Understanding natural dynamics

When making dynamical judgments, people can make effective use of only one salient dimension of information present in the event. People do not make dynamical judgments by deriving multidimensional quantities. The adequacy of dynamical judgments, therefore, depends on the degree of dimensionality that is both inherent in the physics of the event and presumed to be present by the observer. There are two classes of physical motion contexts in which objects may appear. In the simplest class, there exists only one dynamically relevant object parameter: the position over time of the object's center of mass. In the other class of motion contexts, there are additional object attributes, such as mass distribution and orientation, that are of dynamical relevance. In the former class, objects may be formally treated as extensionless point particles, whereas in the latter class some aspect of the object's extension in space is coupled into its motion. A survey of commonsense understandings showed that people are relatively accurate when specific dynamical judgments can be accurately based on a single information dimension; however, erroneous judgments are pervasive when simple motion contexts are misconstrued as being multidimensional, and when multidimensional quantities are the necessary basis for accurate judgments.     

Understanding the language of thought

Conclusions To conclude then, our question was, What determines the content of a thought? This question gets its punch by being set against the background assumption that thoughts are internal physical occurrences. The answer I have proposed is that thoughts can be classified in four importantly different ways. Introspection yields syntactic categories. These are important for cognitive processing, but they do not correspond to contents. A second way of categorizing thoughts is in terms of narrow content. This is determined by the functional role of the thought in rational architecture together with the way in which that rational architecture is tied to the world through input correlations. Narrow contents are indexical, so to get truth bearers we must augment the narrow contents with the values of the indexical parameters. Propositional content can be taken to consist of pairs of narrow contents and values for indexical parameters. Finally, thoughts can be classified in terms of that clauses. This kind of classification does not uniquely determine propositional content, but describes it in a more general way.     

Connectionist natural language parsing

The key developments of two decades of connectionist parsing are reviewed. Connectionist parsers are assessed according to their ability to learn to represent syntactic structures from examples automatically, without being presented with symbolic grammar rules. This review also considers the extent to which connectionist parsers offer computational models of human sentence processing and provide plausible accounts of psycholinguistic data. In considering these issues, special attention is paid to the level of realism, the nature of the modularity, and the type of processing that is to be found in a wide range of parsers.     

Generalization from natural language text

Generalization and memory are part of natural language understanding. This paper describes a model of generalization that is part of a system for language understanding. the Integrated Partial Parser (IPP). The generalization process includes the retrieval of relevant examples from long-term memory so that the concepts to be created can be determined when new stories are read. Generalizations are analyzed as to their critical parts and evaluated in light of later evidence. The knowledge structures used and a number of examples of the system in operation are presented.     

Learning theory and natural language

Formal Learning Theory may be conceived as a means of relating theories of comparative grammar to studies of linguistic development. After a brief review of relevant concepts, the present paper surveys formal results within Learning Theory that suggest correponding constraints on linguistic theory. Particular attention is devoted to the question: How many possible natural languages are there?     

Computer understanding of natural language

The state of research in automatic natural language processing at Yale University is described. Five programs are discussed: SAM, FRUMP, PAM, TALESPIN, and POLITICS.     

Feature integration in natural language concepts

Two experiments measured the joint influence of three key sets of semantic features on the frequency with which artifacts (Experiment 1) or plants and creatures (Experiment 2) were categorized in familiar categories. For artifacts, current function outweighed both originally intended function and current appearance. For biological kinds, appearance and behavior, an inner biological function, and appearance and behavior of offspring all had similarly strong effects on categorization. The data were analyzed to determine whether an independent cue model or an interactive model best accounted for how the effects of the three feature sets combined. Feature integration was found to be additive for artifacts but interactive for biological kinds. In keeping with this, membership in contrasting artifact categories tended to be superadditive, indicating overlapping categories, whereas for biological kinds, it was subadditive, indicating conceptual gaps between categories. It is argued that the results underline a key domain difference between artifact and biological concepts.     

Reference to numbers in natural language

A common view is that natural language treats numbers as abstract objects, with expressions like the number of planets, eight, as well as the number eight acting as referential terms referring to numbers. In this paper I will argue that this view about reference to numbers in natural language is fundamentally mistaken. A more thorough look at natural language reveals a very different view of the ontological status of natural numbers. On this view, numbers are not primarily treated abstract objects, but rather ‘aspects’ of pluralities of ordinary objects, namely number tropes, a view that in fact appears to have been the Aristotelian view of numbers. Natural language moreover provides support for another view of the ontological status of numbers, on which natural numbers do not act as entities, but rather have the status of plural properties, the meaning of numerals when acting like adjectives. This view matches contemporary approaches in the philosophy of mathematics of what Dummett called the Adjectival Strategy, the view on which number terms in arithmetical sentences are not terms referring to numbers, but rather make contributions to generalizations about ordinary (and possible) objects. It is only with complex expressions somewhat at the periphery of language such as the number eight that reference to pure numbers is permitted.     

Thought beyond language: neural dissociation of algebra and natural language

A central question in cognitive science is whether natural language provides combinatorial operations that are essential to diverse domains of thought. In the study reported here, we addressed this issue by examining the role of linguistic mechanisms in forging the hierarchical structures of algebra. In a 3-T functional MRI experiment, we showed that processing of the syntax-like operations of algebra does not rely on the neural mechanisms of natural language. Our findings indicate that processing the syntax of language elicits the known substrate of linguistic competence, whereas algebraic operations recruit bilateral parietal brain regions previously implicated in the representation of magnitude. This double dissociation argues against the view that language provides the structure of thought across all cognitive domains.     

Applying automated deduction to natural language understanding

Very few natural language understanding applications employ methods from automated deduction. This is mainly because (i) a high level of interdisciplinary knowledge is required, (ii) there is a huge gap between formal semantic theory and practical implementation, and (iii) statistical rather than symbolic approaches dominate the current trends in natural language processing. Moreover, abduction rather than deduction is generally viewed as a promising way to apply reasoning in natural language understanding. We describe three applications where we show how first-order theorem proving and finite model construction can efficiently be employed in language understanding.The first is a text understanding system building semantic representations of texts, developed in the late 1990s. Theorem provers are here used to signal inconsistent interpretations and to check whether new contributions to the discourse are informative or not. This application shows that it is feasible to use general-purpose theorem provers for first-order logic, and that it pays off to use a battery of different inference engines as in practice they complement each other in terms of performance.The second application is a spoken-dialogue interface to a mobile robot and an automated home. We use the first-order theorem prover spass for checking inconsistencies and newness of information, but the inference tasks are complemented with the finite model builder mace used in parallel to the prover. The model builder is used to check for satisfiability of the input; in addition, the produced finite and minimal models are used to determine the actions that the robot or automated house has to execute. When the semantic representation of the dialogue as well as the number of objects in the context are kept fairly small, response times are acceptable to human users.The third demonstration of successful use of first-order inference engines comes from the task of recognising entailment between two (short) texts. We run a robust parser producing semantic representations for both texts, and use the theorem prover vampire to check whether one text entails the other. For many examples it is hard to compute the appropriate background knowledge in order to produce a proof, and the model builders mace and paradox are used to estimate the likelihood of an entailment.