See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220202716

Computational Semantics for Monadic Quantifiers

Article in Journal of Applied Non-Classical Logics · January 1998

DOI: 10.1080/11663081.1998.10510934 · Source: DBLP

CITATIONS READS

23 66

1 author:

Marcin Mostowski
Jagiellonian University
38 PUBLICATIONS 241 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Concrete Mathematics View project

All content following this page was uploaded by Marcin Mostowski on 23 May 2014.

The user has requested enhancement of the downloaded file.

 Computational semantics for monadic
quantifiers∗
Marcin Mostowski
e-mail:[email protected]

Abstract
The paper gives a survey of known results related to computa-
tional devices (finite and push–down automata) recognizing monadic
generalized quantifiers in finite models. Some of these results are sim-
ple reinterpretations of descriptive—feasible correspondence theorems
from finite–model theory. Additionally a new result characterizing
monadic quantifiers recognized by push down automata is proven.

The aim of the work is presentation of the state of knowledge and main
research problems in computational approach to finite interpretations of
monadic generalized quantifiers. We will concentrate on purely logical ap-
proach – that is we will consider mainly structures without standard linear
orderings. However proper results for linearly ordered structures will be
mentioned when relevant.
Let us observe that monadic quantifiers – being the first natural class of
quantifiers investigated from computational point of view – are not system-
atically treated in logical literature. In the first attempt [6] of surveying the
subject of generalized quantifiers from computational point of view, monadic
quantifiers are only shortly mentioned.

1 Generalities
By monadic generalized quantifiers we mean Lindström’s quantifiers (see
[5]) of types of the form (1, 1, . . . , 1); if the number of ones in the type
signature is n then we say that the corresponding quantifier is a monadic
∗
The research reported here has been supported by the research grant of Polish National
Commitee of Scientific Research (KBN) 433/H01/95/08.
n-ary quantifier. Syntactically it binds one variable in n formulae. More
precisely for each n–ary monadic quantifier Q we define the logic L(Q) as
an extension of elementary (first order) logic. The set of L(Q)–formulae is
obtained by adding to the standard construction rules the following:
if x is a variable and ϕ1 , . . . , ϕn are formulae then also Qx(ϕ1 , . . . , ϕn ) is
a formula.
Semantically n-ary monadic quantifier Q can be represented by closed
on isomorphisms class KQ of structures of the form (U, R1 , . . . , Rn ), where
U 6= ∅ and Ri ⊆ U , for i = 1, 2, . . . , n.
The definition of the satisaction relation is extended by the following:
for any model M , and a valuation a in M :
M |= Qx(ϕ1 , . . . , ϕn )[a] if and only if (|M |, ϕM,x,a
1 , . . . , ϕM,x,a
n ) ∈ KQ ,
M,x,a
where |M | is the universe of M , and ϕ is the set defined by ϕ in M with
respect to the variable x, under the valuation a – free in ϕ variables different
than x are interpreted according to the valuation a.1
The class KQ can be represented by the set of nonempty words FQ over
the alphabet A = {a0 , a1 , . . . , a2n −1 } such that

α ∈ FQ if and only if there is (U, R1 , . . . , Rn ) ∈ KQ and a linear ordering

U = {b1 , . . . , bk } such that lh(α) = k, and i-th character of α is aj exactly
when bi ∈ S1 ∩ . . . ∩ Sn , where
(
Rl if integer part of j/2(l−1) is odd
Sl =
U − Rl otherwise

So defined intersections S1 ∩ . . . ∩ Sn are called constituents of the proper

model – the above defined constituent is also called j-th constituent. Then
we can think of characters a0 , . . . , a2n −1 as names of constituents. In other
words our definition says that i–th character is aj exactly when bi belongs to
j–th constituent. For identifying a constituent we should know which relation
is taken positively and which is taken negatively. This can be calculated by
writing down j in binary notation, and checking whether l–th digit from the
rigth is 1 (positive) or 0 (negative).
For n = 3 this idea is ilustrated by the following picture:
1
For the discussion of the general definition of Lindström quantifiers see e. g. [5] or [3].

