人物专栏 | Noam Chomsky教授访谈

点击上方蓝字关注我们

编者按

《理论语言学五道口站》（2022年第12期，总第215期）“人物专栏”与大家分享近期Idan Landau教授对Noam Chomsky教授的访谈，访谈以“Language Design”为主题，由本站成员赵欣宇、雷晨、聂简荻、郭思源、丁子意翻译。Noam Chomsky教授，是享誉世界的语言学家、认知科学家、政治批评家、哲学家和历史学家。Idan Landau教授，现任职于以色列本·古里安大学外国文学与语言学系。

本期译介的是Idan Landau教授于2020年7月对Noam Chomsky教授所做的访谈。此次访谈共分为五部分，各个部分采访主题不同，本期译介的是该访谈的第四部分，主题为“语言设计”。在访谈中，Noam Chomsky教授首先表达了自己对语言设计目的的理解，然后从多个方面对核心语言设计进行了深入分析，最后阐述了语言与思维的关系。访谈内容转自网站：CHOMSKY.INFO。

访谈内容

01.

Idan Landau教授：在您的文章及《我们究竟是什么样的生物？》（What Kind of Creatures Are We?, WKC）这本书中多次提到，“语言设计的基本属性”表明语言本质上是一种思维工具，而不是交流工具。那么，我们应该如何看待语言设计中所含有的传达交际意义的特征呢？首先，“话题-焦点”、“主位-述位”或“预设-断言”等概念都预设了一种交际情况，在这种情况下，已知信息应该通过某些语法层面的手段与未知信息区分开来。这些概念具有普遍性，任何语言都有用来标记不同特征的语法手段（如移位、语调或形态变化等）。但就我个人而言，我想不出证明其合理性的方法。另外，许多语言中带有“言据性标记”，用于表明说话人话语的确定性或信息来源，从这一点来看，如果语言设计忽略交流，言据性标记将会变得令人费解。我了解到您区分了“设计”和“使用”，但这种区分并不明显，因此我们需要一些标准来说明为什么这些语法现象（或者其他现象，如事实性、索引性等）属于“使用”而不是“设计”。

Noam Chomsky教授：这些问题十分有趣。我想到了一个简单的答案，但大家仍需进行全面系统的研究。

答案要从合并与前文你所列举的与交际相关的例子讲起。合并是最优的语言运算方法，也是“语言设计”的基础，同运算有效性的一般性原则（即第三要素所涉及的原则）一样，都可以自由地进行操作。如果把合并从语言中移除，语言将不复存在。而从语言中去掉与交际相关的手段则是无关紧要的。这种质的差异是相当大的，证明了语言是一个基于合并的系统，且正如《我们究竟是什么样的生物？》（WKC） 一书中提到的，语言是一个思维系统，而交流是其用途之一。

但我认为应该从以下几个方面对这个问题进行深入探讨。

“语言设计”是一个宽泛的概念。所有人都认为语言是不能被设计的。尽管如此，我仍然认为要对语言（语言设计）的基本原则与语言所具有的特殊性和普遍性进行有效区分。要想解决这一问题，就要将语言学理论提出的目的考虑在内。

提出任何理论的目的都是为了对该领域进行合理解释。就语言而言，我们需要探索语言普遍性。至少在基于WKC及早期的生物语言学研究中，合理的解释需要满足学习条件与进化条件。自从“生成语法”创立以来，学习条件一直占据指导性地位，我们很早就认识到了语言习得刺激贫乏问题的严重性。同时要感谢Charles Yang教授关于数据贫乏的统计学研究及对低龄儿童所拥有知识的大量研究，让我们意识到这一问题远比我们想象的还要严重。另外，进化条件一直被束之高阁，但正如我在WKC中所说，它现在应该被重新纳入讨论范围。

语言理论中的最佳解释与SMT（强式最简主义）理论分不开，SMT提出的语言属性是由最优运算（合并）和“第三要素”（比如最小运算）所决定的。理想情况下，这些独立于语言的原则可以看作是自然法则。只有尽可能地接近SMT，才能获得关于语言的真正解释（具有进化先天性，因而才能具有学习条件）。

