Generative approaches to grammar

In this section, we present an overview of the characteristics of generative approaches to grammar. As we will see, the most prominent generative approach is that of Transformational Grammar developed by Chomsky, but there are also several other broadly generative approaches that are non-transformational, including Kay and Fillmore’s Construction Grammar, Head-driven Phrase Structure Grammar (HPSG) and Lexical Functional Grammar (LFG).

Prior to the emergence of the Chomskyan model, the prominent approach in twentieth-century American linguistics was the behaviourist approach, which viewed linguistics as the study of observable linguistic behaviour. This approach is associated with the American structuralists, such as Leonard Bloomfield (1887–1949), whose work focused upon field linguistics and characterising directly observable linguistic phenomena such as phonological and grammatical form. Bloomfield’s 1933 book Language is regarded by many linguists as a model of careful and precise linguistic description. However, this approach had little to say about unobservable phenomena such as meaning or about the mental representation of language. As we saw in Chapter 4, the behaviourist psychologist B. F. Skinner (1904–90), in his (1957) book Verbal Behaviour, outlined the behaviourist theory of language acquisition, which held that children learnt language by imitation and that language has the status of stimulus-response behaviour, conditioned by positive reinforcement (rather like Pavlov’s dog).

The generative framework has its origins in Chomsky’s (1957) book Syntactic Structures, in which he proposed – contrary to the behaviourist theory of language prevalent at that time – that human beings are predisposed for language acquisition by virtue of a designated cognitive system that later came to be known as Universal Grammar. As we saw in Chapter 4, in his (1959) review of Skinner’s book, Chomsky argued (among other things) that the behaviourist theory failed to explain how children produce utterances that they have never heard before, as well as utterances that contain errors that are not present in the language of their adult caregivers. Chomsky’s theory was the first mentalist or cognitive theory of human language, in the sense that it attempted to explore the psychological representation of language and to integrate explanations of human language with theories of human mind and cognition. For this reason, Chomsky’s early work is often described as one of the catalysts of the ‘cognitive revolution’, coinciding with the birth of cognitive science as a discipline in its own right, uniting through common goals and research questions disciplines such as philosophy, psychology, linguistics and artificial intelligence.

As we saw in Chapter 4, the generative model rests upon the hypothesis that there is a specialised and innate cognitive subsystem that represents unconscious knowledge of language, or competence. The idea that linguistic know ledge arises from ‘drawing out what is innate in the mind’ (Chomsky 1965: 51) is described by philosophers as the rationalist view, and contrasts with the empiricist view, which holds that linguistic knowledge is constructed on the basis of experience and is independent of any specialised cognitive system. Universal Grammar is the model of the initial state of the innate language faculty: in other words, the system of linguistic knowledge that all humans bring to the process of acquiring their first language. In developing this mentalist theory of language, Chomsky asserts that the only revealing object of linguistic study, given the objective of characterising competence, is the system of linguistic knowledge in the mind of the idealised individual speaker. This system of internalised linguistic knowledge is known as I-language (Chomsky 1986: 19–56). From this perspective, the externalised language of the speech community (E-language) is merely epiphenomenal, in the sense that it arises as the output of individual I-languages.

In the generative model, this innate language system is viewed as ‘encapsulated’ or modular and patterns of selective impairment, particularly when these illustrate double dissociation, are often seen as evidence for the encapsulation of such cognitive subsystems. Of course, the interpretations of such patterns are open to a range of interpretations (see the discussion of Tomasello’s review of these issues in Chapter 4). In addition, the language module itself is viewed as a modular system. In other words, the linguistic sub systems such as syntax, semantics and phonology are seen as independent sub modules within the language system. This view rests upon the premise that the principles and processes, and the primitives over which they operate, are different in kind from one area of language (for example, phonology) to another (for example, syntax). In addition, selective impairment within the language system itself is a frequent consequence of acquired left-hemisphere brain damage. For example, certain types of acquired aphasia (or language disorder) such as anomia (loss of content words) or agrammatism (loss of or damage to grammatical units and structures) appear to target different aspects of the language system. This type of selective language impairment is often interpreted as evidence for the plausibility of a model in which subtypes of linguistic knowledge are organised separately within the cognitive system as well as being localised separately within the physical brain. A simple model of the language module is shown in Figure 22.2. The levels of phonological form (PF) and logical form (LF) operate over the output of the syntax (sentence-level structures) with respect to phonological and semantic principles, respectively.

