- 作者：LAWRENCE W. BARSALOU
- 单位：Stanford University, Stanford, California 94305
- 出版：Memory & Cognition 1982, Vol. 10 (1), 82-93
Abstract
It is proposed that concepts contain two types of properties. Context-independent properties are activated by the word for a concept on all occasions. The activation of these properties is unaffected by contextual relevance. Context-dependent properties are not activated by the respective word independent of context. Rather, these properties are activated only by relevant contexts in which the word appears. Context-independent properties form the core meanings of words, whereas context-dependent properties are a source of semantic encoding variability. This proposal lies between two opposing theories of meaning, one that argues all properties of a concept are active on all occasions and another that argues the active properties are completely determined by context. The existence of context-independent and context-dependent properties is demonstrated in two experimental settings: the property-verification task and judgments of similarity. The relevance of these property types to cross-classification, problem solving, metaphor and sentence comprehension, and the semantic-episodic distinction is discussed.
本文提出，概念包含两种类型的属性。上下文无关属性（Context-independent properties）在所有场合下都会被表示该概念的词所激活。这些属性的激活不受上下文相关性的影响。上下文相关属性（Context-dependent properties）则不会脱离上下文被相应的词汇激活。相反，这些属性只有在词汇出现的相关上下文中才会被激活。上下文无关属性构成了词汇的核心意义，而上下文相关属性则是语义编码变异性的来源。这一提议介于两种对立的意义理论之间：一种理论认为概念的所有属性在所有场合下都是活跃的，另一种理论则认为活跃的属性完全由上下文决定。这两种属性（上下文无关属性和上下文相关属性）的存在在两种实验设置中得到了证明：属性验证任务和相似性判断任务。本文还讨论了这些属性类型与交叉分类、问题解决、隐喻和句子理解，以及语义-情节记忆区分之间的相关性。
Some properties in a concept seem to come to mind on all occasions. The word "skunk" usually makes people think of the property "unpleasant smell," and "rattlesnake" usually makes people think of "poisonous." In contrast, other properties in a concept rarely seem to come to mind, and when they do, it is only in relevant contexts. For example, "basketball" rarely makes people think of "floats." However, the sentence frame "Chris used X as a life preserver when the boat sank" would probably bring "floats" to mind for "basketball" when "X" is "basketball." In this paper, I propose there are two important types of properties associated with concepts: context-independent (CI) properties and context-dependent (CD) properties. CI properties are activated by the word for a concept on all occasions (e.g., "unpleasant smell" for "skunk"). CD properties are rarely if ever activated by the word for a concept and are only activated by relevant contexts in which the word appears (e.g., "floats" for "basketball").
概念中的某些属性似乎在所有场合下都会浮现在脑海中。例如，“臭鼬”这个词通常会让人想到“难闻的气味”这一属性，而“响尾蛇”通常会让人想到“有毒”。相比之下，概念中的其他属性似乎很少被想起，而当它们被想起时，也仅仅是在相关的上下文中。例如，“篮球”很少让人想到“可以漂浮”。然而，在“当船沉没时，克里斯把X当作救生圈使用”这个句子框架中，当“X”是“篮球”时，很可能会让人对“篮球”产生“可以漂浮”的联想。在本文中，我提出与概念相关的两种重要属性类型：上下文无关（CI）属性和上下文相关（CD）属性。CI属性在所有场合下都会被表示概念的词激活（例如，“臭鼬”的“难闻气味”）。CD属性很少（如果有的话）被表示概念的词激活，并且只在词汇出现的相关上下文中才会被激活（例如，“篮球”的“可以漂浮”）。
CI properties form the core meanings of words. This is because they are activated by the respective word on all occasions, independent of contextual relevance. Barsalou and Bower (Note 1) have proposed that properties become automatically activated by a word after being frequently associated with it during processing. Frequent pairings of a word and a property cause an automatized relation between them to be established in memory (also see Shiffrin & Schneider, 1977). Barsalou and Bower (Note 1) showed that two types of properties are likely to be frequently active during the processing of a word. First, properties having high diagnosticity may often be active, since they are useful for distinguishing instances of a concept from instances of other concepts. "Gills" becomes CI for "fish" because all fish have gills and no other things do. The second type of property likely to be frequently active during the processing of a word includes properties relevant to how people typically interact with instances of the respective concept. "Edible" becomes CI for "apples" because it is central to how people typically interact with them. As shown by "edible" in relation to "apple," properties frequently relevant to human interaction can become CI even if they have low diagnosticity (i.e., "edible" is true of many other things). CI属性构成了词汇的核心意义。这是因为它们在所有场合下都会被相应的词汇激活，而与上下文相关性无关。Barsalou和Bower（注1）曾提出，当属性在加工过程中频繁地与某个词汇联系在一起时，它们就会被该词汇自动激活。词汇与属性的频繁配对会在记忆中建立它们之间的自动化联系（另见Shiffrin & Schneider, 1977）。Barsalou和Bower（注1）表明，有两类属性在词汇加工过程中很可能频繁处于活跃状态。第一类是具有高诊断性（区分度）的属性，因为它们有助于将某一概念的实例与其他概念的实例区分开来。例如，“鳃”成为“鱼”的CI属性，因为所有的鱼都有鳃，而其他事物则没有。第二类很可能在词汇加工中频繁活跃的属性，包括与人们通常如何与特定概念的实例进行互动的相关属性。例如，“可食用”成为“苹果”的CI属性，因为它是人们通常与苹果互动方式的核心。正如“可食用”与“苹果”的关系所示，频繁与人类互动相关的属性即使诊断性很低（即“可食用”也适用于许多其他事物），也可以成为CI属性。
CD properties are a source of semantic encoding variability. CD properties may be represented in concepts, but they are not usually activated by encoding the respective words. Rather, these properties are activated only by relevant encoding contexts in which a word appears. Semantic encoding variability is the result of different encoding contexts of a word activating different subsets of CD properties in the respective concept. This phenomenon has frequently been observed empirically (e.g., R. C. Anderson & Ortony, 1975; RC. Anderson, Pichert, Goetz, Schallert, Stevens, & Trollip, 1976; Barclay, Bransford, Franks, McCarrell, & Nitsch, 1974; Tulving & Thompson, 1973) and has been incorporated theoretically by Bower (1972) and Estes (1955, 1959). Barsalou and Bower (Note 1) suggest that CD properties are typically inactive because they have rarely, if ever, been processed simultaneously with their respective words. Hence, the associations between these properties and their respective words are weak or nonexistent. When such associations do not exist, various inference processes may be required to compute them. People may not have stored the fact that "fits in a suitcase" is a property of "flashlight," but they can certainly infer it. If a CD property comes to be frequently processed with a word, the property may change status and become CI. Although "wears horseshoes" may be CD for "horse" for people who are rarely around horses, it could become CI for someone during the course of learning to be a horseshoer. CD属性是语义编码变异性的来源。CD属性可能表征在概念中，但它们通常不会通过对相应词汇的编码来激活。相反，这些属性只有在词汇出现的相关编码上下文中才会被激活。语义编码变异性是由于一个词汇在不同编码上下文中激活了该概念中不同子集的CD属性所导致的。这种现象在经验研究中已被频繁观察到（例如，R. C. Anderson & Ortony, 1975; RC. Anderson 等, 1976; Barclay 等, 1974; Tulving & Thompson, 1973），并已被Bower (1972) 和 Estes (1955, 1959) 纳入理论中。Barsalou和Bower（注1）认为，CD属性通常处于非活跃状态，因为它们极少（如果有的话）与其相应的词汇同时被加工。因此，这些属性与其相应词汇之间的联想非常微弱或根本不存在。当这种联想不存在时，可能需要各种推理过程来计算它们。人们可能没有在记忆中储存“能装进手提箱”是“手电筒”的一个属性，但他们肯定能推断出这一点。如果一个CD属性开始频繁地与某个词汇一起被加工，该属性可能会改变状态并成为CI属性。虽然对于很少接触马的人来说，“钉马掌”可能是“马”的CD属性，但对于正在学习成为铁匠的人来说，在学习过程中它可能会变成CI属性。
Some properties are probably neither CI nor CD. These may sometimes be activated by a word, although not on all occasions, and may sometimes be activated by context. Such properties may occasionally be activated by a word because of random fluctuations in the amount of activation the property receives. On other occasions, however, these properties may be activated by relevant contexts. The senses of ambiguous words can similarly not be classified as strictly CI or CD. This is because they often come to mind without context but they are also influenced by sentence contexts in which they occur (Swinney, 1979; Tanenhaus, Leiman, & Seidenberg, 1979). Nevertheless, the senses of ambiguous words can be viewed as concepts that contain CI and CD properties. Once an ambiguous word is disambiguated in context, the distinction between CI and CD properties becomes applicable to the concept converged upon. Consider "bear" in the sentence "The bear caught pneumonia." "Bear" and the sentence frame both converge on the mammalian sense of "bear." However, some of the properties activated for this sense are CI (e.g., "is furry," "can be dangerous") and some are CD (e.g., "can be sick," "has lungs"). Beyond acknowledging their existence, I will not further consider properties and word senses that are neither CI nor CD. Rather, the purposes of this paper are (1) to demonstrate the existence of CI and CD properties in concepts and (2) to consider the roles these property types play in various cognitive phenomena. 有些属性可能既不是CI属性，也不是CD属性。这些属性有时可能会被词汇激活（尽管并非所有场合），有时也可能被上下文激活。这类属性偶尔会被词汇激活，可能是因为它们接收到的激活量存在随机波动。然而在其他场合，这些属性可能会被相关的上下文激活。歧义词的词义同样不能被严格地归类为CI或CD。这是因为它们经常在没有上下文的情况下出现在脑海中，但它们也受其出现的句子上下文的影响（Swinney, 1979; Tanenhaus, Leiman, & Seidenberg, 1979）。尽管如此，歧义词的词义可以被视为包含CI和CD属性的概念。一旦歧义词在上下文中被消除了歧义，CI和CD属性的区分就可以应用于所确定的概念。考虑句子“那只熊（bear）得了肺炎”中的“bear”。“Bear”和句子框架都收敛于“bear”的哺乳动物含义。然而，为这个含义激活的一些属性是CI属性（例如，“长满毛的”、“可能具有危险性”），而另一些则是CD属性（例如，“会生病”、“有肺”）。除了承认它们的存在之外，我将不再进一步考虑既不是CI也不是CD的属性和词义。相反，本文的目的是 (1) 证明概念中存在CI和CD属性，以及 (2) 探讨这些属性类型在各种认知现象中扮演的角色。
This paper addresses a particular aspect regarding the structure of concepts, namely, the accessibility of properties. Two theories of meaning take more extreme views on this aspect. Traditional views of semantics (e.g., Katz & Postal, 1964) assume that the meaning of a word contains a fixed set of semantic features applicable on all occasions on which the word is used. It appears that semantic memory models usually make a similar assumption (e.g., Glass & Holyoak, 1975; McCloskey & Glucksberg, 1979; Smith, Shoben, & Rips 1974). In the terms of this paper, this approach argues that all of a concept's properties are CI. An opposing and more radical view of semantics (e.g., Olson, 1970) argues that the meaning of a word completely depends on the context in which the word is used. Specifically, the meaning of a word in a given context is a function of the distinctions it is supposed to convey in that context. According to this view, there may be no overlap between uses of the same word across contexts. Psychologists who have observed contextual effects on encoding have often reached a similar conclusion (e.g., Jacoby, Craik, & Begg, 1979). In the terms of this paper, this approach argues that all of a concept's properties are CD. The proposal that some of a concept's properties are CI and others are CD lies between these two theories of meaning. Consequently, evidence for the existence of CI and CD properties in concepts would have implications for theories of natural language semantics.¹ 本文探讨了概念结构的一个特定方面，即属性的可访问性。有两种意义理论在这一方面采取了更为极端的观点。传统的语义学观点（例如，Katz & Postal, 1964）假设一个词的意义包含一组固定的语义特征，适用于该词被使用的所有场合。似乎语义记忆模型通常也会做出类似的假设（例如，Glass & Holyoak, 1975; McCloskey & Glucksberg, 1979; Smith, Shoben, & Rips 1974）。用本文的术语来说，这种方法认为概念的所有属性都是CI。一种对立且更为激进的语义学观点（例如，Olson, 1970）认为，一个词的意义完全取决于其使用的上下文。具体而言，一个词在特定上下文中的意义取决于它要在该上下文中传达的区分。根据这种观点，同一个词在不同上下文中的使用可能没有任何重叠。观察到编码上下文效应的心理学家通常也得出了类似的结论（例如，Jacoby, Craik, & Begg, 1979）。用本文的术语来说，这种方法认为概念的所有属性都是CD。认为概念的一些属性是CI而另一些是CD的提议，正处于这两种意义理论之间。因此，概念中存在CI和CD属性的证据将对自然语言语义学理论产生影响。¹
