Читать книгу A Companion to Chomsky - Группа авторов - Страница 32

As the principles of GB were increasingly refined, the theory also became more complex. In lectures towards the end of the 1980s, Chomsky started asking whether the theoretical model could be rationalized in the sense of unifying, including eliminating, principles while maintaining, or possibly improving, empirical coverage (cf. Chomsky 2000, p. 93; Freidin and Vergnaud 2001, p. 642). This became the beginning of the Minimalist Program. Chomsky (1993) is the first paper that is written from an explicit Minimalist perspective and one of the cases discussed there provides a helpful illustration of the underlying logic. Example (18) deals with case, which is not very conspicuous in English, but which can be seen in pronouns.

1 (18)Sara/John likes him/her.She/he likes him/her.Sara proved [him right]

As (18a) and (18b) show, pronoun subjects occur in nominative case and pronoun objects occur in accusative case, which of course are relics of the full‐fledged morphological case system in older varieties of English. (18c) shows that accusative case can also occur in other environments. This raises the interesting question of how case assignment works – what are the configurations in which case licensing occurs?⁹ In GB, the structural representations underlying the bolded items in (18) are all different, though attempts had been made to use one principle (government) to account for all of them (Uriageka 1988, 2012). Before we can consider the argument, we need to do a brief detour to introduce some core concepts of phrase structure (see section 3.6.2 below, and Lasnik and Lohndal 2013 for a much more detailed treatment, including the historical development of phrase structure within generative grammar).

Consider the simplified structure in (19), where we are not illustrating the inner structure of the nominal phrase (called DP; see immediately below).

1 (19)

This is an instance of the X‐bar schema (Chomsky 1970, 1986, Stowell 1981), a structural blueprint for syntactic structure. The schema contains a unified structure for each phrase: The phrase has a head (i.e. the phrase is endocentric), it has a complement as the sister of the head, and then a specifier is the “subject” of the combined unit of head and complement. (19) has three phrases: the verb phrase (VP), the tense/agreement phrase (TP), and the determiner phrase (DP). If we apply this to VP, that means that V is the head of the verb phrase, him is the complement of the verb, and she is the specifier of the verb. The subject then moves to the canonical subject position in English, which is called the specifier position of T, or SpecTP for short (strikethrough marks that the constituent is not pronounced).¹⁰ The tense and agreement morphology starts out as the head of TP and then gets unified in the morphology with the verb stem like. Now we are ready to go back to the problem posed by the data in (18).

In (19), she is in the specifier position of T, which is the locus of nominative case. Rather differently, him is the complement of the verb, which is to say that accusative case is licensed in that position. As for (18c), this is an exceptional case marking construction, since him is licensed in the specifier position of an embedded constituent. These environments for licensing case all look very different. Instead, Chomsky (1993) proposes that case licensing occurs in one type of configuration called a specifier‐head configuration. Just like the subject position seems to be an agreement configuration (as subject–verb agreement shows), Chomsky proposes that there is a similar agreement configuration for the object, except that this agreement is not visible in languages like English. This makes it possible to say that case licensing of the accusative is very similar in (18a) and (18c) despite surface differences suggesting the opposite conclusion. That is, at some point in the derivation the object and verb must stand in a specifier‐head relation so that Case can be assigned. Different variants of the view have proposed different points in the derivation where this would be achieved.¹¹

As Lohndal and Uriagereka (2014, 510) point out, this example makes the logic of the Minimalist Program clear: (i) Assume that the basic theoretical and empirical postulates of GB Case theory are correct: lexical items get case, certain syntactic heads carry the ability to provide case to a lexical item, and case licensing occurs in certain positions. (ii) The domains in which case licensing occurs are very different. (iii) A new theory offers a novel take on what a case licensing configuration is: specifier‐head configurations, which are independently needed to account for subject–verb agreement in languages like English. Put differently, you pick a domain which is fairly well understood, you question some of the core parts of the analysis, and then you seek to develop a new and more principled and economic analysis (true to the Galilean ideal that has shaped much of Chomsky's work; see Allott, Lohndal & Rey, Chapter 33 this volume, for further discussion).