Within the generative model, as we saw in Chapter 4, the existence of a language module is held to account for the rapid acquisition of language by human infants and for the existence of linguistic universals. The system is not open to conscious introspection, nor does it correlate with any single local function in the physical brain. Rather, it is one aspect of a complex model of ‘mind’, and can only be reconstructed on the basis of its output: human language itself. For this reason, native speaker intuition and judgement play a central role in this model. While speakers can rarely explain the rules that govern their native language, they can (often) rapidly judge what is possible in the language and what is not, thereby providing a body of data on the basis of which the linguist can attempt to model the system of knowledge that underlies those judgements. The generative model consists in part of a set of principles of language: statements that account for all possible (grammatical) linguistic structures, and which also rule out impossible (ungrammatical) structures within each of the submodules. This system of principles is described as ‘generative’ because it makes explicit the underlying knowledge that gives rise to the output.

There are a number of current generative theories of language. These theories tend to focus on the directly ‘measurable’ structural aspects of language such as morphology, syntax and phonology, although some approaches (notably Jackendoff’s theory of Conceptual Semantics) attempt to integrate theories of linguistic meaning into a formal generative framework. While all generative theories assume Universal Grammar as a common working hypothesis, they differ in terms of how they model the system. For example, some theories of grammar such as Head-driven Phrase Structure Grammar (HPSG) and Lexical Functional Grammar (LFG) place the burden of explanation on information stored in the lexicon and assume only a single monostratal level of syntactic representation. Others, such as the Transformational Grammar model, place the burden of explanation on the syntax, and therefore assume a multistratal system where ‘underlying’ and ‘surface’ syntactic structures are linked by generalised derivational processes. As we saw in Chapter 19, theories that we might describe as ‘broadly generative’ can differ in significant ways.

We might describe Transformational Grammar and Kay and Fillmore’s Construction Grammar as extreme poles on a continuum of ‘broadly generative’ theories, given the substantial differences between them. Table 22.2 summarises the key characteristics of a generative (‘formal’) approach to grammar. We revisit some of these characteristics in the discussion that follows.

Transformational Grammar

 The Transformational Grammar model was first proposed by Chomsky in the late 1950s, since when it has itself undergone a number of transformations, resulting at various historical stages in models known as Transformational Generative Grammar, Standard Theory, Extended Standard Theory, Revised Extended Standard Theory, Government and Binding Theory, Principles and Parameters Theory and, most recently, the Minimalist Program. The transformational model is not only the most prominent generative model but is also the model against which the cognitive approach to grammar defined itself in the early stages of its development. For this reason, the terms ‘generative model’ and ‘formal model’ have largely been equivalent to ‘transformational model’ for the purposes of our discussion in this book. As we have seen, however, the transformational model is not the only generative model, and generative models are not the only models of language that rely upon a significant amount of formalism.

Within the transformational framework, lexical items are stored in the lexicon together with information about their phonological, semantic and core syntactic properties (such as word class and valence requirements, for example). As a result of its interaction with generalised syntactic principles, this information gives rise to ‘deep structures’: syntactic structures in which the core requirements of the lexical items are satisfied in accordance with the syntactic principles. Deep structures typically correspond to unmarked active declarative sentences, the clause type that is traditionally viewed as the ‘canonical’ or ‘basic’ syntactic structure within any given language. ‘Non-canonical’ clause types such as passives and interrogatives – where these involve syntactic reordering – are then derived by means of syntactic ‘movement’ or ‘transformation’ and give rise to ‘surface structures’. As a simple illustration of these ideas, consider the relationship between the declarative clause in (1a) and the interrogative clause in (1b):

(1) a. Lily has met another friend.

      b. Has Lily met another friend?

In the transformational model, the declarative structure in (1a) corresponds to the ‘deep structure’. If the speaker intends to make a statement, no transformation is necessary and this deep structure is equivalent to the surface structure that the speaker actually produces. However, if the speaker intends to ask a question, the interrogative structure in (1b) is derived from the deep structure in (1a) by a syntactic transformation that raises the auxiliary verb has to a position in front of the subject she. This transformation is illustrated by the tree diagram in Figure 22.3, which shows how the auxiliary verb raises to a clause-initial position created by the transformation. We return to discuss the status of tree diagrams in the generative model in more detail below.

