literal or complex/simple. It struggles with the imperative to reduce data to second-order classifications, which is necessary for statistical analysis ofmultidimensional data. It is axiomatically true that, as Andrew Piper writes, “generalization is a crucial aspect to any scholarly method,” but it is CLS itself that has an issue with the science of generalization. It has to abstract what it is looking for into features that occur enough times that any meaningful data analysis can be had. In raising these structural issues for CLS, it is suggested that I am doubly mistaken—confused about data science and confused about literary studies. After all, Underwood writes, literary scholars themselves “can rarely give all the phenomena that interest them a causal explanation” (p. 903). This is tricky because the art of mere description is a robust and time-honored practice in literary studies. The description of textual data is so powerfully discerning that it becomes new knowledge or a new aesthetic object. Deconstruction, historicism, formal analysis, philological and biographical criticism: exemplary cases share the ability to redescribe that which the critic has seen but others have not seen or not seen and heard described in quite this way. 18. Peter Bruce and Andrew Bruce, Practical Statistics for Data Scientists (Boston, 2017), p. 2. 19. Andrew Piper, Enumerations: Data and Literary Study (Chicago, 2018), p. xi. 922 Nan Z. Da / Critical Response This leadsme to the last kind of redundancy—call it argumentative redundancy or evidentiary repetitiousness—that confuses our ability to evaluate CLS work. In defense of the statistical humanities, Scott Weingart writes: If you claim computational approaches to history (“digital history”) lets historians ask new types of questions, or that they offer new historical approaches to answering or exploring old questions, you are wrong. . . . After decades, or even centuries, of historical work in this area, there will always be examples of historians already having made my claims. My contribution is the bolstering of a particular viewpoint, the expansion of its applicability, the reframing of a discussion. This appears at first glance to be a supremely reasonable plea for the use of quantitative methods in the humanities. After all, most of us are corroborating some known existing truth when we produce scholarship, whether it’s the view of a particular school of thought or the deft turns of a favorite colleague or mentor. We give proof to each other’s intutions. Sometimes literary critics who feel the insights delivered by CLS to be grossly obvious are assured that reductiveness and redundancy are the price we pay for scope and scale—that obviousness in hindsight is different than obviousness period and that unsurprising results ought to be taken seriously because confirming what we already know is a prerequisite for establishing trust. Leaving aside the fact that this makes literary studies a discipline in which reductive obviousness—as defined by disciplinary standards—is not enough to discredit a piece of research, such assurances also conflate various forms of obviousness. “You’re just telling us what you already know” is not an adequate rebuttal to CLS, but only because it is too imprecise. There are many ways to tell people what they already know, some of which are invaluable, some of which are not only useless but misleading or even damaging. Some insights are so intelligent, so counterintuitive, or else so deeply intuitive that they must be told or shown