人物专栏 | Noam Chomsky教授访谈（下）

点击上方蓝字关注我们

编者按

《理论语言学五道口站》（2023年第10期，总第274期）“人物专栏”继续与大家分享Patrick C. Trettenbrein博士对Noam Chomsky教授的访谈，访谈以“50 Years Later: A Conversation about the Biological Study of Language with Noam Chomsky”为题。

访谈中，Noam Chomsky教授首先介绍了生物语言学背景下语言产出和语言理解的研究进展并阐述了语言能力和语言运用的关系，然后解释了递归运算、感觉运动系统和词汇语义的作用，最后对机器翻译方面的研究发表了自己的看法。

Noam Chomsky教授和Patrick C. Trettenbrein博士简介可参考《理论语言学五道口站》“人物专栏”2023年第06期，总第270期。

访谈内容

12.

Patrick C. Trettenbrein博士：在《语言的生物学基础》以及您的附录出版数年后，Lenneberg在他1972年的一篇报告中写下了这么一段话（Michael Arbib发现了这枚沧海遗珠）：

目前只有一种语言结构的科学理论，那就是生成语法。其目的不在于成为生物机制模型，而是旨在创造一个可以充分描写自然语言（如英语）独特关系网络的形式主义理论。生成语法在一段时间内似乎要向着算法或语言分析的方向发展，但还只是一个未实现的梦想，除了在高限制性的话语上。生成语法确实在很大程度上帮助我们定义句子结构复杂性。但由于其建构方式，生成语法会将某些我们直觉上认为最简单的话语归为复杂结构（即：冗杂的转换操作所产生的结构）。例如“Water!”会被解释为“Give me water!”的派生形式，而“Give me water!”又是从“You give me water”这样的更基础的形式派生而来的。因此，我们亟需一种句子产生和理解的理论，一种既具有Chomsky方法的形式准确性又能明确阐释语言能力心理生物学（psychobiological）基础的理论。（Lenneberg 1972: 635–636）

如果从生物语言学的背景下思考这段话，您认为过去几十年间生成语法是否朝着Lenneberg所倡导的句子产生和理解的理论方向发展？最简方案框架下的许多成果似乎与句子产生和加工理论相去甚远，而更侧重语言机制的形式本质，您认为是这样吗？

Noam Chomsky教授：我并不完全理解这段话。语言L的生成语法G寻求的是确定语言L的无限表达结构以及它们在概念-意向（语义-语用）接口和感觉-运动接口的解读（最近的文章表明前者占主要地位）。这是一个关于语言能力的问题。关于语言L运用的理论无疑会调用以G为特征的内在语言能力；相应地，关于语言运用的普遍理论也会调用语言能力的普遍语法特征。这些属于生物学范畴，并具有“心理生物学基础”。

语言加工处理的研究颇有进展，其中对最简语言加工（minimalist parser）的研究范围广、成果多，在Robert Berwick即将出版的新书中（Berwick & Stabler, 已于2019年在Oxford University Press出版——编者注）会有许多介绍。对语言产生的研究也取得了进展，不过对于有意识的言语行为的研究还存在空白。正如这一领域内的两位领军人物Emilio Bizzi和Robert Ajemian （Bizzi & Ajemian 2015）的形象比喻一样，我们对木偶和线已经有了很多了解，但操纵木偶之人是谁却仍旧是个谜团。在语言研究中，正是这个谜团启发了那些创造出巨大科学成果的现代科学奠基者，包括伽利略、笛卡尔以及波尔罗亚尔（Port Royal）学派的逻辑语言学家们。

13.

Patrick C. Trettenbrein博士：无论是语言运用还是语言能力的理论都属于生物学范畴，这一点我当然同意。我对于这段引言的理解是，Lenneberg 某种程度上指出了语言学和生物学之间的差异，这一点我们已经简要讨论过。对语言能力的研究似乎已经将语言学隔离在了心理学和神经科学之外，目前仍然如此，甚至还可能阻碍这些领域的进一步融合，比如心理语言学。您怎么看待语言学与心理学和神经科学的关系？

Noam Chomsky教授：对语言能力的研究不能被隔离在心理学之外，因为它本就是心理学的一部分——除非我们强行从“心理学”的定义中排除语言、算术等存储在人脑内部的知识。在过去的50年里，心理语言学一直与语言能力的研究紧密结合，神经语言学也是如此。语言能力是在大脑中体现的，调用这部分知识储备的语言运用也是如此。

就我个人而言，我始终认为，至少在我感兴趣的方面，语言学是心理学的一部分，因而最终也是生物学的一部分。语言学与心理学是部分与整体的关系，类似于视觉理论与心理学的关系。只要我们关心大脑中发生了什么，就已经与神经科学相结合了。

14.

