Let me start by thanking Professors Santorio and Szabo for their insightful comments and careful readings of Conjoining Meanings (CM). They highlight the main theses and describe the project in helpful ways.1 So instead of multiplying summaries, let me say a little about what motivated the project. Chomsky was influential. But teaching truth-theoretic semantics, for many years, really convinced me of the need for a different conception of linguistic meaning.

Names like ‘Fido’ and ‘Felix’ are said to be entity denoters of a basic semantic type <e>, and declarative sentences are said to be truth-value denoters of another basic type <t>. The idea is that ‘Fido’ and ‘Felix’ denote ordinary entities, perhaps a certain dog and a certain cat, while sentences denote truth or falsity.2 The tensed verb in (1) is then described as an expression—of a non-basic type <e, t>—that denotes a function that maps each entity to truth or falsity depending on whether or not that entity barked. The phrase ‘chased Felix’ is said to be an expression of the same type, corresponding to a function that maps each entity to truth or falsity depending on whether or not that entity chased Felix; and then ‘chased’ is said to be an expression—of the non-basic type <e, <e, t>>—that denotes a dyadic function that maps each entity, x, to a function that maps each entity, x', to truth or falsity depending on whether or not x' chased x.

The quantificational phrase ‘every dog’ gets described as an instance of type <<e, t>, t> that denotes a second-order function that maps each first-order function, F', of type <e, t> to truth or falsity depending on whether or not F' maps every dog to truth. Then ‘every’ is described as an expression of type <<e, t>, <<e, t>, t>> that denotes a dyadic second-order function that maps each function, F, of type <e, t> to a function of type <<e, t>, t> that maps each function, F', of type <e, t> to truth or falsity depending on whether or not every thing that F maps to truth is such that F' also maps that thing to truth.3 From here on, things get less pretty.

As the twists unfold, the value of appealing to Function Application becomes less and less clear, in part because the corresponding hierarchy of types—<e>, <t>, and <α, β> if <α> and <β> are types—seems so needlessly vast and powerful. Moreover, a different picture of composition begins to emerge, at least if one squints a bit; and given this picture, developed in CM, one need not adopt the hypothesis that ordinary sentences like (1-11) have truth conditions. At which point, it seems unreasonable to keep setting aside arguments against this hypothesis. In the book, I try to spell this out in detail. Here, let me just indicate the motivating line of thought.

One can maintain that in (4), ‘every cat’ is of type <<e, t>, t> by treating ‘Fido chased’ as a device for mapping each entity to truth if and only if Fido chased it. The idea is that ‘every cat’ combines with an analog of the relative clause ‘which Fido chased’, which can be treated as an expression of type <e, t> formed by abstracting ‘which’ from the direct object position of ‘chased’; see, e.g., Heim and Kratzer (1998). This highlights the question, discussed in CM, of why ‘every cat which Fido chased’ cannot be understood as a sentence that is true if and only if every cat is such that Fido chased it. But even if there is a good answer, the meanings of (4) and ‘which Fido chased’ are not determined solely by Function Application.

One way or another, neo-Fregean accounts of relative clauses and quantificational direct objects invoke a (syncategorematic) rule that permits conversion of an open sentence of type <t> into a predicate of type <e, t>. Not that there's anything wrong with that. I posit an analogous rule in CM. I also review the reasons for positing a conjunctive rule of composition for phrases formed by combining nouns or verbs with modifiers, as in ‘grey cat (which/that) Fido chased’ and ‘dog that barked loudly today’. It has also become common to posit existential closures of predicates, as suggested by Davidson (1967b) and many others. Function Application is not the only rule invoked these days. So we can ask if it should be invoked at all, given event analyses.

The idea that ‘barked’ is of type <e, t> often gets modified in favor of saying that the intransitive verb is a semantically dyadic expression—of type <e, <e, t>>—that corresponds to a function that maps each entity, x, to a function that maps each event, x', to truth or falsity depending on whether or not x' was a barking by x. In which case, combining the verb with ‘Fido’ yields an expression corresponding to a function that maps events to truth values. This permits conjunctive composition with other such functions, like those corresponding to ‘loudly’ and ‘today’. But to get a plausible truth condition for (5), existential closure is needed.

Moreover, ‘Fido bark’ appears as the direct object of ‘heard’ in (6), which means roughly that Felix heard an event of Fido barking. So it seems that covert existential closure can be triggered by combining a verb with an untensed clause, and without yielding a truth-evaluable sentence; see Higginbotham (1983). There are good reasons for also positing such closure for the noun in (7), as suggested by Kamp (1981) and Heim (1982). The indefinite article ‘a’ is plausibly an English grammatical marker of the singular count noun, as opposed to a quantifier of type <<e, t>, <<e, t>, t>> that needs to combine with ‘cat’ and then ‘Fido chased’.