repeatedly, in different ways, because human minds default to more simplistic truths. In these situations, argumentative redundancy does no disservice to the lesson or the audience. The best literary criticism often tells us what we already felt to be true but hadn’t considered fully or at a particular level of sophistication. Such insights are there, ready to be identified, yet nonetheless difficult to grasp. This is not the obviousness or redundancy in which CLS deals. As literary criticism, CLS describes phenomena that have been reduced to unfalsifiable truisms or unacceptable proxies 20. Scott B. Weingart, “‘Digital History’ Can Never Be New,” The Scottbott Irregular, scottbot.net/digital-history-can-never-be-new/ Critical Inquiry / Summer 2020 923 (quantitative measurements for figuration, character, metaphor, literariness, complexity, and so forth that are not good proxies). Finally, a word on retreating to nomenclature—distant reading or cultural analytics instead of CLS—in the face of immanent critique. First, cultural analytics. I am not a cultural-studies scholar, and I await such a scholar to challenge this appropriation of cultural analysis. Suffice it to say, cultural studies and literary studies are equally vulnerable to CLS’s modes of argumentation. Both have a hard time enforcing standards on something that is strictly literalist when it wishes to claim a method, a term, or the mantle of scientific objectivity but excessively analogical when its quantitative analysis has not effectively turned up any results. One reason that word clouds can pass as topics or character traits—or that the time variation ofmoney words in novels can pass as historicism—is because of their resemblance to theme or motif studies, by which some cluster of words is tracked across literary history. This is a reductive realization of Fredric Jameson’s cultural materialism and Raymond Wil liams’s keywords, two staples of cultural studies that were of course never meant to be inert categories, much less word clouds. Second, distant reading. Even though one often sees nowadays that literary studies will be conducted at different scales of reading as a reflex or as a show of disciplinary etiquette, much of the datawork being done on larger corpora cannot be called a shift in scale (that is, preserving an essential relationship after a change to the parameters). To recognize something as an allegory even though it has none of allegory’s typical motifs and comes in an entirely different length because it preserves the logic of allegory is what constitutes an act of scaling. This is what structuralism (or even just theory) does. By a different logic, the preservation of mathematically essential relationships in typological transformations also constitutes working to or at scale. The use of topological tools—for example, to simply resdescribe the data, analogizing nodes to social connectivity, influence, or variations of the same—is not working at a higher scale. Using a weak or limitedmodel tomake sweeping generalizations about a lot of things is not working at a higher scale. These considerations should carry forward, as CLS inevitably moves to literary history and publishing history as a workaround for the problem of small corpora. My goal in writing “The Computational Case against Computational Literary Studies” was to bring the problems in a disciplinary method heavily invested in cross-application, rhetoric, facile associationism, and semantic fudges to a head, to show how the languages of respective donor disciplines are