You are on page 1of 13

Handout: The antisymmetry of syntax. Richard S. Kayne.

Marlene Guraieb October 2, 2013

A linear order of T
Kaynes attempt: building a theory of syntax from primitive notions (see denition 3). Question to answer, what are the minimum axioms that can generate our insights? Key point: X Theory is not a primitive notion, there is a smaller set of axioms that generates it. To get to that, we need a linear order. Why the fuss about linear orders? Answer will come in two parts: 1. So far, the notion of dominance (see denition 4) does not produce a linear order of the set of objects we are interested in. This is a problem because we need to be able to compare all of them systematically. 2. With a little tweaking, we can produce a denition of dominance that does produce a linear order and explain our insights with more ontological parsimony 1 .

1.1
Tools:

The new ac-command

Bring back category/segment distinction (see denition 14) Exclusion: a category excludes all categories not dominated by all its segments Using this we can say X c-commands Y i these three conditions are met (henceforth cond. 1,2 & 3):
1 This is cool because (perhaps arbitrarily) we always try to select the hypothesis with the fewest assumptions.

1. X and Y are categories; and 2. X excludes Y; and 3. every category that dominates X dominates Y (def. 15). Example: AP1 BP B b AP2 A CP a C c AP1 and AP2 are both segments of a single category. Note that cond. 1 means that AP1 & AP2 do not enter into any c-command relations. AP does not c-command BP because it does not exclude BP (AP1 dominates BP, but AP2 does not, so AP does not exclude BP violating cond. 2). CP does not c-command BP because both segments of AP do not dominate BP (it is not the case that every category that dominates CP dominates BP violating cond. 3). BP c-commands CP and A. A c-commands C.

1.2

Building d: towards a linear order of T

Now lets think about d(X ) see (denition 11) Example: K J L

j M N m P p 2

Dene d(T P ) : Nonterminal Terminal Domain of d(X ) X : {K,J,L,M,N,P} Image of d(X ): {j,m,p } Mapping: d(K) { j,m,p } d(J) {j} d(L) { m,p } ...2 The image under d of < X, Y > In our example: d < K, J >: {< j, j >, < m, j >, < p, j >} d < L, K >: {< m, j >, < m, m >, m, p >, < p, j, < p, m >, < p, p >}
3

Now lets build d({< L, K >, < J, L >, < J, M >}) = i 0 d(< Xi , Yi >):{< m, j >, < m, m >, m, p >, < p, j, < p, m >, < p, p >, < j, j >, < m, j >, < p, j > , < j, m >} Lets build the set A (see denition 13): A(K ) = {< J, M >, < J, N >, < J, P >, < M, P >}

1.3

Proposal

Theorem 1. Linear Correspondence Axiom: d(A) is a linear ordering of T . Proof. More than a proof, lets see how this works in practice: The pairs that constitue A in the tree below are: < J, M >, < J, N >, < J, P >, < M, P >. d(A) =< j, m >, < j, p >, < m, p > Now think of the set {j, m, p} which is what weve been calling T (the set of all terminal nodes). These three ordered pairs constitute a linear order of the set T: 1. Transitivity: jdm mdp jdp.
2 But

notice that it this is not the whole story: d(X )1 :

p { P,N,L,K } j {J,K } ... This is why d(X) is a many-to-many relation. A terminal node is d(X ) ed by many nonterminal nodes & a nonterminal node d(X ) es many terminal nodes. 3 Remember a set is a collection of distinct objects, but the objects in this case are ordered pairs, so < m, m > and < m, m > where in the rst one < m, m >< d(L), d(K ) > and in the second one < m, m >< d(K ), d(L) > are the same object, however < m, j > and j, m > are dierent objects.

2. It is total: x, y {j, m, p} s.t. x = y either xdy or ydx (no pair is not comparable in the set {j, m, p}) 3. It is antisymmetric (xLy yLx). Notice either jdm or mdj , but not both (i.e. d() did not produce pairs such as < j, m > and < m, j > simultaneously). Hence, we can say that d(A) produced a linear order or total order of the set T . Intuitively, it gave us a way to compare all the elements that is complete, antisymmetric and transitive.

2
2.1

Can X be derived entirely from these primitives?


Do our primitives produce the same rules as X ?

You cannot have two heads as sisters: K J j M L P

see John Note that d(K ) does not linearly order the set {j, m, p} because neither < m, p > nor < p, m > is in there. This is why you cant have two heads as sisters. In natural language: the complement John cannot be dominated solely by P, but it needs another node to break the symmetry with M. Results: If two non terminals are sisters: if one of them is a head and the other a nonhead, the phrase marker is admissible; if both are heads, it is inadmissible; if both are nonheads, it is inadmissible. (1) (2) * I saw the boy the girl * The girl the boy were discussing linguistics.

