It is not possible to predict language change, either internal or external. For instance, German has lost the inherited ambidental fricatives from Germanic (contrast Du with thou in English) but English has not. One can say that German got rid of unusual marked segments, but why did English not do the same? English simplified the complex clusters /kn, gn/ at the beginning of words to /n/ (know, gnaw) but German did not.
One can nonetheless offer explanations for why certain changes might have taken place or why marked elements might be retained. Consider the fact that unusual changes can be carried through if the speech community is homogenous or if for some reason they become markers of social class. Icelandic has a distinction between long and short diphthongs which is statistically very rare in the world’s languages. However, the Icelandic speech community is small, closely-knit and aware of its language and the need to preserve it was handed down by previous generations. Nasal vowels are less usual than oral vowels statistically but nasality is often a feature of a class or recognisable groups in a society. This may account for why these vowels developed as phonemes in French, assuming that the better positioned groups in French society of the time favoured definite nasalising of vowels before nasals.
HOW MUCH HOMOPHONY CAN A LANGUAGE HANDLE? The simple answer to this is: quite a lot. Chiefly because language contains a lot of redundancy — information specified more than once — and the context in which something is said usually provides unambiguous clues about what is meant.
Given that language is a set of subsystems, disadvantageous developments in one area can most often be compensated for by some aspect from another area. A simple example of what is meant here is homophony. Hence the homophony which arose in Received Pronunciation due to the loss of syllable-final /r/ did not disturb the system as word-class considerations were sufficient to differentiate the resulting homophones: lead : lead, court : caught. Of course by word-class considerations one means that the homophonous elements cannot occur in the same environment and so are unlikely to be ambiguous in communication.
As long as pragmatic context disambiguates language the speakers would appear not to put restraints on possible language-internal developments. An area which is normally regarded as one where speakers may not be prepared to accept change is that of morphology: think of the development of pronouns of the third person singular in the late Old English/Middle English period. However, even here the limit is not so quickly reached. Consider the following instance from present-day German. The third person singular pronoun is sie, phonetically [zi:] in the standard. The polite form of the pronoun of address (corresponding to vous in French) is Sie, phonetically also [zi:]. Now there are occasions where these forms collide, i.e. where the reference could be to a third female person not present in a conversation or to the person being addressed (assuming that one is using the V-form of address). Here is a real-life situation. A is discussing with B the arrangements for C (female) who is coming to visit both A and B. B wants to ring C later about details but A thinks that a question with regard to this is directed at himself: Wie kann ich [zi:] erreichen? ‘How can I reach her / you (formal) ?’. This is a genuine ambiguity in the pronominal system of German but one which, given a reasonable pragmatic context, does not cause difficulties for speakers or which can at least be resolved by brief clarification. Obviously the number of situations in which this is communicatively ambiguous is not sufficient to motivate speakers towards some alternative which is structurally unambiguous.
HOMOPHONY AND AVOIDING CLASHES Care must be exercised in the area of homophony not to assume that a certain change took place to avoid homophony. Here is an instance of what is meant. The word for ‘barrel for alcoholic drink’ used to have an initial /f-/ (the inherited sound, cf. German Fass) but was replaced by a borrowing from the dialect of Kent which had a voiced initial fricative, hence modern English vat. Before this the word was homophonous with the adjective fat. But it would be a quite unsubstantiated claim to maintain that the Kentish borrowing of the noun took place in order to to avoid homophony with this adjective.