Discussions of causative and passive constructions, as in (9) and (10), play a large role in the relevant literature because such constructions invite a cluster of related questions. Why does the first conjunct of (9) imply the second? Why does each conjunct of (10) imply the other? And why is the subject of a transitive verb so often associated with being the agent or experiencer of an event, as opposed to being the patient or theme? If the verb in the first conjunct of (9) denotes the triadic function λx.λx'.λe.BreakByOf(e, x', x), then why are there no corresponding verbs in other Slangs that denote the “thematically inverted” function λx.λx'.λe.BreakOfBy(e, x', x)?

In many Slangs, causative verbs are formed by adding an overt morpheme to intransitive verbs. Studies of this phenomenon revitalized the old idea that a clause like ‘Felix break a vase’ includes two verbal elements that have combined; see Baker (1988) on incorporation. On this view, the atomic verb 'break' takes a single argument that corresponds to the things that break in events of breaking, and agent positions are associated with a covert morpheme akin to the light verb ‘make’ in ‘make a baby smile’. In which case, the subject and object of ‘Felix break a vase’ are understood as parts of conjoinable event predicates like ‘∃x[AgentOf(e, x) & Felix(x)]’ and ‘TerminatesIn(e, f) & ∃x[BreakOf(f, x) & Vase(x)]’. If ‘Felix’ is understood this way in (9), it is unsurprising if ‘Felix’ is understood the same way in the prepositional phrase in (10). The indirect object of (11) can also be treated like the prepositional phrase in ‘Hubbard gave a bone to Fido’, especially in light of Larson's (1988, 2014) discussion of ditransitive constructions.

Given evidence that these ideas are on the right track, semanticists should at least consider the possibility that while Slang lexical items can be semantically dyadic, polyadicity is unattested—or at least rare—because the operative compositional principles for phrasal syntax are geared to monadic predicates, and relationality is introduced via semantically dyadic lexical items that immediately combine with grammatical arguments (whose positions get existentially closed) to form monadic predicates that can be conjoined with others. CM explores this possibility from both formal and cognitive perspectives.

On the technical front, I show how this typologically spare conception of meaning can handle the standard textbook constructions and more, without appeal to Function Application. This homework exercise doesn't show that the proposal is correct. But advocates of more familiar views are equally in the position of having a proposal that is not the only game in town. So arguments against appeal to Function Application and Fregean typology should be considered without prejudice. I suspect that when people talk about how impressive standard theories of meaning are, they are often thinking about the framework they were initially offered for sentences like (1-3)—while forgetting or downplaying the ways in which that framework creaks, under the pressure of constructions that get considered later—and not considering alternative models of composition. So I offered a specific alternative that can be modified in various ways.

Taken together, these considerations highlight the questions of whether appeals to Function Application are needed or wanted in theories of meaning for Slangs—and if not, whether we should maintain the assumption that sentences like (1-12) have truth conditions. In my replies to Santorio and Szabo, I'll focus more on my general skepticism about attempts to characterize meaning in terms of truth. But I suspect that characterizing truth, even for the restricted range of thoughts that are naturally available to members of our species, will require appeal to many polyadic notions. (Modern logic is, for good reason, deeply relational.) In this sense, my spare conception of natural semantic composition is interwoven with my internalist conception of Slang meanings as instructions for how to assemble concepts of a limited sort.5

Davidsonians, who also eschew appeal to Function Application, may feel unfairly left out of the discussion so far. But as I try to show in CM, the empirical issues—concerning human cognition and semantic composition in Slangs—remain largely the same if talk of verbs denoting (potentially polyadic) functions is replaced with talk of verbs being satisfied by sequences of entities. And with few exceptions (e.g., Larson and Segal 1995), textbook details tend to look more like Lewis (1970) or Montague (1974) than Davidson (1967a) or Tarski (1944).

That said, let me end this introduction with an opinionated reminder about the field's history. Current theories of linguistic meaning descend from Davidson and Lewis, whose proposals emerged in an environment of “Semantic Quineanism”: a form of extensionalism that was motivated by an odd mix—prevalent at Harvard in the 1960s—of externalism, a kind of nominalism, residual behaviorism that often morphed into instrumentalism about psychology, admiration of Tarski combined with a fondness for eliminativism with regard to meanings, and a tendency to treat regimentation as description. Theories born in this setting may need revision.