Now we know not only that the previous examples are ungrammatical, but why they are ungrammatical. Its in the antisymmetry, we need a coordinating conjunction like and to serve as a head to bring coordinate structures in line with the LCA. 4

3
3.1

Adjunction
Adjuncts and heads

Whats wrong with this guy? XP AdvP Adv V VP DP D the report We cant gure out the order between carefully & study and carefully & the report (both < caref ully, study > and < study, caref ully > are in d(A)) 4 Problem: how do we maintain this distinction for when M & P are maximal projections, but at the same time allow for adjuncts? Solution: Lets think of L & P as two segments of one category and apply denition 15. Does this x our problem? VP AdvP Adv V VP DP D the report
4 To

carefully study

carefully study

see this, lets nd d(L):

A : {< AdvP, V >, < AdvP, DP >, < V, D >, < AdvP, D >, < V P, Adv >} d(A) : {< caref ully, study >, < caref ully, thereport >, < study, thereport >, <

caref ully, thereport >, < study, caref ully >, < thereport, caref ully >}

Yes. It eliminates < V P, AdvP > from A and thus gives us the answer: < caref ully, study > < study, caref ully > & < caref ully, thereport > < thereport, caref ully >, appropriately antisymmetric and thus linear.

3.2
3.2.1

Clitics
One clitic L M P

Q M R S q m r T t Lets take our modied notion of c-command (see denition 15). Does Q c-command R? Q c-commands R i: Q and R are categories ( ); Q excludes R (A category excludes all categories not dominated by both its segments) ( ); and Every category that dominates Q dominates R. (In this case, Q is dominated only by L) ( ). So, Q c-commands R (and S and T for the same reasons). General property of adjoined phrases: they always c-command out of the phrase they are adjoined to. Check also: Q c-commands M too! Now lets try it when Q is not a head and is replaced by [U [W w ] ]. Then, as before, U and P enter into no ac-command, but now P c-commands W, but U ac-commands R (and S, and T) so now we have < r, w > and < w, r > violating antisymmetry: A non head cannot be adjoined to a head.

3.2.2

Multiple clitics L M K M P R S r T t

k Q M q m

Same as tree before, but now K (clitic) is adjoined to the same head M. Both K and Q dominate P,R,S and T. But what happens between K and Q? M doesnt dominate either of them, so they do not ac-command each other. So, < k, q > or < q, k >? Neither is in d(A), so our tree is excluded by LCA. Solution: L M Q P M R S r T t

K Q m k q

A : {< K, Q >, < K, M >, < Q, M >, < M, R >, < M, S >, < M, T >, < K, R >, < K, S >, < K, T >, < Q, R >, < Q, S >, < Q, T >, < R, T >} d(A) : . . . trust him, linearly ordered. One clitic per head then . . . Lets think of an actual example and work through it:

TP DP DP John D D NumP NP mother categories: DP, D, NumP Num Np T VP V DP A : {< , >, < , >, < , >, < , >, < , >, < , >, < , >, < , >, < , >, < , >, < , >} d(A) : {< , >, < , >, < , >, < , >, < , >, < , >, < , >, < , > , < , >, < , >, < , >} T e V T VP DP

s Num

likes herself

3.3

Nonhead adjuncts
P L K M k P P

Q R S q r T t

Violation of antisymmetry: L ac-comm Q & M ac-comm K. So, {< L, Q >, < M, K >} A(P ) {< k, q >, < q, k >} d(A) Thus, you cannot adjoin more than one nonhead to a nonhead. A given phrase can only have one specier. 8

3.4

Speciers
L P M P L K J k H h

Q R S q r T t

M ac-comm K,J,H {< M, K >, < M, J >, < M, H >} A(L) {< q, k >, < q, h >} d(A) M ac-comm R,S,T {< M, R >, < M, S >, < M, T >} A(L) {< q, r >, < q, t >} d(A) R ac-comm T {< R, T >} A(L) {< r, t >} d(A) K ac-comm H {< K, H >} A(L) {< k, h >} d(A) P ac-comm K {< P, K >} A(L) {< r, k >, < r, h >, < t, k >, < t, h >} d(A) LCA .

Word order

Recall denition 6: Denition 1. Immediate precedence immediately precedes: A node A immediately precedes a node B if it is immediately to the left of B. In tree (1) on the left, NP1 immediately precedes T and T immediately precedes VP, but NP1 does not immediately precede VP. T also immediately precedes V3 M in tree (1). And T also immediately precedes VP and it also immediately precedes V. Denition 2. Precedence A node A precedes a node B if A immediately precedes B

A node A precedes a node B if there is a node C such C precedes B and A precedes C. Theorem 2. Let X,Y be nonterminals and x,y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c-commands Y, x precedes y. Now we know that speciers ac-command heads, and that heads ac-command complements. It therefore follows that speciers must precede heads and complements must follow them. So, it follows that SOV is strictly impossible without movement, we need the complement to raise to a specier position for this to make sense.