我们可以认为核心语言设计能获得真正的解释，满足SMT。某种程度上，这一概念可以扩展为有更多限制条件的解释。除非有理论要求，否则我们没有必要为像普遍语言设计这样的术语划定一个明确的界限。我们接下来讨论一下核心语言设计的问题。

WKC 的核心部分是书的第一章，因为第一章中回顾了与语言“真正的解释”相关的研究成果，于我而言，这些成果十分重要，在语言和认知系统的研究中是新突破；然而事实上，“解释”还不能进行公式化。最优运算（包括以合并为核心的组合操作）为语言的一般架构提供了一个基本解释，有时称之为语言的“基本属性”：狭义句法在概念-意向（C-I）接口处生成代表项，被解释为思维；外化作为一种辅助现象，将句法结构映射到感知-运动接口，通常是以声音的形式出现。

第一章所提及的基本假设也为语言的其它基本特征提供了真正的解释。以前我一直认为，位移的普遍属性（内部合并（IM））是语言的“不完美”之处，但这其实是语言的基本情况，从各种规则，到重构、复杂的语义结果，再到依存结构特性等等，这些都会阻碍内部合并。然而这同时也意味着儿童会本能地忽略他们所听到的所有线性顺序内容，只关注他们从未听过的内容以及他们大脑中所构建的抽象结构。

还有许多我没有涉及的内容，例如，为什么要使用更复杂的外部合并（EM）操作；内部合并(IM)可能、必须或不适用的条件是什么；语言在外化中的多样性和可变性；参数和可行性等等。

其中一个结果有力地支持了传统观点，即语言设计是一种思维系统，而外化是一个辅助过程，依赖于外化的语言交流仍然不如语言设计重要。

这不但不足为奇，反而都在意料之中。真正令人惊讶的是，这些想法到了现代才再次引人注目，而且现在看来，这些想法本身也令人震惊，这一点从人们的反应中就可见一斑。但外化毕竟是语言本体系统和感知运动系统的融合，通常我们称之为言语。这两种系统在进化史上毫无关联，并有着截然不同的特性。因此，严格来讲，外化不是语言话语部分的全部。这其实一点儿也不奇怪，因为融合过程还有着诸如语⾔的复杂性、语⾔的易变性等其他成分。

语言设计方面取得的一些成果为我们的预测提供了实证支持，此外，对沟通效率和计算效率之间冲突的研究也进一步证明了这一点（计算效率有“第三因素”的特性，因此属于“强式最简主义”）。在所有已知案例中，沟通效率往往被舍弃。神经语言学和心理语言学研究为此提供了支撑，书的第一章中讨论了这一问题。

毫无疑问，语言是用来交流的，虽然WKC中统计的结果显示并非如此。从上述分析中我们只能得到一种可能：一些语言要素主要用于交流，但它们不可能成为语言设计的核心部分。例如：祈使句主要用于社会交际的语境，语言设计也允许如此，然而它在真正的实质性解释中却不起什么作用，你列举的大多数实例也是这样。语言设计为此提供了一些选择（在一些情况下，比如主位-述位理论中，语言设计甚至可能提供部分解释），但实际上还是没有做出真正的解释，而且也不会参与到解释中去。如前所述，如果把你列举的交际实例从语言中剔除，并不会对语言造成太大影响，这就好比从语言中去除颜色词一样，无足轻重。

在我看来，与交际相关的手段也可以用于“内心对话”，语言大部分用途都是进行“内心对话”；而在思维方面，尽管如你所说，其自然场所是社会交际，但从更广的范围看，其实还是交流。

我并不认为这是一种两难境地。

我们可能听说过，语言以交际为功能，从动物的交流系统进化而来，学界也普遍认同这一学说。其中，第二条是经验主义假设，缺乏证据支持，且面临有力反证，而第一条几乎没有事实能够佐证，原因已在WKC中讨论过。反对者的说法是语言以思维为功能。除非把语言的设计解释为一种思维系统，否则即使这一说法存在事实依据，也一定为数不多。在我看来，到目前为止这一提法是实质性的，证据充分且意义重大。