Another area that Chomsky (1993) targeted concerned larger architectural aspects, namely the components of the grammar. In GB, the grammatical architecture had the structure depicted in (17). This architecture contains a certain amount of overlap between D‐structure and LF on the one hand, and S‐structure and PF on the other hand (see Hornstein, Nunes, and Grohmann 2005 for much more discussion). D‐ and S‐structure are clearly more grammar‐internal than PF and LF, given that each sentence needs to receive a semantic encoding and a sound (or sign, as in the case of sign languages) encoding. Chomsky (1993) sets out to investigate whether these grammar‐internal levels in the derivation could be eliminated. He reasons as follows:

Each derivation determines a linguistic expression, an SD [Structural Description], which contains a pair (π, λ) meeting the interface conditions. Ideally, that would be the end of the story: each linguistic expression is an optimal realization of interface conditions expressed in elementary terms (chain link, local X‐bar‐theoretic relations), a pair (π, λ) satisfying these conditions and generated in the most economical way. Any additional structure or assumptions require empirical justification (Chomsky 1993, p. 4).

And since this paper, D‐structure and S‐structure have no longer been part of the grammatical component. Rather, the interfaces, sound and meaning, are regarded as the essential components, as Chomsky (2000, p. 91) argues:

On these assumptions we understand L[anguage] to be a device that generates expressions Exp = <Phon, Sem>, where Phon provides the “instructions” for sensorimotor systems and Sem the “instructions” for systems of thought – information about sound and meaning, respectively, where “sound” and “meaning” are understood in internalist terms, “externalizable” for language use by the performance systems.

The emphasis on interfaces raises many new questions: What is the division of labor between the core syntax and the interfaces? How are the interfaces actually structured (see Uriagereka 2008, Samuels 2012, Lohndal 2014, among several, on the latter)?

Many other examples of minimalist reasoning could be provided as well. An excellent textbook exposition can be found in Hornstein, Nunes, and Grohmann (2005), and we refer the reader to this for more in depth discussion and a rich array of examples.

The minimalist reasoning is in part an application of Occam's Razor, what Martin and Uriagereka (2000) have labeled “methodological minimalism”:

What one might call a “weak minimalist thesis” is nothing new. The drive for simple and nonredundant theories of the world (or Occam's razor) is taken for granted in the core sciences. Even within the more specialized science of linguistics, this working methodology has brought undeniable success. From such a perspective, minimalism is just a new way to refer to what many people have been doing for a long time: seeking the best way to theorize about a particular domain of inquiry. We think of this thesis as methodological minimalism. (Martin and Uriagereka 2000, p. 1)

However, minimalism can also be viewed in a different sense, what Martin and Uriagereka (2000) label “ontological minimalism.” As Chomsky (2000, 92) points out, progress has made it “possible to consider some new questions about [the] F[aculty of]L[anguage]. In particular, we may ask the question, how well is FL designed?” The idea is that language design may be optimal, “approaching a “perfect solution” to minimal design specifications” (Chomsky 2000, 93). The latter is known as the Strong Minimalist Thesis (SMT). Chomsky is the first to stress that it would be surprising if the conclusion holds true, yet it would carry significant implications if true. As such, the reduction and unification that characterize minimalism are also motivated by the SMT. It is not just a question of reducing or dispensing with grammar‐internal levels, but with that comes the gain that the architecture only encompasses the virtually conceptually necessary interfaces, those that a theory of I‐language cannot do without. This is all motivated by the quest to understand why certain principles of the language faculty hold and not others (Chomsky 2000, p. 92; Chomsky 2004). This type of why‐question is characteristic of the Minimalist Program, and Chomsky (2000) emphasizes that they are extraordinarily difficult to answer:

Questions of this kind are not often studied and might not be appropriate at the current level of understanding, which is, after all, still quite thin in a young and rapidly changing approach to the study of a central component of the human brain, perhaps the most complex object in the world, and not well understood beyond its most elementary properties (Chomsky 2000, p. 93).

As a consequence of this quest, other properties of the mind/brain have become more prominent (cf. Boeckx and Uriagereka 2006, p. 542). Chomsky (2005) emphasizes that there are in principle three factors that, in some combination, determine the adult language system: “Assuming that the faculty of language has the general properties of other biological systems, we should, therefore, be seeking three factors that enter into growth of language in the individual” (Chomsky 2005, p. 6). These three factors are provided in (20).