2
 b1 b2 b3 b4 b5 b6
R3 : 1 1 1 0 1 0

R2 : 1 0 1 1 1 1

R1 : 0 1 1 1 1 0

α: a6 a5 a7 a3 a7 a2

Particularly, when b4 6∈ R1 and b4 ∈ R2 ∩ R3 , then the fourth character

of the word α is a3 .

Let C be a class of recognizing devices, and Q be a class of monadic quan-

tifiers. We say that C accepts Q if and only if for each n > 0 and each n–ary
monadic quantifier Q (Q ∈ Q ⇔ ∃M ∈ C(M accepts FQ ).

Originally these concepts were introduced by Büchi (see [4]), who inves-
tigated models with fixed linear orderings. Of course when no ordering is
fixed then classes FQ are closed on permutations in the sense that if α ∈ FQ
and α0 is obtained from α by arbitrary shuffling characters then α0 ∈ FQ .
Interpretations of monadic quantifiers in this spirit were first investigated in
[1] and [2].
we have the following straightforward:

Lemma 1 For each monadic quantifiers Q1 , Q2 of the same type there are
quantifiers being their intersection, union, and the complement of Q1 , defined
by sets of words FQ1 ∩ FQ2 , FQ1 ∪ FQ2 , and A∗ − FQ1 − {ε} respectively.

If the class of quantifiers is closed on all the above operations then we

will say that it is closed on boolean combinations.
Considering monadic quantifiers in presence of linear ordering can be
understood as follows:
Instead of a monadic quantifier Q of type (1, 1, . . . , 1), we consider a gen-
eralized quantifier Q0 of type (2, 1, 1, . . . , 1) (with the same number of ones),
and we assume that in our language we have one binary predicate symbol

3
<, which is always interpreted as a linear ordering — what is equivalent to
saying that we consider only models for the theory of linear ordering in terms
of <. Then we interpret formulae of the form

Qx(ϕ1 , . . . , ϕm )

as
Q0 xy(x < y, ϕ1 , . . . , ϕm ).
A similar remark apply to the Ramsey quantifier Q2 , introduced by Mac-
intyre (see [3]).

2 First Characterizations
The first characterization of quantifiers in finite models from computational
point of view has been given by van Benthem.2

Theorem 1 (straightforward generalization of [2]) Finite automata

with only 1–loops accept the class of all monadic elementary definable quan-
tifiers.

Because of the following

Theorem 2 (see [9]3 ) Monadic quantifiers definable in monadic second or-

der logic are elementary definable.

we have also

Corollary 3 Finite automata with only 1–loops accept the class of all monadic
quantifiers definable in monadic second order logic.

These results can be contrasted with the following:

Theorem 4 (Büchi, Ladner, see [4]4 )

2
Other older results mentioned in this paper are essentially reinterpretations of theo-
rems about descriptive—feasible correspondences, originally formulated in terms of classes
of models.
3
This theorem follows also from the last theorem in [10], which says, among others,
that monadic second–order logic in monadic vocabulary is semantically equivalent to ele-
mentary logic.
4
This theorem follows from works of Büchi. However Ladner gave its reinterpretation
in the spirit of finite–model theory.

4
 1. In presence of linear ordering star–free finite automata accept elemen-
tary definable monadic quantifiers.

2. In presence of linear ordering finite automata accept monadic quanti-

fiers definable in monadic second order logic.

The above theorem is essentially reinterpretation of Ladner’s formula-

tion, which was not given in terms of quantifiers, but definable classes of
models. Reinterpretation of other results from finite–model theory in terms
of quantifiers has been done in the first general survey devoted to generalized
quantifiers in finite models by Makowsky and Pnueli [6].5 In their paper they
give also some new results, particularly, related to mutual interpretations of
languages with generalized quantifiers in finite models.

3 Divisibility Quantifiers and Finite Automata

For n ≥ 2 by divisibility quantifier Dn we mean the quantifier of type (1)
such that

Dn xϕ if and only if the number of x satisfying ϕ is divisible by n.

Divisibility logic L(Dω ) is elementary logic enriched by all quantifiers Dn ,

for n ≥ 2. Divisibility logic is equivalent to the logic with quantifiers counting
modulo (see e. g. [11]).

