Comparative constructions like More people have been to Russia than I have are reported to be acceptable and meaningful by native speakers of English; yet, upon closer reflection, they are judged to be incoherent. This mismatch between initial perception and more considered judgment challenges the idea that we perceive sentences veridically, and interpret them fully; it is thus potentially revealing about the relationship between grammar and language processing. This paper presents the results of the first detailed investigation of these so-called ‘comparative illusions’. We test four hypotheses about their source: a shallow syntactic parser, some type of repair by ellipsis, an incorrectly-resolved lexical ambiguity, or a persistent event comparison interpretation. Two formal acceptability studies show that speakers are most prone to the illusion when the matrix clause supports an event comparison reading. A verbatim recall task tests and finds evidence for such construals in speakers’ recollections of the sentences. We suggest that this reflects speakers’ entertaining an interpretation that is initially consistent with the sentence, but failing to notice when this interpretation becomes unavailable at the than-clause. In particular, semantic knowledge blinds people to an illicit operator-variable configuration in the syntax. Rather than illustrating processing in the absence of grammatical analysis, comparative illusions thus underscore the importance of syntactic and semantic rules in sentence processing. 1 COMPARATIVE ILLUSIONS Presented with the sentence in (1), native English speakers typically report that it is a perfectly acceptable sentence of their language. Yet, upon closer reflection, these same speakers judge that it has no stable, meaningful interpretation. Sentences of this form have come to be called ‘comparative illusions’ (CIs) or ‘Escher sentences’: they have only the appearance of well-formedness. CIs are interesting in that they seem to challenge some of our most basic assumptions about language architecture: that we perceive sentences veridically, that we interpret them fully, and that sentence form and meaning are tightly coupled. (1) More people have been to Russia than I have. © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com Downloaded from https://academic.oup.com/jos/advance-article-abstract/doi/10.1093/jos/ffy014/5065172 by guest on 04 August 2018 2 Alexis Wellwood et al. The phenomenon has been known for some time, but the mismatch between the perception of grammaticality and meaningfulness that characterizes CIs has so far received little systematic investigation. The sentence in (1) was first reported by Montalbetti (1984) as ‘the most amazing */? sentence I’ve ever heard’, attributing it to Hermann Schultze. Importantly, linguists and non-linguists alike experience the phenomenon, but, despite much informal discussion in the linguistics community, formal investigation has so far been limited to preliminary results (Fults & Phillips 2004, Wellwood et al. 2009, O’Connor et al. 2012; O’Connor 2015).1 In this paper, we investigate which properties of sentences like (1) are essential for the initial perception of meaningfulness. Grammatically, the problem with CI-type sentences is in the choice of subject in the than-clause, since superficially similar comparatives succeed in being both