used to hide glaring fallacies and to excuse the absence of import and rigor in the work. These disciplines might be reconvened, I suggest, to tackle what are traditionally called infelicitous arguments—arguments that are not necessarily wrong but for which judgment has been made infelicitous, there 924 Nan Z. Da / Critical Response being no forum, no possible way, given current disciplinary and social formations, to articulate skepticism. These are weak arguments—in principle, abjured by all disciplines—that have been made unfalsifiable. My hope is that I’ve emboldened literary scholars and interested parties to ask questions predicated on non-domain-specific logical reasoning and to ask where these methods are meeting literary-critical or literary-historical arguments.

中文翻译：

---

关键响应 III. 关于 EDA、复杂性和冗余：对 Underwood 和 Weatherby 的回应

文字或复杂/简单。它挣扎于将数据减少到二阶分类的必要性，这对于多维数据的统计分析是必要的。正如 Andrew Piper 所写，“泛化是任何学术方法的一个关键方面”，这是不言而喻的，但 CLS 本身在泛化科学方面存在问题。它必须将它正在寻找的特征抽象为出现次数足够多的特征，以便进行任何有意义的数据分析。在为 CLS 提出这些结构性问题时，表明我是双重错误的——对数据科学感到困惑，对文学研究感到困惑。毕竟，安德伍德写道，文学学者自己“很少能对他们感兴趣的所有现象给出因果解释”（第 903 页）。这很棘手，因为单纯的描述艺术在文学研究中是一种强大而历史悠久的实践。文本数据的描述是如此强大，以至于它成为新知识或新的审美对象。解构主义、历史主义、形式分析、语言学和传记批评：典型案例具有重新描述批评家所看到但其他人没有看到或没有看到和听到的以这种方式描述的内容的能力。18. Peter Bruce 和 Andrew Bruce，数据科学家的实用统计（波士顿，2017 年），第 1 页。2. 19. Andrew Piper，《枚举：数据和文学研究》（芝加哥，2018 年），第 3 页。十一. 922 Nan Z. Da / 批判性回应 这导致了最后一种冗余——称之为论证冗余或证据重复——混淆了我们评估 CLS 工作的能力。Scott Weingart 为统计人文学科辩护：如果你声称历史的计算方法（“数字历史”）可以让历史学家提出新类型的问题，或者他们提供新的历史方法来回答或探索旧问题，那你就错了。. . . 在这一领域进行了数十年甚至数百年的历史工作之后，总会有历史学家已经提出我的主张的例子。我的贡献是支持特定观点，扩展其适用性，重新构建讨论。乍一看，这似乎是在人文学科中使用定量方法的极其合理的请求。毕竟，我们大多数人在产生学术研究时都在证实一些已知的现有真理，无论是特定学派的观点还是最喜欢的同事或导师的灵巧转变。我们证明了彼此的直觉。有时，认为 CLS 所提供的见解非常明显的文学评论家确信，还原性和冗余性是我们为范围和规模付出的代价——事后的显而易见性不同于显而易见的时期，应该认真对待并不令人惊讶的结果，因为确认我们已经知道的是建立信任的先决条件。撇开这样一个事实，即这使得文学研究成为一门学科，其中还原性的显而易见性——按照学科标准的定义——不足以诋毁一项研究，这种保证也将各种形式的显而易见性混为一谈。“你只是告诉我们你已经知道的”并不足以反驳 CLS，但只是因为它太不精确了。有很多方法可以告诉人们他们已经知道的东西，其中一些是无价的，其中一些不仅无用，而且具有误导性甚至破坏性。有些洞察力非常聪明，非常违反直觉，或者非常直观，以至于必须以不同的方式反复讲述或展示它们，因为人类的思想默认了更简单的真理。在这些情况下，论证冗余不会损害课程或听众。最好的文学批评通常会告诉我们什么我们已经觉得是真实的，但还没有完全考虑或在特定的复杂程度。这样的见解就在那里，随时可以被识别，但仍然难以掌握。这不是 CLS 交易的明显性或冗余性。作为文学批评，CLS 描述了已被简化为不可证伪的真理或不可接受的代理的现象 20. Scott B. Weingart，“'数字历史'永远不会是新的，”斯科特不规则，scottbot.net/digital-history-can-never-be-new / 批判性调查 / 2020 年夏季 923（对形象、性格、隐喻、文学性、复杂性等的定量测量，这些都不是很好的代理）。最后，在面对内在的批评时，要退回到命名法——远距离阅读或文化分析而不是 CLS。首先，文化分析。我不是文化研究学者，我等待这样的学者挑战这种文化分析的挪用。可以说，文化研究和文学研究同样容易受到 CLS 的论证模式的影响。当他们希望声称一种方法、一个术语或科学客观性的外衣时，两者都很难对严格的字面主义执行标准，但当其定量分析没有有效地得出任何结果时又过于类比。词云可以作为主题或人物特征传递的一个原因——或者小说中金钱词的时间变化可以作为历史主义传递——是因为它们与主题或母题研究的相似性，通过这种研究可以在整个文学史上追踪一些词组。这是对 Fredric Jameson 的文化唯物主义和 Raymond Wil liams 的关键词的还原认识，这两个文化研究的主要内容当然从来都不是惰性类别，更不用说词云了。第二，远距离阅读。尽管现在人们经常看到，文学研究将在不同的阅读尺度上进行，作为一种反射或作为纪律礼仪的展示，但在更大的语料库上进行的许多数据工作不能称为尺度上的转变（即，保留一个参数更改后的基本关系）。将某物识别为寓言，即使它没有寓言的典型主题并且长度完全不同，因为它保留了寓言的逻辑，这构成了缩放行为。这就是结构主义（甚至只是理论）所做的。按照不同的逻辑，在类型转换中保留数学上的基本关系也构成了按规模工作或按规模工作。拓扑工具的使用——例如，简单地重新描述数据，将节点类比为社会连通性、影响力或相同的变体——并不能在更高的规模上发挥作用。使用弱模型或有限模型对很多事情进行全面概括并不能在更高的范围内发挥作用。这些考虑应该继续下去，因为 CLS 不可避免地转向文学史和出版史，作为解决小语料库问题的方法。我写“反对计算文学研究的计算案例”的目标是将大量投资于交叉应用、修辞、简单联想和语义模糊的学科方法中的问题带到头脑中，以展示各个捐助学科的语言如何被用来隐藏明显的谬误，并为工作中缺乏重要性和严谨性找借口。这些学科可能会重新召集，我建议，为了解决传统上所谓的不恰当的论点——不一定是错误的论点，但对其作出的判断是不恰当的，有 924 Nan Z. Da / 批判性回应是没有论坛，没有可能的方式，鉴于当前的纪律和社会形态，来表达怀疑论。这些都是弱论点——原则上被所有学科所否定——已经被证明是不可证伪的。我的希望是，我鼓励文学学者和感兴趣的团体提出基于非特定领域逻辑推理的问题，并询问这些方法在哪里遇到文学批评或文学历史论点。鉴于当前的学科和社会形态，以表达怀疑态度。这些都是弱论点——原则上被所有学科所否定——已经被证明是不可证伪的。我的希望是，我鼓励文学学者和感兴趣的团体提出基于非特定领域逻辑推理的问题，并询问这些方法在哪里遇到文学批评或文学历史论点。鉴于当前的学科和社会形态，以表达怀疑态度。这些都是弱论点——原则上被所有学科所否定——已经被证明是不可证伪的。我的希望是，我鼓励文学学者和感兴趣的团体提出基于非特定领域逻辑推理的问题，并询问这些方法在哪里遇到文学批评或文学历史论点。