The Ehrenfeucht-Mostowski construction is a construction which produces models in which few types are realized.
Theorem. Let  be a complete theory with infinite models. Then for any 
, there is a model 
 of cardinality 
 such that if 
, then at most 
 types over 
 are realized.
Proof. First suppose that  has definable Skolem functions. Let 
 be a non-constant indiscernible sequence of length 
 in some model 
. We can find such an indiscernible sequence by extracting an indiscernible sequence from a non-constant sequence in an infinite model. Let 
 be 
. Because 
 has definable Skolem functions, 
, so 
 and 
 is still indiscernible within 
. The set 
 is called the spine. Let 
 be a subset of 
. Each element of 
 is in the definable closure of some finite subset of the spine, so we can find some 
 contained in the spine, with 
, and 
. An element’s type over 
 is determined by its type over 
, because 
. So it suffices to show that at most 
 types over 
 are realized.
First we check that at most  types are realized by tuples from the spine. We can write 
 as 
 for some 
 with 
. The indiscernibility of the spine implies that 
 is entirely determined by how the 
 relate to each other and how they relate to elements of 
. That is, 
 is entirely determined by the following pieces of data:
Because  is well-ordered, there are only about 
 choices for the second and third bullet points. All told, there are therefore only 
 possibilities for 
.
Now if  is a 0-definable function, then 
 depends only on 
, so there are at most 
 types over 
 realized by elements of the form 
. But all of 
 is in the definable closure of the spine, so every element of 
 is of this form. Since there are only 
-many 0-definable functions, the total number of types over 
 realized in 
 is at most 
 So at most 
 types over 
 are realized, completing the proof (in the case where we had definable Skolem functions).
Now suppose  is arbitary. We can find a theory 
 expanding 
, which does have Skolem functions. This can easily be done in such a way that 
. By the above argument one gets a model 
 of 
 of size 
 with the property that for every subset 
 of 
, at most 
 types over 
 are realized in 
. Let 
 be the reduct of 
 to the original language. Then 
. If 
, and 
 and 
 have the same 
-type over 
, then they certainly have the same 
-type over 
 within 
, because 
 has fewer definable sets and relations to work with than 
. So there are at most as many 
-types over 
 as there are 
-types over 
, which is at most 
. QED
An important consequence of this result is the following, which is the first step of the proof of Morley’s Theorem.
Corollary. Let  be a complete countable theory which is 
-categorical for some 
. Then 
 is 
-stable (hence totally transcendental).
Proof. Let  be the monster model of 
. Suppose 
 is not 
-stable. Then we can find a countable set 
 over which there are uncountably many types. Realize 
 of these types and let 
 be the set of these realizations. Then 
, so by Löwenheim-Skolem we can find a model 
 of cardinality 
 containing 
. By the Ehrenfeucht-Mostowski construction, we can find a model 
 of cardinality 
 in which at most countably many types are realized over countable sets. By 
-categoricity, 
. So 
 also has the property that over countable sets, countably many types are realized. But over the countable set 
, uncountably many types are realized in 
, a contradiction.
(It is a general fact that -stable theories are totally transcendental. The proof goes as follows: if 
 failed to be totally transcendental, then 
 for some set 
. Then one inductively builds a tree 
 of non-empty 
-definable sets such that 
 is the disjoint union of 
 and 
 for every 
, and such that each 
 has Morley rank 
. This is the same construction used to prove that perfect sets in Polish spaces have cardinality 
. At any rate, there are countably many 
’s, so the 
’s are all definable over some countable set 
. Now each path through the tree yields a different type over 
, so that there are uncountably many types over 
, contradicting 
-stability.) QED