uncontroversially acceptable and meaningful. The meaning of a sentence like (2) just is, ‘the number of people that have been to Russia exceeds the number of elephants that have’. Yet there is no interpretation of (1) suggested by a similar paraphrase, ‘the number of people that have been to Russia exceeds the number of me’. (2) More people have been to Russia than elephants have. Deriving the interpretation of (2) involves mapping the individuals satisfying the matrix and embedded predicates to degrees representing their number, and establishing whether the first number is greater than the second. In the syntactic tradition going back at least to Bresnan (1973) (see also Chomsky 1977), degrees are introduced by the many component of more (i.e., more is underlyingly many and -er). The degree predicates are derived in tandem with a wh-operator that binds a variable in the abstract syntax of the than-clause, as in (3). This operator is akin to how many in (4a). It needs to combine with a bare plural NP, just like how many does, (4b)-(4d). (3) ... than wh-d ... d-many elephants have been to Russia (4) a. How many elephants have been to Russia? b. *How many I have been to Russia? c. *How many the elephant has been to Russia? d. *How many the elephants have been to Russia? Semantically, this binding relation corresponds to a λ-abstraction (see especially Heim & Kratzer 1998) over degrees, (5a). A parallel degree predicate is derived in the matrix clause by quantifier raising the morpheme -er, delivering an LF like that in (5b). Together, these two predicates act as the restrictor and scope arguments for the degree quantifier -er as in (6) (Heim 2000).2 Thus, the LF in (5b) is interpreted as a greater-than comparison between the maximal degrees that satisfy the degree descriptions in the main and than-clauses (i.e., max(Q) and max(P) in (6), respectively). 1 Our early results, reported in Wellwood et al. 2009, inform the present manuscript and have shaped the subsequent literature. A report on these results can be downloaded from https://github.com/alexiswellwood/compillu. 2 For further details concerning the LF syntax of comparative sentences, see Heim 1985, 2000; Bhatt & Pancheva 2004, among many others. There have been several alternative characterizations of the precise semantics of -er , in particular the tradition following Bartsch & Vennemann 1972 and Kennedy 1999. For our purposes these differences are not important. Downloaded from https://academic.oup.com/jos/advance-article-abstract/doi/10.1093/jos/ffy014/5065172 by guest on 04 August 2018 The Anatomy of a Comparative Illusion 3 (5) a. ... than λd ... d-many elephants have been to Russia b. -er [ λd ... d-many people have been to Russia [ (than) λd ... d-many elephants have been to Russia ] (6) [[-er]] = λPdt.λQdt.max(Q) > max(P), where max(R) = ιd[R(d) & ∀d′[R(d′) → d′ ≤ d]] Grammatically, then, there are two problems with (1). For a well-formed comparative sentence, a non-overt wh-operator needs to appear in the than-clause in a position parallel to that of -er in the main clause; this is not possible without a bare plural. Semantically, there is no plurality of individuals that can be compared for their number (see Hackl 2001, Nakanishi 2007, Wellwood et al. 2012; Wellwood 2015 for the semantic ban on singulars in comparatives). Ignoring the syntactic rules for a