人物专栏 | Andrea Moro 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》（2021年第16期，总第150期）“人物专栏”与大家分享本站采编人员陈金玉对Andrea Moro教授进行采访的访谈录。Andrea Moro教授，意大利帕多瓦大学语言学博士，北京语言大学语言学系国际教授委员会成员，曾任美国麻省理工学院“语言学与哲学系”、哈佛大学“语言学系”客座研究员，意大利博洛尼亚大学语言学史副教授，意大利米兰圣拉斐尔大学普通语言学全职教授。

本期访谈中，Andrea Moro 教授首先就生物语言学的两种范式提出了深刻而独到的见解，随后解释了句法中常用的树形图在神经语言学中的应用以及大脑功能偏侧化的研究现状，最后展望了生物语言学的跨学科发展及其对于人工智能会产生的影响，带给我们许多启发。

Andrea Moro 教授简介可参考《理论语言学五道口站》“人物专栏”2021年第10期，总第144期。

访谈内容

05.

陈金玉：在生物语言学的发展过程中，形成了以Lenneberg和Chomsky为代表的两种主要范式。Lenneberg范式侧重于语言的生理基础、认知机制、起源、进化和发展模式，而Chomsky范式侧重于语言功能特异性的语言进化研究。这两种范式您更倾向于哪一种，为什么？

Andrea Moro 教授：在我看来，这两种范式实际上是趋同的，不存在选择其中一个而放弃另一个的说法。我认为Chomsky范式的更准确的定义是探索语言的形式属性，而不考虑它们的功能。事实上，这也影响了我所遵循的Chomsky的语言进化观。[关于此类知识的详细说明，可参考 Friederici et al. 2018. “Language, mind and brain”, Nature Human Behaviour, vol 1, pp. 713-722]

06.

陈金玉：Poeppel & Embick (2005)提出，如果把语言学的术语直接应用于语言的神经与生物基础的研究之中会产生粒度失调（Granularity Mismatch Problem）和本体不可比（Ontological Incommensurability Problem）两个问题。您认为我们在句法中常用的树形图在神经语言学的研究中依然有用吗？

Andrea Moro 教授：我认为粒度问题的提出与语言的神经和生物学基础无关，而是与70年代末乔姆斯基提出的“原则和参数”有关，此理论自提出以来就一直是生成语法的核心。显然，当年的关系化和疑问句构造程序等传统概念只是肤浅的描述性概念，为了理解语法的结构，人们必须把语法拆解为更小单位，这在很大程度上是指布拉格学派按照索绪尔提出的结构主义方法将音素分解为特征。至于有关树形图的具体问题，我认为树形图对于神经语言学来说也十分基础，就像是原子的太阳系模型在物理学中是大有用处的。事实上，有一些方法证明树形图可以表征大脑对不同刺激的反应，甚至可以表示spec-head与head-comp的区别，语言学家对此种现象十分熟悉（见于此前推荐书籍以及实验相关文献）。树形图以及其相关符号表征也大有用处，因为它们使对称这一概念变得十分清晰。这一概念，以及我在动态非对称框架（the Dynamic Antisymmetry framework）中提出的对称性破坏的相关概念，可以看为解决粒度问题的一个例子：这不仅与原始元素的全部内容有关，同时还是基本简单属性之一，这一点至关重要，我不认为这两个潜在的不同问题应该分开。乔姆斯基（2013）实际上展示了如何集成这两个问题，前提是假设动态反对称（Dynamic Antisymmetry）也许是一个很好的克服粒度问题的备选方案（详情见于Moro, A. Dynamic Antisymmetry, MIT Press）。另一个很好的例子是制图项目的发展。从Jean-YvesPollock（Pollock, J.-Y. 1989.Verb Movement, UG, and the Structure of IP, Linguistic Inquiry 20: 365–424）以及我的文章（Moro, A. 1988. Per una teoria unificata delle frasi copulari, Rivista di Grammatica Generativa, 13, pp. 81-110; reprinted in Moro, A. 2013. The Equilibrium of Human Syntax. Symmetries in the brain, Leading Linguists Series, Routledge, New York）开始，运用制图理论对句子结构的分析越来越精细。这两个在经验和理论上截然不同的提议是一个很好的例子：独立的张力作用于句法，以获得一个更好的粒度，而与神经生物学研究无关。

07.

陈金玉：Blumstein (2019)认为，大脑功能偏侧化的研究已经相当成熟，有关语言神经加工的研究主要基于左脑展开。相比之下，右脑的语言功能尚未明晰。因此，在您看来，我们可以进行哪些与语言处理相关的研究来研究右脑？

Andrea Moro 教授：在我看来，现在没有什么研究是相当成熟的，至少就语言而言，大脑偏侧化会被彻底地重新思考。我们可能需要以一种更为清晰的方式来修正两个半球之间的相互作用，以避免新颅相学方法所带来的危险。这种转变不仅仅是由于完全偏侧化理论的失败所导致的，也是由转向“what-problem”方法论的结果所强加的。

08.