The definitions of CI and CD properties lead to several empirical predictions. First, in a given context, all CI properties should be available and irrelevant CD properties should not. This is because CI properties are always activated by their respective words, whereas irrelevant CD properties remain inactive due to lack of contextual activation. The second prediction follows from the definition of CI properties. Since CI properties are always activated by the respective words on all occasions, they should be unaffected by contextual relevance: A CI property should be just as available in an irrelevant context as in a relevant context. This assumes that the activation of a CI property by a word maximally activates that property. The third prediction follows from the definition of CD properties. Since CD properties depend on relevant contexts for activation, they should be available in working memory for processing when the context is relevant and unavailable when the context is irrelevant. The alternative hypotheses are (1) all the properties in a concept are activated by the respective word on all occasions and (2) all the properties active in a concept are determined by context. These hypotheses and their alternatives are contrasted in the two experiments that follow. The first experiment tests these predictions in the property-verification task. The second experiment tests these predictions in judgments of similarity. Evidence from other current work is also brought to bear on these issues. Finally, the relevance of CI and CD properties to cross-classification, problem solving, metaphor and sentence comprehension, and the semantic-episodic distinction is discussed. CI和CD属性的定义引出了几个经验预测。首先，在给定上下文中，所有CI属性都应该可用，而无关的CD属性则不可用。这是因为CI属性总是被其相应的词汇激活，而无关的CD属性由于缺乏上下文激活而保持不活跃状态。第二个预测源于CI属性的定义。由于CI属性在所有场合下总是被相应词汇激活，因此它们应该不受上下文相关性的影响：一个CI属性在无关上下文中的可用性应与在相关上下文中一样。这假设词汇对CI属性的激活达到了对该属性的最大激活水平。第三个预测源于CD属性的定义。由于CD属性的激活依赖于相关的上下文，因此当上下文相关时，它们应该在工作记忆中可用于加工，而当上下文无关时，它们则不可用。备选假设是：(1) 概念中的所有属性在所有场合下都会被相应的词汇激活；(2) 概念中所有活跃的属性都是由上下文决定的。这些假设及其备选方案在接下来的两个实验中进行了对比。实验一在属性验证任务中检验了这些预测。实验二在相似性判断任务中检验了这些预测。本文还引用了其他现有研究的证据来探讨这些问题。最后，讨论了CI和CD属性与交叉分类、问题解决、隐喻和句子理解以及语义-情节记忆区分的相关性。
EXPERIMENT 1
实验一
A version of the property-verification task was used to test the predictions following from the definitions of CI and CD properties. On each trial, subjects read a sentence containing an underlined subject noun. Several seconds later, the label for a property was presented, and subjects indicated whether or not the subject noun in the preceding sentence possessed the property. 实验采用了一种属性验证任务的变体，以检验由CI和CD属性定义引出的预测。在每次试验中，被试阅读一个包含带下划线主语名词的句子。几秒钟后，呈现一个属性的标签，被试指出前一句中的主语名词是否具有该属性。
The logic of the experiment is as follows. If the property for a trial is CI information of the subject noun, then verification time should not vary across sentence contexts. In particular, verification time should be no less when the sentence context is related to the property than when the sentence context is unrelated. This is because the property, being CI, is always activated by the subject noun itself and therefore is not dependent on context for activation. However, if the property for a trial is CD information of the subject noun, then verification time should depend on context. Specifically, verification time should be much less when the sentence context is related to the property than when the sentence context is unrelated. This is because the property, being CD, is not activated by the subject noun and therefore is dependent on context for activation. If the difference between related and unrelated contexts for CD properties is substantial (i.e., on the order of several hundred milliseconds), this would suggest that CD properties are, in fact, inactive in irrelevant contexts. 实验逻辑如下。如果某次试验的目标属性是主语名词的CI信息，那么验证时间不应因句子上下文的不同而变化。具体而言，当句子上下文与属性相关时，验证时间不应比上下文与属性无关时更短。这是因为该属性作为CI属性，总是由主语名词本身激活，因此其激活不依赖于上下文。然而，如果试验的属性是主语名词的CD信息，那么验证时间应取决于上下文。具体而言，当句子上下文与属性相关时，验证时间应远少于上下文无关时的验证时间。这是因为该属性作为CD属性，不会被主语名词激活，因此它的激活依赖于上下文。如果CD属性在相关和无关上下文之间的差异很大（即达到几百毫秒的数量级），这表明CD属性在无关上下文中实际上是处于非活跃状态的。
Method
方法
Procedure. Subjects looked into a modified Siliconix tachistoscope and rested their forefingers on two response buttons 7 cm apart. When prepared for a trial, subjects pressed the "start" button (positioned colinearly and midway between the two response buttons) with the same finger used to press the "true" button. After a 500-msec interval, a context sentence appeared in the top field of the tachistoscope. All sentences began with "The," followed by an underlined subject noun and a predication of the subject noun. Subjects were instructed to fully comprehend the sentence and to read it out loud. The context sentence was removed after 6 sec, and a property label immediately appeared in the bottom field; subjects did not read the label aloud. If the subject noun in the context sentence possessed the property, subjects pressed the "true" response button; otherwise, they pressed the "false" response button. For each trial, time was measured from the onset of the property to the point at which a response was detected. Subjects were instructed to respond as quickly as possible, but to avoid making errors. **程序。**被试通过一台改装的Siliconix速示器进行观察，并将食指放在相距7厘米的两个反应键上。当准备好进行试验时，被试用按“真（true）”键的同一手指按下“开始”键（位于两个反应键之间，并与它们共线）。500毫秒间隔后，速示器的上方视野会出现一个上下文句子。所有句子均以“The”开头，后面跟一个带下划线的主语名词以及该名词的谓语部分。指导语要求被试充分理解句子并大声读出来。6秒后，上下文句子被移除，一个属性标签立即出现在下方视野；被试不需要大声读出该标签。如果上下文句子中的主语名词具有该属性，被试按“真”键；否则，按“假”键。对于每次试验，时间是从属性出现到检测到反应的那一刻进行测量的。被试被要求尽可能快地做出反应，同时避免犯错。
Subjects received 24 practice trials and 60 test trials. There was a short break between the practice and test trials. Subjects could take a break anytime during the test trials, but they rarely did. Following the last test trial, subjects were asked a series of questions concerning their strategies. 被试进行了24次练习试验和60次正式测试试验。练习和正式测试之间有短暂的休息时间。被试在正式测试期间可以随时休息，但他们很少这样做。在最后一次测试试验结束后，主试会询问被试一系列关于他们策略的问题。
Subjects and Materials. The subjects were 19 Stanford students participating for pay or course credit. One subject's data were not used because of an error rate exceeding 15% (the average error rate for the remaining subjects was 2.8%). The materials consisted of context sentences and properties, related as discussed next. Examples of the materials are shown in Table 1. **被试与材料。**被试为19名斯坦福大学学生，他们通过参与实验获得报酬或课程学分。一名被试的数据因错误率超过15%而被剔除（其余被试的平均错误率为2.8%）。实验材料由上下文句子和属性组成，其相关性如下文所述。材料示例见表1。
[Table 1 Placeholder] [表1 占位符]
Trues. Thirty properties were chosen for the "true" trials; 15 were randomly assigned to the CI condition and 15 to the CD condition. The average number of syllables per property did not vary between conditions [4.06 and 4.13 for the CI and CD properties, respectively; \(t(28)=.14\), \(p>.30\)]. For each CI property, three context sentences were constructed. Two of these contained the same subject noun, which was highly related to the property; the predicate for one of these sentences was related to the property (the related-context sentence), and the predicate for the other was unrelated (the unrelated-context sentence). (Degree of relatedness was determined by the ratings of an independent group of subjects, as reported later.) The remaining context sentence contained a subject noun, weakly related to the property, and an unrelated predicate; this sentence served as a control sentence (to be explained in a moment). **“真”项。**为“真（true）”试验选择了30个属性；其中15个被随机分配到CI条件下，15个分配到CD条件下。每个属性的平均音节数在两种条件之间没有差异[CI和CD属性分别为4.06和4.13；\(t(28)=.14\)，\(p>.30\)]。对于每一个CI属性，我们构建了三个上下文句子。其中两个句子包含相同的主语名词，该名词与该属性高度相关；这两个句子中，其中一个的谓语与该属性相关（相关上下文句子），另一个的谓语则与该属性无关（无关上下文句子）。（相关程度由一组独立的被试进行评定来确定，见后文报告。）剩余的一个上下文句子包含一个与该属性弱相关的主语名词，以及一个无关的谓语；这个句子作为控制句子（稍后解释）。
Similarly, for each of the 15 CD properties, three context sentences were constructed. Two of these contained the same subject noun, which was weakly related to the property; the predicate for one of these sentences was related to the property (the related-context sentence), and the predicate for the other was unrelated (the unrelated-context sentence). The remaining sentence (the control sentence) contained a subject noun, highly related to the property, and an unrelated predicate. 类似地，对于15个CD属性中的每一个，也构建了三个上下文句子。其中两个句子包含相同的主语名词，该名词与该属性弱相关；这两个句子中，其中一个的谓语与该属性相关（相关上下文句子），另一个的谓语与属性无关（无关上下文句子）。剩余的一个句子（控制句子）包含一个与该属性高度相关的主语名词，以及一个无关的谓语。
A control sentence in the CI condition (having a weakly related subject noun and an unrelated predicate) served as a contrast to verify that the other two context sentences for the property had a subject noun highly related to the property. The time to verify the unrelated-context sentence (having a highly related subject noun and an unrelated predicate) should be less than that for the control sentence. Similarly, a control sentence in the CD condition (having a highly related subject noun and an unrelated predicate) served as a contrast to verify that the other two context sentences for the property had a subject noun weakly related to the property. The time to verify the unrelated context sentence (having a weakly related subject noun and an unrelated predicate) should be longer than that for the control sentence. CI条件中的控制句子（包含弱相关的主语名词和无关的谓语）作为对比项，用于验证该属性的其他两个上下文句子的主语名词与属性高度相关。验证无关上下文句子（包含高度相关的主语名词和无关谓语）的时间应少于控制句子。类似地，CD条件中的控制句子（包含高度相关的主语名词和无关谓语）作为对比项，用于验证该属性的其他两个上下文句子的主语名词与属性弱相关。验证无关上下文句子（包含弱相关的主语名词和无关谓语）的时间应长于控制句子。
Ratings were obtained to confirm the assumed relations between the subject nouns and properties, and between the predicates and properties. Of primary importance is that context be manipulated equally for the CI and CD conditions. This insures that an effect of context on CD subject nouns but not on CI subject nouns for the latency data cannot be attributed to CD materials having more relatedness for related predicates or less relatedness for unrelated predicates than the CI materials. 通过获取评定分数来确认主语名词与属性之间、以及谓语与属性之间的假设关系。最重要的是，上下文在CI和CD条件下的操纵必须是对等的。这可以确保：在延迟数据中，上下文对CD主语名词产生影响但对CI主语名词没有影响的结果，不能归咎于CD材料在相关谓语上的相关性更高，或者在无关谓语上的相关性低于CI材料。
Four judges rated the 60 subject nouns first (30 for the control and 30 for the noncontrol sentences) and the 90 predicates second (3 for each property). Subjects read either a subject noun or a predicate on one side of an index card and then flipped the card to read the property. Subjects rated how much the subject noun or predicate made them think of the property. Subjects used a scale from 1 to 7, on which 1 meant the property did not come to mind at all and 7 meant the property immediately came to mind. Within each group, the cards were randomly ordered for each subject. 四名评判者首先评定了60个主语名词（30个用于控制句子，30个用于非控制句子），其次评定了90个谓语（每个属性3个）。被试阅读索引卡片一面的主语名词或谓语，然后翻转卡片阅读属性。被试评定该主语名词或谓语在多大程度上让他们想到了该属性。被试使用1到7的量表进行评分，其中1表示该属性完全没有出现在脑海中，7表示立刻想到了该属性。在每个组内，卡片对每位被试进行随机排序呈现。
