requirement

Confronting the two sentences: “every boy loves his mother” and “his mother loves every boy” the
immediate observation is that the quantificational antecedent linearly precedes the pronoun in the
first sentence, and it linearly follows the pronoun in the second sentence. So, an alternative
generalization is the so-called “leftness condition” (Chomsky, Homstein, Barker).
Leftness condition is a very debated topic, mainly because in syntax we do not really rely on the
linear order of words: in fact, syntax is built in a hierarchical way. And this fact brought some
problematic issues in the discussion of this argument.

Free pronouns (unbound pronouns)

A pronoun which is free or unbound is a pronoun that is not subject to syntactic and semantic
binding; therefore, a pronoun that is not c-commanded by a co-indexed argument in the structure.
Pronouns can have a deictic use and an anaphoric use99.
Deictic use
“He is my friend” (pointing to a friend). Accompanied by a demonstrative gesture, it is normally
assumed that the pronoun refers to the entity the speaker is pointing to.
In the deictic use, the pronoun clearly does not have a linguistic antecedent.
Anaphoric use
“Peter got sick. He ate too many candies”: in this case the pronoun is not syntactically bound to its
antecedent (Peter), because there is a full stop that represents a linguistic boundary.
“He” is an anaphoric pronoun, and the relationship that we have between “he” and “Peter” is called
co-reference. This link is not established through syntactic or semantic binding.

99
NOTE: anaphoric refers to the phenomenon of “anaphora” that is not the same as “anaphor”: anaphors are
elements as “himself” or “herself” and they always have a very locally close linguistic antecedent (we saw them
in the previous lesson, “anaphor binding”, pg. 123).

127
In fact, if we try to substitute “Peter” with a quantifier as “no boy”, it does not work.
● *No boy got sick. He ate too many candies
In the case of the quantifier “some”, the substitution is possible. In fact, a DP as “some boy” does not
necessarily have a quantificational interpretation, but it can also have a quasi-referential
interpretation.
● Some boy got sick. He ate too many candies100
To make a long story short, we have two ways to establish a relationship between a
non-quantification antecedent and a pronoun: no syntactic binding and co-reference.

Context dependency in assignment

We said that a pronoun carrying an index must be interpreted according to a specific function
assignment (a), which is determined by context.

⟦his1⟧a = a(1) 🡪 is context dependent. For this reason, truth-conditions become relative to the
context: if the assignment assigns the entity “Mary” to index 1, index 1 will be equivalent to “Mary”;
if the entity assigned to index 1 is “Juan”, then all occurrences of 1 will indicate “Juan”. This happens,
of course, to free pronouns (both in their deictic and their anaphoric use).

⟦he1⟧a = a(1) = Peter101 🡪 In our previous example (Peter got sick. He ate too many candies)
the second occurrence “he” will be bound by its index to the entity “Peter”. In fact, the previous
mention of the noun “Peter” has raised the salience of that individual in the context. And this is
captured by assigning the same entity to the index 1 in the assignment function.

λxe ⟦vP⟧a x/1 🡪 When we are in a situation of syntactic and semantic binding, the index is
captured by the Predicate Abstraction rule: the modification of the assignment function assigns the
variable x to the index (corresponding to the PA rule).
Take a sentence like “he is smart” (pointing to someone). The truth-value of this sentence is
dependent on the context, because the assignment function (that specifies, in fact, the context of
utterance) assigns to the pronoun a particular referent. In fact, free pronouns do not have an
interpretation outside the utterance context: “he” is not referring to anyone unless I specify the
assignment function.
For this reason, the context of utterance becomes crucial for the interpretation process by fixing the
reference of free pronouns.102
In this frame, we are still working on a purely extensional level, without taking intension in account.
This point is completely left aside by H&K.
The mechanism of assigning an entity to a pronoun is called “direct reference”, because there is no
mediation of a descriptive content.

100
Discussed in chapter 11 H&K.
101
Or, we can write: a = [1 🡪 Peter]
102
Full discussion in chapter 9 of H & K

128
 Co-reference and binding: double interpretation
Take as instance a sentence like “John loves his wife”. We have two ways to interpret it, and their
truth-conditions will be equivalent.

John loves his wife

a: λx ·
[1 = x

John] 🡪 co-reference (no syntactic binding) loves x’s wife 🡪 binding (PA binds t1 & his1)

In this analysis, the pronoun his carries index 1, and In this analysis we trigger the PA rule inserting an index
the assignment function contains the information 1 that binds both t1 (John) and D1 (his). The Predicate
that index 1 corresponds to entity John. For this Abstraction rule introduces a lambda that binds index 1,
reason, we can associate John and his to the same and specifies the modification of the assignment
entity (John) without the mediation of binding. function: 1 = x.

Whenever the antecedent is a NP which originally denotes an entity, what we obtain at the end will
be two logical forms that express exactly the same truth-conditions. In fact, the co-reference case
will give the final interpretation “John1 loves John1’s wife”, while the binding process will lead to the
interpretation “x loves x’s wife/John = x”, which is exactly the same.
So, if both of the analysis are valid in this case, why do we distinguish them? Why do we have the
necessity to maintain two different ways of analyzing such structures?
Reinhart proposes a number of empirical arguments to keep this distinction. One of the most salient
concerns the ambiguity in elliptical structures.

