Reasoning about conditional sentences: development of understanding of cues to quantification

The previous literature has reported that when children are asked to judge the truth or falsity of universally quantified conditional sentences of the form If a thing is P then it is Q they typically give responses, e.g., responding "true" whenever there is a case of P and Q even if there are also cases of P and not-Q. Three experiments are reported that address possible sources of this error. Experiment 1 shows that the error survives on sentences that refer to particular things as well as to things of a particular kind, and further shows that articulating the necessity of the consequent (... then it has to be Q) eliminates the error for adults and reduces it for fifth graders, although it does not affect second grade performance. Experiment 2 shows that for second and fifth graders the error survives to problems that are not universally quantified and for second graders to problems that are not conditionals although are otherwise structurally similar. Experiment 3 compares various verbal formulations of such universally quantified conditionals: Second and fifth graders do not make the error when the quantification is expressed with the surface structure that makes its universality most explicit (all things ...); the error tendency is greatest when the indefinite article is used (if a thing ...); and formulations using any fall in between. We argue that such erroneous evaluations of universally quantified conditionals have more to do with the quantificational aspect than the conditional aspect of the problems; children interpret the indefinite article as existential, although they resist the error when the cue to universal quantification is completely clear. The error appears to result more from the surface-structure form of the stimuli than from an inability of children to appreciate the logic of universally quantified conditionals.     

Comprehending quantified sentences: The relation between sentence-picture and semantic memory verification

People were timed as they decided whether quantified sentences like All (some) of the round figures are red were true or false of an accompanying picture. The response latency was a function of the quantifier and the relation between the sets mentioned in the subject and predicate of the sentence. The pattern of latencies was similar to the pattern found for sentences that refer to concepts in semantic memory, e.g., All (some) dogs are animals (Meyer, 1970). This result suggests that the same process may be operating in both domains. Two alternative models of the process are considered. One model assumes that the common process consists of computing the relation between the two sets mentioned in the sentence. The other model assumes that the common process consists of comparing the representation of the sentence to the representation of information computed from the second source. Both models are integrated with broader theories of performance in various comprehension tasks.     

Understanding quantified categorical expressions

A model of categorical inference (Revlis, 1975b) claims that a conversion operation participates in the encoding of quantified, categorical expressions. As a consequence, a reasoner is said to interpret such sentences as “All A are B” in a way that permits it to also be the case that “All B are A.” The present study examines this conception of encoding using a sentence-picture verification task. In two experiments, students were asked to judge whether one of five possible Euler diagrams was true or false of a categorical expression (e.g., All A are B, No A are B, Some A are B, Some A are not B). Verification errors support a three-stage verification model whose major component is access to a “meaning stack” representing the progressive analysis of categorical relations; at the top of that stack is a converted reading of the input sentence. These findings have implications for current conceptions of categorical inference and semantic retrieval.     

Conceptual structure and the structure of concepts: a distributed account of category-specific deficits

We present a new account of the fine-grained structure of semantic categories derived from neuropsychological, behavioral, and developmental data. The account places theoretical emphasis on the functions of the referents of concepts. We claim (i) that the distinctiveness of functional features correlated with perceptual features varies across semantic domains; and (ii) that category structure emerges from the complex interaction of these variables. The representational assumptions that follow from these claims make strong predictions about what types of semantic information are preserved in patients showing category-specific deficits following brain damage. These claims are illustrated with a connectionist simulation which, when damaged, shows patterns of preservation of distinctive and shared functional and perceptual information which varies across semantic domains. The data model both dissociations between knowledge for artifacts and for living things and recent neuropsychological evidence concerning the robustness of functional information in the representation of concepts.     

Favor Referential Representations

Avrutin and Hickok (1993) argue that agrammatic patients have the ability to represent nonreferential or "government" chains ("who... e") but not referential or "binding" chains ("which girl... e"). By contrast, we propose the "referential representation hypothesis," which suggests that agrammatics attempt to cope with their well-known capacity limitations by favoring referential or content-based representations. This predicts that agrammatic patients′ performance should degrade noticeably as task demands increase, and referential demands should take priority over computational ones. In a semantic task, referential phrases should lead to better or more accurate performances. In syntactic tasks, the availability of a referential or content-based representation will interfere with the development of a syntactic representation, resulting in worse syntactic performance on the referential phrases than on nonreferential ones. This predicts that agrammatic patients should incorrectly accept (resumptive) pronoun sentences with a referential wh-phrase because the pronouns will find the semantic or discourse referent of the referential wh-phrase and take it as an antecedent for the pronoun. However, they should reject a (resumptive) pronoun in a sentence with the nonreferential question constituent "who" or "what." "Who" and "what" will remain in syntactic form, since they have only grammatical content and therefore will have only a "nonreferential" syntactic representation. Consequently, they cannot serve as the antecedent of the pronoun. These predictions were largely confirmed by the results of a grammaticality judgement study. Agrammatics performed well on questions with pragmatic biases but failed to distinguish reliably between grammatical and ungrammatical questions where pragmatic biases were neutralized. They assigned especially low ratings to object gap sentences with referential wh-constituents, as predicted. They assigned relatively high ratings to ungrammatical subject pronoun sentences with either type of wh-constituent. The agrammatics accepted ungrammatical reflexive sentences even though syntactic number and gender features alone could have been used to correctly judge the sentences. We attribute this, too, to the unavailability of a reliable syntactic representation of those phrases with referential or extragrammatical semantic content.     