Theorem 5 ([8]) Finite automata accept the class of all monadic quanti-
fiers definable in L(Dω ).

Proof. We will discuss here only the simplest case of unary monadic
quantifier Q such that FQ is regular. Our alphabet A is {0, 1}. Following [8]
we will describe the automaton recognizing FQ .
We define the standard equivalence relation ≈ as follows α ≈ β iff for
every γ ∈ {0, 1}∗ (αγ ∈ FQ iff βγ ∈ FQ ). Let k be the smallest number such
that 0k ≈ 0i , for some i > k, and n be the smallest such that 0k ≈ 0n and
n > k. Numbers p and m are defined in the same way taking 1 in the place
of 0.
Let us consider the following automaton A = (S, {0, 1}, δ, s00 , F )
5
As far as I know the first paper dvoted to generalized quantifiers in finite models is
[10].

5
 s0,0 s s0,m−1
• - - • 0,p
- - •

? ? ?

?sk,0 ?s ?s
• - - • k,p
- - • k,m−1
6 6 6
? ? ?

?sn−1,0 ?s ?s
• - - • n−1,p
- - • n−1,m−1

where all horizontal arrows are labeled by 1, and all vertical arrows are
labeled by 0. We will call this automaton: (n, m, k, p)-automaton.
A priori we cannot expect that A will be identical with the syntactic
automaton A0 defined by ≈. However we can define proper set F of accepting
states in A by constructing a suitable homomorphism f : A → A0 , and
defining F as f −1 (F 0 ), where F 0 is the set of accepting states of A0 . So we
obtain that A accepts FQ .
For every state sij ∈ F we define the formula ψij such that ψij expresses
that:

“there are exactly j x such that ϕ(x) and exactly i x such that ¬ϕ(x)”, if
i < k and j < p,

“the number of x such that ϕ(x) minus p is equivalent to j − p modulo

m − p and there are exactly i x such that ¬ϕ(x)”, if i < k and j ≥ p,

“there are exactly j x such that ϕ(x) and the number of x such that ¬ϕ(x)
minus k is equivalent to i − k modulo n − k”6 , if i ≥ k and j < p,

“the number of x such that ϕ(x) minus p is equivalent to j − p modulo

m − p and the number of x such that ¬ϕ(x) minus k is equivalent to i − k
modulo n − k”, if i ≥ k and j ≥ p.
6
We assume here that for each n, m, n ≡ m(mod 1).

6
 That “there are exactly j x such that ϕ(x)” is expressible in L(Dω ) follows
from that it is expressible in elementary logic. That “the number of x such
that ϕ(x) is equivalent to j modulo s” can be expressed in L(Dω ) by the
following formula:
∃x1 . . . ∃xj (ϕ(x1 )∧. . . ∧ϕ(xj )∧ 1≤a<b≤j (xa 6= xb )∧Ds y(y 6= x1 ∧. . . ∧y 6=
V

xj ∧ ϕ(y))).

We define ψ as the disjunction of all ψij such that sij ∈ F . Observe that
this construction is general and effective.
This ends the proof that each monadic quantifier Q such that FQ is regular
is definable in divisibility logic.
We recall here that a quantifier Q of type (n1 , . . . , nk ) is definable in a
logic L if and only if there is L-formula without free variables ϕ(P1 , . . . , Pk )
with P1 , . . . , Pk as only nonlogical concepts (Pi is ni -ary predicate) such that
Qx(ϕ1 , . . . , ϕk ) is semantically equivalent to ϕ(ϕ1 , . . . , ϕk ), where substitu-
tion of formulae in place of predicates is defined in a natural way respective
to proper variables from the sequence x.
Therefore the second inclusion follows from the fact that each L(Dω )–
formula in monadic vocabulary is equivalent to a boolean combination of
formulae described in the first part of the proof (see [8]).
This ends the proof.