陈金玉：目前，许多语言学家从生物语言学的角度对儿童语言习得进行了研究。在您看来，这种跨学科的研究会给科学发展带来什么好处？

Andrea Moro 教授：我认为这是最基本的，有两个主要原因：第一，定义（不）可能的语言类别的最终目标与儿童大脑在经历世事之前状态的表征相一致，因此语言学家必须不断地用从自发语言习得中所得的数据来检验他们的预测；第二，由于一些不可避免的伦理原因，我们所收集的儿童大脑的神经生物学数据还相对较少:为了获得更优质的数据，我们需要找到新的非侵入性方法，但目前在这方面技术还不够先进。我们还是得依赖临床数据，虽然这些数据是有趣和基本的，但仍然不能获得与健康受试者神经生物学测量数据相结合的整体图像。

09.

陈金玉：现代人工智能的进步离不开现代语言学的发展。能否请您解释一下生物语言学在哪些方面可以促进人工智能的发展？

Andrea Moro 教授：由于下述原因，我认为这是一个很微妙的问题。一方面，我认为许多从事“现代人工智能”的研究者没有意识到语言学的发展，特别是形式语言学的发展。例如，许多自然语言处理论文中相关引文的缺失就证明了这一点。如果人们认为“现代人工智能”是模拟人类行为的努力，那么这未必是个问题，但如果隐含的意图是理解人类语言的形式属性和它们在大脑中的实现方式，那这问题就大了。另一方面，除了模拟和理解之间的必要区别之外，我认为，如果人工智能的研究基于语言的前生成模型，那么我们可以得到一个可信的结论：前文所提到的那些描述性概念，如关系化和疑问句构造程序，远不适合在模拟神经回路中实现。

参考文献

[1] Friederici et al. 2018. “Language, mind and brain”, Nature Human Behaviour, vol 1, pp. 713-722.

[2] Moro, A. 2010. A brief history of the verb “to be”, MIT press.

[3] Moro, A. 2000. Dynamic Antisymmetry, MIT Press.

[4] Moro, A. 2017. Impossible Languages, MIT Press.

[5] Moro, A. 1988. Per una teoria unificata delle frasi copulari, Rivista di Grammatica Generativa, 13, pp. 81-110.

[6] Moro, A. 2015. The Boundaries of Babel II edition, MIT press.

[7] Moro, A. 2013. The Equilibrium of Human Syntax. Symmetries in the brain, Leading Linguists Series, Routledge, New York.

[8] Pollock, J.-Y. 1989. “Verb Movement, UG, and the Structure of IP,”Linguistic Inquiry, 20, 365-424.

English Version

05.

Jinyu Chen: Two paradigms, led by Lenneberg and Chomsky, have been formed in the development of biolinguistics. Lenneberg paradigm focuses on the physiological basis, cognitive mechanism, origin, evolution and development model of language, while Chomsky paradigm focuses on the study of language evolution based on the specificity of language function. Which of these two paradigms do you prefer and why?

Prof. Andrea Moro: I would rather say that the two paradigms converge and one cannot chose one over the other. Actually, I would suggest that a more accurate definition of Chomsky’s paradigm is to explore the formal properties of language independently of their function. In fact, this also impact on Chomsky’s view of evolution of language which I follow. [for this, one can take a look at Friederici et al. 2018 “Language, mind and brain”, Nature Human Behaviour, vol 1, pp. 713-722]

06.

Jinyu Chen: Poeppel and Embick (2005) proposed that utilizing linguistic terminologies directly in researches on neurological and biological basis of language would cause Granularity Mismatch Problem and Ontological Incommensurability Problem. Do you think the tree diagram that we often use in syntax is also useful in neurolinguistics?