129
Elliptical structures and Strict vs Sloppy ambiguity
Elliptical structures
Consider the two sentences:
I. John went to his office
II. Peter did too
The second sentence is considered to be a VP ellipsis: it is parallel to the first one (in a certain sense
it “borrows” the VP from the previous sentence). This phenomenon has been studied since the ‘70s
of the last century.
In the elliptical sentence we do not realize the predicate: its interpretation is parallel to the
preceding clause’s. The parallelism is a constraint on VP-ellipsis: it is impossible to realize an elliptic
clause if it is not parallel to the preceding.

Strict vs Sloppy ambiguity

Here we can note an important thing regarding the elliptical sentence II: it is ambiguous in the
sense that, since it borrows the interpretation of the previous clause’s VP, it can bear two different
interpretations:
a) Peter went to John’s office ("his” is free, co-reference) -STRICT
b) Peter went to Peter’s office (“his” is bound, binding configuration) -SLOPPY

Strict [a : 1 = John]
John went to his office Peter (went to John’s office)

130
Sloppy [λxe ⟦vP⟧a x/1]
John went to his office Peter (went to Peter’s office)

1. ⟦vP**⟧a = ⟦vP*⟧a (⟦NP⟧)a

= [λxe ⟦vP⟧a x/1] (⟦NP⟧)a =
= [λxe ⟦v’⟧a x/1 (⟦t1⟧)] (⟦NP⟧)a =
= [λxe ⟦V⟧a x/1 (⟦PP⟧a x/1) (⟦t1⟧)] (⟦NP⟧)a =
= [λxe · x went to x’s office] (John)
In the Peter’s sentence we do not need to change the index, because index 1 in our modified
assignment refers to x, and not to John. John as input is taken by the vP* in the first structure. In the
second structure, the NP is not John anymore, it is Peter. Therefore, the analysis of the second
structure will be identical to the first one except for the last passage, where the NP “Peter” will be
taken as input by the rest of the vP*, converting the variable x into the NP “Peter”.
2. (same as 1., but) last passage
= [λxe · x went to x’s office] (Peter)

Anaphors (binding configuration)

Considering these following sentences, we will notice that an elliptical configuration can only give
rise to the sloppy interpretation.
- John1 admires himself1
- Peter does too
The meaning of the elliptic sentence is that Peter is admiring himself (Peter), not John.
The point made by Reinhart is that whenever binding is obligatory, the sloppy reading is the only
possible interpretation.

131
Grammatical features on pronouns103
Up to now we have been pretending that the only thing that the interpretive system sees in
pronouns to assign them a reference is the index. In this sense, it seems that the grammatical
features of pronouns are irrelevant for the interpretation process.
but, of course, we can intuitively capture that grammatical features are relevant in this perspective:
a sentence like “John likes her mother” shows a feature mismatch between “John” and “her” such that
we immediately grasp that the pronoun “her” is not referring to John. In fact, “her” is a feminine
pronoun, while “John” is a name that is traditionally used for male individuals.
The solution that was proposed by R. Cooper in 1983 is the so-called presuppositional account: it
is a partial identity function. An identity function is a function that takes as input x and returns as
output exactly the same element x.
⟦[feminine]⟧a = λx : x is female · x
⟦1⟧a = a(1)
⟦[feminine]⟧a (⟦1⟧a) = [λx : x is female · x] (a(1))
In this way, the feature in a partial identity function essentially works as a filter: it throws away all
the possible assignment of values to the index that do not satisfy the requisite of being female. In
this way we attribute a role to the grammatical feature in the assignment function.
We can have a similar treatment for the plural feature.
● ⟦[plural]⟧a = λx : x is not atomic · x
Only if the condition of being non-atomic is satisfied, the output can be the same x of the input.

Personal pronouns
The presuppositional account on personal pronouns is a very interesting topic: lately a crucial issue
has risen about the use of personal pronouns for what concerns gender identity.

🡺 Gender is ultimately a morpho-phonological constraint. Therefore, we can imagine it to be

neutral in the lexical form, and manifested in the phonological form of the clause: PH
(gender)/ LF (neutral).
Another issue related to personal pronouns is the possibility for them to be bound.
I104, you: In an elliptical sentence as: “Mary did her homework. John did not.”, we intuitively grasp that
John did not do HIS (John’s) homework; we are not referring to Mary’s homework. If I have a
presuppositional requirement (λx : x is female · x) in the first sentence, I cannot apply it to the
elliptical one, because it would not satisfy the condition of being feminine. For this reason, we
consider the second sentence’s elliptical construction as referring to John, and not to Mary.
This discussion makes us understand that person features also apply to personal pronouns if they
are bound (ex: only I did my homework. You didn’t. Peter didn’t.).
I/my = λx : x is the speaker · x 🡪 not valid for “you”.

