Transformation
(T) 转换
A formal linguistic operation which enables two levels of structural representation to be placed in correspondence. A transformational rule (T rule, transformation or transform) consists of a sequence of symbols which is rewritten as another sequence, according to certain convention. The ‘input’ to the rule is the structural description (‘structural analysis’ or ‘structure index’), which defines the class of phrase-markers to which the rule can apply. The rule then operates a structural change on this input, by performing one or more of several basic operations. Movement (reordering or permutation) transformations modify an input structure by reordering the elements it contains. When this operation is seen as one of moving elements to adjoining positions in a phrase-marker, it is known as adjunction. Insertion transformation add new structural elements to the input structure (as in element-copying, or the insertion of by in the passive transformation below). Deletion transformations eliminate elements from the input structure.
One of the earliest illustrations of the operation of a transformational rule was the one which converted active sentences into passive ones, which can be formulated as follows:
NP1 - Aux -V - NP2→NP2 - Aux + be + en - V - by + NP1
(where be is a form of the verb to be, and en represents the past-participle ending of the lexical verb). The rule is said to ‘operate’ on the first, underlying phrase-marker, converting it into a second, ‘derived’, phrase-marker. The string produced by the derived phrase-marker may then serve as the underlying string for further transformations, as the analysis of the sentence proceeds. The sequence of phrase-markers assigned to a sentence constitutes its transformational derivation or transformational history.
A grammar which operates using this notion is a transformational grammar (TG) or transformational generative grammar (TGG). This type of grammar was first discussed by Noam Chomsky in Syntactic Structures (1957) as an illustration of a generative device more powerful than finite-state grammars or phrase-structure grammars. In this view, very many sentence types can be economically derived by supplementing the constituent analysis rules of phrase-structure grammars with rules for transforming one sentence into another. Transformational grammars became the most influential type in the development of generative grammatical theory: indeed, the field as a whole for a time came to be variously known as ‘generative grammar’, ‘transformational-generative grammar’ (or simply ‘TG’).
Several models of transformational grammar have been presented since its first outline. The standard model, as presented by Chomsky in Aspects of the Theory of Syntax (1965), consisting of three components: (a) a syntactic component, comprising a basic set of phrase-structure rules (sometimes called the base component), which together with lexical information provides the deep-structure information about sentences, and a set of transformational rules for generating surface structures; (b) a phonological component, which converts strings of syntactic elements into pronounceable utterance; and (c) a semantic component, which provides a representation of the meaning of the lexical items to be used in the sentence. The ways in which these components should be interrelated (especially the relationships between semantics and syntax) have proved to be a source of continuing controversy, since the appearance of Aspects, and alternative models of analysis have developed (compare especially the distinction between generative and interpretive semantics).
As a result of these developments, the status and classification of transformations varied a great deal in the 1960s and 1970s. A distinction introduced early on is that between optional and obligatory transformations, and the former referring to a rule which may apply at a given stage in a derivation, the latter to a rule which must apply, if a well-formed sentence is to result. On the other hand, the classification and terminology of transformations in Syntactic Structures is different in many respects from that encountered in Aspects. In the former, two types of transformation are recognized: singulary (or single-base), where the rule operates on only one terminal string; and generalized, where the rule combines two or more terminal strings, as in conjoining and embedding transformations (which handle co-ordination and sub-ordination respectively). In Aspects, however, other distinctions are introduced, some of which replace those found in the former book. Of particular importance is a distinction drawn in one of the models outlined in Aspects between lexical and ‘non-lexical’ transformations: the former transform pre-lexical structures into deep structures containing complex symbols; the latter transform deep structures into surface structures. A further development is the much increased generality of transformations, culminating in the rule ‘move alpha’ - essentially a licence to move anything anywhere, except that the movement must be an instance of either substitution or adjunction, and must obey subjacency.
指一种形式语言学的操作，它能够在结构表征的两个平面之间建立对应关系。转换规则（或称T规则、转换）是按某种规约将一个符号序列重写为另一个序列。规则的“输入”是结构描写（“结构分析”或“结构标引”），它确定该规则适用于哪一类短语标记。这条规则然后对这一输入作某种结构变换，即实施一种或几种基本操作。移位（重新排序或换位）转换是通过对所含成分的重新排列来修改输入结构。将成分移至短语标记内附接位置的操作称作附接。插入转换是给输入结构增加新的结构成分（如成分复写，或在下述被动式转换中插入by“被”）。删除转换是从输入结构中消除成分。
有关转换规则操作方式的最早示例之一是将主动句转换成被动句，可具体阐述为：
NP1 - Aux -V - NP2→NP2 - Aux + be + en - V - by + NP1
（其中be是动词to be的一种形式，en代表词汇动词的过去分词词尾。）按理论，这条规则“作用”于第一个即底层的短语标记，将其变接成第二个即“派生”的短语标记。由该派生的短语标记产生的语符列然后又可随句子的进一步分析充作底层语符列并作进一步转换。指派给一个句子组构成分的一系列短语标记称作转换派生过程或转换历史。
用转换这一概念进行操作的语法是转换语法。这类语法最初由诺姆·乔姆斯基在《句法结构》一书中加以论述，并以此为例说明生成语法的生成力比有限状态语法或短语结构语法都要强。按这种观点，有许多类型的句子可以经济地派生而成，只要在短语结构语法的组构成分分析规则之外再补充将一个句子转换成另一个句子的规则。转换语法成为生成语法理论发展过程中最有影响的一类语法：实际上整个领域现在有称为“生成语法”的，也有称为“转换生成语法”的。
自其最初的纲要提出以来，已有好几种转换语法的模型陆续出现。标准模型以乔姆斯基的《句法理论要略》（1965）为代表，包括三个组成部分：（a）句法部分，包括一组基本的短语结构规则（有时也称作基础部分）和一组转换规则，前者与词汇信息一起提供句子的深层结构的信息，后者生成表层结构；（b）音系部分，将句法成分语符列变换成可发音的话段；（c）语义部分，提供句子将使用的各词项的意义表征式。这几个组成部分以何种方式互相联系（特别是语义学和句法的关系），实际已成为自《要略》以来不断争论的起因，也因此建立起一些其他分析模型（特别是参看生成语义学和解释语义学之间的区别）。
这些发展的结果导致60年代和70年代转换的地位和分类发生很大变化。较早引入的一种区别是强制性与非强制性转换的区别，前者指为得出合式的句子而必须应用的规则，后者指派生过程中某一阶段可以应用的规则。另一方面，《句法结构》中各种转换的分类和术语在许多方面不同于《要略》，前者区分两类转换：单一转换（或单基转换）是规则只作用于一个终端语符列；概化转换是规则将两个或多个终端语符列组合起来，如并联转换和嵌套转换（分别处理并列关系和主从关系）。而在《要略》中还提出其他一些区分，其中有的取代前书中的区分。最重要的区分是，《要略》勾画的多种模型之一区分词汇转换和“非词汇”转换：前者将前词汇结构转换成包含复杂符号的深层结构；后者将深层结构转换成表层结构。还有一种变化是转换的概括性有很大提高，最终概括为一条“移位α”规则—基本上允许任何成分移至任何位置，唯一的限制是移位必须属于替换或附接两类，并必须满足邻接条件。