Modeling semantic memory: Effects of presenting semantic information in different modalities

How is semantic information from different modalities integrated and stored? If related ideas are encountered in French and English, or in pictures and sentences, is the result a single representation in memory, or two modality-dependent ones? Subjects were presented with items in different modalities, then were asked whether or not subsequently presented items were identical with the former ones. Subjects frequently accepted translations and items semantically consistent with those presented earlier as identical, although not as often as they accepted items actually seen previously. The same pattern of results was found when the items were French and English sentences, and when they were pictures and sentences. The results can be explained by the hypothesis that subjects integrate information across modalities into a single underlying semantic representation. A computer model, embodying this hypothesis, made predictions in close agreement with the data.     

The Advantage of First Mention in Korean The Temporal Contributions of Syntactic,Semantic, and Pragmatic Factors

Using Korean, we investigated how syntactic, semantic, and pragmatic factors influence the representation of a sentence, in particular, the relative accessibility of different components of a sentence representation. In six experiments, participants performed a probe recognition task after reading each of a series of sentences. We manipulated the rate at which each word of the sentence was presented (250 ms and 500 ms) and the interval between the sentence-final word and the probe-recognition test word (immediate, 500 ms delay, and 1000 ms delay). We also manipulated the syntactic position (subject versus object), semantic role (agent versus patient), and order of mention (first-versus second-mentioned participant) of the probed item. Pragmatic factors (the order of mention) strongly influenced accessibility immediately and through the longest delay, whereas syntactic and semantic factors had little effect.     

The effect ofsome andall on reaction time for semantic decisions

Two experiments tested a model proposed by Meyer (1970) to account for the times required to verify semantic-memory statements quantified byall orsome. Each S was presented with bothall andsome statements in a mixed list, and the discriminability of false statements of the two quantifier types was controlled. In Experiment I positive subset statements ("horses are animals") were verified more quickly when quantified byall rather thansome; the reverse ordering occurred for negative subset statements ("horses are not animals"). Sentences with pseudowords in subject or predicate position took longer to reject than false real-word sentences. These findings contradict :Meyer's theoretical predictions and suggest that his earlier results were artifactual. Experiment II replicated the faster verification of positive subset statements quantified byall. This result was further shown to be predictable from the frequency with which Ss gave the predicate as a completion ofAll/Some S are _. The production frequency of predicates which form subset statements was lower when the quantifier wassome rather thanall. However, holding predicate production frequency constant, sentences with different quantifiers were verified equally quickly.     

The Elimination of Self-Reference: Generalized Yablo-Series and the Theory of Truth

Although it was traditionally thought that self-reference is a crucial ingredient of semantic paradoxes, Yablo (1993, 2004) showed that this was not so by displaying an infinite series of sentences none of which is self-referential but which, taken together, are paradoxical. Yablo’s paradox consists of a countable series of linearly ordered sentences s(0), s(1), s(2),... , where each s(i) says: For each k > i, s(k) is false (or equivalently: For no k > i is s(k) true). We generalize Yablo’s results along two dimensions. First, we study the behavior of generalized Yablo-series in which each sentence s(i) has the form: For Q k > i, s(k) is true, where Q is a generalized quantifier (e.g., no, every, infinitely many, etc). We show that under broad conditions all the sentences in the series must have the same truth value, and we derive a characterization of those values of Q for which the series is paradoxical. Second, we show that in the Strong Kleene trivalent logic Yablo’s results are a special case of a more general fact: under certain conditions, any semantic phenomenon that involves self-reference can be emulated without self-reference. Various translation procedures that eliminate self-reference from a non-quantificational language are defined and characterized. An Appendix sketches an extension to quantificational languages, as well as a new argument that Yablo’s paradox and the translations we offer do not involve self-reference.     