首先，我要感谢 Santorio 教授和 Szabo 教授对Conjoining Meanings ( CM ) 的深刻评论和仔细阅读。他们突出了主要论点并以有用的方式描述了该项目。1 因此，让我来谈谈是什么激发了该项目，而不是增加摘要。乔姆斯基很有影响力。但是多年来，教授真论语义学，确实让我相信需要一种不同的语言意义概念。

像“Fido”和“Felix”这样的名字被认为是基本语义类型 <e> 的实体指示符，而陈述句被认为是另一种基本类型 <t> 的真值指示符。这个想法是，“Fido”和“Felix”表示普通实体，也许是某只狗和某只猫，而句子则表示真或假。2 (1) 中的时态动词然后被描述为非-基本类型 <e, t>——表示一个函数根据实体是否吠叫，将每个实体映射到真或假。短语“chased Felix”据说是同一类型的表达式，对应于根据实体是否追逐 Felix 将每个实体映射到真或假的函数；然后 'chased' 被称为非基本类型 <e, <e, t>> 的表达式，它表示将每个实体 x 映射到映射每个实体 x' 的函数的二元函数, 真假取决于 x' 是否追逐 x。

量化短语 'every dog' 被描述为 <<e, t>, t> 类型的实例，它表示将每个 <e, t> 类型的一阶函数 F' 映射到真值的二阶函数或虚假取决于 F' 是否将每只狗映射到真理。然后 'every' 被描述为 <<e, t>, <<e, t>, t>> 类型的表达式，它表示映射每个 <e, t> 类型的函数 F 的二元二阶函数到 <<e, t>, t> 类型的函数，它将每个 <e, t> 类型的函数 F' 映射到真值或假值，这取决于 F 映射到真值的每件事是否满足 F'也将那个东西映射到真相。3 从这里开始，事情变得不那么美好了。

随着曲折的展开，诉诸函数应用的价值变得越来越模糊，部分原因是相应的类型层次结构——<e>、<t> 和 <α, β> 如果 <α> 和 <β> 是类型——看起来如此庞大和强大。此外，一幅不同的构图开始出现，至少如果你稍微眯起眼睛的话；鉴于这张在CM中发展的图片，人们无需采用像 (1-11) 这样的普通句子具有真值条件的假设。在这一点上，继续搁置反对这一假设的论据似乎是不合理的。在书中，我试图详细说明这一点。在这里，让我简单地指出一下激励思路。

人们可以通过将“Fido chased”视为当且仅当 Fido 追逐它时将每个实体映射到真实的设备来维护 (4) 中的“每只猫”都是 <<e, t>, t> 类型。这个想法是'every cat'结合了关系从句'which Fido chased'的类似物，它可以被视为通过从'chased的直接宾语位置抽象'which'形成的<e，t>类型的表达式'; 参见，例如，Heim 和 Kratzer (1998)。这突出了CM中讨论的问题，即为什么“Fido 追逐的每只猫”不能被理解为当且仅当每只猫都是 Fido 追逐它时为真的句子。但即使有一个很好的答案，（4）和“哪个 Fido 追逐”的含义也不仅仅是由函数应用决定的。

不管怎样，新弗雷格对关系从句和量化直接宾语的解释调用了一个（syncategorematic）规则，该规则允许将<t> 类型的开句转换为 <e, t> 类型的谓词。并不是说这有什么问题。我在CM中提出了一个类似的规则。我还回顾了为由名词或动词与修饰语组合而成的短语设定连词构成规则的原因，例如“grey cat (which/that) Fido chased”和“dog that barking loudly today”。正如 Davidson (1967b) 和许多其他人所建议的那样，假设谓词的存在闭包也变得很普遍。函数应用程序并不是最近调用的唯一规则。因此，我们可以根据事件分析询问是否应该调用它。

'barked' 属于 <e, t> 类型的想法经常被修改为支持说不及物动词是语义二元表达式 - 类型 <e, <e, t>> - 对应于映射的函数每个实体 x 到一个函数，该函数将每个事件x' 映射到真或假，这取决于 x' 是否是 x 的吠叫。在这种情况下，将动词与“Fido”组合会产生一个表达式，对应于将事件映射到真值的函数。这允许与其他此类功能进行连词组合，例如对应于“大声”和“今天”的功能。但要获得 (5) 的似是而非的真值条件，需要存在闭包。

此外，'Fido bark'作为(6)中'heard'的直接宾语出现，大致意思是Felix听到了Fido barking的事件。因此，似乎可以通过将动词与不紧张的从句组合来触发隐蔽的存在闭包，而不会产生可评估真值的句子；参见 Higginbotham (1983)。正如 Kamp (1981) 和 Heim (1982) 所建议的那样，也有充分的理由为 (7) 中的名词设置这样的闭包。不定冠词“a”可能是单数可数名词的英语语法标记，而不是需要与“cat”组合的 <<e, t>, <<e, t>, t>> 类型的量词然后是“Fido 追逐”。