103
Section 9.1.2 H&K
104
“I” is an indexical element: it will always refer to the speaker.

132
Lesson 17
14.01
Intensionality
At the very beginning of this course, we established the difference between intensionality and
extensionality, taking as example tautologies like “Rome is Rome”105 (a proper name always denotes
the same entity, this is why we call it “rigid designator”).
We said that this sentence is not very informative, and it is a tautology (namely: it is always true);
while sentences like “Rome is the capital of Italy” are informative and contingent, and they could be
true or false according to the particular context of utterance (for example, “Rome is the capital of
Italy” is true in our actual context, but it is false if I am referring to 1863).
What we will focus on now, is that a natural language expression can yield different possible
truth-values according to different circumstances of evaluation.
Consider a sentence like:
● Rob believes that Rome is the capital of Italy
We consider this sentence to be true for what concerns the capital of Italy in our actual
circumstance of evaluation, and true for the fact that Robs believes it, but now consider:
● Rob believes that Florence is the capital of Italy
In this sentence it is false, under our actual circumstances of evaluation, that Florence is the capital
of Italy, but it is true that Rob believes it (maybe Rob is not well-informed and he really believes that
Florence is the capital of Italy).
So, how do we treat the discordance between the truth-value true referring to Rob’s belief and the
truth-value false referring to our actual circumstance of evaluation of the capital of Italy?
Note that an analysis of only one contextual circumstance is not capable of capturing the complexity
of our language’s possibility of a multiplicity of worlds.
We can represent an attitude like “belief” as an attitude which from a particular circumstance opens
a particular modal horizon, a set of possible worlds that are doxastically106 accessible to a given
individual under a given circumstance of evaluation.
In other words, we can compare the world in which Rob believes that the capital of Italy is Florence
with other worlds: some of them are compatible and some of them are not.
Florence is the capital of Italy
WORLD IN WHICH IT IS TRUE: WORLDS IN WHICH IT IS FALSE:
● a movie in which Florence is capital of ● Year 2021’s reality
Italy ● A well-informed person’s belief
● Rob’s beliefs ● During paleolithic era

105
Pg. 2, “Intensional and extensional semantics”
106
Doxastically derives from ancient Greek. It is a technical term to indicate a belief attitude.

133
 ● Year 1867’s reality
The possible worlds in which the capital of Italy is Florence are compatible with Rob’s belief, while
the worlds in which it is not true are not compatible with Rob’s belief.
So, we can represent the belief state of an individual as a set of circumstances that are compatible
with a given body of information. This can only be analyzed if we refer to a multiplicity of
circumstances of evaluation (possible worlds).

Substitution problem
It has been noticed that, if I am reporting a belief, it is not possible to substitute a particular term
with a term that has the same extensional meaning under different circumstances of evaluation.
For example, Rob believes that the capital of Italy is Florence (very beautiful to him); while Rome,
(which is not the capital to him) is an ugly city.
In this case we cannot say that Rob believes that the capital of Italy is ugly, because under our
circumstances of evaluation the capital of Italy is Rome, but he thinks it is Florence. Therefore, if we
ask him “hey Rob, do you think the capital of Italy is ugly?” he will answer “No, it is a spectacular
city!”, precisely because he attributes to Florence the property of being capital of Italy, and not to
Rome as we do. In this case, our circumstances of evaluation and Rob’s circumstances of evaluation
are not compatible.
This reasoning demonstrates that we cannot substitute “the capital of Italy” with “Rome”:
A: hey Rob, do you think the capital of Italy is ugly?
R: No, it is a spectacular city!
A: hey Rob, do you think Rome is ugly?
R: Yes, I did not like it at all!
The ability to attribute false beliefs and distinguish between one’s own beliefs and other people’s
beliefs, is an important cognitive ability, and it is a very crucial issue that has been accurately
investigated in cognitive development studies.
How can we take in account the possible different worlds in our intensional level of interpretation?
By adding the intensional types to our formal apparatus of analysis that included, until now, only
extensional types107.

Intensional semantic types

1. e is a type
2. t is a type
3. s is a label
4. for any types α and β, <α, β> is also a type
5. for any types α, <s, α> is a type
6. Nothing else is a type.

107
We outlined the extensional semantic types at pg. 16.

134
Where “s” is the type of any possible world D<s, α> : W → Dα
“s” type represents a function that links an entity to its circumstance of evaluation (namely: to the
possible world to which it is linked in context). What does it mean?
We label an entity with s that corresponds to the function that takes as input a specific possible
world and returns as output the entity that it refers to, under that possible circumstance of
evaluation.
In other terms, we keep what we have established until now, and we introduce a further label that
interprets the semantic types we have studied in a particular context.
⟦the capital of Italy⟧ = [λw · Jxe · [x is capital of Italy in w108] (w1) <s, e>
⟦Italy⟧ = [λw · the state of Italy109 in w] (in w1) <s, e>
⟦teacher⟧ = [λw · [λxe · x is a teacher in w] (w1) <s, <e, t>>
⟦every⟧ = [λw · [[λf<e, t> · [λg<e, t> · {x | f(x) = 1} ⊆ {y | g(y) = 1}] (w1) <s, <<e, t>, <e, t> t>>>.

Heim & Kratzer labels

Rome is ugly
Heim and Kratzer use the notation […]w instead of inserting a λw. So, for example, the denotation of
“ugly” for H&K is:
⟦ugly⟧w = [λxe · x is ugly in w]110
⟦TP⟧w = ⟦T’⟧w (⟦NP⟧w)
⟦TP⟧w = [λxe · x is ugly in w] (Rome)
⟦TP⟧w = Rome is ugly in w.

Belief report: an intensional analysis

Let’s see what happens if we try to analyze a sentence that is embedded from a belief report.

