Representational Assumptions Research Articles

Principles of metaphor interpretation and the notion of ‘domain’: A proposal for a hybrid model

Metaphor is widely viewed as the (selective) mapping of properties from one conceptual domain (the ‘source’) onto another (the ‘target’). Two different models of metaphor comprehension are distinguished with respect to this characterization of metaphors: the ‘domain-as-a-schema’ model, and the ‘domain-as-a-taxonomic category’ model. These models differ radically from each other with respect to: 1. Their representational assumptions regarding the way knowledge is organized and represented in memory, and 2. Their interpretation principles, namely, the connectivity vs. diagnosticity principles. On the basis of several counter-examples, it is argued that neither model is sufficient to account for certain phenomena regarding metaphor interpretation. As an alternative, a ‘hybrid model’ of metaphor comprehension is outlined. While preserving the explanatory power of each of the other two models, the ‘hybrid model’ is capable of accounting for those counter-examples. A multiple-stage experiment is described, which provides some initial empirical support for the hybrid model.

Lava flow volume and morphology from digitised contour maps: a case study at Mount Etna, Sicily

The volume and morphology of a lava flow-field can be measured by mapping and comparing the topography before and after lava emplacement. Contour maps are a widely available source of topographic data, containing a record of the geomorphological changes at volcanoes due to lava emplacement. This paper explores the use of 1:25,000 scale digitised contour maps for mapping two lava flows of diverse eruption history and morphology at Mount Etna, Sicily. We find that subtracting DEMs created from these maps gives a good representation of volume and shape for thicker lava flows (>10 m mean thickness) in the case of the 1983 lava flow. However, thinner (<10 m) flows, e.g., the 1981 flow, are not represented on this scale of map and the technique cannot be used. Hence, caution is required in using this technique. The assumption of accurate representation of modified topography in contour maps may not always be valid.

Human navigation ability: Tests of the encoding-errormodel of path integration

This paper tests the generality and implications of an ’’encoding-error‘‘ model (Fujita et al. 1993) of humans‘ ability to keep track of their position in space in the absence of visual cues (i.e., by nonvisual path integration). The model proposes that when people undergo nonvisually guided travel, they encode the distances and turns that they experience, and their errors reflect systematic inaccuracies in the encoding process. Thus when people try to return to the origin of travel, they base their response on mis-encoded values of the outbound distances and turns. The two experiments reported here addressed three issues related to the model: (i) whether path integration is context-dependent and if so, how rapidly it adapts to recently experienced distances and turns; (ii) whether effects of experience can be specifically attributed to changes in the encoding process, and if so, what changes; and (iii) whether the encoding process represents distances and turns in the individual paths without considering their spatial relationship to one another (i.e., an object-centered representation). Testing these issues allows us to evaluate and develop the model. Subjects who were blindfolded or had restricted vision were led through two legs of a triangle and the turn between, then tried to return to the origin. Paths varied in whether experienced legs and turns were small or large (Experiment 1) and in variability of return and outbound course (Experiment 2). Response turn, distance and course were determined. The assumption of immutable encoding functions was not supported; encoding processes were context dependent, although they did not adapt within a block of trials. Although effects of experience could be accounted for by the model, the affected parameters were not always as predicted, and in some cases additional parameters were necessary. Results of manipulating variability in return course were consistent with the model‘s assumption of object-centered representation.

Mechanising shared configuration and diagnosis theories through constraint logic programming

Configuration and diagnosis problem-solvers are commonly championed as successes of applied artificial intelligence techniques. A common problem is that problem-solvers typically encode task-specific representation assumptions and simplifications in their domain theories, hindering the reuse of the domain theories between the problem-solvers. While model-based reasoning techniques have been shown to provide an interesting approach to sharing component and device specifications, their respective mechanisations are generally too inefficient. We show how constraint logic programming languages provide a flexible environment in which constraint-based specifications can be effectively shared and efficiently mechanised by exploiting constraint solving and propagation techniques tightly integrated with the backtracking search mechanism of logic programming languages. A component specification language is presented and the mappings from the language to the constraint system and strategies for guiding the search are defined for the respective problem-solvers.

Alignment and category learning.

Recent research shows that similarity comparisons involve an alignment process in which features are placed into correspondence. In 6 studies, the authors showed that alignment is involved in category learning as well. Within a category, aligned matches (feature matches occurring on the same dimension) facilitate learning more than nonaligned matches do (matches on different dimensions), although nonaligned matches still facilitate learning relative to nonmatches. Analogously, feature matches that cross category boundaries hurt learning more if they occur on the same versus a different dimension, and cross-category feature matches on different dimensions hurt learning relative to nonmatching features. Representational assumptions of category learning models must be modified to account for the differences between aligned and nonaligned feature matches.