By the above theorem monadic quantifiers definable in L(Dω ) will be

called regular quantifiers.
Let us observe that theorem 1 can be obtained as a simple corollary from
the above reasoning. When considered quantifier Q is elementary definable
we obtain the above described automaton A such that n = k + 1 and m =
p + 1. In this case obviously A has only 1–loops.
The last theorem can be contrasted with the following

Theorem 6 (easily following from [11]) In presence of linear ordering

finite automata having syntactic monoids not containing non solvable groups
accept the class of all monadic quantifiers definable in L(Dω ).

Of course there are regular sets having syntactic monoids, which contain
nonsolvable groups. Then, in presence of linear ordering, monadic second
order logic is stronger than divisibility logic. Without ordering we have the
opposite relation, divisibility logic is stronger than monadic second order
logic.

7
4 Push–Down Automata
We can describe monadic quantifiers also in another way. Let Q be n-ary
monadic quantifier, then Q is uniquely determined by 2n -ary relation RQ on
ω defined as

RQ = {(s0 , . . . , s2n −1 ) : ∃M = (U, R1 , . . . , Rn ) ∈ KQ ∀j ∈ {0, . . . , 2n − 1}

j-th constituent of M has cardinality sj }

On the other hand each 2n -ary relation R on ω not containing the tuple
(0, 0, . . . , 0) uniquely determines n-ary monadic quantifier Q such that R =
RQ . Therefore we can consider classes of monadic quantifiers Q determined
by some classes R of relations on ω, in the sense that Q = {QR : R ∈ R}.
Van Benthem in [1] considered the class of quantifiers determined by rela-
tions elementary definable in the structure (ω, +), which he called additively
definable quantifiers.

Theorem 7 ([1]) Push Down Automata accept additively definable quanti-

fiers of type (1).

The proof of this theorem is based on the fact that additively definable
relations are exactly semilinear relations, that is finite unions of so called
linear relations of the form:
S = {(x1 , . . . , xm ) ∈ ω m : ∃z1 , . . . , zk (x1 , . . . , xm ) = (a1 , . . . , am )+
z1 (b11 , . . . , b1m ) + . . . + zk (bk1 , . . . , bkm )}.

Then quantifiers determined by semilinear relations will be called also

semilinear quantifiers.
Van Benthem observed also that some semilinear (additively definable)
quantifiers of arity > 1 cannot be recognized by PDA.
In this section we will consider also quantifiers of type (1) accepted by
deterministic PDA. However firstly we will recall some basic concepts related
to Push–Down Automata (PDA).
A push–down automaton is of the form A = (A, Σ, S, s0 , F, δ), where A
is an input alphabet, Σ – stack alphabet, S – a set of internal states, s0 ∈ S
is an initial state, F ⊆ S is a set of accepting states, and finally

δ : (A ∪ {ε}) × S × Σ −→ P (S × Σ∗ )

is a transition function.
The function δ describes the behaviour of A in the following sense:

8
 When reading a word the automaton A pops an element σ from the top of
the stack, reads a character a ∈ A (or nothing in a case of so called ε–moves),
and is in a state s ∈ S, then it chooses (s0 , σ 0 ) ∈ δ(a, s, σ), and changes its
internal state to s0 , it shifts the head into the next character, and pushes the
word σ 0 into the top of the stack.
If δ(ε, s, σ) = ∅, for each s ∈ S, σ ∈ Σ (there are no ε–moves), and
δ(a, s, σ) always contains not more then one element, then we say that A is
deterministic push down automaton, or shortly deterministic PDA, otherwise
it is called nondeterministic PDA.
When A is deterministic PDA then the transition function δ can be de-
scribed as a partial function

δ : A × S × Σ −→ S × Σ∗

An input word α ∈ A∗ is accepted by A if and only if starting its work

with its head on the first character of α, being in the iternal state s0 , and
having the stack empty, the automaton A ends reading α in a state belonging
to F . If additionally the stack is empty at the end, then we say that A accepts
α by empty stack.
A set X ⊆ A∗ is recognized by A if for each α ∈ A∗ :

α ∈ X if and only if A accepts α.