John believes that Rome is ugly

…⟦TP⟧w = ⟦T’⟧w (⟦NP⟧w)

108
We added this dependence that is bound by λw.
109
Remember: Italy (N) is a rigid designator; it always denotes the same entity (constant function).
110
Instead of ⟦ugly⟧ = [λw · [λxe · x is ugly in w]]

135
The CP contributes an intension: in fact, in our last structure (Rome is ugly in w) we analyzed a
generic circumstance of evaluation. But here, with the CP, we want to underline that this
circumstance of evaluation we have considered is relative to John’s belief.
For this reason, we have obtained at the TP level a truth-value for the sentence “Rome is ugly”. Now
we want to link that truth-value to a multiplicity of worlds including John’s belief, to bind the
truth-value of TP (Rome is ugly) to the fact that it is a John’s belief (a set of possible worlds <s, t>.
This is what we call a proposition).
We want to have this CP contributing to the whole clause conveying a proposition, a set of possible
worlds. In fact, <s, t> is structurally analogous of type <e, t>. If <e, t> is the characteristic function of
a set of entities, <s, t> is the characteristic function of the set of possible worlds (in which that
particular clause has the value true).
⟦CP⟧w = λw · ⟦TP⟧w 🡪 input: a function of possible worlds; output: extension of TP at those possible
worlds.
We can consider that the C node triggers the introduction of a λ prefix and the insertion of the
intensional parameter.
H&K do provide another kind of analysis of this phenomenon. We will talk about it later, but first we
will introduce the dentation of a verb like “believes”:
⟦believes⟧w = a verb like this111 expresses a relation between an individual and a proposition, or, the
attitude that a subject has regarding a certain preposition
Some worlds are compatible with John’s beliefs, some others
do not. John’s beliefs can be relativized to the compatible
possible worlds.
We can say that w2, w3 and w4 are doxastically accessible to J in
w1 if they conform to everything that J believes in w1.
In this sense a binary relation is established between the
worlds: <w2, w1>; <w3, w1>.
DOXJOHN (W1) = set of accessible worlds (w2, w3, w4).
What these compatible worlds have in common with w1 is the
informational content of the believe state of John: John is in believe relation with P (where P is
“Rome is ugly”).
For this reason, ⟦believes⟧w1 = takes as input a proposition and returns the set of entities <<s, t>, <e,
t>>
⟦believes⟧w1 = [λP<s, t> · [λxe · ∀ w2 ∊ DOXx (w1), P (w2) = 1
All words w2 that belong to the set of worlds that are doxastically accessible to x (John) in w1 are worlds that belong in
the proposition P; therefore: P applied to w2 returns the value true.

Or
⟦believes⟧w1 = [λP<s, t> · [λxe · DOXx (w1) ⊆ {w3 | P(w3) = 1}]]

111
Verbs like “think”, “believe”, “guess” are called propositional attitude verbs.

136
Compositional analysis: at the TP level we have truth-condition respect to the sentence “Rome is
ugly”. And at the CP level we want to have a function that maps possible worlds with the
truth-condition of “Rome is ugly”. So, one way to analyze it is the one we previously mentioned
assuming that the C node triggers the insertion of a lambda prefix such that TP must be interpreted
according to w1 (⟦CP⟧w = λw · ⟦TP⟧w). We also said that H&K do not follow this procedure.

Heim & Kratzer analysis

Heim & Kratzer propose a rule to be applied precisely when we have a context of an intensional
operator:112
The rule states that ⟦β⟧w requires an input of type <s, t>
(with t as semantic type of γ),
then ⟦α⟧w1 = ⟦β⟧w1 (λw2 · ⟦γ⟧w2) 🡪 intensional function
application
⟦v’⟧w1 = ⟦v⟧w1 (λw2 · ⟦CP⟧w2)
This rule does two things:

🡺 Function application
🡺 Intensionalizes the CP (γ)

⟦v’⟧w1 = [λP<s, t> · [λxe · ∀w3 ∊ DOXx (w1), P(w3) = 1]] (λw2 · Rome is ugly in w2) =
⟦v’⟧w1 = [λxe · ∀w3 ∊ DOXx (w1), (λw
2 · Rome is ugly in w2) (w3) = 1] =

⟦v’⟧w1 = [λxe · ∀w3 ∊ DOXx (w1), Rome is ugly in w3 = 1]

In this analysis:
● w1 stands for the evaluation world of the whole sentence “John believes that Rome is ugly”;
● w2 is the abstraction operator introduced by the intensional function application at the CP
level.
● w3 is a variable bound by a universal quantifier which defines the worlds that are
doxastically accessible to x in w1. It is relative to the verb “believe” (v node), and restricts
the set of possible accessible worlds.

Montague’s solution
As we already said, Montague had the idea of a complete isomorphism between syntax and
semantics: one and only one specific semantic type could correspond to a single syntactic node.
⟦T’⟧ = ⟦AP⟧ = ⟦ugly⟧ = [λw1 · [λxe · x is ugly in w1]]
⟦NP⟧ = ⟦Rome⟧ = [λw2 · Rome]
At this point we notice that the type of T’ and NP do not match. So,
how do we compose them according to Montague?

112
Chapter 12 H&K.

137
⟦TP⟧ = λw3 · ⟦T’⟧(w3) (⟦NP⟧(w3))
In a few words, before we analyze the nodes respectively in a possible world, then we link them to
world 3 through a rule that is somehow similar to the rule that we use to link the subject and the
object to the same event, in event semantics.

⟦TP⟧ = λw3 · [λw1 · [λxe · x is ugly in w1]] (w3) ([λw2 · Rome] (w3 ))
⟦TP⟧ = λw3 · [λxe · x is ugly in w3] (Rome)
⟦TP⟧ = λw3 · [Rome is ugly in w3]
In this example we have no node that wants as input an intensional type.

Montague’s solution with intensional operators

⟦v⟧ (intension) = [λw1 · [λP<s, t> · [λxe · ∀w2 ∊ Doxx (w1), P(w2) = 1]

⟦v’⟧ = λw3 · ⟦v⟧ (w3) (⟦CP⟧)

(similar to the example with no intensional type, but the CP is
not applied to w3 here).