A version of the model in Figure 22.2 that incorporates a transformational syntax is shown in Figure 22.4. This corresponds to the model of transformational syntax assumed most recently within Government and Binding Theory and within Principles and Parameters Theory.

Within the syntactic component, there are phrase structure rules that build syntactic structures. X-bar syntax is one approach to the statement of generalised phrase structure rules that was introduced into the transformational model during the 1970s, and a version of this approach remains in the current model. The X-bar model replaces category-specific phrase structure rules (separate sets of rules for building NPs, VPs and so on) with a small set of category neutral rules, where hierarchical (head-dependent) relationships are universal but linear precedence (word order) relations are subject to cross linguistic variation. The existence of a small set of category-neutral rules within Universal Grammar is motivated on the basis of economy of representation: a small set of category-neutral rules eliminates redundancy and thus accounts for the efficiency of the language system both in terms of how it is acquired and in terms of how it underlies language use. A small set of category-neutral rules is also motivated on the basis of learnability: the fewer the rules, the more rapidly the child will fully acquire the grammatical system of his or her native language. The tree diagram in Figure 22.5 represents the structure that is built by X-bar rules, where X0 is the head of a phrase and XPits phrasal level. An important constraint on this structure is that it is (maximally) binary branching. This constraint is also motivated on the basis of learnability: the fewer the structures the grammar can build, the more rapidly the child can fully acquire the system of his or her native language.

In the X-bar model, X is a variable that can be instantiated by any word class. For example, if X is a noun, XP is a noun phrase; if X is a verb, XP is a verb phrase, and so on. The structure in Figure 22.5 is used to model the relation ships between heads and dependents. As we have seen, specifiers, complements and modifiers are types of dependent. In principle, the phrase is limited to a single specifier (e.g. the determiner in a noun phrase), head and complement, but may contain an unlimited number of modifiers. Of course, the existence of ditransitive verbs has proven a challenge to this highly constrained syntactic model. It is important to point out that certain parts of this structure are ‘optional’ in the sense that not every phrase will contain some, all or any dependents, and some phrases will contain more that one adjunct (or modifier). The minimal requirement for a phrase is the head.

An important development within the transformational framework was the extension of the X-bar structure from content phrases such as the noun phrase (NP) and the verb phrase (VP) to grammatical units such as the determiner phrase (DP) and the clause or ‘tense phrase’ (TP). This means that the same basic X-bar structure is used to model clauses as well as phrases; indeed, the extension of the X-bar model to a range of functional categories was one of the defining features of the Principles and Parameters framework. According to this model, the universal properties of human language are attributable to the shared principles of Universal Grammar, while cross-linguistic variation relates to ‘parameter setting’: the typological characteristics of each language arise from ‘options’ within a set of well-defined parameters of variation. Since the early 1990s, Chomsky has proposed some radical changes to the transformational model, which together constitute the basis for the ongoing research framework known as the Minimalist Program. Figure 22.6 represents the Minimalist model of the grammar.

An important difference between the Minimalist model and the model assumed within the Principles and Parameters framework concerns the elimination of ‘deep structure’ and ‘surface structure’ as distinct levels of syn tactic representation. Instead, a single syntactic component described as the ‘computational system’ derives syntactic structures from sets of lexical items (including both lexical and functional categories) and maps these structures onto two distinct ‘interface’ levels: the phonological level (PF), which inter faces with the articulatory-perceptual performance system (phonology), and the semantic level (LF), which interfaces with the conceptual-intentional performance system (meaning). In principle, the lexical items themselves, consisting of phonological, semantic and formal features, encode all the information required for the derivation, so that principles operating over the derivation remain maximally simple and general.