Alternative conceptions of semantic theory

It is argued that theories of semantic memory have diverged in a manner that parallels current linguistic controversy concerning the representation of meaning. The feature-comparison model (Smith, Shoben & Rips, 1974) applies the linguistic theory of Lakoff (1972) to predict people's reaction times to verify sentences, while the marker-search model, described here, uses the type of semantic representation outlined by Katz (1972) to explain a similar range of data. The two models are described and the evidence for each is reviewed. Available evidence supports the marker-search model, but disconfirms a major prediction of the feature-comparison model. It is argued that the feature-comparison model is in principle inadequate as a model of semantic representation, unless its conception of semantic components is substantially alatered.     

When “All the Five Circles” are Four: New Exercises in Domain Restriction

The domain of a quantifier is determined by a variety of factors, which broadly speaking fall into two types. On the one hand, the context of utterance plays a role: if the focus of attention is on a particular collection of kangaroos, for example, then “Q kangaroos” is likely to range over the individuals in that set. On the other hand, the utterance itself will help to establish the quantificational domain, inter alia through presuppositions triggered within the sentence. In this paper, we concentrate our attention on constructions like the following, in which “the square to which ... ” is the critical presupposition trigger:

1. (i)
   Q circles ...
    
2. (ii)
   Q of these circles ...
    
3. (iii)
   Q of these five circles ... ... have the same colour as the square to which they are connected.
    

Many theories predict that all instances of these schemata will give rise to the presupposition that every circle is connected to a square. We present an analysis which predicts that these sentences should generally be accepted in a context in which not all the circles are connected to a square, with one exception only: if a quantified sentence is of type (iii) and Q is non-intersective, then the sentence should be more likely to be rejected. Furthermore, we predict that manipulating the context so as to make the connected circles more or less salient should have an effect on statements with non-intersective quantifiers only. These predictions were tested in a series of experiments.     

Measuring text similarity based on structure and word embedding

The problem of finding the similarity between natural language sentences is crucial for many applications in Natural Language Processing (NLP). An accurate calculation of similarity between sentences is highly needed. Many approaches depend on word-to-word similarity to measure sentence similarity. This paper proposes a new approach to improve the accuracy of the sentence similarity calculation. The proposed approach combines different similarity measures in the calculation of sentence similarity. In addition to traditional word-to-word similarity measure, the proposed approach exploits sentence semantic structure. Discourse representation structure (DRS) which is a semantic representation for natural sentences is generated and used to calculated structure similarity. Furthermore, word order similarity is measured to consider the order of words in sentences. Experiments show that exploiting structural information achieves good results. Moreover, the proposed method outperforms the current approaches on a standard benchmark dataset achieving 0.8813 Pearson correlation with human similarity.     

祈使句和反问句中双重否定的加工机制

采用句-图匹配范式,以可预测对立意义的句子和图片为实验材料,考察汉语双重否定的加工方式以及句子的形式与语义对句子加工的影响,包括2个实验。研究发现：（1）双重否定是整体加工的;（2）在理解初期（250ms）,否定反问句、双重否定祈使句与肯定祈使句的加工模式一致,都是直接表征事件的实际状态;（3）当句子形式与语义不一致时,读者更多地受到句子语义的影响。这表明,在双重否定加工机制的理论解释中,命题表征理论忽略语义,注重句子形式结构,而基于经验模拟理论的两步模拟假设则立足语义,忽略句子形式结构,两者都有一定的合理性,但也都存在一定的片面性。     

Recognition memory for elements of sentences

Although several theoretical positions and a variety of empirical tasks indicate the importance of verbs to sentences, nouns are generally recalled and recognized better in memorial tasks. Three main models can be identified to explain this discrepancy ("Fillenbaum's paradox"). To try to resolve this paradox, several experiments explored the efficiency of various sentence elements as cues in recognition memory. In Experiment I, concreteness of the stimuli did not interact with the type of distractor; however, verb phrase changes were harder to recognize than noun phrase changes when synonym distractors were used. This result was replicated in a forced-choice recognition paradigm (Experiment II) and with whole sentences where the derivational similarity of verbs and nouns was controlled (Experiment IV). The effect could not be attributed to characteristics of the English language (Experiment III) or to superior memory for form information in nouns (Experiment V). The total results are interpreted as suggesting that subjects process different parts of a sentence to different semantic levels, with verbs receiving more semantic representation and nouns more orthographic or phonological representation. The results are taken as support for a "semantic encoding model" of Fillenbaum's paradox.     