Patrick C. Trettenbrein博士：我当然同意语言能力是在大脑中体现的。我想说的是，比如这种储存于人脑内部的知识可能只在语言加工的过程中才被调用，因为它的构造贴合了语言运用系统的工作方式。您认为这种看法有道理吗？

Noam Chomsky教授：I-语言或语言能力在每次使用中都会被调用，不论是加工语言输入还是进行语言产出（包括脑内的思想构建）。如果说语言能力的构造“贴合了语言运用系统的工作方式”，那么它一定会在每个运用系统都被复写，但这似乎不太合理。这样的观点无论用何种方式表达，实质上都没有理解到I-语言是语言运用系统所调用的中心系统这一点，正如计算过程会调用代数知识那样。

15.

Patrick C. Trettenbrein博士：您最近与Angela Friederici共同撰写了一篇关于语言加工的神经基础的论文（Friederici et al. 2017），想必您对这一领域的实际进展抱有比较乐观的态度。对于那些尝试将语言学理论与心理学或神经科学更紧密地结合起来的研究者，您有什么建议吗？

Noam Chomsky教授：我并没有什么特别的建议。在她的那本重要著作（Friederici 2017a）中，我曾撰写过一些介绍性的评论。其次，我认为这个问题就好比在问“您对那些尝试将视觉感知理论与心理学及神经科学更紧密地结合起来的研究者有何建议”一样，并起不到什么建设性的作用。

在我看来，这种表述问题的方法并不恰当。语言（或者视觉等领域）的研究有各种各样的手段，这些手段相互结合会产生一些有趣的问题，我的建议是去探究这个方面。

16.

Patrick C. Trettenbrein博士：还有一个与此相关的问题。最简方案最大的成就之一是降低了理论假设中认知机制的复杂度，比如只采用合并（Merge）操作，并将合并视为语言官能的核心。但目前，您本人却认为合并是否是语言特有的机制这一点至少在系统发生学上尚不明确。而至于人脑中执行合并操作的方式，现阶段我们充其量只能做一些有根据的猜测（如Friederici 2017b)，因为我们还缺少将运算/算法层次与执行层次联系起来的理论。在我看来似乎有这样一种可能，即能够进行递归运算的脑回路可能是为了适应运动计划或音乐等其他用途而进化出来，随后再拓展到语言。您目前是如何看待这些问题的呢？

Noam Chomsky教授：人们时常认为运动计划涉及到递归运算，但在我看来，这似乎是对递归运算和运动计划的误解。递归运算适用于数字无限性的系统，而运动计划并非这种系统（尽管人们可以在连续系统上强加任意的数字网格，但这并不能解决问题）。如果忽视这种最原始、最重要的区别，就会造成严重的分歧。

至于音乐和语言的关系，音乐只涉及非常有限的递归运算。在我看来，音乐是由语言拓展而来的假设比相反的假设要合理得多。我们也可以参考Jeffrey Watumull和Marc Hauser在他们最近的研究（Hauser & Watumull 2016）中的提议，认为递归运算涌现后，被那些具有数字无限性的、语言的和算术（可能也包括音乐）的认知系统加以运用。

17.

Patrick C. Trettenbrein博士：有同事让我向您提几个问题。其中一位想让我问您，您为何如此确信感觉运动系统，尤其是从语言加工的角度来看，在语言中只起到辅助作用？当然，在运算层面上，这种观点非常合理；但考虑到执行层面时，这位提问者就不那么肯定了。她想到了具身认知，而且她认为在最简方案的背景下，这些辅助系统在接口上发挥的作用比先前所假设的更大。

Noam Chomsky教授：我已经在其他地方从概念和实证的角度解释了其中的原因（部分内容的回顾参见《我们是谁？》（What Kind of Creatures are We? ）Chomsky 2016）。这些理由在我看来很可靠、很有说服力。从定义上看，感觉运动系统在语言加工中确实发挥着核心作用，但剩下的部分我就不太能理解了。

18.

