Distributional Paradoxes
1Simplicity and Restriction in Grammar
For any linguistic phenomenon, if there are a number of possible linguistic processes that could apply to account for that phenomenon, the fewer of these processes that are used by a grammar, the simpler the grammar is.
It could be argued that if the adoption of a single principle of type X allows the discarding of a number of principles of type Y, then it is simpler to assume that the two types of principles operate together instead of basing the grammar on one type
Against this are the following:
1.it is clearly difficult to calculate ‘simplicity’ when comparing two grammars just by counting the number of operations needed to generate a language: do we apply weighting to one type of operation as opposed to another?; if so, on what basis?; how much?; etc.
2.Simplicity is only really useful from a theoretical point of view if it increases understanding and explanation – e.g. adding constraints to early transformational grammars allowed transformations to become more general and the constraints themselves ultimately were made general, so overall the grammar became more general and more explanatory
3.The kind of simplicity we are referring to here has to do with restriction – restriction always adds to explanation: if a grammar is restricted to one type of process it is more explanatory to say that this reflects something basic about the system than if two or more processes are used – this will always lead to the question: why these and not the others? ‘Because that’s how language is’ is a better answer to the question: why this principle as opposed to all the other? than it is to: why these set of principles as opposed to other possible combinations?
Therefore we conclude that it would be better to assume a more restrictive theory than a less restricted one.
2Distribution is a linguistic phenomenon which has a number of possible grammatical accounts
The notion of structure is a very successful way of defining positions in an expression which can be occupied by linguistic elements. It works like this:
If X is an element, its distribution is given relative to the constituent it is immediately part of
Y ... X ...
then the distribution of this constituent, which therefore determines wider the distribution of X, is given relative to the constituent it is immediately part of:
Z ... Y ...
and so forth until the root is reached
In other words:
If X has distribution, this is given with respect to the root or with respect to Y which has distribution
What ‘given with respect to X’ means is ‘placed in a position relative to the other immediate constituents of X’
It is also possible to give a linear account of distribution:
If X is an element, its distribution is given relative to Y that it is dependent on
X precedes Y, X follows Y
then the distribution of Y, which therefore determines the wider distribution of X, is given relative to Z which it is dependent on
Y precedes Z, Y follows Z
and so forth until the root (element) is reached
In other words:
if X has distribution, this is given with respect to the root or with respect to Y which has distribution
What ‘given with respect to X’ means is placed in a position relative to X
It should be noted that the linear approach may involve conflict
if the linear relations are not just simply X preceded or follows Y at any distance (it would be difficult to see how this could provide an account of distribution) then we are bound to get conflicts such as X precedes Y and Z precedes Y or X precedes Y and Y follows Z
The structural account avoids such conflict by placing limitation on the elements that any particular element is ordered with respect to (X will only precede or follow the elements which are its co-immediate constituents within a constituent).
3Given this, we can ask is it possible to give an account of distribution in purely structural terms?
4To determine this we need to recognise certain properties of structurally determined distribution
It follows from the lack of conflicting requirements in the structure approach that its conditions are absolute
Thus we can say that X and Y will be in complementary distribution if they conform to the same distributional requirements
Note that this is not so in a system which allows conflict as if X and Y have the same distributional requirements (and hence are in conflict with each other) then they are not necessarily in complementary distribution depending on how the conflict is resolved
It also follows that if X has the same distributional requirements as Y and Y has the same distributional requirements as Z, then X and Z will be in complementary distribution as the relationship ‘has the same distributional requirements’ (= is the same category as) is transitive
Again, this is not necessarily so if conflict is allowed: if X, Y and Z all have the same distributional requirements, it follows that the interaction between them will be more complex, but not that the appearance of one will restrict the appearance of any of the others.
Another consequence of the structural approach to distribution is that if two elements are in complementary distribution with each other in one place, then they should be in complementary distribution with each other in all places
This is because if distribution is defined in terms of structure alone, then there can be no other explanation for why two elements are in complementary distribution than that they have the same distributional requirements
5Here are some paradoxical distributions which cannot be given a purely structural account
1.Wh phrases are in complementary distribution with complementisers
2.complementisers are in complementary distribution with inverted auxiliaries
3.wh phrases are not in complementary distribution with inverted auxiliaries
If structure were the only determining fact for distribution then this situation cannot arise.
Obviously some other explanation is called in to account for this:
Wh-phrases and complementisers cannot co-occur not because they occupy the same structural position, but because “the specifier of CP and the head of C cannot be filled simultaneously by overt elements” (Doubly filled COMP filter). This looks suspiciously like a linear restriction, not a structural one.
Determiners and Possessors are in complementary distribution
This rule introduces something more than structure to account for the observation: the curly brackets is not something that can be rendered in purely structural terms. Note, it isn’t the same as:
Because these say that a determiner and a possessor are in front of a noun in an NP, but not that they occupy the same structural position. For this you would need something like:
But this then classifies determiners and possessive NPs as the same thing
1.Hungarian negative is in complementary distribution with the pre-verb
2.pre-verb is in complementary distribution with foci
3.foci are not in complementary distribution with the negative
1.Hungarian pre-verbal foci are in complementary distribution with fronted Negative phrases
2.pre-verbal foci are in complementary distribution with each other
3.fronted negative phrases are not in complementary distribution with each other