An ANOVA was performed on the ratings for the subject nouns. The two factors of interest were condition (i.e., CI vs. CD) and relatedness (i.e., weakly vs. highly related). Note that for the CI materials, the nouns in the related and unrelated-context sentences were supposed to be highly related and the nouns in the control sentences were supposed to be weakly related to their respective properties. For the CD materials, the nouns in the related- and unrelated-context sentences were supposed to be weakly related and the nouns in the control sentences were supposed to be highly related to their respective properties. The mean ratings from this analysis are shown in Table 2. There was no effect of CI/CD [\(F(1,3) = 1.74\), \(p>.25\)], there was an effect of relatedness [\(F(1,3)=204.54\), \(p<.001\)], and there was no interaction between these two factors (\(F<1\)). Thus the assumed difference in relatedness was substantial and equivalent for the CI and CD materials. A similar ANOVA was performed for the predicates. Again, there was no effect of CI/CD [\(F(1,3)=1.71\), \(p>.25\)] and no interaction of this factor with relatedness [\(F(1,6)=1.19\), \(p>.25\)]. The predicates for the control and unrelated-context sentences did not differ in relatedness (\(F<1\)). However, the predicates for the related-context sentences were higher in relatedness than those for the unrelated-context sentences [\(F(1,6)=89.97\), \(p<.001\)] and those for the control sentences [\(F(1,6)=79.28\), \(p<.001\)]. Thus the assumed difference in relatedness was again substantial and equivalent for the CI and CD materials. Crucial to the interpretation of the latency results are the findings that (1) the related predicates for the CD materials were not higher in relatedness than those for the CI materials (\(F<1\)) and (2) the unrelated predicates for the CD materials were not lower in relatedness than those for the CI materials (\(F<1\)). 对主语名词的评定进行了方差分析（ANOVA）。两个感兴趣的因素是条件（即CI对比CD）和相关度（即弱相关对比高度相关）。注意，对于CI材料，相关和无关上下文句子中的名词应与其各自的属性高度相关，而控制句子中的名词应与其弱相关。对于CD材料，相关和无关上下文句子中的名词应与其相应的属性弱相关，而控制句子中的名词应与其高度相关。该分析的平均评分结果如表2所示。结果显示没有CI/CD的主效应 [\(F(1,3) = 1.74\)，\(p>.25\)]，有相关度的主效应 [\(F(1,3)=204.54\)，\(p<.001\)]，这两个因素之间也没有交互作用（\(F<1\)）。因此，所假设的相关度差异在CI和CD材料中既是实质性的，也是对等的。对谓语进行了类似的方差分析。结果再次显示没有CI/CD主效应 [\(F(1,3)=1.71\)，\(p>.25\)]，该因素与相关度也没有交互作用 [\(F(1,6)=1.19\)，\(p>.25\)]。控制句子和无关上下文句子的谓语在相关度上没有差异（\(F<1\)）。然而，相关上下文句子的谓语在相关度上高于无关上下文句子的谓语 [\(F(1,6)=89.97\)，\(p<.001\)]，也高于控制句子的谓语 [\(F(1,6)=79.28\)，\(p<.001\)]。因此，所假设的相关度差异再次在CI和CD材料中被证明是实质性且对等的。对反应延迟结果解释至关重要的两项发现是：(1) CD材料中相关谓语的相关度并不比CI材料高（\(F<1\)）；(2) CD材料中无关谓语的相关度也不比CI材料低（\(F<1\)）。
[Table 2 Placeholder] [表2 占位符]
Falses. Thirty context sentence/property pairs were constructed, each context sentence having a subject noun that clearly did not possess the property. The context sentences and properties used were similar in nature to those for the "true" materials. Five of the 30 "true" context sentences presented to a subject (as discussed in the Design section) contained a subject noun and a predicate both highly related to the same property (i.e., the CI related-context sentences). Therefore, 5 of the 30 "false" context sentences also contained a subject noun and a predicate both highly related to some property; however, this was not the property actually tested (i.e., for the "false" items, the subject noun could not possess the property). Creating some "false" items in this manner made it impossible for subjects to discriminate the "true" from the "false" items on the basis of subject-predicate-property relations. **“假”项。**构建了30个上下文句子/属性对，每个上下文句子都有一个明确不具备该属性的主语名词。使用的上下文句子和属性在性质上与“真”材料相似。展示给被试的30个“真”上下文句子中（如设计部分所述），有5个句子的主语名词和谓语都与同一个属性高度相关（即CI相关上下文句子）。因此，30个“假”上下文句子中有5个句子也包含一个主语名词和一个谓语，它们都与某个属性高度相关；但这并非实际测试的那个属性（也就是说，对于“假”项目，主语名词不能具有被测属性）。通过这种方式创建部分“假”项目，可以防止被试根据“主语-谓语-属性”关系来区分“真”项与“假”项。
Practice items. Twenty-four context sentence/property pairs were constructed; half were true and half were false. These items were similar in nature to the "true" and "false" test items. Also, the distribution of item types was similar to that found in the set of test items. **练习项。**构建了24个上下文句子/属性对；一半为真，一半为假。这些项目在性质上类似于“真”和“假”的正式测试项目。此外，项目类型的分布情况也与正式测试集中的相似。
Design. Three lists were constructed. Each contained the same 30 context sentence/property pairs for the "false" items and the same 30 properties for the "true" items. The lists differed only with respect to the context sentences for the "true" properties, as discussed next. The 15 CI properties for the "true" items were randomly divided into three groups of five properties each; the 15 CD properties were also randomly divided into three groups of five properties each. The 30 "true" context sentences in a given list consisted of (1) the control sentences from one CI group and one CD group, (2) the unrelated-context sentences from a second CI group and a second CD group, and (3) the related-context sentences from the remaining CI group and the remaining CD group. Each of the three context sentence types for each property group was instantiated in one and only one of the lists. This rotation of context sentence type through property group and list was done as randomly as possible, given the necessary constraints of a Latin square. **实验设计。**构建了三个列表。每个列表包含相同的30个针对“假”项的上下文句子/属性对，以及30个针对“真”项的属性。列表之间的差异仅在于“真”属性对应的上下文句子，如下文所述。15个“真”项的CI属性被随机分为三组，每组5个；15个CD属性也被随机分为三组，每组5个。在给定列表中的30个“真”上下文句子由以下几部分组成：(1) 来自第一组CI属性和第一组CD属性的控制句子；(2) 来自第二组CI属性和第二组CD属性的无关上下文句子；(3) 来自剩余一组CI属性和剩余一组CD属性的相关上下文句子。每个属性组的三种上下文句子类型分别仅在其中一个列表中实例化。在拉丁方设计的必要约束下，这种上下文句子类型在属性组和列表间的轮换过程尽可能地随机化。
The 24 practice items were presented in the same random order to all subjects. The 60 test items were presented to each subject in a different, computer-generated, random order. Half the subjects used their right forefingers to press the "start" and "true" buttons and their left forefingers to press the "false" button; the other subjects had the inverse assignment. Subjects were assigned randomly to one of the six lists by hand assignment cells of the design, three subjects per cell. 24个练习项目以相同的随机顺序呈现给所有被试。60个正式测试项目以不同的、计算机生成的随机顺序呈现给每位被试。一半被试用右手食指按“开始”和“真”按钮，用左手食指按“假”按钮；另一半被试的按键设置相反。被试被随机分配到设计单元格的六个列表之一，每个单元格分配三名被试。
Results
结果
Latencies for the correct true trials were analyzed as follows. Averages were computed separately across subjects and across items (i.e., properties). For each subject, the average latency was determined for each of the six subject relation by predicate relation conditions. For each property in the CI and CD conditions, the average latency was determined for each of the three predicate relations. The results for the subject averages are shown in Table 3. Separate subject relation by predicate relation ANOVAs were performed on the subject averages and item averages. The results of both analyses were combined to compute min \(F^{\prime}\) planned comparisons of interest (H. H. Clark, 1973). 对正确反应“真”试验的延迟时间进行如下分析：分别计算了基于被试和基于项目（即属性）的平均值。对于每个被试，确定了在“主语关系”与“谓语关系”交叉生成的六种条件下的平均延迟时间。对于CI和CD条件下的每个属性，确定了在三种谓语关系下的平均延迟时间。被试平均值的结果如表3所示。对被试平均值和项目平均值分别进行了“主语关系×谓语关系”的方差分析。两项分析的结果被合并计算了感兴趣的最小 \(F^{\prime}\) （min \(F^{\prime}\)）计划比较（H. H. Clark, 1973）。
[Table 3 Placeholder] [表3 占位符]
For the CI items, the control sentences led to longer latencies than the unrelated-context sentences [min \(F^{\prime}(1,89)=15.80\), \(p<.001\)]. For the CD items, the control sentences led to shorter latencies than the unrelated-context sentences [min \(F^{\prime}(1,90)=24.80\), \(p<.001\)]. These two results show that (1) the subject nouns in the CI noncontrol sentences were in fact highly related to their respective CI properties, and (2) the subject nouns in the CD noncontrol sentences were in fact weakly related to their respective properties.² 对于CI项，控制句子的延迟时间长于无关上下文句子 [min \(F^{\prime}(1,89)=15.80\)，\(p<.001\)]。对于CD项，控制句子的延迟时间短于无关上下文句子 [min \(F^{\prime}(1,90)=24.80\)，\(p<.001\)]。这两个结果表明：(1) CI非控制句子中的主语名词实际上与其相应的CI属性高度相关，以及 (2) CD非控制句子中的主语名词实际上与其相应的属性弱相关。²
The remaining results pertain only to the noncontrol sentences. For the CI items, there was no difference between related and unrelated-context sentences (min \(F^{\prime}<1\); the subject's F and item's F were also less than 1). For the CD items, related-context sentences led to shorter latencies than unrelated-context sentences [min \(F^{\prime}(1,90)=5.97\), \(p<.025\)]. For the unrelated-context sentences, the latencies were less for the CI items than for the CD items [min \(F^{\prime}(1,90)=22.13\), \(p<.001\)]. For the related-context sentences, there was a marginal difference between the CI and CD items [min \(F^{\prime}(1,89)=3.16\), \(.10>p>.05\)]; however, the subject's F was significant [\(F(1,34)=9.66\), \(p<.01\)], as was the item's F [\(F(1,56) = 4.70\), \(p<.05\)]. There was a significant Subject Relation by Predicate Relation interaction for the noncontrol sentences [min \(F^{\prime}(1,90)=4.19\), \(p<.05\)]. 其余结果仅涉及非控制句子。对于CI项目，相关与无关上下文句子之间没有差异（min \(F^{\prime}<1\)；被试层面的F值和项目层面的F值也小于1）。对于CD项目，相关上下文句子导致的延迟时间短于无关上下文句子 [min \(F^{\prime}(1,90)=5.97\)，\(p<.025\)]。对于无关上下文句子，CI项目的延迟时间短于CD项目 [min \(F^{\prime}(1,90)=22.13\)，\(p<.001\)]。对于相关上下文句子，CI和CD项目之间存在边缘差异 [min \(F^{\prime}(1,89)=3.16\)，\(.10>p>.05\)]；然而，被试的F检验具有显著性 [\(F(1,34)=9.66\)，\(p<.01\)]，项目的F检验也同样显著 [\(F(1,56) = 4.70\)，\(p<.05\)]。非控制句子的主语关系与谓语关系之间存在显著的交互作用 [min \(F^{\prime}(1,90)=4.19\)，\(p<.05\)]。
These data indicate that context had no effect on the CI items but had an effect on the CD items. More specifically, related contexts did not increase the priming of properties when the subject noun was highly related to the target property. However, related contexts did increase the priming of properties when the subject noun was weakly related to the property. It is not clear whether the facilitation caused by related contexts for the CD subject nouns was equivalent to the facilitation caused by the CI subject nouns themselves. 这些数据表明，上下文对CI项目没有影响，但对CD项目有影响。更具体地说，当主语名词与目标属性高度相关时，相关的上下文并未增加对属性的启动效应。然而，当主语名词与该属性弱相关时，相关的上下文确实增加了对属性的启动。目前尚不清楚相关上下文对CD主语名词所产生的促进作用，是否等同于CI主语名词本身所产生的促进作用。
Mean latency for the correct "true" trials was 1,226 msec, and for the correct "false" trials, it was 1,253 msec. The average "true" latency for 13 of the 18 subjects was less than the average "false" latency. The average error rate per subject for all 60 test trials was 2.8%. The average error rate per subject for the 30 "true" trials was 5%, and for the 30 "false" trials, it was .6%. The average error rates per subject for the six subject relation by predicate relation cells of the design are shown in Table 3. These data, in conjunction with the latencies, indicate there was no speed-accuracy tradeoff. Notably, the most errors occurred for sentences having weakly related subject nouns and unrelated contexts. When questioned at the end of the experiment about errors on these sentences, subjects said they believed the correct response was "true" in all cases. They also indicated they had realized this almost immediately after responding "false." This suggests that CD properties in irrelevant contexts are normally inactive and that errors for these sentences occur when subjects decide to respond prior to this information's becoming active. 正确的“真”试验的平均延迟时间为1,226毫秒，而正确的“假”试验的平均延迟时间为1,253毫秒。18名被试中有13人的“真”试验平均延迟时间小于“假”试验平均延迟时间。在所有60次测试试验中，每名被试的平均错误率为2.8%。30次“真”试验中每名被试的平均错误率为5%，而30次“假”试验中的平均错误率为0.6%。表3展示了六个由“主语关系×谓语关系”构成的设计单元格中，每位被试的平均错误率。这些错误率数据结合延迟数据表明，在此任务中不存在速度-准确率权衡（speed-accuracy tradeoff）。值得注意的是，主语名词弱相关且上下文无关的句子出现了最多的错误。当实验结束被问及这些句子上的错误时，被试表示他们其实在所有情况下都认为正确的反应应该是“真”。他们还表示，在按下了“假”的反应键后，几乎立刻就意识到了这一点。这表明无关上下文中的CD属性通常是不活跃的，当被试决定在这些信息变为活跃状态之前做出反应时，就会在这些句子上发生错误。