1 (20)Genetic endowment specific to languageExperiencePrinciples not specific to the faculty of language

The first part is what is often referred to as Universal Grammar (UG): it is whatever the genetic endowment is for acquisition of human language to take place. Furthermore, it is obvious that experience plays a role, given that a child growing up in Beijing acquires Mandarin Chinese and a child growing up in Athens acquires Greek. The last factor encompasses principles that are not specific to Universal Grammar. Chomsky argues that principles of data analysis may be among such principles, and also:

[…] principles of structural architecture and developmental constraints that enter into canalization, organic form, and action over a wide range, including principles of efficient computation, which would be expected to be of particular significance for computational systems such as language (Chomsky 2005, p. 6).

As Lohndal and Uriagereka (2017, p. 117) point out, these factors are very similar to the three factors that hold for organisms more generally (cf. Gould 2002, p. 259 on the “adaptive triangle”). That said, it is not trivial to develop specific proposals for what the third factors actually are; see Lohndal and Uriagereka (2017) for comprehensive discussion. However, this focus on third factors has also led to a reduced emphasis on the role of Universal Grammar, hence, for example, Chomsky (2007) has the title “approaching UG from below.” This is a marked shift compared to GB, where concerns about the size of Universal Grammar were nonexistent. If a property is innate, it does not have to be somehow learned from the environment, so a big UG is compatible with taking arguments from the poverty of the stimulus seriously (see Crain, Giblin and Thornton, Chapter 29 of this volume). In recent minimalism, the goal is to reach explanatory adequacy while attributing as little as possible to UG. Another reason for this shift is the concern for biological and evolutionary adequacy, which characterizes minimalism. It is hard to see how all the separate parts (and their interrelationships) of the GB system could or would have evolved. How did the uniquely human ability for language arise in the species? The evolutionary literature is by now very rich and multiple proposals have been advanced (for reviews, see Fitch 2010; Reboul, Chapter 30 of this volume). Even though minimalism does not offer a single answer to the question, the question itself has shaped a lot of research within minimalism, e.g. the increased emphasis on third factor effects.

The architecture of the grammar in minimalism is illustrated in (21).

1 (21)

Let us consider this computational system in a bit more detail.

There is a lexicon, just as in GB and previous approaches. Based on whatever is in the lexicon, there is one basic mechanism that creates structure: Merge puts two items together, say, α and β, and that generates the set {α, β}, which typically then has α or β as its label: {α, {α, β} or {β, {α, β}}(Chomsky 1995). This, then, is the way to implement labeling in terms of (unordered) sets. More traditional tree structure representations are provided in (22).

1 (22)

Further items can be added and merged with α or β to generate binary‐branching structures. This accounts for the creative linguistic ability that all humans possess. Merge comes in two different “flavors”: external merge, which is the first time an item is merged into a structure, and internal merge, which means that, say, α is merged again later in the structure (hence accounting for displacement phenomena) (Chomsky 2004; see Svenonius, Chapter 9 of this volume, for extensive discussion). Importantly, both ways of applying merge are manifestations of the same operation. The Extension Condition is a principle that holds that all trees are extended at the top, meaning that new lexical items from the lexicon have to be merged on top of the existing structure, they cannot be “tucked into” existing structure (Chomsky 1995).

In addition to merge, there is at least the operation Agree, which is an abstract mechanism of dependency‐formation. More specifically, the role of Agree is to ensure that syntactic features are checked or licensed. This is important in order to achieve descriptive adequacy, notably when it comes to constraining the grammar, since we know that dependency‐formation is heavily constrained across languages. A simple example of dependency‐formation when it comes to features can be seen in English subject–verb agreement as in (23).

1 (23)They are happy.She is happy.I am happy.