Alignment and category learning.

Recent research shows that similarity comparisons involve an alignment process in which features are placed into correspondence. In 6 studies, the authors showed that alignment is involved in category learning as well. Within a category, aligned matches (feature matches occurring on the same dimension) facilitate learning more than nonaligned matches do (matches on different dimensions), although nonaligned matches still facilitate learning relative to nonmatches. Analogously, feature matches that cross category boundaries hurt learning more if they occur on the same versus a different dimension, and cross-category feature matches on different dimensions hurt learning relative to nonmatching features. Representational assumptions of category learning models must be modified to account for the differences between aligned and nonaligned feature matches.

Knowledge digraph contribution analysis of protocol data

A knowledge digraph defines a set of semantic (or syntactic) associative relations among propositions in a text (e.g., the conceptual graph structures of Graesser & Clark, 1985, and the causal network analysis of Trabasso & van den Broek, 1985). This article introduces the Knowledge Digraph Contribution (KDC) data analysis methodology for quantitatively measuring the degree to which a given knowledge digraph can account for the occurrence of specific sequences of propositions in recall, summarization, talk‐aloud, and question‐answering protocol data. KDC data analysis provides statistical tests for selecting the knowledge digraph that “best fits” a given data set. KDC data analysis also allows one to test hypotheses about the relative contributions of each member in a set of knowledge digraphs. The validity of specific knowledge digraph representational assumptions may be evaluated by comparing human protocol data with protocol data generated by sampling from the KDC distribution. Specific concrete examples involving the use of actual human recall protocol data are used to illustrate the KDC data analysis methodology. The limitations of the KDC approach are also briefly discussed.

APPLYING CONSTRAINT SATISFACTION TECHNIQUES TO JOB SHOP SCHEDULING

In this paper, we investigate the applicability of a constraint satisfaction problem solving (CSP) model, recently developed for deadline scheduling, to more commonly studied problems of schedule optimization. Our hypothesis is twofold: (1) that CSP scheduling techniques provide a basis for developing high-performance approximate solution procedures in optimization contexts, and (2) that the representational assumptions underlying CSP models allow these procedures to naturally accommodate the idiosyncratic constraints that complicate most real-world applications. We focus specifically on the objective criterion of makespan minimization, which has received the most attention within the job shop scheduling literature. We define an extended solution procedure somewhat unconventionally by reformulating the makespan problem as one of solving a series of different but related deadline scheduling problems, and embedding a simple CSP procedure as the subproblem solver. We first present the results of an empirical evaluation of our procedure performed on a range of previously studied benchmark problems. Our procedure is found to provide strong costyperformance, producing solutions competitive with those obtained using recently reported shifting bottleneck search procedures at reduced computational expense. To demonstrate generality, we also consider application of our procedure to a more complicated, multi-product hoist scheduling problem. With only minor adjustments, our procedure is found to significantly outperform previously published procedures for solving this problem across a range of input assumptions.

The functional meaning of reverberations for sensoric and contextual encoding

AbstractAmit argues that the local neuronal spike rate that persists (reverberating) in the absence of the eliciting stimulus represents the code of the eliciting stimulus. Based on the general argument that the inferred functional meaning of reverberation depends in part on the type of representational assumptions, reverberations may only be important for the encoding of contextual information.

What is learned during automatization? The role of attention in constructing an instance.

Seven experiments were conducted to examine the role of attention in automatization. Ss searched 2-word displays for members of a target category in divided-atten tion, focused-attention, and dual-task conditions. The main issue was whether attention conditions would affect what Ss learned about co-occurrences of the words in the displays. The attention hypothesis, derived from the instance theory of automaticity, predicts learning of co-occurrences in divided-attention and dual-task conditions in which Ss attend to both words but not in focused-attention conditions in which Ss only attend to 1 word. The data supported the attention hypothesis and therefore the instance theory. This article concerns what is learned during automatization. This is an important question in the automaticity literature, especially from the perspective of memory-based theories, such as the instance theory of automaticity (Logan, 1988,1990, 1992). Memory-based theories assume that automatic performance is based on retrieval of representations of past solutions from memory. What those representations during learning and what is taken out of them during automatic performance are central questions in memory-based theories. The purpose of this article is to evaluate the answers offered by the instance theory of automaticity. The answers are important because they are derived from two of the three main assumptions of the theory: obligatory encoding and instance representation. The obligatory encoding assumption says that attention determines what gets into the representation. The instance representation assumption says that the representations in memory are instances—separate representations of cooccurrences. Attention determines what is in an instance; attention determines which co-occurrences are remembered. The experiments tested this hypothesis.

