There comes a time in every metaphysician’s life when she finally thinks—sure, maybe I should learn more about woodworking. She might then find herself reading something like Christopher Schwarz’s The Anarchist’s Workbench. In a chapter titled “All The Mistakes” Schwarz (2020: 47) reflects on attempts to design “a perfect bench,” starting with the honestly named $175 Workbench: The poor bench has changed so much during the last 19 years that it’s almost unrecognizable. Despite its oddness, I still love it…. If I had to rebuild this bench, it would be much the same, except I would remove all the knockdown hardware…. I wish the benchtop was longer (it’s 70″ long now—I don’t know why).This is a review of Bounds of Possibility—not of The Anarchist’s Workbench. But like Schwarz, the authors of Bounds of Possibility begin honestly: “This book is long. It could have been somewhat shorter. But it couldn’t have been just one sentence long.”They aren’t wrong. It is a long book, and perhaps it could have been written with less knockdown hardware. But, as it stands, Bounds of Possibility is a master class in dogged dedication to a puzzle, a meticulous and fantastically rewarding investigation that easily justifies the length.The focus is a family of arguments that purport to show that if some object (a workbench, a book) could have been different in some respect (longer, shorter), then it also could have been vastly different in that respect (700″, a single sentence). These Tolerance Arguments fit a canonical schema: Tolerance:a is tolerant.Non-contingency:If a is tolerant, then it is necessary that a is tolerant.Iteration:Whatever is possibly possible is possible.Persistent Closeness:When properties are close, they are necessarily close.Hypertolerance:a is hypertolerant. We generate instances by supplying some definition of ‘closeness’ for properties. An object is tolerant if it could have had any property close to a property it has, and it is hypertolerant if it could have had any property ancestrally close to a property it has. Since we can supply any binary relation between properties for ‘closeness’, some Tolerance Arguments aren’t interesting: sometimes Tolerance is obviously false, or Hypertolerance immediately plausible. The most puzzling arguments of this form—Tolerance Puzzles—challenge entrenched modal judgments, threatening the stability of a picture on which familiar bits of the world could have been otherwise but couldn’t have been just anywise.Take Woody. Woody the workbench is tolerant with respect to close lengths—it could have been slightly longer or shorter. But that doesn’t seem to turn on anything special about how Woody is; it is difficult to imagine how we could have made Woody so that it wasn’t tolerant! If Woody is necessarily tolerant, then for any length that Woody might have had it is possibly possible for Woody to have been slightly longer or shorter than that. Given Iteration, we are led to the uncomfortable conclusion that Woody is also hypertolerant: our trusty workbench could have been too long to fit in the driveway or too short to balance a toolbox.Bolstered by a primer in the first chapter (“Logical Tools”), Bounds of Possibility is an