Patrick C. Trettenbrein博士：接下来，另一位同事想让我问您怎么看待词汇性元素的起源。在当前的句法理论中，词汇性元素发挥了重要的作用。Lenneberg提出了一种非指称的、完全内涵性的词汇语义研究方法，您也反复强调了这一方法的重要性。在您的《为什么只有人类？》（Why Only Us? Berwick & Chomsky 2016）一书中，有关词汇性成分的讨论几乎完全被搁置。有趣的是，Lenneberg似乎认为人类的分类方式与其他动物并没有本质区别。此外，他认为“词”不是存储在记忆中的概念的标签，而是分类过程的标签。目前您对这些问题有何看法？“词”的进化起源会继续保持神秘，还是会最终变得易于研究？

Noam Chomsky教授：我认为有非常有力的证据说明人类的概念/词汇语义与动物世界中已知的任何事物有本质区别，这一点我在其他地方也讨论过。词汇的进化起源是一个谜，正因如此，我们讨论语言官能的进化时会将其搁置。据我所知，目前对于这个谜团还没有可行性建议。

19.

Patrick C. Trettenbrein博士：最后，如果我没记错的话，您在MIT着手参与了一个机器翻译的项目，但您最后并没有真正投入工作，因为您认为它毫无意义，是这样吗？虽然与本期特辑的主题无关，但我的一位对翻译很感兴趣的同事想让我问您，您认为在生成语法的框架下，人们应该如何研究翻译？是否有“翻译官能”一说？我个人还想补充一点：这个问题可能涉及到我们如何建构和理解语义-语用接口而非感觉运动系统的映射，以及这一映射所容许的变化范围。

Noam Chomsky教授：我碰巧承担过一个机器翻译的研究项目，但我并未参与这方面的课题。一开始我就感觉，就实用性的用途来说，采用暴力算法的手段是最可行的。也许有一天，人们对语言的理解会对这个项目做出实质性的贡献，但那一天仍然很遥远。在我看来，对机器翻译的研究并不能推进对语言本质的理解。据我所知，这些想法已经得到了证实。我不觉得有什么理由能支持“翻译官能”的存在。语义-语用/概念意象接口的变化问题是一个有趣的问题，属于边缘研究（我个人猜测这方面的研究并不多）。但以目前对语言的理解还远远无法对改进自动化翻译做出长足的贡献。

Patrick C. Trettenbrein博士：Chomsky教授，非常感谢您回答我的问题。

English Version

12.

Dr. Patrick C. Trettenbrein: Against this background, consider the following quote from a report by Eric Lenneberg written in 1972, rescued from oblivion by Michael Arbib (cf. Arbib, this issue), a few years after Biological Foundations of Language including your appendix had been published：

At present there is only one type of scientific theory of language structure available, generative grammar, and this was never intended to serve as a model for biological mechanisms. The intent was to create a formalism that would adequately describe the web of relationships that characterizes a natural language such as English. For a time it looked as if it might turn into an algorithm or language analysis, but this is yet an unrealized dream (except for highly restricted discourse). Generative grammar does help us here and there to define the meaning of complexity in the structure of sentences, but because of the way it has been set up, it will also classify some utterances as complex (i.e., products of a long and complex history of transformations) that are intuitively the simplest of all—e.g., “Water!” which would have to be accounted for as a derivative of “Give me water!” which, in turn, is derived from a more basic form, “*You give me water.” What is most urgently needed is a theory of sentence production and comprehension that has the formal precision of Chomsky’s approach but is explicitly intended to explicate the psychobiological underpinnings of language capabilities. (Lenneberg 1972: 635–636)

Do you think that developments in generative grammar in the past decades have brought it closer to the theory of sentence production and comprehension that Lenneberg had called for? In many ways work within the Minimalist Program (MP) seems even further removed from a theory of language production and processing, instead focusing on the formal nature of the involved machinery?