During the postexperimental interviews, most subjects reported not trying to guess properties before their presentation. The most common strategy involved focusing attention on the subject noun and forgetting the predicate while waiting for the property. All subjects reported either rehearsing the subject noun or focusing on it during the waiting period. Several subjects said they imaged referents of the subject nouns; several said they rehearsed the subject nouns once and then focused on them until presentation of the property. All but one subject reported that the psychological status of the predicate was either peripheral or gone from consciousness. Some subjects indicated that trying to maintain the predicate interfered with the task. In general, subjects appeared to be focusing only on the subject noun, believing this would maximize their ability to perform the verification task. Interestingly, the predicates still had an effect, as shown by the results for the CD properties. 在实验后的访谈中，大多数被试报告说他们没有试图在属性呈现之前去猜测它。最常见的策略是在等待属性出现的过程中将注意力集中在主语名词上，并忘掉谓语。所有被试都报告说在等待期间复述了主语名词或将注意力集中在它身上。几名被试说他们在脑海中对主语名词所指代的事物进行了表象；几名被试说他们将主语名词复述了一遍，然后就一直专注于它，直到属性呈现。除了极个别的一名被试外，其他所有被试都报告说，谓语的心理状态要么处于边缘地位，要么从意识中消失了。一些被试指出，试图维持谓语的记忆会干扰任务的执行。总体而言，被试似乎只将注意力集中在主语名词上，相信这能使他们最大程度地完成验证任务。有趣的是，从CD属性的结果来看，谓语仍然发挥了作用。
Discussion
讨论
These data are consistent with the distinction between CI and CD properties. Some properties are CI because their verification is unaffected by contextual relevance. Others are CD because their verifications are faster in relevant contexts than in irrelevant contexts (a facilitation of 145 msec in this experiment). These data suggest that CD properties are not activated in irrelevant contexts. Specifically, properties weakly related to subject nouns were verified 237 and 291 msec more slowly than properties highly related to subject nouns for control and unrelated sentences, respectively. It seems unlikely that differences of this size could occur if the weakly related properties were activated by their respective words. Rather, these differences may largely reflect the time it takes to activate these properties. 这些数据与CI和CD属性之间的区分是一致的。有些属性是CI，因为它们的验证不受上下文相关性的影响。另一些则是CD，因为它们在相关上下文中的验证速度快于在无关上下文中的验证速度（在本实验中起到了145毫秒的促进作用）。这些数据表明，CD属性在无关的上下文中是不被激活的。具体而言，对于控制句子和无关句子，与主语名词弱相关的属性其验证速度分别比与主语名词高度相关的属性慢了237毫秒和291毫秒。如果弱相关的属性也是由其相应的词汇激活的话，似乎不太可能出现如此之大的差异。相反，这些差异可能在很大程度上反映了激活这些属性所需的时间。
These results provide a functional account of property availability: Highly accessible properties of a concept are available independent of context, whereas less accessible ones are available only in relevant contexts. Conrad (1978) has also found results consistent with this account. Her task employed interference in a color naming task as the dependent variable. On each trial, subjects read a sentence and reported the ink color of a subsequent word. For "true" trials, the word in colored ink was either a highly related or weakly related property of the last word in the sentence. This factor was crossed with whether or not the sentence context made the property in colored ink relevant to the last word in the sentence. The results were analogous to those in this experiment. The amount of interference for the highly related properties was independent of contextual relevance. For the weakly related properties, however, interference occurred only when the context made the critical property relevant to the final sentence word. 这些结果为属性的可用性提供了一个功能性的解释：概念中高度可访问的属性是独立于上下文而可用的，而可访问性较低的属性则仅在相关上下文中可用。Conrad (1978) 也发现了与该解释一致的结果。她的任务以颜色命名任务中的干扰量作为因变量。在每次试验中，被试阅读一个句子，并报告随后出现的词汇的墨水颜色。对于“真”试验，带有颜色的词汇或者是句子最后一个词的高度相关属性，或者是弱相关属性。这个因素与句子上下文是否使彩色词属性与句子最后一个词相关进行了交叉设计。其结果与本实验的结果类似。高度相关属性的干扰量独立于上下文的相关性。然而，对于弱相关的属性，只有当上下文使关键属性与句子最后一个词相关时，才会发生干扰。
Tabossi and Johnson-Laird (1980) also found results similar to those reported here. In a property-verification task, in which only predicate relatedness was systematically varied, subjects were faster to verify properties in relevant than in irrelevant contexts. This indicates that some of the properties must have been CD. If they had all been CI, this effect would not have occurred, given the results of the current experiment and those of Conrad (1978). Besides using contexts that primed the target property and contexts that primed no property of the target noun, Tabossi and Johnson-Laird also used contexts that primed a property of the target noun other than the target property. This third type of context led to the longest verification times. But since strength of association between the target noun and the target property was not controlled, it is not clear whether this interference effect occurred for CI properties, CD properties, or both. Nevertheless, this effect further constrains a functional account of property availability: Contexts can inhibit the activation of properties, although this may not be true of all properties. Tabossi和Johnson-Laird (1980) 也发现了与本文报告相似的结果。在一个只系统地改变了谓语相关性的属性验证任务中，被试在相关的上下文中验证属性的速度快于在无关上下文中。这表明其中一些属性必须是CD属性。如果它们全都是CI属性，考虑到当前实验和Conrad (1978) 的结果，这种效应就不会发生。除了使用启动目标属性的上下文和不启动目标名词任何属性的上下文外，Tabossi和Johnson-Laird还使用了一种上下文，它启动的是目标名词除目标属性之外的其他属性。第三种类型的上下文导致了最长的验证时间。但是，由于没有控制目标名词与目标属性之间的联想强度，目前尚不清楚这种干扰效应是发生在CI属性上、CD属性上，还是两者皆有。尽管如此，这种效应进一步约束了对属性可用性的功能性解释：上下文可以抑制属性的激活，尽管这可能并不适用于所有属性。
In the current experiment and in Tabossi and Johnson-Laird's (1980) Experiment 2, the context sentences were presented 6 sec before the target properties (Conrad, 1978, did not report the details of her procedure). These experiments, therefore, are not informative at any level more specific than a functional one. This is because both automatic and conscious priming effects have been shown to occur well within 2 sec (Neely, 1977; Posner & Snyder, 1975; Swinney, 1979; Tanenhaus et al., 1979). However, Posner and Snyder's (1975) theory of attention may be an interesting way to think about property availability. They propose two types of attentional processes: (1) unconscious, automatic processes that are the result of past learning, and (2) conscious processes that are subject to capacity limitations. Although both types cause priming, conscious processes do so more slowly and interfere with other processing. Viewing property availability in this framework, the perception of a word may automatically activate its CI properties. In contrast, conscious attention may be responsible for activating relevant CD properties and for keeping both types of property active via rehearsal. (It is also possible that some CD properties are automatically activated.) Finally, focusing conscious attention away from automatically activated properties may eventually inhibit their verification. 在当前的实验和Tabossi和Johnson-Laird (1980) 的实验二中，上下文句子在目标属性出现前6秒就展示了（Conrad, 1978 没有报告其程序的细节）。因此，除了功能层面，这些实验在更具体的层面上并没有提供更多的信息。这是因为已经证明，自动启动效应和有意识的启动效应在2秒内就能充分发生（Neely, 1977; Posner & Snyder, 1975; Swinney, 1979; Tanenhaus et al., 1979）。然而，Posner和Snyder (1975) 的注意力理论可能是思考属性可用性的一种有趣方式。他们提出了两种类型的注意过程：(1) 无意识的自动化过程，它是过去学习的结果；(2) 受能力限制的有意识过程。虽然这两种过程都会引起启动，但有意识过程启动较慢，并且会干扰其他加工。在这个理论框架下看待属性的可用性：感知一个词可能会自动激活其CI属性；相比之下，有意识的注意可能负责激活相关的CD属性，并通过复述保持这两类属性处于激活状态。（也有可能某些CD属性是自动激活的）。最后，将有意识的注意力从自动激活的属性上转移开，可能最终会抑制对它们的验证。
This information processing account of property availability is consistent with the results reported here and with those of Conrad (1978): CI properties are always available because they are automatically activated by their respective words and are kept active by conscious rehearsal; CD properties are available only in irrelevant contexts because they are either automatically or consciously activated via contexts and are kept active by conscious rehearsal. This account also explains the Tabossi and Johnson-Laird (1980) interference effect: The verification of an automatically activated property may be inhibited if context focuses attention away from it. 这种从信息加工角度对属性可用性的解释，与本文报告的结果以及Conrad (1978) 的结果是一致的：CI属性之所以总是可用，是因为它们被相应的词汇自动激活，并通过有意识的复述保持激活状态；CD属性仅在无关上下文中（译注：此处原文可能笔误，按上下文逻辑应为“仅在相关上下文中”可用）可用，因为它们通过上下文被自动或有意识地激活，并通过有意识的复述保持活跃状态。这种解释也说明了Tabossi和Johnson-Laird (1980) 实验中的干扰效应：如果上下文将注意力从某个自动激活的属性上转移开，那么对该属性的验证可能会被抑制。
Priming in this experiment appears at first glance to be nonadditive (cf. Foss, Cirilo, & Blank, 1979). For the CI properties, priming from the word and from the context did not add, since CI related-context sentences did not lead to faster verifications than CI unrelated-context sentences (i.e., there was no additional priming from the contexts). But in the Posner and Snyder (1975) framework, this pattern could well be additive. CI properties may receive their initial activation from encoding their respective words, this automatic activation dissipating within a few hundred milliseconds. But once these properties become active, they may receive conscious attention, which increases as automatic activation decreases. The activation of CI properties may therefore be additive in the sense that different processes are responsible for maintaining a high level of activation. 乍一看，本实验中的启动似乎是非叠加的（参见 Foss, Cirilo, & Blank, 1979）。对于CI属性来说，来自词汇的启动和来自上下文的启动并没有叠加，因为CI相关上下文句子并没有比CI无关上下文句子带来更快的验证（即没有来自上下文的额外启动）。但在Posner和Snyder (1975) 的框架中，这种模式很可能是叠加的。CI属性可能在编码相应的词汇时获得了初始激活，这种自动激活在几百毫秒内消散。但是一旦这些属性变得活跃，它们可能会接收有意识的注意，这种有意识的注意会随着自动激活的衰减而增加。因此，在不同过程负责维持高激活水平的意义上，CI属性的激活可能是叠加的。
Finally, it is necessary to comment on the activation of CD properties. Functionally speaking, these properties are available in relevant but not in irrelevant contexts. But trying to explain this in information processing terms quickly becomes complex. CD properties may become available in two ways. First, they may actually be stored in a concept and be activated by contexts containing similar or associated information. Certain noun-property relations in this experiment appear to have been of this type (e.g., "snake-can be a pet"; "fingers-can be used for eating"; "frog-can be eaten"). Just how contexts activate these properties is a topic worthy of future interest. The second way CD properties can become available is via inference. Certain CD properties may not be stored in a concept but may be computed with various inference procedures (e.g., cognitive economy; Collins & Quillian, 1969; Conrad, 1972). Certain noun-property relations in this experiment may have been of the inference type for certain subjects (e.g., "basketball-can float"; "pencil-can pierce something"; "zebra-has ears"). The range and nature of these inference processes are other topics worthy of future interest. In particular, they appear to present a problem for theories of semantics, which usually try to characterize word meanings with finite sets of properties. 最后，有必要对CD属性的激活进行评论。从功能上讲，这些属性在相关上下文中可用，而在无关上下文中不可用。但试图用信息加工的术语来解释这一点会很快变得复杂。CD属性的可用性可能源于两种方式。首先，它们可能实际上存储在某个概念中，并被包含相似或关联信息的上下文激活。本实验中的某些“名词-属性”关系似乎属于这种类型（例如，“蛇-可以做宠物”；“手指-可以用来吃东西”；“青蛙-可以被吃掉”）。上下文具体是如何激活这些属性的，是一个值得未来关注的话题。CD属性变得可用的第二种方式是通过推理。某些CD属性可能并没有存储在概念中，而是通过各种推理过程计算出来的（例如，认知经济性原则；Collins & Quillian, 1969; Conrad, 1972）。在本实验中，对于某些被试来说，特定的“名词-属性”关系可能是推理类型的（例如，“篮球-能漂浮”；“铅笔-能刺穿东西”；“斑马-有耳朵”）。这些推理过程的范围和性质是另一个值得未来关注的主题。特别是，它们似乎对通常试图用有限的属性集合来描述词义的语义学理论提出了难题。