advertisement-by-example for the usefulness of higher-order languages in first-order metaphysics. Throughout, the authors are eager to put to rest “any nagging concern that one somehow needs to learn a lot of metatheory in order to do higher-order metaphysics” (52). The beginning of the book, however, is largely nontechnical. The authors introduce the puzzles (in chapter 2), paying special attention to the crucial but underexamined Non-contingency premise (in chapter 3) and to the relationship between Tolerance Puzzles and nearby “Coincidence Puzzles” (in chapter 4). Even taken alone, the first act of Bounds of Possibility is a clarifying intervention in the literature on identity, persistence, and material constitution. The authors are also careful throughout to quarantine likely distractions. They are, for example, emphatic about distinguishing Tolerance Puzzles from the Sorites, offering motivations for each of the key premises that “have nothing to do with the deeply problematic ‘small differences can’t matter’ idea” (79). The remainder of Bounds of Possibility is an original and richly detailed exploration of responses to Tolerance Arguments—accepting Hypertolerance, denying Iteration, and denying Non-contingency. The four chapters focused on Hypertolerance and Iteration are particularly ambitious, and together systematize a mess of threads through debates about essentialism, supervenience, and the nature of metaphysical necessity. It is no surprise, then, that this part of the book is challenging. The chief downside of this is that we don’t get to the authors’ own positive proposal until chapter 11. (Impatient readers should take the advice in the introduction: read chapters 2, 3, and 11 before returning to the exploration of alternative strategies.)Happily, the resolution is worth the wait. In the end the authors resist any fully general treatment of Tolerance Puzzles. Instead, they defend a two-piece account that throws the brakes on Non-contingency while allowing principled instances of Tolerance-denial or Hypertolerance. The account combines a dramatically plenitudinous ontology with an equally dramatic metasemantic proposal. On the metasemantic side, they argue for widespread semantic plasticity in terms like ‘this’, ‘Woody’, and ‘workbench’. Many of our expressions have semantic profiles that are massively sensitive to external factors, so what we assert using them can differ across modal circumstances in fine-grained ways. On the metaphysics side, they defend a form of plenitude that guarantees that wherever there is any material object, there is a “dizzying variety of other, coinciding material objects” (266). These coincidents also vary modally in fine-grained ways and so supply referents for semantically plastic expressions.To see how this helps with Tolerance Puzzles, we’ll have to dig into the details. There is far more in this book (in the footnotes alone!) than I can helpfully engage with here. Instead, I will focus on drawing out some key ideas from the positive account.If the carpenter had planned a bit better, Woody could have been slightly longer. But if Tolerance is contingent, had she done so, there might have been some ways of making a