Prof. Noam Chomsky: I don’t entirely understand the quote. A generative grammar G of language L seeks to determine the structure of the infinite class of expressions of L and their interpretation at the conceptual-intentional (semantic-pragmatic) and sensorimotor interface (the role of the former primary, so recent work suggests). It is a theory of competence. Theories of performance for L will of course access the stored competence characterized by G; and general theories of performance will, correspondingly, access general properties of competence grammars. It all falls within biology, and all has “psychobiological underpinnings”.

There has been a great deal of progress in the study of language processing, including extensive and productive work on minimalist parsers, much of it appearing in a forthcoming book on minimalist parsers edited by Robert Berwick (Berwick & Stabler, in press). The study of language production has also progressed, though with a huge gap that holds for all voluntary action. As described figuratively by two leading researchers of voluntary action, Emilio Bizzi and Robert Ajemian (Bizzi & Ajemian 2015), we are learning a good deal about the puppet and the strings, but the role of the puppeteer remains a mystery. In the case of language, it was essentially this mystery that inspired the awe and wonder of some of the great founders of modern science, including Galileo, Descartes and the logician-linguists of Port Royal.

13.

Dr. Patrick C. Trettenbrein: Of course I agree that it all falls within biology, theories of performance as well as competence. In my understanding of the quote, Lenneberg was—in part—pointing to the discrepancy between linguistics and biology that we already briefly discussed. It seems that the study of competence has isolated and still isolates linguistics from psychology and neuroscience; maybe even preventing a closer integration of these respective fields, as in parts of psycholinguistics. How would you describe the relation of linguistics to psychology and neuroscience?

Prof. Noam Chomsky: The study of competence can’t be isolated from psychology because it is part of psychology—unless we (perversely) define “psychology” to exclude internally-stored knowledge of language, arithmetic, etc. Psycholinguistics, for the past 50 years, has been closely integrated with the study of linguistic competence. Same with neurolinguistics. Linguistic competence is represented in the brain and the same is true of performances that access this stored knowledge of language.

Speaking personally, I’ve always regarded linguistics, at least the aspects that interest me, as part of psychology, hence ultimately biology. The relation of linguistics to psychology is similar to the relation of the theory of vision to psychology: part to whole. And insofar as we are concerned with what is happening in the brain, it’s integrated with neuroscience.

14.

Dr. Patrick C. Trettenbrein: Of course I agree that linguistic competence is represented in the brain. What I was trying to get at is that, for example, it could be the case that this internally stored knowledge is only accessed during processing in the sense that it is built into the way in which the performance system works. Does that make any sense?

Prof. Noam Chomsky: The I-language—linguistic competence—is accessed in every use of language: in processing linguistic input, but also production (including internal construction of thought). If it is “built into the way in which the performance system works,” then it must be duplicated in each performance system, which does not seem a reasonable proposal. I don’t see any way of reformulating this idea that does not reduce to the assumption that I-language is a central system accessed by performance systems, much as knowledge of arithmetic is accessed in calculating.

15.

Dr. Patrick C. Trettenbrein: Okay. You have recently co-authored a paper on the neural basis of language processing with Angela Friederici (Friederici et al. 2017), so you must be somewhat optimistic that real progress can be made in this area. What is your advice for researchers trying to bring linguistic theory and psychology or neuroscience closer together?

Prof. Noam Chomsky: Not quite. I contributed some introductory remarks to her very important book (Friederici 2017a). On the rest, I don’t think the formulation of the questions is helpful, any more than the question of how to advise researchers trying to bring the theory of visual perception and psychology and neuroscience closer together.

It doesn’t seem to me the right way to formulate the issues. There are different approaches to the study of language (vision, etc.), and intriguing problems where they intersect. The advice is to pursue them.

16.

Dr. Patrick C. Trettenbrein: On a related note, one of the biggest successes of the MP has been to reduce the complexity of the postulated cognitive machinery, for example, by relying only on a single operation which we think is at the core of the language faculty, that is Merge. Now, you yourself have speculated that it is not at all clear whether Merge is specific to language, at least in phylogenetic terms. With regard to how an operation like Merge may be implemented in wetware we are still in a situation where we can at best make educated guesses (e.g., Friederici, this issue), as we lack a linking theory between the computational/algorithmic and the implementational level. It seems to me that a scenario where circuitry capable of recursive computation may have evolved for another purpose (e.g., motor planning or music) and later was exapted for language seems quite likely. What is your current take on these questions?