这些思考引出了一个我一直避而不谈但耐人寻味的问题。诚然，外化涉及两个不同系统的交融：语言本体系统和感知运动系统。但是，狭义句法所产生的思维表征也是一个外部系统，这难道不是相同的情况吗？或者说它真的是一个外部系统吗？这一问题会引发多个亟需重视而又难以阐明的议题。

传统观点认为语言和思维是密切相关的：对威廉·冯·洪堡（Wilhelm von Humboldt）来说，二者是完全相同的；在威廉·德怀特·惠特尼（William Dwight Whitney）看来，语言是“思维听得到的信号”；而作为深受结构主义/行为主义潮流冲击的传统主义最后一人，奥托·叶斯泊森（Otto Jespersen）认为揭示“所有语言的语法规则背后的普遍原则”将会引发“对人类语言和人类思维最深层本质更深刻的洞察”（引用自WKC）。

思维并没有明确的定义，我们只知道它是由语言构成的。这就是为什么艾伦·图灵（Alan Turing）在他那篇关于机器智能的人工智能领域开山之作中写道，机器是否能思考的问题“没有任何意义，不值得讨论”。C-I接口在一定程度上构成了外部系统（洪堡则认为该接口完全是外部系统）。而感知-运动接口却不然，它完全独立于语言，这就是另一个关键的非对称性所在。

我略过了很多材料（尤其是一般语法中被称为“形式语义学”的部分，这是一个很大的话题）。当然，关于这些话题还有很多内容要讲，但上述内容应该足以答复提出的问题了。

English Version

01.

Prof. Idan Landau: A recurrent theme in your writings, and also in this book, is that “fundamental properties of language design” indicate that it is essentially an instrument of thought—not of communication. How then should we think of features of this design whose sole purpose seem to be communicative? First and foremost—topic-focus articulation, theme-rheme or presupposition-assertion, all variations on the same idea. These concepts presuppose a communicative situation where shared assumptions should be distinguished from unshared ones by some grammatical means; they are universal, in the sense that any language has some grammatical means of marking the distinction (by displacement, prosody, or morphology). Yet I can't think of a way of rationalizing them from a strictly private point of view. One also wonders about evidential markers in many languages, which indicate the level of certainty or source of information for one's utterance—again, a puzzling feature if language design ignores communication. Now, I know you distinguish between “design” and “use”, but that distinction isn't self-evident, and we need criteria to tell why the mechanisms underlying these grammatical phenomena (and others, like factivity, indexicality, etc.) fall under “use” rather than “design”.

Prof. Noam Chomsky: These are interesting questions, which have never been carefully discussed. A simple answer comes to mind, but it may be worth looking into it more fully.

The simple answer derives from comparing Merge with the set S of examples of the kind you mention. Merge is the optimal combinatorial operation and the basis for what I've been calling “language design”—along with general (“third factor”) principles of computational efficiency, which come free. Remove Merge from language and there’s nothing left. Remove S from language and nothing changes—some options aren’t used, which happens all the time. The qualitative difference is vast, justifying the conclusion that language is designed as a Merge-based system fundamentally—and it follows, as discussed in WKC, as a system of thought, with communication one of its uses.

But let’s look at the issues more carefully. I think we can approach them along the following lines.

“Language design” is a loose descriptive term. None of us believes that languages are designed. Nonetheless, I think we can make a useful distinction between fundamental principles of language (language design) and properties that languages happen to have, some universally. A principled way to approach this question is to begin by considering the goals of linguistic theory in the first place.

The goals of any theory are to find genuine explanations for significant properties of the domain under investigation. For language, that means genuine explanations for significant universal properties of language. A genuine explanation, at least within the biolinguistics program assumed in WKC and earlier work, has to satisfy the conditions of learnability and evolvability. The former has been a guiding concern since the beginning of the “generative enterprise”, hence the focus on poverty of stimulus, understood from the earliest days to be a serious problem and now known to be far more severe than had been supposed thanks to Charles Yang’s statistical studies of sparsity of data and extensive research into what is known by very young children. The latter has been put on the shelf until recently but is now I think within the range of discussion for reasons noted in WKC.