slightly longer workbench that wouldn’t have made Woody at all.Why is that so hard to face? One source of discomfort is the apparent threat to the security of ordinary tolerance judgments. When the carpenter remarks “I could have made this a bit longer,” she’s right to be confident—of course she could have! But if Woody very easily could have been intolerant, couldn’t the carpenter easily have been wrong? Given how slight the changes are that could have taken us “to the edge,” “if one takes Tolerance to be a contingent truth, one seems forced to think that one is just lucky not to be mistaken about it.” But our Tolerance judgments don’t “feel like a risky bet in the way this picture suggests” (89).This is the idea behind the security argument, which the authors argue in chapter 3 is the primary obstacle to denying Non-contingency. In chapter 11, we find that the problematic premise is Independence: Independence:If Tolerance could easily have been false, we could easily have falsely believed it. The difference between Woody being tolerant and Woody being intolerant can come down to minor or distant differences—differences in the position of the carpenter’s saw or the quantity of metal used months ago at the screw factory (92–94). Independence is motivated by the thought that those can’t be differences that our modal beliefs are sensitive to (90–91, 262–63).Here’s where semantic plasticity comes in. Independence is false: the carpenter couldn’t have easily been mistaken, because in the worrying circumstances she would have asserted some other proposition by saying “I could have made this a bit longer.” (Plenitude already guarantees a wealth of appropriately tolerant candidates for ‘this’.) Of course, not all of our tolerance judgments crucially involve demonstratives, so this can’t be the whole story. The much harder part is making sense of more general speeches, like ‘Every table you made today could have been made smoother by sanding it more carefully’ (6).Here, the authors distinguish Tolerance Puzzles from what they call Quantified Tolerance Puzzles (see esp. 2.2 and 11.5). These are structurally parallel arguments that instead begin with premises like: Table Tolerance:Every table is tolerantly a table. Where something is tolerantly a table if it could have had any ‘close’ property while still being a table. The authors favor a unified treatment of the main Tolerance Puzzles and the quantified versions. In this case: Table Tolerance is true but contingent. Again, we’re in no danger of making a mistake in nearby worlds. If there had been intolerant tables, the predicate ‘table’ in utterances like “This could have been a better table if we’d made it longer” would have picked out a different property. The authors readily admit that plasticity on this scale takes some getting used to. Thankfully, the convincing discussion of choice-points and alternatives in chapters 12 and 13 does a lot to build fluency.Still, I’m unsettled about the treatment of quantified puzzles. This isn’t because of any unwillingness to accept the more dramatic parts of their account—it may be that I’m tempted by something slightly more dramatic.When we reflect only on paradigm instances of