Deterministic PDA–s recognize by arbitrary stack larger class of sets then

the class recognized by empty stack, e. g. let L = {0n 1k : k ≤ n, k, n ∈ ω}.
L is acceptable deterministically, but not by empty stack. The reason is that
each PDA after reading many symbols 0 and making the stack empty cannot
memorize how many symbols 1 can be still accepted.
Let us observe that all the above concepts are defined in such the way that
forgeting about stacks and stack alphabets we obtain adequate definitions of
corresponding concepts related to finite automata.
Let a, b, c, d ∈ ω, by (a, b, c, d)–linear quantifier Lin(a,b,c,d) we mean the
quantifier of type (1) such that Lin(a,b,c,d) = QR , where

(∗) R = {(x, y) ∈ ω 2 : ∃z ∈ ω (x, y) = (c, d) + z(a, b)}

We say that a quantifier Q of type (1) is almost linear iff for some
a, b, c1 , . . . , ck , d1 , . . . , dk ∈ ω and some finite relation R0 ⊆ ω 2 :

Q = QR0 ∪ Lin(a,b,c1 ,d1 ) ∪ . . . ∪ Lin(a,b,ck ,dk ) .

In other words almost linear quantifiers are those semilinear quantifiers,

which have at most one nontrivial vector (a, b) in its defining relation.

9
 The ratio a/b will be called the ratio of the corresponding almost linear
relation.
For geometrical illustration of what does it mean almost linear, let us
consider a relation S ⊆ ω 2 being a finite union of relations of the form (∗)
with the same ratio a/b. Allowing z to vary over the set of reals we can
think of S as a set of points on the euclidean plane. Then almost linear
relation S can be defined (modulo a finite set) as a finite union of straight
lines R1 , . . . , Rm parallel each to other.
y
6

···

b
a -x
R1 R2 Rm

The above picture suggests that complements of almost linear relations

are never almost linear, their finite unions are almost linear only if they have
the same ratio, and their intersections are either almost linear or finite.
Of course deterministic PDA–s recognize by empty stack all regular quan-
tifiers, but what more? We will show that they recognize also almost linear
quantifiers. However the intersection of a regular set and a set recognized by
deterministic PDA (by empty stack) is still recognized by deterministic PDA
(by empty stack). We claim that nothing more (between permutation closed
sets) is recognized by deterministic PDA–s by empty stack.

Theorem 8 The class of all quantifiers recognized by deterministic PDA–s

by empty stack is the union of the following:
1. regular quantifiers,
2. intersections of regular and almost linear quantifiers.

Proof Firstly we will show that almost linear quantifiers are recognized
by deterministic PDA–s. We start with the simplest linear case without

10
exceptions. Let R be defined as above for some fixed a, b, c, d ∈ ω. Let Q be
QR . Now we will describe the deterministic PDA recognizing FQ .
We define the stack alphabet S as {<0–mile>, <1–mile>}.

<0–mile> contains a steps, and <1–mile> contains b steps.

The behaviour of our PDA can be described by the following picture:

<1–mile>
-
•
@
@
@
@
@
@
@
@
@
@
F

?
<0–mile> @
@
@
@
@
@
@
@
@
@

Reading a word α we go to right – when 1 is read b times, and down –

when 0 is read a times.
We count miles poping and pushing them as follows:

• when we have passed one mile and the stack is empty then we push it,

• when we have passed one mile and the stack contains a mile of the
same kind then we push it,

• when we have passed one mile and the stack contains a mile of another
kind then we pop the mile from the stack.

0–miles can be counted after reading 0 c–times, 1–miles can be counted

after reading 1 d–times.
The word α is accepted if after reading it we are staying at the line F —
it means that the stack is empty.

11
 The general case allowing exceptions can be obtained by enriching our
construction by some states and instructions not modifying the stack.

Remark Let us observe that a similar picture can be used to justify Van
Benthem’s theorem 7.

Let Q0 = QR0 be defined by a semilinear R0 , that is being a finite union

of linear relations S such that for some a1 , b1 , . . . , ak , bk , c, d ∈ ω

S = {(x, y) ∈ ω 2 : ∃z1 , . . . , zk (x, y) = (c, d) + z1 (a1 , b1 ) + . . . + zk (ak , bk )}.