Confronting Representation(s)

The crisis of representation highlighted by recent postmodern works in the social sciences has exposed the ethnocentric nature of (re)presentations of the world. Although it is recognised that the self-reflexivity which this project introduces has undoubtedly opened up the possibilities for more culturally sensitive work, its rejection of any notion of hierarchy could be read as precluding the deployment of political projects beyond this opening. It is the authors' contention that this need not necessarily be the case, that the postmodern challenge can be met without having to discard a political standpoint. This move is important because the power structures inherent in colonial discourses largely remain in place. Instances are explored where conflicting representations can be brought into collision so that the naturalized assumptions of hegemonic representation can be challenged. This meeting of discourses can facilitate the undermining of certain cultural assumptions in the representative schema which have, until now, remained unquestioned.

Functional and teleological knowledge in the multimodeling approach for reasoning about physical systems: a case study in diagnosis

The basic concepts of the multimodeling approach to the representation of physical systems are presented. Emphasis is placed on the exploitation of many, diverse models of a system for the execution of complex problem solving tasks, such as interpretation, diagnosis, design, simulation, etc. The considered models are based on different ontologies, representational assumptions, epistemological types, and aggregation levels. After a brief survey of the techniques adopted for representing structural and behavioral knowledge, attention is focused on function and teleology. A novel approach is proposed for defining, representing, and using these two types of knowledge which play a fundamental role both from the representation and reasoning perspectives. The fundamental claim is that while teleological knowledge concerns the specific purposes for which the system has been designed, functional knowledge is devoted to bridge the gap between such abstract purposes and the actual structure and behavior of the system, through the concepts of phenomena, processes, and functional roles. A clear definition is provided of all the various epistemological and ontological links existing between the different models. >

What's in a Model? Issues in the Use of Simulation Models to Analyze Student Understanding: A Reaction to Ohlsson, Ernst, and Rees

Ohlsson. Ernst, and Rees (this issue) have produced a wonderfully lucid description of their paradigmatic approach to issues of cognition and instruction. They illustrate their approach by presenting the details of a well worked out computational model. Then, on the basis of simulation runs on the model, they derive some implications for prac tice. The authors have also laid down some rather stringent constraints for commentary. Do not critique our paradigms, they say, unless you can offer a replacement that does better. Do not critique the choice of knowledge representation (production systems) or modeling assumptions (e.g., limitations on working memory) unless you have compelling data to offer in service of your argument and in contradiction of our assumptions. Argument about details is useful, they say, but that won't change the conclusions we draw. So what's a reviewer to do?

Stimulus Generalization and Representation in Adaptive Network Models of Category Learning

An exponential-decay relationship between the probability of generalization and psychological distance has received considerable support from studies of stimulus generalization ( Shepard, 1958 ) and categorization ( Nosofsky, 1984 ). It is shown here how an approximate exponential generalization gradient emerges from stimulus representation assumptions isomorphic to a special case of Shepard's (1987) theory of stimulus generalization in a “configuralcue” network model of human learning that represents stimulus patterns in terms of elementary features and pairwise conjunctions of features ( Gluck &amp; Bower, 1988b ; Gluck, Bower, &amp; Hee, 1989 ). The network model can be viewed as a combination of Shepard's theory and an associative learning rule derived from Rescorla and Wagner's (1972) theory of classical conditioning.

Penguins can make cake

This article is a reply to Matthew Ginsberg's article entitled Universal Planning: An (Almost) Universally Bad Idea. Ginsberg argues that uni-versal plans are infeasible for reasons of compu-tational complexity and concludes that classical planning -- or something like it -- is the appropri-ate basis for activity. He also argues that a number of other systems, including Pengi, are approximately universal plans and subject to the same criticisms. I think that this extension is incorrect. I illustrate my reasoning with a description of Blockhead, a Pengi-like system that efficiently solves the fruitcake problem which Ginsberg argues is infeasible for universal plans. The structure of Blockhead elucidates the relationship between planning, universal plans, and Pengi. I conclude that planning and universal plans are computationally intractable because of the representational assumptions they make.

The psychological reality of linguistically defined gaps

In five experiments, recognition times for an adjective from a sentence-initial noun phrase were examined following the processing of coreferential gaps in syntactic structure as a means of determining whether linguistically defined gaps access their antecedents. Recognition latencies and/or error rates following the processing of gaps were found to be lower than a nonanaphoric control in a number of constructions, including two instances of constructions with NP-movement gaps—namely, the passive and raising-to-subject—and constructions with a PRO gap in an infinitival clause. Additionally, NP-movement gaps were found to produce lower latencies and error rates than PRO-gaps. The results are interpreted as evidence for the psychological reality of representational assumptions of Government and Binding Theory (Chomsky, 1981).