predicates like ‘table’, it’s easy to conclude that part of what it is to be a table is to be modally flexible in certain unrepresentative respects. The authors thus warn that we should “steer clear of Tolerance premises about all tables which would seem plausible if we were thinking only of paradigmatic tables” but are still friendlier to generalizations in the neighborhood than I would expect (64). Here, Yablo (forthcoming) provides an especially helpful foil. He proposes an account that is similarly plenitudinous, but which would reject Table Tolerance. On that view, although every table could have been slightly longer, if it had been, it might have lost the modal flexibility required for tablehood. Yablo and the authors agree that every table is tolerant; they disagree about whether the tablehood of (actual) tables is modally robust (285–87).My hesitation is that plenty of familiar objects—tables, workbenches, books—already seem to be intolerantly “on their edge.” Imagine a raw-edged table made from a distinctively knotty piece of cherry wood. We might naturally insist: “That table could have been made shorter, but only if you’d kept that lovely pin knot on the edge.” The authors acknowledge similar examples at key points in their discussion but call them “rare” or “marginal and exotic” (83–84, 162–63). I worry that they aren’t exotic at all. When we think about particular tables (your first handmade workbench, your grandmother’s dining table, your ex’s wobbly IKEA Melltorp) it becomes much more plausible that ‘table’ fails to fix on objects with any obvious modal uniformity—undermining even the simpler claim that every table is tolerant. (The authors themselves float a number of interesting proposals in this vein. They’re tempted, for example, to say that ‘house’ expresses a property of objects with “attitude-sensitive” modal profiles, so which differences houses could tolerate depends on the attitudes of relevant agents [300–301].)Although this suggests a more cautious approach to instances of Quantified Tolerance Puzzles, I think it is very much in the spirit of the core account. Given that our modal judgments are in good standing and that (as I suspect) interesting generalizations of those judgments are very hard to come by, the combination of plenitude and widespread semantic plasticity seems by far the most promising story on offer. Of course, there’s far more to say. Unsurprisingly, given a source as rich as Bounds of Possibility: this review could have been much longer.

中文翻译：

---

可能性的界限：模态变化之谜

每个形而上学家的一生中都会有这样一个时刻，她最终会想——当然，也许我应该更多地了解木工。然后她可能会发现自己在读克里斯托弗·施瓦茨（Christopher Schwarz）的《无政府主义者的工作台》之类的书。在题为“所有错误”的一章中，Schwarz (2020: 47) 反思了设计“完美长凳”的尝试，从诚实命名的 175 美元工作台开始：可怜的长凳在过去 19 年里发生了很大变化，几乎无法辨认。尽管它很奇怪，但我仍然喜欢它......如果我必须重建这个长凳，那会是一样的，除了我会移除所有可拆卸的硬件……。我希望工作台更长一些（现在有 70 英寸长——我不知道为什么）。这是对《可能性的界限》的评论，而不是对《无政府主义者的工作台》的评论。但和施瓦茨一样，《可能性的界限》的作者也诚实地开头：“这本书很长。它本来可以更短一些。但它不可能只有一句话那么长。”他们没有错。这是一本很长的书，也许它可以用更少的硬件来写。但是，就目前而言，《可能性的界限》是一堂大师班，顽强地致力于解决一个难题，是一项细致且非常有益的调查，很容易证明其篇幅的合理性。重点是一系列论点，旨在表明如果某个对象（一个工作台） ，一本书）可能在某些方面有所不同（更长，更短），那么它也可能在这方面有很大的不同（700英寸，一个句子）。这些容忍参数符合规范模式： 容忍：a 是容忍的。非偶然性：如果 a 是容忍的，那么 a 是容忍的。迭代：任何可能的都是可能的。持久接近：当属性接近时，它们必然接近。 超耐受性：a 是超耐受性。我们通过为属性提供一些“接近度”定义来生成实例。