使役结构和被动结构的讨论，如 (9) 和 (10) 中，在相关文献中起着重要作用，因为此类结构引发了一系列相关问题。为什么（9）的第一个连词暗示第二个？为什么 (10) 的每个连词都暗示另一个？为什么及物动词的主语经常与事件的代理人或经历者联系在一起，而不是被访者或主题？如果(9)的第一个连词中的动词表示三元函数λx.λx'.λe.BreakByOf(e, x', x)，那么为什么其他俚语中没有相应的动词表示“主题反转”函数λx.λx'.λe.BreakOfBy(e, x', x)？

在许多俚语中，使役动词是通过向不及物动词添加显性语素而形成的。对这种现象的研究使旧观念重新焕发活力，即像“Felix break a vase”这样的从句包含两个组合的动词元素；参见 Baker (1988) 关于合并的内容。按照这种观点，原子动词“break”采用单个参数，对应于 breaking 事件中的 break 事物，并且代理位置与类似于“make a baby smile”中的轻动词“make”的隐蔽语素相关联. 在这种情况下，“Felix break a vase”的主语和宾语被理解为可连接事件谓词的一部分，例如“∃x[AgentOf(e, x) & Felix(x)]”和“TerminatesIn(e, f) &” ∃x[BreakOf(f, x) & Vase(x)]'。如果在 (9) 中这样理解 'Felix'，如果在 (10) 中的介词短语中以相同方式理解“Felix”也就不足为奇了。(11) 的间接宾语也可以像“Hubbard gave a bone to Fido”中的介词短语一样对待，特别是根据 Larson (1988, 2014) 对双及物结构的讨论。

鉴于有证据表明这些想法是正确的，语义学家至少应该考虑这样一种可能性，即虽然俚语词项在语义上可以是二元的，但多元性是未经证实的——或者至少是罕见的——因为短语句法的操作组合原则适用于单元谓词，关系性是通过语义二元词汇项引入的，这些词项立即与语法参数（其位置在存在上封闭）结合形成可以与其他谓词结合的单子谓词。CM从形式和认知的角度探索这种可能性。

在技​​术方面，我展示了这种类型学上的备用意义概念如何处理标准教科书结构等，而无需诉诸功能应用。这个家庭作业并没有表明这个建议是正确的。但是，更熟悉的观点的拥护者同样处于提出一个不是镇上唯一游戏的提案的位置。因此，应该在不带偏见的情况下考虑反对诉诸函数应用和弗雷格类型学的论点。我怀疑当人们谈论意义的标准理论是多么令人印象深刻时，他们通常会考虑最初为 (1-3) 等句子提供的框架——同时忘记或淡化该框架在压力下崩溃的方式稍后考虑的结构——而不是考虑替代的组合模型。

综上所述，这些考虑突出了以下问题：在俚语的意义理论中是否需要或想要诉诸功能应用——如果不需要，我们是否应该维持像 (1-12) 这样的句子具有真值条件的假设。在我对 Santorio 和 Szabo 的回复中，我将更多地关注我对用真理来描述意义的尝试的普遍怀疑。但我怀疑，即使对于我们物种成员自然可用的有限范围的思想，表征真理也需要诉诸许多多元概念。（现代逻辑，有充分的理由，是密切相关的。）从这个意义上说，我对自然语义组合的备用概念与我对俚语含义的内在主义概念交织在一起，作为如何组合语言概念的说明有限排序.5

戴维森主义者也回避诉诸函数应用，他们可能会觉得到目前为止被排除在讨论之外是不公平的。但正如我试图在CM中展示的那样，如果将表示（可能是多元的）功能的动词替换为被实体序列满足的动词，那么经验问题——关于人类认知和俚语中的语义构成——在很大程度上保持不变。除了少数例外（例如，Larson 和 Segal 1995），教科书的细节往往看起来更像 Lewis（1970）或 Montague（1974），而不是 Davidson（1967a）或 Tarski（1944）。

也就是说，让我以一个关于该领域历史的自以为是的提醒来结束本介绍。当前的语言意义理论源于戴维森和刘易斯，他们的提议出现在“语义奎因主义”的环境中：一种外延主义形式，其动机是一种奇怪的混合——在 1960 年代哈佛流行——外在主义，一种唯名论，残留的行为主义经常演变成关于心理学的工具主义，对塔斯基的钦佩加上对意义取消主义的喜爱，以及将军团化视为描述的倾向。在这种情况下诞生的理论可能需要修改。