A connectivity model for semantic processing

A connectivity model for semantic processing, based on two major assumptions, is proposed and tested. The representation assumption states that semantically related features are represented by an interconnected structure of features. A network is termed interconnected if each node is linked to at least two other nodes. According to the processing assumption, indirect activation becomes increasingly effective as the number of interconnected nodes increases. The connectivity model thus predicts that more complex concepts will be processed faster than less complex concepts. Experiment 1 was performed to determine the number of features for a set of 72 words and to show that subordinate concepts share common features with their natural superordinate concept. In Experiments 2 and 3 subjects had to judge whether or not a word belonged to a previously defined natural superordinate concept. The results confirm the connectivity model and show that concepts with many features can be classified faster than concepts with only a few features.

On the demand generated by a smooth and concavifiable preference ordering

It is shown that if a consumer's preference ordering is strictly convex and is representable by means of a concave, twice continuously differentiable utility function, then the partial derivative of a demanded commodity with respect to its price is bounded from above in a neighborhood of a price vector at which the demand fails to be differentiable. In the case of two commodities, if the demand does not possess finite derivatives with respect to prices at a certain point, then the partial ‘derivative’ of a commodity with respect to its price is equal to minus infinity. The same result holds for n commodities under ‘almost every’ choice of coordinates in the commodity space. If preferences are weakly convex but the same representation assumption holds, demand may not be single-valued but own-price difference quotients are still bounded from above.

Stochastic Adaptive Control for Exponentially Convergent Time-Varying Systems

This paper shows that the standard stochastic adaptive control algorithms for time-invariant systems have an inherent robustness property which renders them applicable, without modification, to time-varying systems whose parameters converge exponentially. One class of systems satisfying this requirement is those having non-steady-state Kalman filter or innovations representations. This allows the usual assumption of a stationary ARMAX representation to be replaced by a more general state space model.

Relative-Permeability Estimation From Displacement Experiments: An Error Analysis

Summary Parameter estimation can be a superior method of estimating relative permeabilities from displacement (unsteady-state) coreflood experiments. To use the parameter-estimation method, one must choose functional representations of the relative-permeability curves. In previous work, only a simple exponential function was used, which can result in very large errors in estimation of relative permeabilities. In this work, we extend the parameter-estimation method to consider the use of several alternative functional forms, including the highly flexible cubic splines, which should be better able to represent a wide variety of relative-permeability curves. In this way, the error associated with the assumption of a simple functional representation can be greatly decreased. We also develop a procedure to analyze the accuracy with which relative permeabilities may be estimated by parameter estimation. The analysis is used to evaluate the performance of different functional representations in the parameter-estimation method. Furthermore, it can be used to assess quantitatively the effect of experimental design on the accuracy of estimates of relative permeability. Introduction Relative permeabilities frequently are estimated from flow data gathered from laboratory displacement-type coreflood experiments in which one fluid (e.g., water) is pumped through a sample of the reservoir rock that is saturated with another fluid or fluids (e.g., oil and connate water). The flow data typically consist of measurements of two time-dependent quantities - the pressure drop across the core and the volume of displaced fluid recovered. There are two basic approaches for estimating relative permeability curves from displacement data. The methods of Johnson et al.1 and Jones and Roszelle2 are explicit methods in which saturation values and corresponding relative-permeability values at the end of the core sample are estimated directly. Alternatively, an implicit approach may be used. In this approach, relative-permeability curves are chosen so that the quantities simulated with the mathematical model. of the experiment match, in some sense, the experimental data. The explicit methods are suitable only for systems that are described adequately by the Buckley-Leverett model3; thus they are not appropriate for low-flow-rate experiments in which the effects of capillary pressure are important. Furthermore, they require the numerical or graphical differentiation of experimental data. It is well known that inaccuracies in data measurement become amplified by the process of differentiation. Tao and Watson4,5 have quantified errors in relative-permeability estimates caused by differentiation of data in explicit methods and found that they can be quite large. The use of an implicit method overcomes the two limitations of explicit methods - that is, differentiation of data is not required, and general mathematical models of the experimental process can be used (such as those that include the effects of capillary pressure). While trial-and-error adjustment6 could be used to choose estimates for the relative-permeability curves, a more appropriate way is to make use of parameter estimation (or nonlinear regression) techniques. Because relative permeabilities are curves (Le., functions of saturation), parameter estimation is applied by representing the relative-permeability curves by a set of parameters. This can be done by assuming a functional representation, in terms of a set of parameters (or adjustable coefficients), for each relative permeability curve. Then, parameter estimation may be used to estimate the parameters in those functional representations of the relative-permeability curves.