Prof. Noam Chomsky: It is often claimed that recursive computation is involved in motor planning, but that seems to me a misunderstanding of both recursive computation and motor planning. Recursive computation holds for systems of digital infinity. Motor planning is not a system of digital infinity (though one can impose an arbitrary digital grid on continuous systems, leaving the issues where they were). And as one looks beyond this initial (and crucial) distinction, divergences proliferate.

As for music and language, it seems to me far more reasonable to suppose that music (to the very limited extent that it involves recursive computation) was exapted from language than the converse. Or, perhaps, as suggested by Jeffrey Watumull and Marc Hauser in recent work (Hauser & Watumull 2016), that recursive computation emerged and was applied in cognitive systems of digital infinity, language and arithmetic, maybe music.

17.

Dr. Patrick C. Trettenbrein: Well, some colleagues have asked me to pose some questions to you. One of them wanted me to ask you why you are so convinced that sensorimotor systems only play an ancillary role in language, that is, especially when taking a processing perspective? On the computational level this view is, of course, very plausible; but when we look at implementation she is not so sure. Think embodied cognition. Also, she thinks that in the context of the MP these ancillary systems are thought to do a lot more work at the interfaces than previously assumed?

Prof. Noam Chomsky: I’ve explained the reasons elsewhere, both conceptual and empirical (for review of some of them, see What Kind of Creatures are We?; Chomsky 2016). They seem to me sound and compelling. In processing, sensorimotor systems play a central role, by definition. I don’t understand the rest.

18.

Dr. Patrick C. Trettenbrein: Okay. Next, another colleague wanted me to ask you about your take on the origin of lexical elements which, incidentally, do a lot of the work in current syntactic theory. Lenneberg already put forward an approach to lexical semantics that was non-referential and completely intensional, the importance of which you have repeatedly emphasised. In Why Only Us? (Berwick & Chomsky 2016), lexical elements were almost completely put aside. Interestingly, Lenneberg seems to have thought that the way in which humans categorise is not qualitatively different from other animals. Also, he considered “words” not as labels for concepts stored in memory but as labels for categorisation processes. What is your current outlook on these questions? Will the evolutionary origin of “words” remain mysterious or eventually turn out to be susceptible to study?

Prof. Noam Chomsky: I think there is very strong evidence, which I’ve discussed elsewhere, that human concepts/lexical semantics are radically distinct from anything known elsewhere in the animal world. Their evolutionary origin is a mystery, which is why they were “put aside” in our discussion of the evolution of the faculty of language. For the moment, there are, to my knowledge, no useful ideas as to how to investigate this mystery.

19.

Dr. Patrick C. Trettenbrein:  Lastly, if I am not mistaken, you started out at MIT as part of a machine translation project that you ended up never really working on because you thought it quite pointless—is that correct? A colleague of mine is interested in translation and, while not at all related to the topic of this special issue, he wanted me to also ask you about how you think one might approach the study of translation within a framework such as that of generative grammar? Is there something like a “translation faculty?” Personally, I might add that this question probably touches upon how we conceive of and understand the mapping to the semantic-pragmatic interface and the degree of variation that this mapping permits as opposed to the mapping to sensorimotor systems?

Prof. Noam Chomsky: I happened to be appointed in a machine translation research project, but never worked on the topic. My feeling from the start was that for practical purposes, brute force approaches would be the most feasible. While some day understanding of language might contribute materially to this project, that time was still remote. And research on machine translation did not seem to me the way to advance the project of understanding the nature of language. These expectations have been borne out, as far as I know. I don’t know of any reason to suppose that there is a “translation faculty”. The question of variation at the semantic-pragmatic/conceptual intentional interface is an interesting one, at the border of research (my own guess is: not much). But understanding is far from contributing much to improving automatic translation.

Dr. Patrick C. Trettenbrein: Professor Chomsky, thank you very much for answering my questions.

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：雷晨 赵欣宇 何姝颖 董泽扬

排版：雷晨 赵欣宇 何姝颖 董泽扬

审校：时仲 田英慧

英文编审责任人：董泽扬