EXPERIMENT 2
实验二
A much different task was used in this experiment to further demonstrate the distinction between CI and CD information. Subjects judged the similarity of instance pairs drawn from various categories (e.g., "desk-sofa" from "furniture"). Two types of categories, common and ad hoc, were used. Common categories are highly conventional categories, such as those studied by Rosch, Smith, and their colleagues (e.g., Rips, Shoben, & Smith, 1973; Rosch, 1975; Rosch & Mervis, 1975; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976; Smith et al., 1974). Examples of these categories are "birds," "furniture," and "vegetables." In contrast, ad hoc categories are highly unusual categories that are rarely, if ever, used (Barsalou, Note 2). As a result, they are not well established in memory. Examples of these categories are "things that have a smell," "things that float," and "things that can be thrown." 本实验使用了一个截然不同的任务来进一步证明CI和CD信息之间的区别。被试判断从各种类别中抽取的实例对的相似性（例如，来自“家具”类的“书桌-沙发”）。使用了两种类型的类别：常见类别（common categories）和特设类别（ad hoc categories，又称临时类别）。常见类别是高度常规的类别，例如Rosch、Smith及其同事研究的类别（例如，Rips等, 1973; Rosch, 1975; Rosch & Mervis, 1975; 等等）。这些类别的例子有“鸟类”、“家具”和“蔬菜”。相比之下，特设类别是极不寻常的类别，很少（如果有的话）被使用（Barsalou，注2）。因此，它们在记忆中并没有被稳固地建立起来。这类范畴的例子有“有气味的东西”、“能漂浮的东西”和“能被扔的东西”。
Half the subjects received the category name prior to judging the similarity of each pair (the context condition); the remaining subjects did not receive the category names (the no-context condition). The predictions for this experiment are derived from Barsalou's (Note 2) finding that the properties shared by common category instances are usually CI, whereas the properties shared by ad hoc category instances may often be CD. For example, it is fairly obvious that carrots and broccoli share properties common to vegetables. However, it is not obvious that basketballs and logs share properties common to things that float. It follows that the similarity of pairs from ad hoc categories should be greater when these pairs are preceded by their category names than when they are not. This is because the category names activate shared properties that are normally inactive. Thus, there should be more common properties active in the context than in the no-context condition for ad hoc categories. In contrast, the similarity of pairs from common categories should not be increased by the addition of category names. This is because the shared properties are equally active with and without context. Combining the different patterns for ad hoc and common categories, the central prediction for this experiment is that there should be a Context by Category Type interaction. The difference in similarity between pairs from common and ad hoc categories should be less with context than without. 一半被试在判断每一对实例的相似性之前接受了类别名称（上下文条件）；其余被试没有收到类别名称（无上下文条件）。本实验的预测源自Barsalou（注2）的发现：常见类别实例所共享的属性通常是CI，而特设类别实例所共享的属性通常是CD。例如，胡萝卜和西兰花共享蔬菜共有的属性，这是相当明显的。然而，篮球和原木共享能漂浮事物的共有属性，这就不那么明显了。由此推论，特设类别的实例对在前面加上其类别名称时，其相似度应该高于未加名称时。这是因为类别名称激活了通常处于不活跃状态的共享属性。因此，对于特设类别而言，在上下文中活跃的共同属性应该比无上下文条件下多。相比之下，常见类别的实例对的相似性不应因增加类别名称而提高。这是因为共享属性无论有没有上下文都一样活跃。结合特设类别和常见类别的不同模式，本实验的核心预测是：应该存在一个“上下文×类别类型”的交互作用。有上下文条件时，常见类别和特设类别实例对之间的相似度差异应该小于无上下文时的差异。
One other prediction for this experiment also follows from Barsalou (Note 2). The similarities should generally be greater for common than for ad hoc categories. This is because common categories are some of the categories having the highest intraclass similarity, whereas the exemplars of ad hoc categories often have much less in common. This effect is not relevant to the purpose of the experiment, but it is expected to occur. 本实验的另一个预测同样基于Barsalou（注2）。常见类别的相似度通常应大于特设类别。这是因为常见类别是具有最高组内相似度的一些类别，而特设类别的范例往往共同点少得多。这一效应虽然与本实验目的无关，但预计是会发生的。
Method
方法
Materials and Design. Twenty common categories were selected from Battig and Montague (1969) and Rosch (1975). These categories intuitively appeared to be well-known and often used. Twenty ad hoc categories were selected that appeared to be atypical and infrequently used. Two instances were chosen from each category. The common category instances were selected such that the category properties shared by these instances appeared to be CI. The ad hoc category instances were selected such that the category properties shared by these instances appeared to be CD. Examples of the materials are shown in Table 4. **材料与设计。**从Battig和Montague (1969) 以及Rosch (1975) 中选取了20个常见类别。直觉上，这些类别似乎众所周知且经常使用。选取了20个似乎是非典型的、且不常使用的特设类别。从每个类别中选择了两个实例。所选常见类别的实例确保了它们所共享的类别属性表现为CI属性。特设类别实例的选择则确保了它们所共享的类别属性表现为CD属性。材料示例见表4。
[Table 4 Placeholder] [表4 占位符]
Eight judges verified that the CI properties were indeed more accessible for the common category instances than the CD properties were for the ad hoc category instances. The judges read the name of an instance on one side of an index card and then read the category name on the other side. They rated how much reading the instance name brought to mind the properties associated with the category name. The judges used a scale from 1 to 7, on which 1 meant the properties did not come to mind at all and 7 meant the properties immediately came to mind. Each judge rated only one instance per pair, to avoid priming effects between instances. So, four judges rated each instance, and each judge rated 20 common category instances and 20 ad hoc category instances. The 40 cards were randomly ordered for each judge. The mean accessibility rating for common category instances was 5.52, and for ad hoc category instances, it was 2.32 [\(F(1,6)=164.39\), \(p<.001\)]. This indicates that there was a substantial difference in property accessibility between the common and ad hoc category materials. 八名评判者验证了常见类别实例的CI属性确实比特设类别实例的CD属性更易于访问。评判者阅读索引卡片一面的实例名称，然后阅读另一面的类别名称。他们评定阅读该实例名称在多大程度上会使人联想到与该类别名称相关的属性。评判者使用1到7的评分量表，1表示完全没有想到该属性，7表示立即想到了该属性。每位评判者对每对实例只评定其中一个，以避免实例间的启动效应。因此，四名评判者评定了每个实例，并且每位评判者评定了20个常见类别实例和20个特设类别实例。40张卡片对每位评判者进行随机排序。常见类别实例的平均可访问性评分为5.52，而特设类别实例为2.32 [\(F(1,6)=164.39\)，\(p<.001\)]。这表明，在常见类别和特设类别材料之间，属性的可访问性存在实质性的差异。
Two versions of the pairs were constructed. In each version, the 40 pairs were randomly ordered, as were the two words in each pair. The pairs were typed onto two pages, 20 per page. In the context condition, the category name appeared to the left of each pair; in the no-context condition, the pairs appeared in isolation. Thus, there were four lists: two versions of the context list and two versions of the no-context list. To the right of each pair appeared the integers from 1 to 9. At the top of the page, above this block of response scales, appeared labels for the scale. Above 1 appeared "not similar at all," above 9 appeared "very similar," and above the remaining integers appeared "increasing similarity." 构建了两个版本的实例对。在每个版本中，40个对偶以及对偶中的两个单词都被随机排序。这些对偶被打印在两页纸上，每页20对。在上下文条件中，类别名称出现在每一对的左侧；在无上下文条件中，这些对孤立出现。因此，共有四种列表：两种版本的上下文列表和两种版本的无上下文列表。在每对的右侧显示了1到9的整数。在页面顶部、这块反应量表上方，出现了量表的标签。在1的上方是“完全不相似”，在9的上方是“非常相似”，在其余的整数上方是“相似度逐渐增加”。
Subjects and Procedure. The subjects were 28 Stanford students participating to earn course credit. Fourteen subjects were randomly assigned to the context condition and 14 to the no-context condition. Within each of these groups, half the subjects received each version of the list. Subjects were asked to think of the thing to which each word in a pair referred. They were then to judge the similarity of these two referents. Subjects were told about the scale and asked to circle one of the numbers for each pair to indicate their judgment. Subjects in the context condition were told that each pair was preceded by the name of a category to which the words in the pair belonged. **被试与程序。**28名参与获取课程学分的斯坦福学生担任被试。14名被试被随机分配到上下文条件，另外14名分配到无上下文条件。在这些组内，一半被试接收每个版本的列表。指导语要求被试想出对偶中每个单词所指代的事物，然后判断这两个指代物的相似度。被试被告知了量表的含义，并被要求为每一对圈出一个数字来代表他们的判断。处于上下文条件下的被试被告知，每一对词的前面都有一个类别名称，这对词均属于该类别。
Results
结果
The reliability of the mean ratings for the pairs was computed using the intraclass correlation for averages (Guilford & Fruchter, 1973). The reliability of the mean ratings was .96 for the no-context condition and .88 for the context condition. 配对平均评分的信度是使用平均数的组内相关系数计算的（Guilford & Fruchter, 1973）。无上下文条件的平均评级信度为0.96，上下文条件为0.88。
A four-way ANOVA, context by category type by categories by subjects, was performed on the data. Since categories and subjects were both random factors, it was necessary to compute quasi-F's (H. H. Clark, 1973; Winer, 1971). The relevant means from this analysis are shown in Table 5. There was no main effect of context (\(F^{\prime}<1\)). However, there was a main effect of category type [\(F^{\prime}(1,62)=52.80\), \(p<.001\)], common categories exhibiting more similarity than ad hoc categories, as predicted. Most important, there was a Context by Category Type interaction [\(F^{\prime}(1,48)=12.50\), \(p<.001\)]. The difference between common and ad hoc categories was less with context than without, as predicted by the definitions of CI and CD properties. Planned comparisons were computed by performing separate ANOVAs on only the relevant data. For the ad hoc categories, the similarities were higher with context than without [\(F^{\prime}(1,35)=8.31\), \(p<.01\)]. In contrast, the context manipulation had no effect on the similarity of common category pairs (\(F^{\prime}<1\)). 对数据进行了一个四因素方差分析：上下文 × 类别类型 × 类别 × 被试。由于类别和被试都是随机因素，因此有必要计算准F值（quasi-F's; H. H. Clark, 1973; Winer, 1971）。该分析的相关均值如表5所示。没有上下文主效应（\(F^{\prime}<1\)）。然而，类别类型的主效应显著 [\(F^{\prime}(1,62)=52.80\)，\(p<.001\)]，如预测的那样，常见类别比特设类别表现出更高的相似性。最重要的是，存在“上下文×类别类型”交互作用 [\(F^{\prime}(1,48)=12.50\)，\(p<.001\)]。正如CI和CD属性定义所预测的那样，有上下文时常见类别和特设类别之间的差异小于无上下文时。通过仅对相关数据分别进行方差分析计算了计划比较。对于特设类别，有上下文时的相似度高于无上下文时 [\(F^{\prime}(1,35)=8.31\)，\(p<.01\)]。相比之下，上下文操纵对常见类别对的相似度没有影响（\(F^{\prime}<1\)）。
[Table 5 Placeholder] [表5 占位符]
Discussion
讨论
