源於組合的非透明現象

Opacity from Constituency

早在1980 年代的Lexical Phonology，推演的非透明度與結構的複雜度是對應的。可是1990 年代後的優選論不再承認推演的過程，所以結構與非透明度的關係模糊了。本文以普通話連上變調為例，在優選論框架內，提出層級對應論 (ICT)，假設結點之間須信息對應，重新將結構納入理論範圍。底層結點信息來自輸入項，其他結點由按結構逐層與底層結點相應。這麼一來標記性制約條件的適用範疇就轉到了結點上，得出標記性乃因同屬一個音系單位而起，非相鄰之罪的結果。

Structural opacity often has derivation histories as deep as the number of structural tiers. This is implicit since the days of Lexical Phonology in the 1980s. However, with the advent of Optimality Theory (OT), the relationship between structural depth and opacity has become obscured due to the output-orientation of OT. This paper uses Mandarin third tone sandhi as a case study of structural opacity to develop, within OT, the Inter-tier Correspondence Theory (ICT) where structural configurations are taken into account by allowing for nonterminal nodes to reconstruct information of subordinate nodes. Since structures are built by GEN and selected by EVAL, information at the terminal nodes must come directly from the input string; information at non-terminal nodes would be derived by inter-tier correspondences. This asymmetry of terminal and non-terminal nodes allows ICT to marry the insights of both the Containment Approach and the Correspondence Approach to Optimality Theory. ICT further predicts that mere adjacency does not result in markedness, and consequently does not trigger alternation. Rather, it is to the constituency of units that markedness constraints apply. This is why derivation histories are rarely deeper than there are structural tiers.