= λw3 · [λw1 · [λP<s, t> · [λxe · ∀w2 ∊ Doxx (w1), P(w2) = 1] (w3)
(⟦CP⟧) =
= λw3 · [λP<s, t> · [λxe · ∀w2 ∊ Doxx (w3), P(w2) = 1] (⟦CP⟧)
= λw3 · [λxe · ∀w2 ∊ Doxx (w3), ([λw4 · Rome is ugly in w4]) (w2) = 1] =
= λw3 · [λxe · ∀w2 ∊ Doxx (w3), (Rome is ugly in w2) = 1] = <s, <e, t>>.
At the end of our analysis what we obtain to say is that our circumstance of evaluation belongs to
the preposition (in this case, Rome is ugly) 🡪 w@ ∊ ⟦TP⟧, or, the world we are, is a world that
belongs to that proposition.
It is also possible to extend this treatment to other intensional operators. For example, in cases like
the adjective “alleged”, as in “(someone is an) alleged murdered”
⟦AP⟧ (intension) = λw1 · λq<s, <e, t> · λxe · for some worlds in w2
compatible with relevant evidence that x is a murderer in w2.
Q (w2) (x) = 1

An intentional operator is defined so because it takes as input the intension of its sister node, and
not just the extension of it.

138
Week 7
Exercise 5: Bound and free pronouns
She introduced every boy1 to his1 trainer
(assuming an utterance context in which, by using “she”, the speaker intends to refer to Lucy).
Remember:
● “Introduce” is a three place predicate and “to” is semantically empty.
● “Trainer” will be analyzed as a relational noun (relation between two entities: the
trainer and the trainee)

Compositional analysis

∀ y | y boy, v introduced y to z | z is y’s

trainer.

⟦she2⟧a = a(2) = Lucy

⟦vP**⟧a = (⟦QP⟧a) ⟦vP*⟧a

⟦vP**⟧a = (⟦QP⟧a) λx · ⟦vP⟧a x/1

⟦vP**⟧a = (⟦QP⟧a) λx · ⟦v’⟧a x/1 (⟦DP⟧a x/1)

⟦vP**⟧a = (⟦QP⟧a) λx · ⟦VP⟧a x/1 (⟦DP⟧a x/1)

⟦vP**⟧a = (⟦QP⟧a) λx · ⟦V’⟧a x/1 (⟦PP⟧a x/1) (⟦DP⟧a x/1)

⟦vP**⟧a = (⟦QP⟧a) λx · ⟦V⟧a x/1 (⟦t1⟧a x/1) (⟦PP⟧a x/1) (⟦DP⟧a x/1)

139
 ⟦DP⟧a x/1 = ⟦the⟧a x/1 (⟦trainer⟧a x/1 (⟦his⟧a x/1))113
⟦DP⟧a x/1 = ⟦the⟧a x/1 [λte · [λre · r is the trainer of t]] (x)
⟦DP⟧a x/1 = ⟦the⟧a x/1 [λre · r is the trainer of x]]
⟦DP⟧a x/1 = [λf<e, t> · f characterizes singleton set · Jue : f(u) = 1] ([λre · r is the trainer of x]])
⟦DP⟧a x/1 = Jue : ([λre · r is the trainer of x]]) (u) = 1]
⟦DP⟧a x/1 = [Jue : u is the trainer of x = 1]

(⟦QP⟧a) λx · [λy · [λz · [λv · v introduced y to z]]](x) ((⟦P⟧a x/1) ⟦DP⟧a x/1) (she2)

(⟦QP⟧a) λx · [λz · [λv ·v introduced x to z]]] (⟦DP⟧a x/1) (Lucy)

(⟦QP⟧a) λx · [λz · [λv ·v introduced x to z]]] ([Jue : u is the trainer of x]) (Lucy)

(⟦QP⟧a) λx · [λv · v introduced x to [Jue : u is the trainer of x]]] (Lucy)

(⟦QP⟧a) λx · [Lucy introduced x to [Jue : u is the trainer of x]]

⟦QP⟧a = ⟦every⟧ (⟦boy⟧) = ⟦Q⟧a (⟦NP⟧a) =

⟦QP⟧a = [λf<e, t> · {x | f(x) = 1} ⊆ {y | g(y) = 1}]] ([λue · u is a boy]) =
⟦QP⟧a = [λg<e, t> · {x | [λue · u is a boy] (x) = 1} ⊆ {y | g(y) = 1}]] =
⟦QP⟧a = [λg<e, t> · {x | x is a boy} ⊆ {y | g(y) = 1}]]

[λg<e, t> · {x | x is a boy} ⊆ {y | g(y) = 1}]] (λx· [Lucy introduced x to [Jue : u is the trainer of x]]) =

{x | x is a boy} ⊆ {y | (λx · Lucy introduced x to[Jue : u is the trainer of x]) (y) = 1} =

{x | x is a boy} ⊆ {y | Lucy introduced y to Jue : u is the trainer of y)}

Week 8
Lesson 18
19.01
Intensional composition interpretation
Last time we analyzed the case of propositional attitudes: verbs or predicates that express a
relation between an individual and a proposition. And we saw that we can characterize this relation
by defining an attitude state of an individual (a doxastic state, as “believing”) as a set of possible
worlds that are accessible to that particular individual under certain circumstances of evaluation

113
We analyzed this DP as if there was a covert determiner that conveys the meaning of “the unique entity”
(supremum operator), and the function application of the two place relation entity “trainer” that takes as
input “his”. Reference: “possessor raising”.