Genuine explanation within the theory of language is closely related to the Strong Minimalist Thesis (SMT), which proposes that the properties of language are determined by the optimal computational operation (Merge) and “third factor” principles such as minimal computation, principles that hold independently of language and in the best case might be considered natural laws.To the extent that SMT can be approached, we have genuine explanations (with evolvable innateness, hence learnability).

We can take core language design to be what receives genuine explanation, satisfying SMT. The concept can be extended by degree to more limited kinds of explanation. There is no need to draw a sharp boundary for a useful descriptive term like general language design unless some theoretical reason arises to do so. Let’s keep here to core language design.

The central part of WKC, from my perspective, is chapter 1, which reviews recent results that qualify as genuine explanations. To me at least, these seem to be results of real significance, new in the study of language and cognitive systems generally; in fact, hardly even formulable until fairly recently. Optimal computation (including choice of Merge as the core combinatorial operation) provides a basic account for the general architecture of language, what is sometimes called the “Basic Property” of language: narrow syntax generates representations at the conceptual-intentional (C-I) interface interpreted as thoughts, with externalization an ancillary phenomenon mapping syntactic structures to the sensory-motor interface, typically sound.

In addition to the basic property, the same elementary assumptions, as discussed in chap. 1, provide genuine explanations for other fundamental features of language: the ubiquitous property of displacement (IM), always previously regarded—by me in particular—as an anomalous “imperfection” of language but now shown to be the simplest case, which would have to be barred by stipulation; reconstruction and its intricate semantic consequences; the strange property of structure-dependence, implying that children reflexively ignore 100 percent of what they hear (linear order) and attend only to what they never hear and that their minds construct (abstract structure).

There’s more that I didn’t go into, for example, an explanation for why the more complex operation external merge (EM) is used at all; the conditions under which IM may, must, and does not apply; the locus of diversity and mutability of language in externalization; the status of parameters and feasible search spaces; and considerably more to a greater or lesser degree.

One consequence of these results is strong support for the traditional view that language design is a system of thought, and with externalization understood as an ancillary process, the use of language for communication, which depends on externalization, is still more peripheral to language design.

None of this should be very surprising. On the contrary, it should be expected. What is surprising is that such ideas took so long to resurface in the modern period and that they now seem surprising (if not downright offensive or virtually incomprehensible, as reactions often indicate). Externalization, after all, involves an amalgam of two systems, unrelated in evolutionary history and with quite different properties: language proper and sensory-motor systems, usually speech. For that reason externalization is not strictly speaking part of language alone. Nor should it be surprising that the locus of variation, complexity, and mutability of language turns out to be mostly, maybe entirely, in this amalgam.

Such expectations receive empirical support from the results concerning language design. And still further support from investigation of conflicts between communicative and computational efficiency (the latter a “third factor” property, hence conforming to SMT). In all known cases communicative efficiency is sacrificed. There is also neurolinguistic and psycholinguistic support, discussed in chap. 1.

There is no doubt that language is used for communication, though statistically speaking, as noted in WKC, it is rare, for whatever that is worth. There is nothing in the above analysis that excludes the possibility that some elements of language will be useful primarily for communication, but we would not expect them to be part of core language design. For example, imperative expressions are used primarily in socially interactive contexts, but while language design allows for them, they play no role in genuine explanation. I think that is true of the kinds of examples you raise generally, the set S. Language design provides options for these (and in some cases, like theme-rheme, may even provide partial explanation). But they do not have genuine explanations and do not participate in such explanations. And as noted, if they are dropped from language, nothing changes, just as nothing changes if languages were to drop color words.

It seems to me that the items of S can also be used in “internal dialogue”, the bulk of language use, and in this respect for thought, though their natural place, as you say, is social interaction, communication in some broad sense.

I don't see then that they pose a dilemma.

We might take note of the widely held doctrine that the function of language is communication and that language evolved from animal communication systems. The latter is an empirical hypothesis lacking support and facing strong counter-evidence. The former assertion has little substance, for reasons discussed in WKC. There would also be little if any substance to the counter-claim that the function of language is thought, unless interpreted as a way of saying that language design is a system of thought. That formulation is substantive and seems to me by now well grounded and of no slight significance.