The presence of context reduced the difference in similarity between common and ad hoc categories by one-half. Without context, the difference was 3.21, whereas with context, the difference was 1.61. This is further support for the existence of CI and CD information. The category properties shared by ad hoc category instances were CD, since the similarity of these pairs was greater with relevant context than without. Relevant context was necessary to activate shared properties not activated by the words themselves. In contrast, the category properties shared by common category instances were CI, since the similarity of these pairs did not change across context. Relevant context was not required to activate shared properties activated by the words themselves. 上下文的存在将常见类别和特设类别之间的相似度差异减少了一半。在没有上下文时，差异为3.21，而在有上下文时，差异为1.61。这进一步支持了CI和CD信息的存在。特设类别实例共享的类别属性是CD，因为这些实例对在相关上下文中的相似度大于没有上下文时。相关上下文对于激活那些不被词汇本身激活的共享属性是必要的。相比之下，常见类别实例共享的类别属性是CI，因为这些对的相似度并没有随上下文的改变而改变。不需要相关的上下文去激活那些早已由词汇本身激活的共享属性。
Again, these results only provide support for a functional account of property availability: Some properties of a concept are available independent of context, since they are activated by the respective word, whereas others become available only in relevant contexts. Since subjects had as much time as they needed to perform their judgments, it was not possible to observe the time course of property activation. For this reason, it is not possible to test explanations based on the concepts of automatic and conscious attentional processes. However, the application of the Posner and Snyder (1975) framework to property availability, as discussed for Experiment 1, also makes sense in the context of the current experiment. 同样地，这些结果仅为属性可用性的功能解释提供了支持：概念的某些属性可独立于上下文被访问，因为它们是由相应的词汇激活的；而其他属性仅在相关的上下文中变得可用。由于被试有足够多的时间来进行判断，因此无法观察属性激活的时间进程。由于这个原因，无法检验基于自动和有意识注意过程概念的理论解释。然而，像实验1中讨论的那样，将Posner和Snyder (1975) 的框架应用于属性的可用性，在当前实验的上下文中也是有意义的。
GENERAL DISCUSSION
总讨论
These experiments demonstrate the existence of CI and CD properties. CI properties were shown to be activated by their respective words independently of context. In Experiment 1, the verification of CI properties was unaffected by the relevance of sentence frames. In Experiment 2, the similarity of two concepts was not increased when a context relevant to shared CI properties was presented. In contrast, CD properties were shown not to be activated by their respective words, but only by relevant contexts in which the words appeared. In Experiment 1, the verification of CD properties was faster in relevant than in irrelevant contexts. In Experiment 2, the similarity of two concepts increased when a relevant context activated shared CD properties. 这些实验证明了CI和CD属性的存在。实验表明，CI属性是由相应的词汇独立于上下文而激活的。在实验1中，CI属性的验证不受句子框架相关性的影响。在实验2中，当提供与共享CI属性相关的上下文时，两个概念的相似度没有增加。相反，实验表明，CD属性不会被其相应的词汇独立激活，而仅由词汇出现的相关上下文激活。在实验1中，CD属性在相关上下文中的验证速度比在无关上下文中快。在实验2中，当相关上下文激活了共享的CD属性时，两个概念的相似度增加了。
These findings have implications for theories of natural language semantics and for semantic memory models. Given the existence of CD properties, the meaning of a word is not a fixed set of properties that is activated as a whole every time the respective word is encoded. Rather, the meaning of a word also contains weakly associated and inferable properties that are inactive in irrelevant contexts and active in relevant contexts. Given the existence of CI properties, the meaning of a word is not completely determined by context. Rather, certain properties appear to be automatically activated by a word independently of context. These findings indicate that accounts of natural language semantics should include assumptions regarding (1) the accessibility of semantic properties and (2) the impact of context on the accessibility of these properties. 这些发现对自然语言语义学理论和语义记忆模型具有重要意义。鉴于CD属性的存在，一个词的意义不再是每次对该词进行编码时就作为一个整体被激活的固定属性集。相反，词的意义还包含了弱联想和可推断的属性，这些属性在无关的上下文中处于不活跃状态，而在相关的上下文中处于活跃状态。鉴于CI属性的存在，一个词的意义也不是完全由上下文决定的。相反，某些属性似乎独立于上下文，被一个词自动激活。这些发现表明，对自然语言语义的解释应包括以下假设：(1) 语义属性的可访问性；(2) 上下文对这些属性可访问性的影响。
The remainder of this paper addresses the roles of CI and CD properties in the following cognitive phenomena: cross-classification, problem solving, metaphor, and sentence comprehension. Also discussed are implications for the semantic-episodic distinction. 本文的其余部分探讨了CI和CD属性在以下认知现象中的作用：交叉分类、问题解决、隐喻和句子理解。还讨论了它们对语义-情节记忆区分的影响。
Cross-Classification
交叉分类
Any concept is potentially cross-classifiable into an indefinitely large number of categories (see Barsalou, Note 2). For example, "chair" belongs to "furniture," "gifts," "things to sell at a garage sale," "things that can be used to hold a door open," and so on. Some of the classifications of a concept may be explicitly represented in memory (e.g., a robin is a bird) such that they can be directly accessed from the word for the concept. Many cross-classifications, however, may be implicit, in that they are not prestored but are computed by various inference processes when necessary (cf. Camp, Lachman, & Lachman, 1980). For example, there could be a process that takes any property, X, associated with a concept and infers that instances of the concept belong to the category of things that exhibit X. "Bear" can be cross-classified into things that have fur because fur is associated with bears. 任何概念都有潜力被交叉分类到数量不限的各种类别中（参见 Barsalou，注2）。例如，“椅子”属于“家具”、“礼物”、“在车库甩卖会上出售的东西”、“可以用来抵住门的物品”等等。一个概念的某些分类可能会显式地表征在记忆中（例如，知更鸟是一只鸟），使得它们可以直接通过该概念的词汇访问。然而，许多交叉分类可能是隐式的，因为它们并非预先存储，而是在必要时通过各种推理过程计算得出（参见 Camp, Lachman, & Lachman, 1980）。例如，可能存在一个过程提取与一个概念相关的任何属性 X，并推断该概念的实例属于表现出 X 属性的事物类别。因为皮毛与熊相关联，所以“熊”可以被交叉分类为长有皮毛的动物。
Barsalou (Note 2) proposed a model of how implicit cross-classifications are computed. The model's first assumption is that for each possible classification, there is a set of criterial properties (coupled with a decision rule) used to discriminate category instances. The model's second assumption is that the properties active for a concept on a given occasion are a subset of the properties in that concept, this subset containing CI and CD properties. It follows that the implicit cross-classifications of a concept computable in a given context are those whose criterial properties are contained in the concept's active subset. Consequently, cross-classifications based on CI properties should be possible on any occasion. The category "things that are round" should be computable on all occasions for "basketball" if "round" is CI for "basketball." Cross-classifications based on CD properties should similarly be possible on occasions when these properties are active, but they should not be possible when these properties are inactive. "Things that float" should only be computable for "basketball" if a relevant context (e.g., a need for a life preserver) activates "floats" in the concept for "basketball." Barsalou (Note 2) reports data consistent with this view of cross-classification. Barsalou（注2）提出了一个关于隐式交叉分类如何计算的模型。该模型的第一个假设是，对于每种可能的分类，都有一组用于区分类别实例的标准属性（结合决策规则）。该模型的第二个假设是，在特定场合下一个概念活跃的属性，是该概念中所有属性的一个子集，该子集包含CI和CD属性。由此推论，在特定上下文中能够计算出一个概念的隐式交叉分类的前提是，这些分类的标准属性包含在该概念处于激活状态的属性子集中。因此，基于CI属性的交叉分类在任何场合都应该是可能的。如果“圆形”是“篮球”的CI属性，那么在所有场合下对于“篮球”，“圆的东西”这个类别都应该是可计算的。基于CD属性的交叉分类同样应该在这些属性处于激活状态的场合下可能发生，而在这些属性不活跃时不应发生。只有当相关上下文（例如，需要救生圈）激活了“篮球”概念中的“可漂浮”时，“能漂浮的东西”对于“篮球”才是可计算的。Barsalou（注2）报告的数据与这一关于交叉分类的观点相一致。
Problem Solving
问题解决
The account of cross-classification just discussed bears upon functional fixedness in problem solving (Duncker, 1945). Functional fixedness is the phenomenon of an object's typical function preventing insight into other, less typical functions that might be more useful in a particular situation. In one problem, subjects are presented several objects and asked to use them to support a board (Duncker, 1945). Crucial to solving this problem is using a pair of pliers as a support. But since "can provide support" is not a salient function of pliers, subjects often have difficulty solving the problem. In many such cases, the salient function may be CI, whereas the less salient function may be CD. Perception of the less salient cross-classification may depend on attending to the critical object in the appropriate mental context such that the CD-based classification can be inferred. For example, the CI properties in the concept for things that could support the board might activate relevant CD properties in the representation of pliers if these two concepts were simultaneously active in memory. These CD properties could then be used to infer that "pliers" belongs to "things that could support the board." 刚讨论的关于交叉分类的解释与问题解决中的功能固着（functional fixedness）有关（Duncker, 1945）。功能固着是指一个物体的典型功能会阻碍人们洞察其他在特定情况下可能更有用的非典型功能这一现象。在一个问题中，被试得到几个物体，并被要求用它们来支撑一块木板（Duncker, 1945）。解决这个问题的关键在于将一把钳子用作支撑物。但由于“可以提供支撑”并非钳子的显著功能，被试通常很难解决这个问题。在许多此类情况中，显著功能可能是CI的，而不太显著的功能可能是CD的。对不太显著的交叉分类的感知可能依赖于在适当的心理上下文中注意关键物体，从而推断出基于CD的分类。例如，如果“可以支撑木板的东西”的概念和钳子的表征在记忆中同时活跃，前者概念中的CI属性可能会激活钳子表征中相关的CD属性。然后，这些CD属性可被用于推断“钳子”属于“可以支撑木板的东西”。
Finding a solution to this problem may be delayed because subjects are misled by the CI properties of the critical object. For example, "to grasp something" is probably CI for "pliers." Once this property is automatically activated, subjects may rule out pliers as a possible support. Consequently, the object is not attended to in the context necessary for activating the relevant properties. When subjects get desperate, this initial classification may be discarded such that the object is more carefully scrutinized and properly classified. In support of this, Duncker (1945) and Glucksberg and Danks (1968) have found that it takes longer to solve a problem when attention is drawn to the interfering CI properties. Duncker (1945) distracted subjects by having them use the pliers to grasp something before solving the problem. Glucksberg and Danks (1968) either mentioned the word for the critical object or labeled the object with a nonsense syllable. Mentioning the word delayed solutions, presumably because hearing the word automatically activated interfering CI properties. 解决这个问题的时间可能会被推迟，因为被试被关键物体的CI属性误导了。例如，“夹住东西”可能是“钳子”的CI属性。一旦这个属性被自动激活，被试可能会排除将钳子作为潜在支撑物的可能。因此，在这个足以激活相关属性的必要语境中，该物体就不会被注意到了。当被试一筹莫展时，最初的分类可能就会被抛弃，使得该物体受到更仔细的审视并被正确地重新分类。为了支持这一点，Duncker (1945) 以及 Glucksberg 和 Danks (1968) 发现，当注意力被吸引到具有干扰性的CI属性上时，解决问题需要花费更长的时间。Duncker (1945) 在被试解决问题之前，让他们用钳子去夹东西来分散他们的注意力。Glucksberg 和 Danks (1968) 则是提到关键物体的词汇，或者用无意义的音节来标记它。提到表示该物体的词汇延迟了问题的解决，这大概是因为听到这个词会自动激活具有干扰性质的CI属性。
Sometimes the activation of CI properties may facilitate finding a solution. In the candle problem, subjects are given a candle, some matches, and a box of tacks; their task is to attach the candle to the wall and light it (Duncker, 1945). Usually it takes subjects a while to cross-classify the box as something that could contain the candle. However, Glucksberg and Weisberg (1966) found that having the experimenter label the box as "box" resulted in faster solutions than when the box of tacks was simply labeled "tacks." They argued that using "box" drew attention to an object that was otherwise obscured by what it contained. However, another factor may be involved as well. Assuming that "contains things" is CI for "box," it follows that this property should become available when subjects hear the experimenter say "box." Having this property available should then facilitate cross-classifying the box as something that could contain the candle. 有时候CI属性的激活可能会促进解决方案的发现。在“蜡烛问题”中，被试得到一根蜡烛、一些火柴和一盒图钉；他们的任务是将蜡烛固定在墙上并点燃（Duncker, 1945）。被试通常需要一段时间才能将“盒子”交叉分类为“可以装蜡烛的东西”。然而，Glucksberg 和 Weisberg (1966) 发现，当主试将盒子标记为“盒子（box）”时，被试解决问题的速度比仅仅将这盒图钉标记为“图钉（tacks）”时要快。他们认为，使用“盒子”一词将注意力吸引到了一个原本被其内容物掩盖的物体上。然而，可能还涉及另一个因素。假设“装东西”是“盒子”的CI属性，那么当被试听到主试说出“盒子”时，这一属性就会变得可用。有了这个可用的属性，就会促进被试将盒子交叉分类为“可以装蜡烛的东西”。