In this case for recognizing QS we can use the same idea constructing
nondeterministic PDA with k kinds of 0–miles, and k kinds of 1–miles. Our
PDA will choose nondeterminstically what kind of miles have to be used for
measuring distances.
Therefore FQ0 as being a finite union of context free sets is itself context
free.
end of Remark

Now let us assume that Q is accepted by some deterministic push down

automaton A. By Parikh’s theorem there is a semilinear R0 , as described in
the above remark, such that Q = QR0 .

Firstly we assume that modulo finite number of exceptions R0 is linear,

that is R0 = S ∪ R0 , where R0 is finite, and for some a1 , b1 , . . . , ak , bk , c, d ∈ ω

S = {(x, y) ∈ ω 2 : ∃z1 , . . . , zk (x, y) = (c, d) + z1 (a1 , b1 ) + . . . + zk (ak , bk )}.

Then we obtain two possible cases:

1. for some i = 1, . . . , k either ai = 0 and bi 6= 0 or bi = 0 and ai 6= 0, in

this case Q is definable by divisibility quantifiers;
2. all a1 , . . . , ak , b1 , . . . , bk are different than 0, in this case we show that
Q is almost linear.

Let us consider in details only the case 2. The case 1 is simpler, and it
can be analyzed similarly. We consider the simplest case when the set S is
defined by three vectors:

S = {(x, y) ∈ ω 2 : ∃z1 , z2 (x, y) = (c, d) + z1 (a1 , b1 ) + zk (a2 , b2 )},

12
and no one parameter is equal to 0. The general argument differs only by
straightforward technicalities.
Let us consider words:

αx = 0c 1d 0(xa1 )a2 1(xa1 )b2

βx = 0c 1d 0(xa2 )a1 1(xa2 )b1

Now we will show that a1 b2 = a2 b1 .
Let us assume that a1 b2 < a2 b1 . Then for arbitrary large n we can chose
x such that
k = x(a2 b1 − a1 b2 ) > n.
Then βx = αx 1k .
The automaton A after reading αx is in accepting state with empty stack.
It means that it cannot memorize x, and from this moment, not reading any
zeros, it behaves like finite automaton (without stack). Chosing sufficiently
large x we obtain the contradiction.
Therefore in the case 2 we prove that there are natural numbers i, j > 0
such that for some x1 , . . . , xk :

(a1 , b1 ) = x1 (i, j)
..
.
(ak , bk ) = xk (i, j)

Moreover we can choose x1 , . . . , xk being relatively prime and i, j being

greatest possible. Then we take a = yi, b = yj, and c0 = c+ix, d0 = d+jx, for
some suficiently large x, e. g. x = x1 . . . xk z, where y is the least multiplier of
(i, j) greater than all exceptions and z is used to make c0 and d0 greater than
all exceptions. For differentiating between particular (ai , bi ), for i = 1, . . . , k,
we need only counting modulo y, what can be done without using the stack.
Therefore Q is Lin(a,b,c0 ,d0 ) ∩ Q0 modulo some finite number of exceptions,
for some regular Q0 .
The general case can be obtained by a similar argument.
This ends the proof.

As the result we obtain the following:

Corollary 9 The complement of a quantifier Q recognized by a deterministic

PDA by empty stack is also recognized by deterministic PDA by empty stack
if and only if Q is regular.

13
 Of course, by a trivial argument, the complement of a quantifier recog-
nized by a deterministic PDA is also recognized by a deterministic PDA by
arbitrary stack.

Corollary 10 The class of quantifiers recognized by deterministic PDA–s by

empty stack is closed on intersections.

Let us observe that the assumption that accepting is by empty stack is

essential here, because the class of quantifiers described above is not closed
on complements. Additionally the union of an almost linear and a regular
quantifier can be still recognized by a deterministic PDA but not by empty
stack.
Similarly as in the case of nondeterministic PDA–s no natural descrip-
tive class of all monadic quantifiers recognized by deterministic PDA–s is not
known.

Finally let us mention another open problem. In the paper [7] it was
observed that in the theory of (ω, +) divisibility quantifiers are eliminable.
We do not know any other monadic quantifiers being eliminable in arithmetic
of addition, which are not definable in divisibility logic. Are there any such
quantifiers?