These reflections bring up an intriguing issue that I avoided. True, externalization involves an amalgam of two distinct systems: language proper and sensory-motor systems.But isn’t that also true of narrow syntax, generating representations of thought, an external system? Or is it truly an external system? That question raises serious and rather obscure issues.

The traditional view was that language and thought are intimately related: for Wilhelm von Humboldt, they are literally identical; for William Dwight Whitney, language is “audible signs of thought”; for Otto Jespersen, arguably the last figure in the tradition that was swept aside in structuralist/behaviorist currents, unearthing “the great principles underlying the grammars of all languages” will yield “a deeper insight into the innermost nature of human language and of human thought” (quoted in WKC).

We have no clear concept of thought apart from linguistically constituted thought. That is why Alan Turing, in his famous paper on machine intelligence that initiated the field of artificial intelligence, wrote that the question of whether machines can think is “too meaningless to deserve discussion”. The C-I interface constitutes the external system to a certain extent (for Humboldt, completely). That is not true of the sensory-motor interface, which is entirely independent of language, another crucial asymmetry.

I am skipping over a lot of material (in particular, the parts of general syntax that are called “formal semantics”, a large topic). And there is, of course, a good bit more to say about these topics, but this should be enough to deal with the issues that arise here.

采访人物简介

Noam Chomsky教授

诺姆·乔姆斯基（Noam Chomsky）教授全名Avram Noam Chomsky，是享誉世界的著名语言学家、认知科学家、政治批评家、哲学家和历史学家。乔姆斯基1928年12月7日出生于美国宾夕法尼亚州费城的一个中产阶级犹太家庭。1945年，16岁的乔姆斯基进入宾夕法尼亚大学学习哲学、逻辑学与语言，后来在Harris的指导下学习研究生课程，并在Harris的推荐下，向Nelson Goodman和Nathan Salmon学习哲学，向Nathan Fine学习数学。1951年，他完成硕士论文《现代希伯来语语素音位学》，1951年至1955年，他在哈佛大学以学术协会会员的身份从事语言学研究工作，在此期间以《语言理论的逻辑结构》(LSLT)中的部分章节作为其博士论文，取得博士学位。乔姆斯基在该论文中采用了Harris的语言研究方法以及Nelson Goodman对形式系统和科学哲学的观点。但在研究过程中发现了结构主义语言学的局限性，转而探索新的方法，逐步建立起“转换生成语法”。1957年，他在其博士论文的基础上做出了新的发展，完成了《句法结构》，成为“乔姆斯基革命”开始的标志。他的开创性著作包括《句法结构》、《语言与心智》、《句法理论的若干问题》以及《最简方案》等。

乔姆斯基提出了“转换生成语法”，其理论在美国被称为理论语言学领域主流的学问。他倡导生物语言学理论，认为人生来就有与普遍语法相关的语言官能，普遍语法由原则和参数组成。这一观点解释了为什么儿童可以很容易地习得母语。乔姆斯基针对行为主义的批判性研究开创了语言学和心理学的“认知革命”，他在很大程度上将语言学建立成一门正式的自然科学。他建立了数理科学和计算机科学领域的“乔姆斯基等级”，根据方法生成力的不同对形式语言进行了分类，这一理论目前仍是计算机科学中的有机组成部分。

1961年乔姆斯基获聘MIT语言学系教授，1976年被聘为MIT“校聘教授”，2002年作为“荣休教授”从MIT退休，但是他并没有因为退休而停止研究的脚步。2017年秋，他又加入亚利桑那大学，受聘为社会和行为科学学院语言学系“桂冠教授”。他曾获京都奖——“基础科学”类、亥姆霍兹奖和本杰明·富兰克林计算机与认知科学奖。McGilvray认为乔姆斯基开创了语言学的“认知革命”，他在很大程度上将语言学建立成一门正式的自然科学。Micheal Egnor认为乔姆斯基是过去半个世纪中最伟大的科学家。也有人称乔姆斯基为“现代语言学之父”。语言学家John Lyons曾指出，在几十年的出版中，乔姆斯基语言学已经成为该领域“最具活力和影响力的”学派。他出版有150多部专著，根据艺术和人文引文索引，在1980年到1992年，乔姆斯基是被文献引用数最多的健在学者，并是有史以来被引用数排名第八多的学者。

