Featural Cyclicity and (Anti-)Superiority Norvin Richard (MIT) Pesetsky's (1982) Path Containment Condition, which requires movement paths to nest rather than cross, has been argued by Kitahara (1994) to follow from Cyclicity and Shortest Move. In the well-formed derivation in (1a), at the point in the derivation at which Y is introduced into the structure, it attracts AP, the nearest available mover; next, X is introduced into the structure and attracts the nearest available mover again, which is BP (AP is no longer available, since it has already moved). The structure in (1b) has two possible derivations, both of which are ruled out; either movement 1 occurs first, which is ruled out by Shortest Move, or 2 occurs first, which obeys Shortest Move but is ruled out by Cyclicity. 1. a. [BP X [AP Y [ta tb ]]] b.*[AP X [BP Y [ta tb ]]] ^ ^_____| | ^ ^--1--|--' ---------------| ---------2-- Similarly, the English Superiority paradigm in (2) follows from Shortest Move; the attracting C attracts _who_ first, in the overt syntax, since this is the shortest move. 2. a. Who t bought what? b.* What did who buy t? There is a class of cases in which paths must cross, rather than nest, in violation of the PCC; cf. the Bulgarian examples in (3). On our standard assumptions about adjunction, this is the opposite of the Superiority facts in (2); the object seems to move first, followed by the subject. Similarly, Japanese exhibits an "Anti-Superiority" effect, given in (4), on the assumption that _naze_ 'why' must occupy the lowest Spec CP in order to antecedent-govern its trace (Saito 1982). ----------| v v-----|--------| 3. a. Koj kogo e t vidjal t who whom is seen 'Who saw whom?' ------------------| v v----| | b. * Kogo koj e t vidjal t ----------------| v v-------------|------| 4. a. John-ga nani-o naze katta no? John NOM what ACC why bought Q 'What did John buy why?' ---------------| v v------------|----| b. * John -ga naze nani -o katta no? John NOM why what ACC bought Q I will show that this effect and its distribution are predicted by the considerations that derived the facts in (1) and (2). The account will use Chomsky's (1995) version of Cyclicity, in which Cyclicity follows from a requirement that strong features be checked immediately upon being introduced into the derivation. Featural Cyclicity Consider the countercyclic derivation in (1b), in which movement 2 precedes movement 1. This derivation is ruled out because the strong feature introduced by Y is not checked immediately, but waits until after X and its specifier are introduced into the derivation. Now suppose we consider a case in which the landing sites are multiple specifiers of a single head, as in (5), rather than of multiple heads, as in (1). Here Cyclicity fails to distinguish between the derivation in (5a), where move 2 lands in a specifier to the left of the existing one, and that in (5b), where move 2 is to a specifier to the right of the existing one; in either case, no new structure is introduced before movement checks a strong feature. In fact, Shortest Move prefers the derivation in (5b), since Step 2 of (5b) is a shorter move then Step 2 of (5a). Featural Cyclicity and Shortest Move thus conspire to predict nested paths in cases like (1), and crossing paths in cases like (5). 5. a. Step 1: [AP Y [ta BP ]] Step 2: [BP AP Y [ta tb]] ^-----| ^ |-----| | |-----------| b. Step 1: [AP Y [ta BP ]] Step 2: [AP BP Y [ta tb]] ^-----| | ^-----|--| |--------| We expect, then, to see Anti-Superiority effects in cases of multiple movement into a single checking domain at a single point in the derivation. Such effects are indeed found in Japanese (all movement covert) and Bulgarian (all movement overt), as predicted. In cases where only a single XP moves on a particular level, it must be the highest XP, as before; this is the English case. The distinction between Superiority and Anti-Superiority thus reduces to a matter of timing. (Anti-)Superiority and Timing In fact, there is evidence internal to Japanese that this is the correct approach. Takahashi (1993) notes that long-distance scrambling of wh-words in Japanese obeys Superiority, rather than Anti-Superiority (6, from Takahashi 1993, 664). Long-distance scrambling must attract the higher of the two wh-words, as Shortest Move predicts. 6.a.John -ga [Bill -ga dare-ni [Mary-ga nani-o tabeta to] itta to] omotteiru no John NOM Bill NOM who DAT Mary NOM what ACC ate that said that thinks Q 'Who does John think that Bill told that Mary ate what?' v----------------------| b.Dare-ni John-ga [Bill-ga t [Mary-ga nani-o tabeta to] itta to] omotteiru no v-----------------------*---------------| c.*Nani-o John-ga [Bill-ga dare-ni [Mary-ga t tabeta to] itta to]omotteiru no On the other hand, multiple long-distance scrambling to a single position exhibits Anti-Superiority again, again as predicted. 7. a. John-ga [Tanaka-sensee -ga dare-ni nani-o yomaseta to] itta no? John NOM Tanaka teacher NOM who DAT what ACC read-CAUS that said Q 'Who did John say that Professor Tanaka made read what?' -------------------------------------- v v------------------------------|--| b. Dare-ni nani-o John-ga [Tanaka-sensee-ga t t yomaseta to] itta no? -----------------------------------------| v v-------------------------------| | c.* Nani-o dare-ni John-ga [Tanaka-sensee-ga t t yomaseta to] itta no? When the timing of the movements parallels English, then, Superiority effects appear, as in English, and when the timing is as in Bulgarian, Anti-Superiority effects resurface, as predicted. A-Movement An account of (Anti-)Superiority effects based on Cyclicity and Shortest Move predicts that all movements should in principle obey (Anti-)Superiority. Local Japanese scrambling arguably involves movement to multiple A-specifiers. Such scrambling interacts, as is well known, with a variety of semantic properties, including focus and quantifier scope. If such effects are controlled for, however, the (Anti-)Superiority paradigm emerges. Scrambling of multiple idiom chunks, for instance, behaves as predicted: if a single chunk is scrambled, it must be the higher of the two (8.b-c), while if both are scrambled the paths must cross (8.d-e). 8. a. Taroo -ga hi -ni abura-o sosoida Taroo NOM fire DAT oil ACC poured 'Taroo made things worse' v-------------| b. Hi-ni Taroo-ga t abura-o sosoida v----------------*----| c. * Abura-o Taroo-ga hi-ni t sosoida -------------------- v v-------------|--| d. Hi-ni abura-o Taroo-ga t t sosoida -----------------------| v v-------------| | e. * Abura-o hi-ni Taroo-ga t t sosoida Scrambling of multiple quantifiers behaves similarly when it has no effect on quantifier scope (9b-c, from Yatsushiro 1996); scrambling which disobeys (Anti-)Superiority does affect quantifier scope (9d-e). Thus, semantically vacuous scrambling (i.e., scrambling of nonreferential elements, or scrambling of quantifiers with no effects on scope) does indeed obey (Anti-) Superiority, as predicted. I assume that semantically contentful scrambling must involve multiple attractors. 9. a. John-ga dareka -ni daremo -o syookaisita John NOM someone DAT everyone ACC introduced 'John introduced everyone to someone' unambiguous: someone>everyone v---------------| b. Dareka-ni John-ga t daremo-o syookaisita unambiguous: someone>everyone -----------------------| v v--------------|--| c. Dareka-ni daremo-o John-ga t t syookaisita unambiguous: someone>everyone v-----------------------| d. Daremo-o John-ga dareka-ni t syookaisita ambiguous: some>every, every>some --------------------------| v v--------------| | e. Daremo-o dareka-ni John-ga t t syookaisita ambiguous: some>every, every>some