moment, the interpretation we would expect for (1) would have the schematic LF in (7b), which should make as little sense as those underlying (4b)-(4d). This stands in contrast to the interpretation of (2) in (7a). (7) max(d-many people have been to Russia) > a. max(λd.d-many elephants have been to Russia) b. max(λd.d-many I have been to Russia) * The claim that CIs are ungrammatical is not incompatible with an initial perception of acceptability, as acceptability and grammaticality have often been seen to diverge (cf. Lewis & Phillips 2015). Garden path sentences (8a) and sentences with multiple centerembedding (8b) are often perceived to be unacceptable, yet are nonetheless grammatical (see Bever 1970 and Lewis 1996, respectively). Conversely, in some cases ungrammatical sentences are judged acceptable, as in cases of plural attraction (8c) and NPI illusions (8d) (see Bock & Miller 1991, Clifton et al. 1999, Vasishth et al. 2008, Wagers et al. 2009, Xiang et al. 2009, Parker & Phillips 2016). (8) a. The horse raced past the barn fell. b. The man the woman the child kissed knows jumped. c. * The key to the cabinets are on the table. d. * The bills that no senator voted for will ever become law. Other well-known examples of divergence involve grammatical sentences that are perceived to have meanings starkly different from their literal meanings. If a man has a widow, then that man is dead, and no dead man can marry; yet, 30% of respondents answer ‘yes’ when presented with (9a) (Sanford & Sturt 2002). Similarly, (9b) is said to be literally equivalent to ‘All head injuries are trivial enough to ignore’; nevertheless, speakers routinely understand (9b) as equivalent to ‘Any head injury is too important to ignore’ (Wason & Reich 1979, O’Connor 2015). In these cases, comprehenders construct and linger on a certain misinterpretation that prevents them from recognizing the error. (9) a. Can a man marry his widow’s sister? b. No head injury is too trivial to ignore. CIs appear to present a different sort of case from all of these examples, however. Sentences like (1) strike speakers as well-formed, unlike (8a) and (8b). That perception can persist, unlike the easily detectable problems with (8c) and (8d). Furthermore, one never arrives at a specific, grammatically-licensed interpretation—there doesn’t seem to be a single Downloaded from https://academic.oup.com/jos/advance-article-abstract/doi/10.1093/jos/ffy014/5065172 by guest on 04 August 2018 4 Alexis Wellwood et al. sort of misinterpretation that speakers eventually converge on, unlike (9a) and (9b). Rather, informal reports by colleagues, friends, and audiences at professional meetings suggest that speakers tend to believe sentences like (1) are acceptable and have a coherent interpretation, even while they struggle to articulate that interpretation. These considerations implicate online processing in the effect. When things go right, comprehenders can infer at more people that they are likely to require an operatorvariable configuration of a certain sort in a dependent than-clause. Encountering than elephants, the variable can be posited in the

中文翻译：

---

比较幻觉的剖析