140
(or, in other words, the set of worlds which conform with everything that that individual believes).
This research was initially carried out by Hintikka114.
Hintikka characterized verbs as “believe”, relative to a particular circumstance of evaluation, as:
⟦believes⟧w1 =
● <<s, t>, <e, t>> type
● λP<s, t> · λxe · Vw2 ∊ DOXx (w1), P (w) = 1
● input is a preposition (that Rome is ugly); output is entity x (Rob); such that all worlds w2
that are doxastically accessible (or, compatible) to x in w1, are worlds that belong to the
preposition P
● INPUT preposition (Rome is ugly) · OUTPUT entity (Rob) · believes… (P)
He extended a similar treatment to verbs of knowledge, with the difference that, instead of being
doxastically accessible to some worlds, he considered these knowledge verbs to be epistemic
accessible.
⟦knows⟧w1 =
● <<s, t>, <e, t>> type
● λP<s, t> · λxe · Vw2 ∊ EPISTx (w1), P (w) = 1
In this sense, the possible worlds w2 are not doxastically compatible with x’s belief in w1; but
they are epistemically compatible with x’s knowledge in w1.
There is a rich literature about this distinction. Let’s briefly explain it:
When we have belief, this belief is surely true in our world of evaluation, but it can be false with
respect to the world of reality: it John believes that Florence is the capital of Italy, it will be true for
him, it will be true in other doxastically accessible worlds that share this proposition (a movie,
another epoch), but it is false in our actual circumstance of evaluation of Italy 2021, in which the
capital of Italy is Rome and not Florence.
On the contrary, the epistemically accessible worlds are bound to a knowledge that reflects the
actual circumstance of evaluation.
So, in other words, doxastic accessibility is a non-reflexive relation (it is not necessarily
guaranteed in our effective circumstances of evaluation); while epistemic accessibility is reflexive:
a factual attitude (because it is always compatible with our current circumstances of evaluation).

“De dicto” vs “de re” attitudes

Suppose we have these two sentences:
1. “Rob believes that the capital of Italy is wonderful”
2. “Rob believes that Rome is ugly”.

114
Hintikka, 1962, “Knowledge and belief”. He gave an explicit characterization of propositional attitudes in
terms of possible-worlds semantics. He was the first one.

141
De re interpretation
There is an interpretation in which these two sentences are contradictory (if Rome is the capital of
Italy, and Rob thinks it is ugly, how can he also think it is wonderful?).
In fact, if we assume that we must interpret “the capital of Italy” in our actual circumstances of
evaluation, we will assume that the capital of Italy is Rome. It would lead to a contradiction.
We interpret the definite description “the capital of Italy” as if it was outside the scope of the
propositional attitude verb “believes”.
Our full interpretation in this case is “Rob believes that the city that is capital of Italy in our world,
namely Rome, is not ugly”.

De dicto interpretation
But, let’s see what happens if we interpret the definite description within the scope of the
propositional attitude verb. We will see that the two sentences will not be contradictive anymore.
In fact, our full interpretation is now: “Rob believes that the city that he thinks is the capital of Italy,
namely Florence, is not ugly”, and this is perfectly compatible with the idea that Rome is ugly in Rob’s
opinion.
The De dicto interpretation rises when we interpret the description according to the worlds that
are doxastically accessible to the subject’s belief.
If Rob believes that Rome is the capital of Italy, we say that Rob has a true belief, but the doxastic
relation is still not reflexive (because its truth is possible, but not guaranteed in any case).

Propositional attitude: children acquisition

There is a rich discussion about the topic of propositional attitudes. One interesting aspect concerns
the acquisition and the cognitive development of this kind of predicates in children (both typical
and atypical development). And there have been evidences that the syntactic and linguistic mastery
of such structures with subordination has a crucial role in children on the development of the ability
to distinguish their own mental states from other person’s mental states.

Modal predicates
● (It) is possible [that Valentina is tired].
Here we have an adjective “possible” and the rest of the
proposition “that Valentina is tired”.
A modal adjective like “possible” is an intensional operator too: in
other words, it predicates something about a proposition (in fact,
it takes as input a proposition). For this reason, its semantic type is
<<s, t>, t>.
This semantic type is structurally analogous to the type <<e, t>, t>.

142
<<e, t>, t> is the semantic type of a quantifier over entity, like “everyone”, or “every boy”. For this
reason, we say that “possible” can be conceived of as a quantifier over possible worlds, instead of
entities.
⟦possible⟧w1 = λP<s, t> · ∃w2 · P(w1) = 1 🡪 existential quantifier over possible worlds.
⟦necessary⟧w1 = λP<s, t> · ∀w2 · P(w1) = 1 🡪 universal quantifier over possible worlds.
The necessity and possibility modals, in natural languages, are quantifiers over a restrictive subset
of possible worlds, exactly as in propositional attitudes the universal quantifier quantifies over a
restrictive set of doxastically accessible worlds.
In other words, using the possibility modal means “there are some worlds in which this preposition
can be true”. For this reason, it is considered as a restrictive type of quantification.115

115
Reference: Kratzer, 2012, “Modals and conditionals”: a series of paper written during the ‘80s about this
topic, where she added her latest studies; SEP (Standford Enciclopedia of Philosophy), entry on modality by
Kratzer; Portner, “Modality”.

143
Lesson 19
20.01
Context in intensions
The next topic we will discuss about has its reference in a study published by Kaplan in 1989 (even
though this topic was circulating from the late ‘70s), and it concerns the relationship between
context dependence, extension and intension.
Kaplan analyzed the following sentence:

I am here now
In this sentence there are precisely three deictic expressions that are anchored in their
interpretation to the context of utterance:

I 🡪 refers to the speaker of the utterance in w

here 🡪 refers to the place in w
now 🡪 refers to the time in w
When we analyze something, we can specify under which circumstances of evaluation, or to which
particular assignment function we refer. All of these specifications are expressible through an index
of evaluation: an index that notates all elements that contribute to determine a certain preposition.