Metaphor and Sentence Comprehension
隐喻与句子理解
Ortony (1979) has proposed that metaphoricity depends on a particular type of salience imbalance. Specifically, the property brought to mind by a metaphor should have low salience for the subject and high salience for the referent of the metaphor. For "sermons are sleeping pills," the property "induces drowsiness or sleep" has low salience for "sermons," but high salience for "sleeping pills." In many metaphors, the shared property may be CD in the subject and CI in the referent. In these cases, the CI property in the referent may automatically activate the corresponding CD property in the subject (see Glucksberg, Dial, & Bookin, Note 3). That is, the referent serves as context for the subject, activating relevant CD properties. It follows from Ortony's (1979) analysis that the best metaphors should be those in which the shared property is CD in the subject and CI in the referent. This is because these are the cases in which salience imbalance is maximized. Metaphors in which the shared property is not CD in the subject should not appear as metaphorical, since the property may come to mind for the subject outside the context of the referent. Ortony (1979) 提出，隐喻的性质取决于一种特定类型的显著性失衡（salience imbalance）。具体而言，由隐喻引发在脑海里的属性，在隐喻的“主词（subject）”中应当具有较低的显著性，而在隐喻的“喻词（referent）”中应当具有较高的显著性。例如，在“布道是安眠药”中，“导致困倦或睡眠”这一属性对“布道”而言显著性很低，但对“安眠药”而言显著性很高。在许多隐喻中，这个共享的属性可能是主词的CD属性和喻词的CI属性。在这些情况下，喻词中的CI属性可能会自动激活主词中相应的CD属性（参见 Glucksberg, Dial, & Bookin, 注3）。也就是说，喻词充当了主词的上下文，激活了相关的CD属性。根据Ortony (1979) 的分析推论，最好的隐喻应该是那些共享属性是主词的CD属性、同时是喻词的CI属性的隐喻。这是因为在这些情况下，显著性失衡被最大化了。如果共享属性不是主词的CD属性，那么这类比喻看起来就不会像隐喻，因为即便在没有喻词上下文的情况下，该属性也可能为主词联想到。
In general, the mechanism of CI properties in one word activating CD properties in other words may be central to sentence comprehension. As shown in Experiment 1, the predicate in a sentence can bring to mind properties of the subject (e.g., the predicate in "The rag was used to start the fire" brings to mind "is flammable" for "rag"). There appear to be many other ways in which CI properties of one sentence word activate CD properties of another sentence word. For example, CI properties can bring to mind the appropriate senses of ambiguous words. This occurs from direct objects to verbs. For "John ate X," the instantiation of X determines the sense of "eat" that comes to mind (e.g., consider X = soup, a sandwich, a steak, and so on). Similarly, the CI properties of a noun serve to disambiguate modifiers. For "the broken X," the instantiation of X determines the sense of "broken" that comes to mind (e.g., consider X = bowl, truck, plan, and so on). Similarly, the CI properties of an object in a prepositional phrase determine the sense of the preposition that comes to mind. For "on the X," the instantiation of X determines the sense of "on" that comes to mind [e.g., consider X = table, television, roof (where the subject is a person vs. a fly), and so on]. In general, converging on the intended meaning of a sentence may often involve selecting the properties associated with individual words that result in the most coordinated interpretation. This selection mechanism can be characterized, at least to some extent, as the activation of relevant CD properties in some words by CI properties in other words. This mechanism serves to minimize the number of words necessary for communicating all possible intended meanings. This is because it allows words to be used in many different ways, rather than requiring a different word for every possible meaning. 总的来说，一个词的CI属性激活其他词CD属性的机制可能是句子理解的核心。如实验1所示，句子中的谓语可以让人联想到主语的属性（例如，“用这块抹布来生火”中的谓语会让人联想到“抹布”具有“易燃”的属性）。似乎还有许多其他方式可以使句子中一个词的CI属性激活另一个词的CD属性。例如，CI属性可以使人联想到歧义词在当下合适的词义。这可以从直接宾语对动词的影响中看出。对于“约翰吃（ate）了 X”，X的实例化决定了浮现在脑海中的“吃”的意思（例如，考虑 X = 汤，一个三明治，一块牛排等情形）。同样，名词的CI属性可用于消除修饰语的歧义。对于“破损的（broken） X”，X 的实例化决定了脑海中“破损”的含义（例如，考虑 X = 碗、卡车、计划等情形）。同样，介词短语中介词宾语的CI属性决定了脑海中介词的含义。对于“在 X 的上面（on）”，X的实例化决定了脑海中“on”的含义 [例如，考虑 X = 桌子、电视、屋顶（主语是一个人还是一只苍蝇）等情形]。总的来说，要收敛并确定一个句子的预期意义，通常需要选择与各个词相关的属性，从而产生最协调的解释。这种选择机制，至少在某种程度上可以被描述为通过其他词的CI属性去激活某些词的相关CD属性的过程。这种机制有助于最大限度地减少传达所有可能预期意义所需词汇的数量。因为这就允许词汇以多种不同的方式被使用，而不是为每种可能的意义创造一个不同的词汇。
The Semantic-Episodic Distinction
语义-情节记忆区分
Tulving (1972) proposed a distinction between episodic and semantic memories. Episodic memories represent autobiographical experiences, that is, events coded by space and time. In contrast, semantic memories represent our knowledge of the world and the meanings of words. Although most investigators have not argued for physically separate memories in the brain, many have agreed that there may be different representations and processes associated with each memory type. Recently, this view has come under attack. J. R. Anderson and Ross (1980) and McKoon and Ratcliff (1979) have argued that episodic and semantic memories may be similarly represented and subject to the same processes. Barsalou and Bower (Note 1) further argue that the CI-CD distinction is problematic for the semantic-episodic distinction. If CI properties are automatized, and if practice results in automaticity, then particular processing episodes determine the accessibility of semantic memories. Similarly, CD properties are CD because there have not been many episodes in which the property and the respective word were simultaneously processed. Since the availability of semantic information depends directly on episodic information, it is not clear that two types of memories are needed when one would probably be sufficient. Tulving (1972) 提出了情节记忆（episodic memories）和语义记忆（semantic memories）的区分。情节记忆代表自传式体验，即由空间和时间编码的事件。相比之下，语义记忆代表了我们对世界的知识以及词汇的意义。虽然大多数研究者并未主张在大脑中存在物理上分离的两种记忆系统，但许多人同意不同记忆类型可能存在相关的、不同的表征和过程。最近，这种观点受到了质疑。J. R. Anderson和Ross (1980) 以及 McKoon和Ratcliff (1979) 认为，情节记忆和语义记忆可能是以类似方式表征的，并受相同过程的支配。Barsalou和Bower（注1）进一步提出，CI-CD属性的区别对语义-情节记忆的区分提出了挑战。如果CI属性是自动化表征的，并且如果重复练习导致了自动化，那么特定的加工情节（episodic）就决定了语义记忆的可访问性。同样，CD属性之所以是CD，是因为没有发生过足够多的同时加工该属性及相应词汇的情节。由于语义信息的可用性直接依赖于情节信息，因此在可能一种记忆类型就足够的情况下，并不清楚是否真的需要两种记忆类型。
Barsalou and Bower (Note 1) discuss specific ways in which particular processing episodes may affect the psychology of lexical semantics. To start with, a word can refer to different kinds of instances. "Car" can refer to cars with or without air conditioning. Consequently, the accessibility of "air conditioning" should depend on the type of car someone is used to. In general, properties of a concept not typically found for familiar referents may become CD through disuse. In contrast, properties typically encountered are more likely to become CI. Analogously, particular uses of an object may vary in accessibility. Someone who has just been to a circus may be more likely to categorize "chair" as something to fight lions with. In general, encoding a particular episode in which an object is put to atypical use may make that use more accessible, at least temporarily. Barsalou和Bower（注1）讨论了特定加工情节可能影响词汇语义学心理学机制的具体方式。首先，一个词可以指代不同种类的实例。“汽车（car）”可以指带空调的汽车，也可以指不带空调的汽车。因此，“空调”这一属性的可访问性应该取决于一个人习惯接触哪种类型的汽车。总的来说，那些对熟悉的指代物并不典型的概念属性，可能会因为较少使用而变成CD属性。相比之下，经常遇到的属性更可能成为CI属性。类似地，一个物体的特定用途在可访问性上也会有所不同。一个刚去过马戏团的人可能更有可能将“椅子”归类为用来对抗狮子的工具。通常而言，如果对一个物体被进行了非典型使用的特定情节进行编码，可能会使该用途变得更易于访问（至少在暂时阶段是如此）。
It should be pointed out that CI properties are not necessarily more semantic than CD properties. Episodes can be CI (e.g., "doberman pincher" may always activate a particular, well rehearsed episode of being bitten by one of these dogs), and semantic properties can be CD (e.g., properties that are usually irrelevant for an object, such as "floats" for "basketball"). The primary difference between CI and CD information is simply the means by which they are activated: CI information is activated by the word for a concept, whereas CD information is activated by relevant contexts in which the word is encoded. As suggested by Barsalou and Bower (Note 1), this difference in accessibility is a function of the frequency and recency of processing episodes, regardless of whether the information is an episode, a semantic feature, an affect, or some other type of information. 应当指出的是，CI属性并不一定比CD属性更具有“语义”性质。情节可以成为CI（例如，“杜宾犬”可能会总是激活某人曾经被这类狗咬伤过的那个特定且被反复回忆的情节），而语义属性也可以是CD（例如，通常与某个物体无关的属性，如“篮球”能“漂浮”）。CI和CD信息之间的主要区别仅仅在于它们被激活的方式：CI信息由表示概念的词汇激活，而CD信息则在词汇被编码的相关上下文中被激活。正如Barsalou和Bower（注1）所提出的，这种可访问性的差异是加工情节频率（frequency）和近因效应（recency）的函数，无论这些信息是一个情节、一个语义特征、一种情感，还是其他任何类型的信息。
Finally, E. V. Clark and H. H. Clark (1979) have shown that certain innovative uses of words can result in new meanings for those words. Computing these novel meanings often requires retrieving a particular episode. Consider their example of "teapotting." Suppose someone named Max has a strange habit of rubbing a teapot on the backs of people's legs. Imagine that two people had seen Max do this before, and one of them said, "Max is in trouble, he just teapotted a policeman." The listener would compute the meaning of "teapotted" by retrieving the relevant episodes, even though he or she has never heard the word used that way. Clearly, this example illustrates the necessity of using episodic information to arrive at the speaker's intended meaning. There is no linguistic rule that could generate the exact meaning intended by the speaker in this situation. Instead, the specifics of the meaning are derived from the structure of the relevant episodes. With recurrent uses of "teapot" in this manner, however, the new meaning could eventually be abstracted away from the particulars of episodes and become CI. Thus, particular processing episodes not only enable comprehension of certain linguistic innovations but are also responsible for the respective word senses' becoming well established in memory. In general, changes in word meanings over time may often be the result of changes in the accessibility of CI and CD properties. 最后，E. V. Clark 和 H. H. Clark (1979) 已经表明，对词汇的某些创新性使用可能会使这些词产生新的含义。计算这些新含义通常需要检索某个特定的情节。看看他们关于“teapotting（把茶壶当动词用）”的例子。假设一个名叫Max的人有一个奇怪的习惯，喜欢用茶壶在人们的腿肚子上摩擦。想象一下，有两个人以前见过Max这么做，其中一个人说：“Max有麻烦了，他刚刚‘茶壶（teapotted）’了一个警察。”听者会通过提取相关的情节记忆来计算“teapotted”的含义，即使他或她以前从未听过这个词被这样使用。显然，这个例子说明了使用情节信息去获知说话者意图含义的必要性。在这种情况下，没有任何语言规则能够生成说话者所期望表达的确切含义。相反，这种含义的具体细节是直接从相关的情节结构中衍生出来的。然而，随着以这种方式不断重复使用“茶壶（teapot）”，这种新含义最终可能会从特定的情节细节中抽象出来，并成为一个CI属性。因此，特定的加工情节不仅有助于理解某些语言创新，而且对于相应词义在记忆中牢固确立也负有责任。总的来说，词汇意义随着时间的推移而发生改变，往往可能是CI和CD属性可访问性发生变化的结果。
REFERENCE NOTES
参考注记（Reference Notes）
Barsalou, L. W., & Bower, G. H. A priori determinants of a concept's highly accessible information. Paper presented at the annual meeting of the American Psychological Association, Montreal, September 1980.
