【摘要】 本文按照最简方案框架下的理论探索思路,尤其是Chomsky（1999）的《语段推导》,论证句法推导可以完全按自下而上、由右向左的方式进行。由于附加语的推导在现有理论中有不同的处理并且带来许多理论上的问题（Larson　1988,Kayne 1994,Chomsky 1995b,Epstein et al 1998）,作者以此为研究突破口,试图取消右向附加。大量的例句分析显示,以往通过附加手段处理的副词、介词短语、状语从句等可以通过合并的方式进入句法推导。它们的表层位置均是其它语类经过相关移动后推导出来的结果。 通过分析英汉语的附加语,我们发现附加语有A、B、C、D四种合并位置,均在VP和TP的指示语位置,即[TP AdvD[TP Spec[TP AdvC[TP T[v＊P AdvB[v＊PSpec[v＊Pv＊[AgroP Spec[AgroP Agro[VP AdvA[VP[VDP]]]]]]]]]]]]。例如,句末附加语的基础生成在下层VP的指示语位置,动词和宾语移动后,停留在原位的附加语位于句末。所有附加语都在成分统制被修饰对象的位置进入句子的推导。 作者将Chomsky的语段推广至带谓词的介词短语（关联词也是介词,因此状语从句是介词短语）,认为这类结构是CP强语段。介词的内在属性[+P]决定T是否有EPP特征,或者说是否有主语。有些介词没有带主语是因为在此无法删除特征的DP通过CP语段的边沿移动至主句的相更多还原

【Abstract】 Categories such as DP (D and N), V, T, Neg, C, Agr enter syntactic derivations by means of bottom-up, leftward merger. However, adjuncts, including adverbs, prepositional phrases, adverbial clauses, are adjoined to XP, leftward or rightward, thus violating strict cyclicity in derivation and causing chaos in order of elements. These inconsistencies concerning adjunction are obviously theoretical flaws and have cause many problems, such as determination of the linear order of syntactic objects and the maintenance of strict successive cyclicity in derivation (Larson 1988, Kayne 1994, Chomsky 1995b, Epstein et al 1998).This paper proposes the Principle of Generalized Leftward Merger to guarantee a strictly successive cyclic derivation of any syntactic configuration in a bottom-up, leftward manner.Principle of Generalized Leftward Merger (PGLM)In the whole process of syntactic derivation from base generation to convergence, all categories enter theconfiguration in a leftward manner and any moving/remerger categories move leftward.Full evidence from adjuncts has been found to prove the explanatory adequacy and simplicity of PGLM. Adjuncts including adverbs, prepositional phrases and adverbial clauses, are not adjoined to categories they modify. Instead, an adjunct c-commands the modified in base generation and stays in situ unless it is topicalized or focus-topicalized. Its final position is the result of movements of movable categories.PGLM adopts Derivation by Phase proposed by Chomsky (1999) and extends his analysis to complex preposition phrases, which are analyzed as strong phases CP in PGLM. Adverbial clauses and prepositional phrase (PP) containing predicates are treated as equals in that both are headed by a preposition with an inherent property [±P], the choice of which determines whether the clause in PP has tense or not, or whether the subject DP (usually analyzed as obligatory PRO) moves out or not, as suggested in Hornstein (1999).Our study shows that adjuncts are generated in modifying positions of VP or TP and are only liable to be topicalized or focus-topicalized. Their surface positions are the results of derivation, that is, overt and covert feature checking. Evidence from adjuncts in both English and Chinese proves that adjuncts are merged in four different positions (A, B, C, D) as specifiers of iterated VP or TP, depending on what domain they modify.Post-verbal adjuncts are not adjoined to VP rightward. Instead they are merged in AdvA (Spec of VP) and occur post-verbally when V and its complement move out of base VP to have their features checked. After derivation, adjuncts merged in AdvB positions occur between light verbs and main verbs while AdvC adjuncts are between subjects and light verbs. AdvD adjuncts are sentential adverbials.Moreover, we have found evidence supporting the following conclusions:Some C-type adverbials in English have subject-oriented reading because they c-command subjects in base positions while in Chinese post-verbal adjuncts derived from A positions may have subject-oriented reading if they are headed by "de" and the subjects are raised from object positions which are c-commanded by "de" adjuncts in base generation.Adjuncts of the same property are not allowed in A and (or) B positions unless they are conjoined.更多还原