⟦…⟧a, w 🡨 It seems that we can pack together these indexes of evaluation (index: -tuple that
contributes to fully determine a full propositional content).
If we consider a pronoun like “I” to be interpreted as the speaker, “here” as the place, “now” as the
time; we will end up not being completely able to fully explain the logical status of this sentence.
In fact, we note that this sentence is a priori true: when I say it, it is always true that “I am here now”,
it is not possible for this sentence to be false in context, but not with respect to all the other possible
worlds of evaluation: it is true a priori and contingent (not a necessary proposition: it is a
contingent statement).

Two-dimensional system
Kaplan proposes that we should not collapse the contribution of the context into the circumstances
of evaluation. We can individuate two layers of meaning that contribute to the final interpretation of
a proposition in order for it to be contextually valid:
1. Character of the sentence: it takes as input a context (C) and returns as output an intension
(proposition). This output can be called “Content”
2. Content (namely, the output of Character): it takes as input a specific world of evaluation
(or, a pair of world and time), and returns as output the extension (truth-value of the
clause).
In this way, we can clearly distinguish a two-dimensional system.

144
Heim & Kratzer skip the intermediate step, linking directly context to extension, but in fact, there is
an intermediate step (intensional level) that Kaplan calls “content”.
Context is extremely important in our analysis, and the index of evaluation contributes to enrich our
model in context. In this sense, the context can be represented as a tuple of parameters116. For
this reason, even though we have largely talked only about two particular indexes (g, w), in fact,
there are many:
Context = <a, h, t, l w, g>
Respectively: agent, hearer, time, location, world of evaluation, assignment function…
The pronoun “I” (agent) has the function of assigning an entity “agent” without passing through the
interpretation of the descriptive content. This is called direct anchoring: the definition of “I” is not
collapsing the element of interpretation to the intensional layer, precisely because deictic
expressions do not have deictic content: they just directly refer to one particular entity. This concept
is, under certain aspects, similar to rigid designator, but there are some important difference:

● Rigid designator: ⟦Rome⟧ = [λw = Rome] 🡪 constant character; constant intension.

● Deictic: ⟦I⟧ = I 🡪 no constant character, constant intension.

Local contribution of context

If I read Kaplan’s sentence “I am here now” in a diary I found today in Siena, I will know that the
present deictic elements will refer to particular circumstances that can change with respect to the
context it was originally written: maybe, the writer wrote those words while he was in India in
1970, or he could have written them in Rome in 2016. In these cases, the words will not, of course,
refer to out actual circumstances of evaluation (Siena, 2021).
This shows us that the contribution of the context is local: it fills some “holes” relative to a
particular uttered clause. If the clause is “I am here now”, the context will provide to assign to the
agent the identity of the speaker (could be Valentina Bianchi during lesson, could be me); to the
location the place where the speaker is (could be Valentina Bianchi’s house, could be India if the
speaker uttered this sentence while he was in India); to the time the precise moment when the
sentence was uttered (yesterday, ‘70s). It is a matter of substitution of the deictic elements with
their referent.

Ability to interpret different states of the world

The cognitive ability of interpreting states of the world and imagine alternative states of the world is
proper of very small children and also to some animals, as we have already mentioned in pg. 2.

Deictic expressions: context-bound

Another case we will mention is relative to deictic expression anchored to invented worlds. In these
cases:

116
Kaplan’s definition.

145
 ● or we change the deictic expression;
● or we assume a fictional context.
Because it is necessary (according to Kaplan) to specify the context of utterance when deictic
expressions are used. Deictic expressions can also be bound to the internal thoughts of a fictional
character (as in the quote “what she loved was this, here, now” by Virginia Woolf in Mrs. Dalloway).

De se attitudes
The character has a particular importance at a cognitive level.
Imagine the following scenario: I see my reflection on a window and I wrongly assume that my
reflection is another person that is on the other side of the window. Then, I notice that her pants are
on fire. Will my reaction be different if I realize that what I see is my reflection and therefore my
pants, and not someone else’s, are on fire? Of course, yes.
This funny imaginary scene demonstrates that formulating a particular thought in 1st person has
incredible consequences. This thinker’s attitude of thinking about the thinker itself has become
known as “de se” thought.117
Another level of analysis shows us that some kinds of thoughts necessarily involve the thinker as
part of them. Think about the following example:
A person is in the supermarket, pushing his shopping trolley around. Suddenly he realizes that some
sugar in on the floor, and what he thinks is “mmh, someone should have made it fall without
noticing it”. So, he starts looking for the person that made the sugar fall down; but, as he goes
around in the supermarket, he realizes he is the only present person, therefore, it was surely him
that made the sugar fall.
There are a variety of de se thoughts. Perceptions are things to which only the person who feels
them has a direct access.
Various approaches to context-dependence has generalized the kaplanian strategy of the
two-dimensional system, because it allows to keep the relation between the extensional and the
intensional level, and add the context “on the top” of it, on a second level which is above the first one
(content). But, also here some issues rise: in fact, pragmatics cannot come completely after the
semantic level: the context of interpretation enters before.