如果一个对象可以拥有任何与其拥有的属性接近的属性，则该对象是宽容的；如果它可以拥有与其拥有的属性相近的任何祖先属性，则该对象是超宽容的。由于我们可以提供“接近”属性之间的任何二元关系，因此某些容忍度论证并不有趣：有时容忍度显然是错误的，或者过度容忍立即是合理的。这种形式中最令人费解的论点——宽容之谜——挑战了根深蒂固的模态判断，威胁到了一幅画面的稳定性，在这幅画面上，世界上熟悉的部分本来可以是其他的，但无论如何也不可能是这样。以伍迪为例。 Woody 工作台可以容忍较短的长度——它可以稍微长一些或短一些。但这似乎并没有说明伍迪有什么特别的地方。很难想象我们怎么能让伍迪变得不宽容！如果伍迪一定是宽容的，那么对于伍迪可能拥有的任何长度，伍迪都有可能比该长度稍长或稍短。给定迭代，我们得出了一个令人不安的结论，伍迪也是过度宽容的：我们值得信赖的工作台可能太长，无法适应车道，或者太短，无法平衡工具箱。由第一章（“逻辑工具”）中的入门知识支持，边界《可能性》是对高阶语言在一阶形而上学中有用性的举例广告。自始至终，作者都渴望消除“任何令人烦恼的担忧，即人们需要以某种方式学习大量元理论才能进行更高阶的形而上学”（52）。然而，本书的开头大部分都是非技术性的。作者介绍了这些谜题（在第 2 章中），特别关注关键但未被充分检验的非偶然前提（在第 3 章中）以及宽容谜题和附近的“巧合谜题”之间的关系（在第 4 章中）。即使单独来看，《可能性的界限》的第一个行动就是对有关身份、持久性和物质构成的文献进行澄清干预。作者自始至终都小心翼翼地隔离可能的干扰。例如，它们强调将宽容难题与算术题区分开来，为每个关键前提提供动机，这些前提“与存在严重问题的‘小差异无关紧要’的想法无关”（79）。 《可能性的界限》的其余部分是对容忍论据的回应的原创且详细的探索——接受过度容忍、否认迭代和否认非偶然性。专注于超宽容和迭代的四章尤其雄心勃勃，通过关于本质主义、附带性和形而上必然性本质的辩论，将混乱的线索系统化。因此，本书的这一部分具有挑战性也就不足为奇了。这样做的主要缺点是，直到第 11 章我们才得到作者自己的积极建议。（不耐烦的读者应该听取引言中的建议：在返回探索替代策略之前阅读第 2、3 和 11 章） .）令人高兴的是，该决议值得等待。最后，作者抵制对宽容难题进行任何完全一般化的处理。相反，他们捍卫一个由两部分组成的账户，即对非偶然性踩刹车，同时允许原则性的拒绝宽容或过度宽容的情况。该叙述结合了戏剧性的丰富本体论和同样戏剧性的元语义建议。在元语义方面，他们主张“这个”、“伍迪”和“工作台”等术语具有广泛的语义可塑性。我们的许多表达都具有对外部因素非常敏感的语义配置文件，因此我们断言使用它们的内容可能会在不同的模态环境中以细粒度的方式有所不同。在形而上学方面，他们捍卫一种丰富性形式，保证只要有任何物质对象，就存在“令人眼花缭乱的其他、重合的物质对象”（266）。这些重合也以细粒度的方式模态变化，因此为语义可塑表达提供了指称。要了解这对公差难题有何帮助，我们必须深入研究细节。本书中的内容（仅在脚注中！）远远超出了我在这里所能提供的帮助。相反，我将专注于从积极的叙述中提取一些关键想法。如果木匠计划得更好一点，伍迪可能会稍微长一点。但如果宽容是偶然的，如果她这样做了，可能会有一些方法来制作一个稍微长一点的工作台，而这根本不会让伍迪成为现实。为什么这这么难以面对呢？令人不安的根源之一是对普通宽容判断的安全性的明显威胁。当木匠说“我本可以把它做得更长一点”时，她的自信是正确的——她当然可以！但如果伍迪很容易不宽容，那么木匠就不会很容易犯错吗？考虑到这些微小的变化可能会把我们带到“边缘”，“如果一个人将宽容视为一个偶然的事实，那么人们似乎不得不认为，一个人没有犯错只是幸运的。”但我们的宽容判断并不“像这张图所暗示的那样，感觉像是一个冒险的赌注”（89）。这是安全论证背后的想法，作者在第三章中认为这是否认非偶然性的主要障碍。在第11章中，我们发现有问题的前提是独立：独立：如果宽容很容易是错误的，我们也很容易错误地相信它。伍迪的宽容和伍迪的不宽容之间的差异可以归结为微小或遥远的差异——木匠锯的位置或几个月前螺丝工厂使用的金属数量的差异(92-94)。独立性的动机是这样的想法：这些不可能是我们模态信念敏感的差异（90-91, 262-63）。这就是语义可塑性的用武之地。独立性是错误的：木匠不可能轻易被误解，因为在令人担忧的情况下，她可能会通过说“我可以把这个时间写得长一点”来提出其他主张。 （充足性已经保证了“这个”有大量适当宽容的候选者。）当然，并不是所有的宽容判断都关键地涉及指示词，所以这不可能是全部。更困难的部分是理解更一般性的演讲，比如“你今天制作的每一张桌子都可以通过更仔细地打磨而变得更光滑”(6)。在这里，作者将公差谜题与他们所谓的量化公差谜题区分开来（参见尤其是 2.2 和 11.5）。这些是结构上并行的论点，而是以以下前提开始： 表容差：每个表都是容差的表。如果某个东西在仍然是一张桌子的同时可以具有任何“关闭”属性，那么它就可以是一张桌子。作者赞成对主要的宽容难题和量化版本进行统一处理。在这种情况下： 表公差是真实的，但是偶然的。再说一次，我们在附近的世界中没有犯错误的危险。如果存在不宽容的表，那么“如果我们把它做得更长，这可能会是一个更好的表”这样的话语中的谓词“表”会选择不同的属性。作者欣然承认，这种规模的可塑性需要一些时间来适应。值得庆幸的是，第 12 章和第 13 章中对选择点和替代方案的令人信服的讨论对于提高流畅性有很大帮助。尽管如此，我对量化难题的处理仍然感到不安。这并不是因为不愿意接受他们的叙述中更具戏剧性的部分——可能是我被一些稍微更戏剧性的东西所诱惑。当我们只反思像“table”这样的谓词的范式实例时，很容易得出结论表格的一部分是在某些非代表性方面具有模态灵活性。因此，作者警告说，我们应该“避开所有表格的公差前提，如果我们只考虑范式表格，这似乎是合理的”，但对附近的概括仍然比我预期的更友好（64）。在这里，Yablo（即将出版）提供了一个特别有用的陪衬。他提出了一种同样丰富的解释，但拒绝了表容差。按照这种观点，虽然每个表格都可以稍微长一些，但如果是的话，它可能会失去表格所需的模式灵活性。 Yablo 和作者一致认为，每张桌子都是宽容的；他们对于（实际）桌子的桌子性是否在模态上稳健存在分歧（285-87）。我的犹豫是，许多熟悉的物体——桌子、工作台、书籍——似乎已经不宽容地“处于边缘”。想象一下一张由一块明显多节的樱桃木制成的毛边桌子。我们可能会自然地坚持：“那张桌子本来可以做得更短，但前提是你要在边缘保留那个可爱的小结。”作者在讨论的关键点上承认了类似的例子，但称它们为“罕见”或“边缘和外来”（83-84、162-63）。我担心它们根本不具有异国情调。当我们考虑特定的桌子（你的第一张手工制作的工作台、你祖母的餐桌、你前任的摇摇晃晃的宜家梅尔托普）时，“桌子”无法固定在具有任何明显模态一致性的物体上就变得更加合理——甚至破坏了每个简单的主张表是宽容的。 （作者自己在这方面提出了许多有趣的建议。例如，他们很想说“房子”表达了具有“态度敏感”模态轮廓的对象的属性，因此房子可以容忍的差异取决于相关主体的态度[300-301]。）尽管这表明对量化容忍难题的实例采取更加谨慎的方法，但我认为这非常符合核心帐户的精神。鉴于我们的模态判断具有良好的信誉，并且（正如我怀疑的那样）很难对这些判断进行有趣的概括，因此丰富性和广泛的语义可塑性的结合似乎是迄今为止最有希望的故事。当然，还有很多话要说。毫不奇怪，考虑到《可能性的界限》这样丰富的来源：这篇评论本来可以更长。