- Barsalou, L. W., & Bower, G. H. 概念高度可访问信息的先验决定因素。论文发表于美国心理学协会（APA）年会，蒙特利尔，1980年9月。
Barsalou, L. W. Ad hoc categories and cross-classification. Unpublished manuscript, 1981. 2. Barsalou, L. W. 特设类别与交叉分类。未发表的手稿，1981年。
Glucksberg, S., Dial, P. G., & Bookin, H. B. On understanding nonliteral speech: Can people ignore metaphors? Manuscript in preparation, 1980. 3. Glucksberg, S., Dial, P. G., & Bookin, H. B. 论理解非字面语言：人们能忽略隐喻吗？准备中的手稿，1980年。
REFERENCES
(学术论文的 Reference 为保障检索精确性通常只保留英文版，不作中译)
ANDERSON, J. R., & Ross, B. H. Evidence against the semantic-episodic distinction. Journal of Experimental Psychology: Human Learning and Memory, 1980, 6, 441-466. ANDERSON, R. C., & ORTONY, A. On putting apples into bottles-A problem of polysemy. Cognitive Psychology, 1975, 7, 167-180. ANDERSON, R. C., PICHERT, J. W., GOETZ, E. T., SCHALLERT, D. L., STEVENS, K. V., & TROLLIP, S. R. Instantiation of general terms. Journal of Verbal Learning and Verbal Behavior, 1976, 15, 667-679. BARCLAY, J. R., BRANSFORD, J. D., FRANKS, J. J., MCCARRELL, N. S., & NITSCH, K. Comprehension and semantic flexibility. Journal of Verbal Learning and Verbal Behavior, 1974, 13, 471-481. BATTIG, W. F., & MONTAGUE, W. E. Category norms for verbal items in 56 categories: A replication and extension of the Connecticut category norms. Journal of Experimental Psychology, 1969, 80(Whole No. 3, Pt. 2). BOWER, G. H. Stimulus-sampling theory of encoding variability. In A. W. Melton & E. Martin (Eds.), Coding processes in human memory. Washington, D.C: Winston, 1972. CAMP, C. J., LACHMAN, J. L., & LACHMAN, R. Evidence for direct-access and inferential retrieval in question-answering. Journal of Verbal Learning and Verbal Behavior, 1980, 19, 583-596. CLARK, E. V., & CLARK, H. H. When nouns surface as verbs. Language, 1979, 55, 767-811. CLARK, H. H. The language-as-fixed-effect fallacy: A critique of language statistics in psychological research. Journal of Verbal Learning and Verbal Behavior, 1973, 12, 335-359. COLLINS, A. M., & QUILLIAN, M. R. Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behavior, 1969, 8, 240-247. CONRAD, C. Cognitive economy in semantic memory. Journal of Experimental Psychology, 1972, 92, 149-154. CONRAD, C. Some factors involved in the recognition of words. In J. W. Cotton & R. L. Klatzky (Eds.), Semantic factors in cognition. Hillsdale, N.J: Erlbaum, 1978. DUNCKER, K. On problem solving. Psychological Monographs, 1945, 58(Whole No. 270). ESTES, W. K. Statistical theory of spontaneous recovery and regression. Psychological Review, 1955, 62, 145-154. ESTES, W. K. The statistical approach to learning theory. In S. Koch (Ed.), Psychology: A study of a science (Vol. 2). New York: McGraw-Hill, 1959. Foss, D. J., CIRILO, R. K., & BLANK, M. A. Semantic facilitation and lexical access during sentence processing: An investigation of individual differences. Memory & Cognition, 1979, 7, 346-353. GLASS, A. L., & HOLYOAK, K. J. Alternative conceptions of semantic memory. Cognition, 1975, 3, 313-339. GLUCKSBERG, S., & DANKS, J. Effects of discriminative labels and of nonsense labels upon availability of novel function. Journal of Verbal Learning and Verbal Behavior, 1968, 7, 72-76. GLUCKSBERG, S., & WEISBERG, R. W. Verbal behavior and problem solving: Some effects of labeling in a functional fixedness problem. Journal of Experimental Psychology, 1966, 71, 659-664. GUILFORD, J. P., & FRUCHTER, B. Fundamental statistics in psychology and education. New York: McGraw-Hill, 1973. JACOBY, L. L., CRAIK, F. I. M., & BEGG, I. Effects of decision difficulty on recognition and recall. Journal of Verbal Learning and Verbal Behavior, 1979, 18, 585-600. KATZ, J. J., & POSTAL, P. An integrated theory of linguistic descriptions. Cambridge, Mass: M.I.T. Press, 1964. MCCLOSKEY, M., & GLUCKSBERG, S. Decision processes in verifying category membership statements: Implications for models of semantic memory. Cognitive Psychology, 1979, 11, 1-37. MCKOON, G., & RATCLIFF, R. Priming in episodic and semantic memory. Journal of Verbal Learning and Verbal Behavior, 1979, 18, 463-480. NEELY, J. H. Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 1977, 106, 226-254. OLSON, D. R. Language and thought: Aspects of a cognitive theory of semantics. Psychological Review, 1970, 77, 257-273. ORTONY, A. Beyond literal similarity. Psychological Review, 1979, 86, 161-180. POSNER, M. I., & SNYDER, C. R. R. Attention and cognitive control. In R. L. Solso (Ed.), Information processing and cognition: The Loyola symposium. Hillsdale, N.J: Erlbaum, 1975. RIPS, L. J., SHOBEN, E. J., & SMITH, E. E. Semantic distance and the verification of semantic relations. Journal of Verbal Learning and Verbal Behavior, 1973, 12, 1-20. Rosch, E. H. Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 1975, 104, 192-233. Rosch, E. H., & MERVIS, C. B. Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 1975, 7, 573-605. Rosch, E. H., MERVIS, C. B., GRAY, W. D., JOHNSON, D. M., & BOYES-BRAEM, P. Basic objects in natural categories. Cognitive Psychology, 1976, 8, 382-439. SHIFFRIN, R. M., & SCHNEIDER, W. Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 1977, 84, 127-190. SMITH, E. E., SHOBEN, E. J., & RIPS, L. J. Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review, 1974, 81, 214-241. SWINNEY, D. A. Lexical access during sentence comprehension: (Re)Consideration of context effects. Journal of Verbal Learning and Verbal Behavior, 1979, 18, 645-659. TANENHAUS, M. K., LEIMAN, J. M., & SEIDENBERG, M. S. Evidence for multiple stages in the processing of ambiguous words in syntactic contexts. Journal of Verbal Learning and Verbal Behavior, 1979, 18, 427-440. TABOSSI, P., & JOHNSON-LAIRD, P. N. Linguistic context and the priming of semantic information. Quarterly Journal of Experimental Psychology, 1980, 32, 595-603. TULVING, E. Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization and memory. New York: Academic Press, 1972. TULVING, E., & THOMPSON, D. M. Encoding specificity and retrieval processes in episodic memory. Psychological Review, 1973, 80, 352-373. WINER, B. J. Statistical principles in experimental design. New York: McGraw-Hill, 1971.
NOTES
注释（NOTES）
It should be pointed out that this proposal regarding the existence of CI and CD properties is not a theory of meaning. It simply addresses one aspect of concepts, namely, the accessibility of properties.
- 应当指出，关于存在CI和CD属性的这一提议本身并不是一种“意义理论”。它仅仅探讨了概念结构的一个方面，即属性的可访问性。
The CI control and CD unrelated-context items both have weakly related subject relations and unrelated predicate relations; analogously, the CD control and CI unrelated-context items both have highly related subject relations and unrelated predicate relations. It is not possible, however, to pool the latencies within these two sets of items, since this would make comparisons between unrelated- and related-context items impossible. The proper way to compare related and unrelated contexts is to observe latencies for the same properties under different context conditions. Pooling violates this design, since latencies for the control properties would be included in the unrelated-context conditions but not in the related-context conditions. 2. CI控制项目和CD无关上下文项目均具有弱相关的主语关系和无关的谓语关系；类似地，CD控制项目和CI无关上下文项目均具有高度相关的主语关系和无关的谓语关系。然而，不能将这两组项目中的延迟数据合并（pool）计算，因为这会使得无关上下文项目与相关上下文项目之间的比较成为不可能。比较相关和无关上下文的正确方法是在不同的上下文条件下观察同一属性的延迟时间。合并数据违反了这一设计初衷，因为控制属性的延迟数据会被包含在无关上下文条件中，却不会包含在相关上下文条件中。
As discussed earlier, these senses may not be strictly CD. Even though context is required to converge on a particular sense, many may easily come to mind in no context (Swinney, 1979; Tanenhaus et al., 1979). Consequently, the primary senses of an ambiguous word may lie in the middle ground between CI and CD properties. These senses are CD in the weaker sense that they are attenuated or strengthened by context once their linguistic form has automatically activated them. 3. 如前所述，这些词义可能并不严格属于CD属性。即使需要上下文来收敛于某一个特定的词义，许多词义也很容易在没有上下文的情况下就浮现于脑海（Swinney, 1979; Tanenhaus et al., 1979）。因此，歧义词的主要词义可能处于CI和CD属性之间的中间地带。这些词义是较弱意义上的CD，一旦它们的语言形式将它们自动激活，它们就会受到上下文的削弱或增强影响。
This use of "semantic" is nonstandard, since "semantics" is typically used to refer only to the meanings of words. 4. 这里对“semantic（语义的）”一词的使用是非标准的，因为“语义学”通常只用来指代词汇的意义。