限制共時與歷時的變異和演變：中心詞後置短語須居其上之約束條件

Limiting Synchronic and Diachronic Variation and Change: The Final-over-Final Constraint

本文探討共時語法的一個約束條件──「中心詞後置短語須居於中心詞後置短語之上之約束條件」（the Final-over-Final Constraint，以下簡稱為FOFC）(Holmberg 2000, Biberauer, Holmberg & Roberts 2007, 2008)，以及此約束條件對縱時變化的預測。作為一個絕對原則，FOFC 排除了中心詞後置結構居於隸屬相同範疇的中心詞前置結構之上的可能性。這預示著從中心詞後置詞序發展到中心詞前置詞序必須經歷由上至下的發展過程（從標句詞開始），而由中心詞前置詞序發展到中心詞後置詞序必須經歷由下至上的發展過程（從動詞開始）。英文、南非荷蘭語以及衣索比亞閃族語的個案研究似乎支持這些預測。基於Biberauer, Holmberg & Roberts 文中提出的理論，本文提出FOFC 是階段和諧產生的效應。由此，如若階段中心詞中含有線形辨識符（標誌著其中心詞後置屬性），那麼所有該中心詞可見的，與之範疇相同的中心詞也必須含有該辨識符。根據階段論的標準版本，該理論預示著FOFC 不受階段界線的制約。然而這一事實引發出的推斷與上面所提出的詞序變化理論明顯是相矛盾的，後置結構變前置或者前置結構變後置似乎都是不可能的。本文重點提供實例證據，以顯示語言是如何避開這一詞序變化的潛在障礙的，而詞序變化的發展順序確實與FOFC 的預測相一致。

This paper explores a synchronic grammatical constraint, the Final-over-Final Constraint (henceforth: FOFC; Holmberg 2000, Biberauer, Holmberg & Roberts 2007, 2008), and its predictions for diachronic change. As an absolute principle, FOFC rules out the possibility of a head-final phrase dominating a categorially non-distinct head-initial phrase. This predicts that change from head-final to headinitial order must proceed from the top down (starting with the complementizer), whereas change from head-initial to head-final order must proceed bottom-up (starting with the lexical verb). Case studies from English, Afrikaans, and Ethiopian Semitic appear to support these predictions. Building on the account in Biberauer, Holmberg & Roberts (2007, 2008), it is proposed that FOFC is an effect of phaseharmony, whereby if a phase head bears a linearization diacritic (signalling headfinality), then all the categorially non-distinct heads visible to it must also bear this diacritic. According to standard versions of phase theory, this account predicts that FOFC should hold across phase boundaries. This fact, however, creates an apparent ‘paradox’ for the theory of word-order change proposed, making it appear impossible for final structures to become initial or vice versa. The paper highlights empirical evidence to show how languages avoid this potential barrier to wordorder change, which does indeed seem to progress as FOFC would predict.