10

Formal Denitons

Denition 3. Somewhat ironically: Primitive notion A primitive notion is an undened concept. In particular, a primitive notion is not dened in terms of previously dened concepts. Formal theories cannot dispense with primitive notions, under pain of innite regress. When we set out to construct a given discipline, we distinguish, rst of all, a certain small group of expressions of this discipline that seem to us to be immediately understandable; the expressions in this group we call primitive terms or undened terms, and we employ them without explaining their meanings. At the same time we adopt the principle: not to employ any of the other expressions of the discipline under consideration, unless its meaning has rst been determined with the help of primitive terms and of such expressions of the discipline whose meanings have been explained previously. The sentence which determines the meaning of a term in this way is called a denition. For example, in Euclidean geometry, the primitive notions are point, segment and motion. Denition 4. Dominance Theoretical primitive represented graphically in terms of top-to-bottom order, i.e. if a node A dominates a node B, A appears above B in the tree. The root node (or start symbol) is the only node that dominates every other node and is itself not dominated by any node. Dominance is transitive. It is also reexive (xDxx X ). Note. The formal denition of constituency can be expressed in relation to dominance. A constituent is a set of nodes exhaustively dominated by a single node. Denition 5. Immediate dominance A immediately dominates B i: 1. A dominates B, and 2. there is no node C, distinct from A and B, such that A dominates C and C dominates B. Immediate dominance is not transitive. Also, immediate dominance is equivalent to motherhood. Denition 6. Precedence The second primitive relation. Left to right order on the trees. Dominance and precedence are mutually exclusive, i.e. if A dominates B, A cannot precede B and conversely, if A precedes B A cannot dominate B. Precedence is transitive (but just as with dominance, there is immediate precedence which is not transitive). In more simple terms, node A precedes node B i: 1. A is to the left of B 2. neither A dominates B nor B dominates A 11

3. every node dominating A either precedes or dominates B? Properties of precedence: Antisymmetric: if xP y , then (yP x) Transitive: if xP y , and yP z xP z Preference and dominance are mutually exclusive: if xP y yP x xDy yDx Lines dont cross! xP y i u, v s.t. u, v are terminal nodes, (xDu yDv ) uP v . Denition 7. c-command Node A c-commands node B if every branching node dominating A also dominates B, and A does not itself dominate B. Note: Symmetric C-Command sisterhood. Asymmetric c-command aunts to nieces. Denition 8. Linear order A relation on a set S that is transitive, antisymmetric and total. 1. Transitivity: xLy & yLz xLz . 2. It is total: x, y X s.t. x = y either xLy or yLx. 3. It is antisymmetric (xLy yLx). Denition 9. Many-to-many relationship A type of cardinality that refers to the relationship between two entities A and B in which A may contain a parent row for which there are many children in B and vice versa. For instance, think of A as Authors, and B as Books. An Author can write several Books, and a Book can be written by several Authors. Denition 10. Head A nonterminal node that dominates no other nonterminal node. Every other node is a nonhead. Denition 11. d (dominance with a small caps d) For a given nonterminalX , d(X ) is the set of terminals that X dominates, d(X ) is the image under d of X . d(X ) is a many to many mapping from non-terminals to terminals. Denition 12. The image under d of < X, Y > The Cartesian product of d(X ) and d(Y ): d < X, Y >= X where X = {< a, b >, < c, d >, . . . , < e, f >} s.t. a, c, e d(X ) b, c, f d(Y ). Denition 13. Set A The maximal set of ordered pairs < Xj , Yj > s.t. j, Xj asymmetrically ccommands Yj .

12

Denition 14. Category/segment A category is an extended node: if two directly connected nodes in a tree have the same label, these two nodes are both segments of a single category. Ccommand is dened in terms of categories using the notion of exclusion. A category excludes all categories not dominated by both its segments. A ccommands B if every category that dominates A also dominates B, and A excludes B. Example: AP1 BP B b AP2 A CP a C c AP1 and AP2 are both segments of a single category. AP does not c-command BP because it does not exclude BP (AP1 dominates BP, but AP2 does not, so AP does not exclude BP). CP does not c-command BP because both segments of AP do not dominate BP (it is not the case that every category that dominates CP dominates BP). BP c-commands CP and A. A c-commands C. Denition 15. c-command 2 X c-commands Y i X and Y are categories and X excludes Y and every category that dominates X dominates Y. Note. You cannot strand a clitic. In the denition 15 the idea that Y too must be a category has potential signicance: if a segment cannot be c-commanded and if antecedent government strictly has c-command as a necessary component, then a segment cannot be antecedent-governed and thus cannot be moved. In other words, a phras ethat has something adjoined to it cannot be moved by itself.

13

You might also like