Brief Introduction of Interviewee

Noam Chomsky, in full Avram Noam Chomsky, is a distinguished linguist, cognitive scientist, political dissident, philosopher and historian. He was born into a middle-class Jewish family, on December 7, 1928, in Philadelphia, Pennsylvania, US. In 1945, at the age of 16, Chomsky entered the University of Pennsylvania to study philosophy, logic and language. He took graduate courses with Harris and, at Harris’s recommendation, studied philosophy with Nelson Goodman and Nathan Salmon and mathematics with Nathan Fine. In his 1951 master’s thesis, The Morphophonemics of Modern Hebrew. In The Logical Structure of Linguistic Theory (LSLT), written while he was a member of Academic Association at Harvard (1951–55), Chomsky adopted aspects of Harris’s approach to the study of language and of Goodman’s views on formal systems and the philosophy of science. However, during the research, he found the limitations of structural linguistics. In this way, he explored something novel and gradually established “Transformational-Generative Grammar”. In 1957, he made a new development on the basis of his doctoral thesis and completed Syntactic Structures, which marked the beginning of “Chomskyan Revolution”. Among his groundbreaking books are Syntactic Structures, Language and Mind, Aspects of the Theory of Syntax and The Minimalist Program.

Chomsky introduced “Transformational-Generative Grammar” to the linguistics field, whose theory is considered as the mainstream knowledge in theoretical linguistics in the United States. According to the biolinguistic theory he advocated, children are born with language functions related to universal grammar, which consists of principles and parameters and explains why children could acquire their native languages easily. Based on his critical analysis on Behaviorism, Chomsky inaugurated the “cognitive revolution” in linguistics and psychology, and that he is largely responsible for establishing the field as a formal, natural science. He established the “Chomsky hierarchy” in the fields of mathematical science and computer science, and classified formal languages according to the different generative power of methods. This theory is still an integral part of computer science.

He was appointed full professor at MIT in 1961, Institute Professor in 1976 and retired as professor emeritus in 2002, but his researches are still continuing. He joined the UA faculty in fall 2017, is a laureate professor in the Department of Linguistics in the College of Social and Behavioral Sciences. He has received the Kyoto Prize in Basic Sciences, the Helmholtz Medal and the Ben Franklin Medal in Computer and Cognitive Science. McGilvray observes that Chomsky inaugurated the “cognitive revolution” in linguistics, and that he is largely responsible for establishing the field as a formal, natural science. Micheal Egnor believes that Chomsky is the best scientist of the past half-century. As such, some have called Chomsky “the father of modern linguistics”. Linguist John Lyons further remarked that within a few decades of publication, Chomskyan linguistics had become “the most dynamic and influential” school of thought in the field. He has published more than 150 monographs. According to the Art and Humanities Citation Index, from 1980 to 1992, Chomsky was the most cited living scholar in the literature and the eighth most cited scholar in history.

采访者简介

Idan Landau教授

伊丹·兰多 (Idan Landau)，1999年获得美国麻省理工大学语言学博士学位，目前担任以色列本·古里安大学外国文学与语言学系教授。他的研究兴趣包括生成语法、语法模态性、句法、控制理论、论元结构等。

Brief Introduction of Interviewer

Idan Landau, obtaining his Ph.D. from MIT in 1999, is a Professor of Linguistics in the Department of Foreign Literatures and Linguistics at Ben Gurion University, Israel. His research interests include generative grammar, grammatical modularity, syntax, theory of control, argument structure and so on.

往期推荐

本文版权归“理论语言学五道口站”所有，转载请联系本平台。

编辑：闫玉萌 赵欣宇 雷晨

排版：闫玉萌 赵欣宇 雷晨

审校：陈旭 李芳芳 田英慧