Epistemic Contextualism and the Semantics-Pragmatics Distinction

Contextualism, in its standard form, is the view that the truth conditions of sentences of the form ‘S knows that P’ vary according to the context in which they are uttered. One possible objection to contextualism appeals to what Keith DeRose calls a warranted assertability maneuver (or WAM), according to which it is not our knowledge sentences themselves that have context-sensitive truth conditions, but what is pragmatically conveyed by the use of such sentences. Thus, proponents of WAMs argue, the context sensitivity of knowledge attributions is not a semantic phenomenon but a pragmatic one. I examine a number of WAMs and show that each (i) is seriously flawed, or (ii) undercuts standard contextualism if we hold a minimalist conception of semantic content. I propose an alternative form of contextualism that accommodates minimalism and is immune to the second type of WAM, and show that this new form of contextualism shares the virtues of standard contextualism.     

Hemispheric asymmetries in interpreting forms vs colors in ambiguous patterns

Prior to receiving briefly lateralized presentations of ambiguous visual patterns (Rorschach inkblots), 28 normal right-handed males heard two words, each describing a different interpretation of the pattern, e.g., one word might describe an interpretation based on attending to shapes or forms (form choices) and the other might describe an interpretation based on attending to chromatic color (e.g., color choices). Color choices were more frequent than form choices on left-hemisphere (LH) presentations of the patterns, and form choices were more frequent than color choices on right-hemisphere (RH) presentations. These results were interpreted as consistent with the theory that the LH's favored mental representation is semantic (i.e., frequent choices of concepts associated with names of colors in the patterns) and the RH's favored representation is imaginal (i.e., frequent choices of concepts associated with difficult-to-label forms in the patterns).     

The effects of context and content on immediate processing in reading

Four experiments were conducted to examine the processes by which fluent readers comprehend prose. In each study the material was presented a few words at a time on an on-line visual display and the subject pressed a button to move on from one display to the next. The inter-reponse time was used as an index of local processing difficulty. The results of Experiment I indicated that readers pause considerably at the ends of clauses and sentences, and that they show no tendency to speed up across sentences. This pattern of results questions the role of prediction in reading. In Experiments II and III immediate processing was found to be unaffected by two types of syntactically-predictive clue and the effect of a third (semantic) clue was equivocal. Experiment IV replicated the results of Experiment I and showed, in addition, that pausing at the ends of clauses and sentences is a function of the difficulty of the content of the text. More detailed analyses showed that reading rate is modulated by the frequency of the words and by the number of characters in the display. Taken together the results suggest that reading rate is largely determined by the speed with which a reader can access the meanings of words and construct a representation of the text rather than by the speed with which they can formulate and test successive predictions about it.     

Interpretation-based processing: a unified theory of semantic sentence comprehension

We present interpretation-based processing—a theory of sentence processing that builds a syntactic and a semantic representation for a sentence and assigns an interpretation to the sentence as soon as possible. That interpretation can further participate in comprehension and in lexical processing and is vital for relating the sentence to the prior discourse. Our theory offers a unified account of the processing of literal sentences, metaphoric sentences, and sentences containing semantic illusions. It also explains how text can prime lexical access. We show that word literality is a matter of degree and that the speed and quality of comprehension depend both on how similar words are to their antecedents in the preceding text and how salient the sentence is with respect to the preceding text. Interpretation-based processing also reconciles superficially contradictory findings about the difference in processing times for metaphors and literals. The theory has been implemented in ACT-R [Anderson and Lebiere, The Atomic Components of Thought, Lawrence Erlbaum Associates Publishers, Mahwah, NJ, 1998].     

Quantification and semantic memory

Two models are considered for how people verify explicitly quantified sentences, such as All fathers are parents and Some fathers are parents. The models share the same second stage, but have different first-stage mechanisms. In the Predicate Intersection Model, suggested by Meyer (1970), the first stage involves a serial, self-terminating search among names of categories that intersect the predicate category. In the Feature Comparison Model (Smith, Shoben & Rips, 1974a), the first stage involves evaluating the overall relatedness between the subject and predicate categories by comparing their semantic features. To test the models, three reaction time experiments required subjects to verify statements quantified by Some or All. In the first experiment, the semantic relatedness of categories in false Some-statements was varied. Contrary to predictions of the Predicate Intersection Model, related categories increased reaction time for both true and false Some-statements. While the first experiment revealed that All-statements took longer to verify than comparable Some-statements, the second experiment demonstrated that the All-Some difference can be eliminated by presenting both statement types in the same block of trials, also disconfirming the Predicate Intersection Model. Finally, Experiment III examined the meaning of Some-statements in more detail by having subjects interpret the quantifier Some as “some but not all.” With this interpretation, Some-statements took longer to verify than All-statements. Overall the results support the Feature Comparison Model.