Prof. Andrea Moro: I think that the problem of granularity is raised independently of the research in neurological and biological basis of language and lies at the very heart of generative grammar ever since the late seventies when Chomsky proposed the so-called “Principles and parameters model”. In those years it was clear that such traditional notions as relativisation or question formation could be maintained only as superficial descriptive notions but that in order to understand the architecture of grammar one had to decompose them in simpler unites, much in the sense that phonemes have been decomposed into features in the Prague School following the structuralist method inaugurated by Saussure. As for the specific question of the tree diagram I think it is fundamental for neurolinguistics much in the same sense as a solar system representation of the atom was useful in physics. In fact, there are ways to show that the tree is a representation of the brain response to specific stimuli which can even lead to the distinction of the spec-head vs. head-compl. representation linguists are familiar with [see again the books suggested before including references to this experiment. Tree diagrams and their notational equivalents are also very helpful and useful because they make the notion of symmetry very clear. This notion, and the correlated notion of symmetry-breaking I proposed in the Dynamic Antisymmetry framework, is a clear example of how the granularity problem may be solved: it’s not just a matter of the repertoire of the primitive elements that counts it’s also the one of he basic simple properties involved which is crucial and I do not think that these two potentially distinct issue should be kept as separated. Chomsky 2013 in fact shows how these two can be integrated assuming that Dynamic Antisymmetry may be a good candidate to overcome the granularity problem. [For a detailed illustration of Dynamic Antisymmetry see: Moro, A. Dynamic Antisymmetry, MIT Press]. Another good example is the development of the cartographic project. The sparkle of this new approach ultimately yielding to a more fine grained analysis of the spine of clause structure started independently from the seminal work by Jean-Yves Pollock (see Pollock, J.-Y. 1989. “Verb Movement, UG, and the Structure of IP,” Linguistic Inquiry 20: 365–424) and my own (Moro, A. 1988. "Per una teoria unificata delle frasi copulari", Rivista di Grammatica Generativa, 13, pp. 81-110; reprinted in Moro, A. 2013. The Equilibrium of Human Syntax. Symmetries in the brain, Leading Linguists Series, Routledge, New York). These two empirically and theoretically distinct proposals constitute a good example of the independent tension acting within syntax to arrive at a better granularity, independently of any confrontation with neurobiological studies.

07.

Jinyu Chen: Blumstein (2019) holds that the study of the lateralization of the brain function is quite mature, and researches on language neural processing are mainly around left brain. By contrast, the language function of the right brain is not very clear. Therefore, from your point of view, what kinds of researches related to language processing can we conduct to study the right brain?

Prof. Andrea Moro: I actually think that, at least insofar as language is concerned, lateralisation will be radically rethought and nothing is quite mature right now. We probably need to revise the interaction between the two hemispheres in a much more articulated way avoiding the danger of what one may dub a neo-phrenological approach to brain studies. This shift is not only suggested by the failure of fully lateralised theories but it is also, so to speak, imposed by the result of the shift toward the what-problem methodologies.

08.

Jinyu Chen: At present, many linguists have studied children's language acquisition from the perspective of biolinguistics. In your opinion, what benefits do you think this interdisciplinary research will bring to scientific research?

Prof. Andrea Moro: I still think it is fundamental for two major reasons: first, we must not forget that the final goal of defining the class of (im)possible languages coincides with the representation of the state of a child’s brain prior to any experience and as such linguist must continuously test their predictions against data taken from spontaneous language acquisition; second, due to correct and unavoidable ethical reasons we still have a comparatively less rich repertoire of neurobiological data of the child’s brain: to get a better one we need to find new non-invasive methodologies but technology is not too advanced with respect to this right now. One must still rely on clinical data which are surely interesting and fundamental but still cannot get the same overall picture that result from combining it with neurobiological measures on healthy subjects.

09.

Jinyu Chen: The progress of modern artificial intelligence cannot be separated from the development of modern linguistics. Could you please explain in what ways can biolinguistics promote the development of artificial intelligence?

Prof. Andrea Moro: This is a delicate issue for at least the following reasons. On the one hand, I don’t think that many people doing research in “modern artificial intelligence” are aware of the development of linguistics in general and in particular of formal linguistics, witness for example the scarcity of citations in many NLP papers: this is not necessarily a problem if by “modern artificial intelligence” one thinks of the effort to simulate human behaviour but it is so if the implicit intention is to understand both the formal properties of human languages and the way they are implemented in the brain. On the other, though, beside this necessary distinction between simulation and comprehension I think that if research in AI is based on pre-generative models of language the conclusion that descriptive notions such as those mentioned before, like relativisation or question formation, are far less adapt to be implemented in artificial circuitry is more than plausible.

References:

[1] Friederici et al. 2018. “Language, mind and brain”, Nature Human Behaviour, vol 1, pp. 713-722.

[2] Moro, A. 2010. A brief history of the verb “to be”, MIT press.

[3] Moro, A. 2000. Dynamic Antisymmetry, MIT Press.

[4] Moro, A. 2017. Impossible Languages, MIT Press.

[5] Moro, A. 1988. Per una teoria unificata delle frasi copulari, Rivista di Grammatica Generativa, 13, pp. 81-110.

[6] Moro, A. 2015. The Boundaries of Babel II edition, MIT press.

[7] Moro, A. 2013. The Equilibrium of Human Syntax. Symmetries in the brain, Leading Linguists Series, Routledge, New York.

[8] Pollock, J.-Y. 1989. “Verb Movement, UG, and the Structure of IP,”Linguistic Inquiry, 20, 365-424.

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：马晓彤 闫玉萌 陈金玉 王平

排版：马晓彤 闫玉萌 高洁

审校：王丽媛 李芳芳