1 ISLAND DOMAINS IN ADYGHE AND THE TYPOLOGY OF SUBORDINATION Dmitry V. Gerasimov Institute for Linguistic Studies, St. Peterburg, Russia It has been generally assumed, since the groundbreaking study by Ross [1], that major types of subordinate clauses behave differently with respect to instances of A’-movement, like wh- extraction and relativization. While complement clauses are “transparent” for argument ex- traction, other types of subordinate constructions, like adverbial clauses, relative clauses and complex subjects are not; they form syntactic “islands” out of which no extraction is possible. There have been many attempts to explain these in one way or another, mostly within the generative framework ([1–5], inter alia). It must be noted, however, that in many world’s lan- guages the situation is not that straightforward. Such is the case with Adyghe, a highly polysynthetic language of West Caucasian fam- ily. While no extraction is possible out of relative clauses, particular subtypes of adverbial and complement clauses exhibit deviations from the pattern predicted by Ross and his followers’ theory. Island effects are not observed in adverbial clauses marked with adverbial suffix -ew (1a-b), just as in case of complement clauses of modal and phasal predicates marked with the same suffix (1c-d). This seems particularly surprising given that NPs marked with -ew gener- ally behave like adjuncts and not arguments. (1) í. se sE -f aj som-jE -I qE -s -a -tE -n -ew. I 1SG-want rouble-100 INV-1SG-3PL-give-FUT-ADV ‘I want them to give me 100 roubles’. b. tHapSa wE -zE -f aj -er qE -w-a -tE -n-ew? how.much 1SG-REL-want-ABS INV-2SG-3PL-give-FUT-ADV ‘How much do you want them to give you?’ c. se sE -qe -KWa -R somjEI qE -s -a -tE -n -ew. I 1SG-INV-come-PERF rouble-100 INV-1SG-3PL-give-FUT-ADV ‘I came so that they give me 100 roubles’. d. tHapSa wE -zE -faje -r wE -zE -(fe )-KWa -R -er how.much 1SG-REL-want-ABS 2SG-REL-purp-COME-PERF-ABS qE -w-a -tE -n -ew? INV-2SG-3PL-give-FUT-ADV lit. *‘How much did you come so that they give you?’ Other types of adverbial clauses for which we have data do not allow extraction, even if similarly marked complement clauses do. Finally, argument extraction out of another major type of Adyghe complement clauses—factive complements headed by verbs bearing zere- prefix,—is impossible: (2) a. se se -I& e Hasan qE -zere -KWa -R -er. I 1SG-know Hasan INV-zere-come-PERF-ABS ‘I know that Hasan came’. b.*heta wE -zE -I& e -re -r qE -zere -KWa -R -er? who 2SG-REL-know-BCG-ABS INV-zere-come-PERF-ABS ‘Whom do you know to have come?’ This may provide additional evidence in favour of our hypothesis that zere-complements are in fact relative clauses, formed by relativization of a covert “factive argument” [6]. It is