Common ground
Kaplan’s view is completely static in the sense we first make a full-fletched analysis of a proposition,
and then we add the context layer of interpretation.
Stalnaker, instead, introduced a more dynamic view118, as the title of the paper itself suggest: “What
happens when a speaker makes an assertion”.
First, if the speaker is making an assertion, he is trying to make its interlocutor evaluate the content
of his assertion, relative to one particular circumstance of evaluation.
The main point of Grice is that we must think about a conversation where there is a cooperative
activity involving rational agents. This idea influenced Stalnaker in his studies about the topic.

117
D. Levis, 1979, “Attitudes de dicto and de se”; Perry, 1979.
118
R. Stalnaker, 1978, Assertion; P. Grice, 1975, Logic and conversation.

146
By rational agents we mean a person who carries a particular belief state (characterized by a set of
worlds that are compatible with the agent’s point of view) and enters in a cooperative activity with
some particular aims. The idea is that we can represent this agent through a set of possible worlds.
Here is where the common ground of conversation enters in account.
The common ground of conversation is a set of propositions that represent all the information that
the participants have in common, share and take for granted (Stalnaker).
The information contained in the common ground are not necessarily true in our actual world of
evaluation, but all participants act as if they are in fact (for example, I am talking with a group of
friends, and one of these friends says that she left her boyfriend because he cheated on her. Maybe it
is not true, maybe she just did not want to take the responsibility for having left him without a
reason, but from that moment onwards, me and all of my friends will take for granted that her poor
ex-boyfriend cheated on her, and this information will become part of our common ground).
Each preposition that contributes to the common ground represents a portion of the logical space (a
set W of all possible worlds; and any subset of W will be a proposition). For this reason, in a certain
sense, the information is limited in common ground.

Context set 🡪 intersection P ⋂ Q ⋂ R 🡪 truth-conditions of P, Q and R, are all satisfied.

The context set is a subset of possible worlds which is candidate to be the actual world (because it
results from the intersection of all true propositions in our conversation).

147
 P: Valentina Bianchi is Italian
Q: Valentina Bianchi works in Siena
R: Valentina Bianchi teaches semantics
If these three propositions belong to the common ground, the context set (their intersection) will
contain only worlds in which these truth-conditions will be satisfied.
The subset resulting from the intersection of all of these three propositions satisfies all of the
information, but there is some other information that is missing from the common ground. For
example, I might not know whether she has sons or daughters, how many, if she owns pets, and so
on. The worlds included in the context set can diverge between then with respect to these missing
pieces of information, but not with respect to the information given by P, Q, R.
The preposition that are taken for granted in the common ground are called presuppositions (this
kind of presupposition is pragmatic, in contrast with the semantic presupposition that we talked
about in definite descriptions).

Assertions
Now that I have this common ground, any new assertion that I make can increment the information
of the common ground. For example, I can say “Valentina Bianchi has three children”, and if this new
assertion is accepted as true, from this moment all of the participants will behave as if this
information is true, because it will become part of the common ground.
Of course, an assertion can also be rejected from the other participants: this happens especially
when a new assertion contrasts with some piece of previous common ground. For example, if I say
“this Sunday we have lesson”, this assertion will be rejected from the other participants because the
common ground contains the information that there is never lesson on Sunday.

● C + S119 = {W | W ∊ C & W ∊ S} = C ⋂ S 🡪 update effect of assertion of the declarative

sentence (reducing context set).
This means that when someone makes an assertion and it is accepted by the participants, he
adds another preposition to the common ground, reducing the context set: in fact, adding a
fourth proposition S “Valentina Bianchi has three children” to our previous common ground,
we will automatically exclude from the context set all the possible worlds in which Valentina
has no children, in which she has only one or two children, and so on.
🡺 The larger the common ground, the smaller the context set.

This notion brings us from a stative to a dynamic interpretation 🡪 it leads from an input context to
an output context: now the meaning of a declarative sentence is still a proposition, but this
proposition dynamically applies to context set via intersection to reduce the context set.
In a conversation it is also important to represent the individual states of participants (beliefs, for
example).

119
Context set + proposition

148
Requirements for felicity assertion
Stalnaker 🡪 when a speaker asserts a proposition in a context, the proposition, in order to
be felicitous must be informative in the following ways:
- should contain some but not all the worlds of the context set;
- must be logically compatible with the common set.
For any assertion of P in C, in order to be felicitous, ¬C ⊆ P (the common set must not already
contain the preposition) & ¬[C ⋂ P = ∅] (the proposition must be logically compatible with the
common set)

If these requirements are satisfied:

● Update effect will not be vacuous
● We will have no empty set

Second condition:
Stalnaker 🡪 The uttered sentence should convey the same P of all the worlds of the
context set.
In other words, we do not want to assert a sentence which in some of the worlds of the context set
has no truth-value.
These are the conditions (informativeness condition and condition of truth-value in all the worlds
of the context set) for an assertion to be felicitously informative in context.

Mechanism of accommodation (David Lewis)

If the speaker takes for granted some non-controversial information conveyed by P, this mechanism
allows him to automatically incorporate the information in the common set: if, during the lesson,
the professor says “Sorry, my dog is snoring”, our mind will process the information that that person
owns a dog.
This mechanism is required, because if we were not able to process information in this indirect way,
the entity “my dog” would refer to the empty set (if, when the professor talks about her dog, I do not
automatically compute the fact that she owns a dog, that entity would not exist in the context set,
and this is not possible, because the sentence would be nor true nor false).

Conclusion
What Stalnaker tried to do was to find a common point between the formal apparatus of language
and the ordinary use of language, that for many years were considered to be two distinct things that
were not, somehow, totally connected one to each other.

THE END

149