 Indeed, according to Chomsky (2000b) there are only two basic operations occurring within the computational system: Merge and Agree. Merge is a basic structure-building operation that is driven largely by the lexical properties of the predicational item(s) within the set of lexical items. This operation assembles phrase markers (tree structures) from pairs of syntactic objects, beginning with the head-complement structure, then merging the resulting structure with its specifier, and finally combining the resulting phrase markers into larger structures. The second operation Agree matches the morphosyntactic features of two elements within the structure. This process involves features like cate gory selection features, phi-features (person, number, gender), case, tense aspect, and interrogative or wh-features. The ‘matching’ of these features has to take place within a local configuration, and it is this requirement that motivates syntactic transformations. Consider the examples in (2) by way of illustration.

In example (2a), the embedded clause selected by the verb ask has an interrogative feature because of the semantics of ask. This explains why it takes a complementiser with a wh-feature (whether as opposed to that). The transformational model assumes that wh-expressions like what in (2b) have a wh-feature that needs to be locally ‘matched’ with a functional head (complementiser) in the left periphery of the clause. This explains why the wh-expression, which ‘originates’ as the object of want, raises to clause-initial position. Because the wh-feature only needs to be spelled out or made explicit by one unit in the clause (the fronted wh-expression), the wh-complementiser remains implicit in (2b), which explains why the sentence *George asked what whether Lily really wanted is ungrammatical in English, although other wh-fronting languages allow both components to be spelled out. This is an example of parametric variation. This analysis is the transformational equivalent of the filler-gap analysis of wh-dependencies that we saw in our discussion of Construction Grammar in Chapter 19.

The transformational model is open to a range of interpretations. At one extreme, it can be interpreted as underlying a literal step-by-step process of (unconscious) sentence construction in the mind of the speaker, where syntactic trees are planted and pruned during the processing of each sentence uttered. There is little evidence to support this view: indeed, the transformational model is not intended as a model of language processing, but as a model of linguistic knowledge that interfaces with performance (production and comprehension) systems. At the other extreme, the syntactic transformation can be viewed as a metaphor that attempts to capture similarities between related constructions both within and between languages, and attempts to model the invisible and mysterious aspects of human cognition that underlie those similarities.

 It follows that both the generative approach and the cognitive approach are ‘cognitive’ in the sense that they seek to model the psychological representation of language. However, the two frameworks approach this in radically different ways, as we have seen. While the formal model views language as an innate, encapsulated and computational system, the cognitive model views language as an emergent system, inextricably linked with general processes of communication and conceptualisation, with meaning at its core. We look in more detail at the key differences between the two models in the next section.

اخر الاخبار

اخبار العتبة العباسية المقدسة

أكاديمية التطوير الإداري تقيم ورشة عن منهجية الإشراف الأكاديمي الفعّال على بحوث الطلبة

قسم الشؤونِ الخدمية يوفر ثلاثة آلاف صندوق أمانات لزائري مرقد أبي الفضل العباس (عليه السلام)

خلال عيد الفطر المبارك.. قسم الشؤون الدينية يضاعف جهوده للإجابة عن الاستفتاءات الشرعية للزائرين

قسم الشؤون الفكرية يقيم ندوةً ثقافية حول فوائد تشريع العيد في ذي قار

المزيد

الآخبار الصحية

كوب قبل النوم.. مشروب طبيعي يحارب الأرق

دون أدوية.. طريقة طبيعية لخفض ضغط الدم بفعالية مذهلة

المعكرونة ليست عدوك.. كيف تصبح جزءا من وجبة صحية متوازنة؟

دراسة: تناول القهوة يوميا يخفض خطر الإصابة بالاضطرابات النفسية

المزيد

الآخبار التكنلوجية

علماء روس يخططون لبناء منشأة للكشف عن المادة المظلمة

ثاني أكسيد الكربون ليس تهديدا للمناخ فقط.. بل لصحة الإنسان أيضا!

ديدان الأرض تكشف تلوث التربة بالمعادن الثقيلة

خبير أرصاد جوية: ظاهرة النينيو القادمة قد تكون الأقوى خلال 10 سنوات

المزيد

مواضيع ذات صلة

Loss of recursion

Cognitive approaches to grammar

Theories of grammar: assumptions, objectives, methodology

Cognitive approaches to grammar in context

Comparison of the three approaches: be going to

Case study: the evolution of auxiliaries from verbs of motion or posture

Animal cognition

On the rise of recursion Discussion

Are there languages without recursion?

Treatment of recursion in linguistic description

Embedding, iteration, and succession

Simple vs. productive recursion