5 Examples
¿From our work and known proprieties of finite and push–down automata
we obtain the following facts:

1. Definable in L(Dω ), regular quantifiers:

• even, odd;
• counting modulo ∃pq , for q < p, where ∃p,q xϕ iff #{x : ϕ} ≡
q(mod p);7
• Qx(ϕ, ψ) iff #{x : ϕ} ≡ #{x : ψ}(mod p).
2. Recognized by deterministic PDA–s by empty stack, but not
regular quantifiers:
• Half, every other, where Half xϕ iff #{x : ϕ} = #{x : ¬ϕ};
• every third, every fourth, . . . , every n–th, where every n–th
xϕ iff n · #{x : ϕ} = #{x : x = x};
7
For a set A by #A we denote the cardinal number of A.

14
 • Härtig quantifier, QH x(ϕ, ψ) iff #{x : ϕ} = #{x : ψ}.

3. Recognized by deterministic PDA–s by arbitrary stack, but

not by deterministic PDA–s by empty stack:

• majority, where majority xϕ iff #{x : ϕ} > #{x : ¬ϕ}.

4. Recognized by non deterministic PDA–s, but not by deter-

ministic PDA–s:

• two or three times more, where the quantifier Q is definned

by disjunction of Lin(1,2,0,0) and Lin(1,3,0,0) ,
Qxϕ iff #{x : ϕ} = 2 · #{x : ¬ϕ} or #{x : ϕ} = 3 · #{x : ¬ϕ}.

5. Semilinear, but not recognized by PDA–s:

• Qx(ϕ, ψ) iff #{x : ϕ ∧ ¬ψ} = #{x : ϕ ∧ ψ} = #{x : ¬ϕ ∧ ψ}.

References
[1] J. van Benthem, Essays in Logical Semantics, D. Reidel Publish-
ing Company 1986.

[2] J. van Benthem, Towards a Computational Semantics, in P. Gar-

denförs (ed.), Generalized Quantifiers, D. Reidel Publishing Com-
pany 1987, pp. 31–71.

[3] M. Krynicki and M. Mostowski Quantifiers, Some Problems and

Ideas, in M. Krynicki, M. Mostowski, and L.W. Szczerba (eds.)
Quantifiers: Logics, Models and Computation Volume I, Kluwer
Academic Publishers, 1995, pp. 1–19.

[4] R. E. Ladner, Application of Model Theoretic Games to Discrete Lin-

ear Orders and Finite Automata, Information and Control 33 (1977),
pp. 281–303.

[5] P. Lindström, First order predicate logic with generalized quantifiers,

Theoria 32 (1966), pp. 186–195.

[6] J. A. Makowsky and Y. B. Pnueli Computable quantifiers and logics

over finite structures, in M. Krynicki, M. Mostowski, and L.W.
Szczerba (eds.) Quantifiers: Logics, Models and Computation
Volume I, Kluwer Academic Publishers, 1995, pp. 313–357.

15
 [7] M. Mostowski, Divisibility quantifiers, in Bulletin of the Section
View publication stats

of Logic 20, no. 2 (1991), pp. 67–70.

[8] M. Mostowski, The Logic of Divisibility, to appear in The Journal

of Symbolic Logic.

[9] M. Mostowski, Kwantyfikatory rozgalȩzione a problem formy log-

icznej, (in Polish, Branched Quantifiers and the Problem of Logical
Forms) in Nauka i jȩzyk, Wydzial Filozofii i Socjologii Uniwersytetu
Warszawskiego, Warszawa 1994, pp. 201 – 241.

[10] J. Väänänen, Remarks on generalized quantifiers and second–order

logic, in Set Theory and Hierarchy Theory, J. Waszkiewicz, A.
Wojciechowska, and A. Zarach (eds.), Prace Naukowe Instytutu
Matematyki Politechniki Wroclawskiej, Wroclaw 1977, pp. 117–123.

[11] H. Straubing, D. Thérien, W. Thomas, Regular Languages De-

fined with Generalized Quantifiers, Proc. 15th ICALP 88, T. Lep-
istö and A. Salomaa (eds.) Lec. Notes in Comput. Sci. 317, Springer-
Verlag, Berlin 1988, pp. 561–575.

16