诸如英语这样的比较结构被更多的人去过俄罗斯，据报道，英语为母语的人是可以接受的和有意义的。然而，经过仔细的思考，他们被认为是不连贯的。最初的感知与更深思熟虑的判断之间的这种不匹配挑战了我们对句子进行真实地感知并对其进行充分解释的想法。因此，它可能揭示语法和语言处理之间的关系。本文介绍了对这些所谓的“比较错觉”的首次详细研究的结果。我们测试关于它们来源的四个假设：一个浅层的语法分析器，某种形式的省略号修复，错误解决的词汇歧义或持久性事件比较解释。两项正式的可接受性研究表明，当矩阵子句支持事件比较阅读时，说话者最容易产生错觉。逐字召回任务测试并在说话者对句子的回忆中找到此类解释的证据。我们建议，这反映出说话者的娱乐性最初与句子相符，但没有注意到这种解释何时无法使用。特别是，语义知识使人们对语法中非法的操作员变量配置视而不见。比较错觉不是在没有语法分析的情况下说明处理，而是强调句法和语义规则在句子处理中的重要性。1个比较错觉伴随（1）中的句子出现，以英语为母语的人通常会说这是他们语言中完全可以接受的句子。然而，经过进一步的反思，这些发言者认为它没有稳定而有意义的解释。这种形式的句子已被称为“比较错觉”（CIs）或“埃舍尔句子”：它们仅具有格式正确的外观。CI很有意思，因为它们似乎挑战了我们对语言体系结构的一些最基本的假设：我们能真实地感知句子，我们能完全理解它们，并且句子的形式和含义紧密结合。（1）去俄罗斯的人数比我多。©作者2018.牛津大学出版社出版。版权所有。有关许可，请发送电子邮件至：journals.permissions@oup.com从https：//academic.oup下载。com / jos / advance-article-abstract / doi / 10.1093 / jos / ffy014 / 5065172在2018年8月4日的访客2 Alexis Wellwood等。这种现象早已为人所知，但到目前为止，表征CI的语法性和有意义性之间的不匹配尚未得到系统的研究。（1）中的句子由Montalbetti（1984）首次报道为“最惊人的* /？”。我听过的一句话”，归功于Hermann Schultze。重要的是，语言学家和非语言学家都经历过这种现象，但是，尽管在语言学界进行了许多非正式的讨论，但到目前为止，正式的调查仅限于初步结果（Fults＆Phillips 2004，Wellwood等，2009； O'Connor等。 。2012; O'Connor 2015）.1在本文中，我们研究了句子（1）的哪些属性对于有意义的初始感知至关重要。在语法上，CI型句子的问题在于如何选择比从句更重要的主题，因为表面上相似的比较成功地获得了毫无争议的接受和意义。像（2）这样的句子的意思是，“去过俄罗斯的人数超过了已经走过大象的人数”。然而，没有类似的措词对（1）的解释：“去过俄罗斯的人数超过了我的人数”。（2）到俄罗斯去的人多于大象。推导（2）的解释涉及将满足矩阵和嵌入谓词的个体映射到代表其数量的程度，并确定第一个数字是否大于第二个。在至少回溯至布雷斯南（Bresnan，1973）的句法传统中（另请参见乔姆斯基，1977），学位是由更多的许多成分引入的（即，更多的是潜在的许多和更多）。程度谓词是与wh运算符串联而来的，该wh运算符以非子句的抽象语法绑定变量，如（3）所示。该运算符类似于（4a）中的运算符。它需要与一个裸复数NP组合，就像有多少（4b）-（4d）。（3）...比wh-d ... d多头大象去过俄罗斯（4）a。去过俄罗斯的大象多少？b。*我去过俄罗斯多少人？C。*多少头大象去过俄罗斯？d。*去过俄罗斯的大象有多少？从语义上讲，此绑定关系对应于λ-抽象（尤其请参见Heim＆Kratzer 1998）度以上（5a）。通过增加量词的量词，在矩阵子句中得出并行度谓词，并提供与（5b）中相同的LF。这两个谓词一起充当程度量词-er的限制条件和作用域参数，如（6）（Heim 2000）。2因此，（5b）中的LF被解释为大于满足主要和非主要条款中的程度描述（即分别在（6）中的max（Q）和max（P））。1我们的早期结果，报道于Wellwood等。2009年，提供了本手稿并形成了后续文献。可以从https://github.com/alexiswellwood/compillu下载有关这些结果的报告。2有关比较句的LF语法的更多详细信息，请参见Heim 1985，2000； Him 1985，2000。Bhatt＆Pancheva 2004，等等。对-er的精确语义有几种替代的表征，特别是遵循Bartsch＆Vennemann 1972和Kennedy 1999的传统。就我们的目的而言，这些区别并不重要。客人于2018年8月4日从https://academic.oup.com/jos/advance-article-abstract/doi/10.1093/jos/ffy014/5065172下载了《比较幻觉的解剖学》 3（5）a。...比λd... d多头大象去过俄罗斯b。-er [λd... d-许多人去过俄罗斯[（比λd... d-许多大象已经去过俄罗斯]）（6）[[-er]] =λPdt.λQdt.max（Q） > max（P），其中max（R）=ιd[R（d）＆dd [[R（d'）→d'≤d]]]，从语法上讲，（1）存在两个问题。对于格式比较好的句子，非公开的运算符需要出现在比从句中与主句中-er平行的位置；没有仅有的复数，这是不可能的。语义上，没有可以比较的个体数目（参见Hackl 2001，Nakanishi 2007，Wellwood等人2012； Wellwood 2015有关比较中单数的语义禁令）。暂时忽略语法规则，我们对（1）的解释将具有（7b）中的示意图LF，其含义与（4b）-（4d）的含义一样少。这与（7a）中对（2）的解释相反。（7）max（d个人去过俄罗斯）> a。max（λd.d-许多大象去过俄罗斯）b。最大（λd。d-很多人去过俄罗斯）*声称CI不合语法与最初对可接受性的看法并不矛盾，因为人们通常认为可接受性和语法存在差异（参见Lewis＆Phillips 2015）。通常认为花园路径句子（8a）和具有多个中心嵌入的句子（8b）仍然是不可接受的，但它们还是语法上的（分别见Bever 1970和Lewis 1996）。相反，在某些情况下，如复数吸引（8c）和NPI幻觉（8d），则认为不合语法的句子可以接受（参见Bock＆Miller 1991，Clifton等1999，Vasishth等2008，Wagers等2009）。 ，Xiang等人，2009年； Parker＆Phillips，2016年）。（8）那匹马跑过谷仓摔倒了。b。孩子亲吻的男人知道的男人跳了起来。C。*机柜钥匙在桌子上。d。*没有参议员投票通过的法案将成为法律。其他众所周知的差异示例还包括语法句子，这些句子被认为具有与文字含义完全不同的含义。如果一个男人有一个寡妇，那么那个男人已经死了，死人也不会结婚。但是，当有（9a）出现时，有30％的受访者回答“是”（Sanford＆Sturt 2002）。同样，（9b）在字面上等同于“所有头部受伤都足以忽略不计”。然而，演讲者通常会理解（9b）等同于“任何头部受伤都太重要而不能忽略”（Wason＆Reich 1979，O'Connor 2015）。在这些情况下，理解者会构建并徘徊在某种误解上，这会阻止他们识别错误。（9）一个男人可以嫁给寡妇的妹妹吗？b。没有头部受伤是微不足道的。但是，配置项似乎呈现出与所有这些示例不同的情况。与（8a）和（8b）不同，像（1）这样的句子能很好地打击说话者。与（8c）和（8d）容易发现的问题不同，这种感觉可以持续存在。此外，从来没有人会得到一种经过语法许可的特定解释-似乎并没有一种解释是从https://academic.oup.com/jos/advance-article-abstract/doi/10.1093/jos/ffy014下载的客人在/ 5065172/2018年8月4日4 Alexis Wellwood等 与（9a）和（9b）不同，说话者最终会产生误解。相反，同事，朋友和听众在专业会议上的非正式报告表明，说话者倾向于相信（1）之类的句子是可以接受的，并且具有连贯的解释，即使他们努力表达这种解释也是如此。这些考虑暗示了在线处理的效果。当事情进展顺利时，理解者可以向更多的人推断他们可能需要从属子句中的某种操作员变量配置。与大象相比，该变量可以位于