Subject–verb agreement is the morphological expression of an abstract syntactic relationship between the verb and the subject, since, as (23) illustrates, they have to agree in the features [NUMBER] (singular and plural) and [PERSON] (1, 2, 3). Agree is the mechanism that ensures that, say, the features of they match the features on the verb (see Adger 2003 for a textbook illustration). Notably, it is assumed that pronouns (and other nominals) carry number and person features that are interpreted at the C‐I interface, whereas similar features on verbs are not interpreted by the C‐I interface. For that reason, such features are uninterpretable, even though they can be morphologically realized, as in the case of English subject–verb agreement.

As structures are generated, they are at various points transferred to the two interfaces: the sensorimotor interface (“sound”) and conceptual‐intentional interface (“meaning”). That is, pieces of syntactic structure (called “phases” within minimalism) are transferred to the interfaces and there given the necessary interpretation at both interfaces. Space does not allow us to elaborate on phases, but see Gallego (2012) for an extensive presentation.

The Minimalist Program is a program, meaning that it comes in different guises. As Chomsky (2000, 92) puts it: “[t]here are minimalist questions, but no minimalist answers, apart from those found in pursuing the program, perhaps that makes no sense, or that it makes sense but is premature.” Chomsky has never provided a more theoretical foundation of the notion of a program, but Boeckx (2006) does so by pointing to work by Lakatos (1968, 1970) and others. Lakatos argues that research programs have a core, a set of principles that the scholar has to adhere to. In addition, auxiliary assumptions are made, which often vary from one timeslice of the theory to another. The latter are often oversimplifications that the researcher knows are not true, but yet are made in order to be able to provide sufficiently detailed analyses of a set of phenomena. Because of this, Lakatos argues that the rigor of a research program may not at first be great, and that maturation is often slow. Relevantly for some of the criticism leveled against generative work, namely that there are so many counterexamples to proposed principles or generalizations, Lakatos has a specific view on the role of falsification: “A corroborated falsifying hypothesis does not have sufficient power to enable a counterexample to eliminate a theory. If it had, we would eliminate all science instantly” (Lakatos 1968, 163). And he continues:

A counterexample, in order to reject, to eliminate a theory, needs more powerful support than that which a lower‐level falsifying hypothesis can provide: it needs the support of a theory with more corroborated content, with wide explanatory power. There must be no elimination without the acceptance of a better theory (Lakatos 1968, 163; his italics).

In this regard, Lakatos differs from Popper: “[…] in my conception criticism does not – and must not – kill as fast as Popper imagined. Purely negative, destructive criticism, like “refutation” or demonstration of an inconsistency does not eliminate a programme” (Lakatos 1968, 183; his italics). This conception fits very well with how research within the Minimalist Program has taken place: It has been exploratory and continued despite obvious empirical or theoretical issues that needed to be fixed.

Boeckx (2006) and others argue that this programmatic nature of the Minimalist Program is a good thing. It allows researchers to pursue a multitude of approaches within the larger umbrella of minimalism. Put differently, it is possible to assume lexicalism (e.g. Chomsky 1995: chapter 4, Williams 2007), distributed morphology (Embick and Noyer 2007, Embick 2015), cartography (Rizzi 1997, Shlonsky 2010), or an exo‐skeletal approach (Borer 2005a, b, 2013, Lohndal to appear). Within each version, a substantial body of work has emerged and the versions are developing fairly independently of each other. At the same time, it makes it hard to pin down exact criteria that can be used to label a particular scholar a minimalist. It also differs to what extent a particular scholar considers his or her work minimalist in nature, in particular in the ontological sense identified by Martin and Uriagereka (2000). However, it should be pointed out that some argue that minimalism should start producing what they label “theories,” here summarized by Lohndal and Uriagereka (2014, 520):

Some scholars have argued that it is time to produce theories that can be evaluated and falsified (see, e.g. Hornstein 2009). Soon after the [minimalist program] was initiated, Lasnik (1999: p. 6) stated that “there is not yet anything close to a Minimalist theory of language.” Differences in the underlying philosophy of science will in part determine whether a scholar regards programs as sufficient or whether theories are required for genuine progress (Lohndal and Uriagereka 2014, p. 520).

This raises obvious and thorny questions about what the difference between a program and a theory may be, issues that we cannot go into here.

That concludes our fairly brief overview of the general nature of the Minimalist Program. In the next section, we continue our discussion of this program by turning to some current tendencies.