Katja Wiemer-Hastings

Situating abstract concepts

Situating Abstract Concepts Lawrence W. Barsalou (barsalou@emory.edu) Department of Psychology Emory University, Atlanta, GA 30322 USA Katja Wiemer-Hastings (katja@niu.edu) Department of Psychology Northern Illinois University, DeKalb IL 60115 USA Roughly speaking, abstract concepts such as TRUTH refer to entities that are neither purely physical nor spatially constrained (Wiemer-Hastings, Krug, & Xu, 2001). Such concepts pose a classic problem for theories that ground knowledge in modality-specific systems (e.g., Barsalou, 1999, 2003a,b). Abstract concepts also pose a significant problem for traditional theories that represent knowledge with amodal symbols. Surprisingly, few researchers have attempted to specify the content of abstract concepts using feature lists, semantic networks, or frames. It is not enough to say that an amodal node or a pattern of amodal units represents an abstract concept. It is first necessary to specify the concept’s content, before beginning the task of identifying how this content is represented. concepts. For all concepts, participants tended to describe background situations, including information about entities, settings, events, and mental states. Indeed the similarities between concepts were more striking than the differernces. Both analyses also offered support for Hypothesis 2, namely, concrete and abstract concepts differed in their situational foci. Whereas concrete concepts focused more on objects, locations, and behaviors, abstract concepts focused more on social aspects of situations (e.g., people, communication, social institutions) and mental states (e.g., beliefs. complex relations). Intermediate concepts lay in between. Consistent with Hypothesis 3, conceptual structures were most complex for abstract concepts. Abstract concepts were most likely to contain deep hierarchies of large conceptual clusters organized by complex relations. Regarding Hypothesis 4, we see no reason that the content of abstract concepts cannot be represented in simulations. Because their content is perceived in the situations that involve abstract concepts, it could, in principle, be reenacted later when representing them. Clearly, much further research beyond this exploratory study is necessary. Hypotheses A common assumption is that abstract and concrete concepts have little conceptual content in common, if any. Alternatively, we propose that concrete and abstract concepts share important similarities. In particular, we propose that they share common situational content, namely, information about agents, objects, settings, events, and mental states (Hypothesis 1). Where concrete and abstract concepts differ is in their specific foci within background situations. Whereas concrete concepts focus on objects and settings, abstract concepts focus on events and mental states (Hypothesis 2). As a result of these different foci, the representations of abstract concepts are more complex, being less localized in situational content (Hypothesis 3). Finally, because the content of abstract concepts is grounded in situations, modality-specific simulations could, in principle, represent it (Hypothesis 4). Acknowledgement This work was supported by National Science Foundation Grants SBR-9905024 and BCS-0212134 to Lawrence W. Barsalou. References Barsalou, L.W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-609. Barsalou, L.W. (2003a). Abstraction in perceptual symbol systems. Philosophical Transactions of the Royal Society of London: Biological Sciences, 358, 1177-1187. Barsalou, L.W. (2003b). Situated simulation in the human conceptual system. Language and Cognitive Processes, Barsalou, L.W., & Wiemer-Hastings, K. (in press). Situating abstract concepts. In D. Pecher and R. Zwaan (Eds.), Grounding cognition: The role of perception and action in memory, language, and thought. New York: Cambridge University Press. Wiemer-Hastings, K., Krug, J., & Xu, X. (2001). Imagery, context availability, contextual constraint, and abstractness. Proceedings of the 23 rd Annual Conference of the Cognitive Science Society, 1134-1139. Mahwah, NJ: Erlbaum. Method In an exploratory study, we assessed the content of three abstract concepts: TRUTH, FREEDOM, and INVENTION. These concepts were compared to three concrete concepts— BIRD, CAR, and SOFA—and also to three intermediate concepts—COOKING, FARMING, and CARPETING. We first asked participants to produce properties typically true of these concepts. We then analyzed these properties using two coding schemes, one that coded small protocol units, and a second that coded large ones. Results For the complete results, see Barsalou and Wiemer-Hastings (in press). Both coding analyses offered support for Hypothesis 1, namely, common situational content was produced across concrete, intermediate, and abstract

AutoTutor: A simulation of a human tutor

AutoTutor is a computer tutor that simulates the discourse patterns and pedagogical strategies of a typical human tutor. AutoTutor is designed to assist college students in learning the fundamentals of hardware, operating systems, and the Internet in an introductory computer literacy course. Most tutors in school systems are not highly trained in tutoring techniques and have only a modest expertise on the tutoring topic, but they are surprisingly effective in producing learning gains in students. We have dissected the discourse and pedagogical strategies these unskilled tutors exhibit by analyzing approximately 100 hours of naturalistic tutoring sessions. These mechanisms are implemented in AutoTutor. AutoTutor presents questions and problems from a curriculum script, attempts to comprehend learner contributions that are entered by keyboard, formulates dialog moves that are sensitive to the learners contributions (such as short feedback, pumps, prompts, elaborations, corrections, and hints), and delivers the dialog moves with a talking head. AutoTutor has seven modules: a curriculum script, language extraction, speech act classification, latent semantic analysis, topic selection, dialog move generation, and a talking head.

Content differences for abstract and concrete concepts.

Concept properties are an integral part of theories of conceptual representation and processing. To date, little is known about conceptual properties of abstract concepts, such as idea. This experiment systematically compared the content of 18 abstract and 18 concrete concepts, using a feature generation task. Thirty-one participants listed characteristics of the concepts (i.e., item properties) or their relevant context (i.e., context properties). Abstract concepts had significantly fewer intrinsic item properties and more properties expressing subjective experiences than concrete concepts. Situation components generated for abstract and concrete concepts differed in kind, but not in number. Abstract concepts were predominantly related to social aspects of situations. Properties were significantly less specific for abstract than for concrete concepts. Thus, abstractness emerged as a function of several, both qualitative and quantitative, factors.

Changes in Reading Strategies as a Function of Reading Training: A Comparison of Live and Computerized Training

The purpose of this study was to compare the relative effectiveness of live (SERT) and computer-based (iSTART) reading strategy training. Prior to and after training, participants read scientific texts and self-explained after each sentence. They also answered comprehension questions. Students showed improvement in the quality of their self-explanations and in the performance on the comprehension questionsasa function of both live and computer-based training. However, there were some differences in response to iSTART training as a function of reading skill. Specifically, less skilled readers improved their performance on text-based questions, but not bridging questions. The opposite was found for skilled readers. These results indicate that computer-based, reading-skills training is effective, but different readers may improve at different levels of comprehension.

Using latent semantic analysis to assess reader strategies

We tested a computer-based procedure for assessing reader strategies that was based on verbal protocols that utilized latent semantic analysis (LSA). Students were given self-explanation—reading training (SERT), which teaches strategies that facilitate self-explanation during reading, such as elaboration based on world knowledge and bridging between text sentences. During a computerized version of SERT practice, students read texts and typed self-explanations into a computer after each sentence. The use of SERT strategies during this practice was assessed by determining the extent to which students used the information in the current sentence versus the prior text or world knowledge in their self-explanations. This assessment was made on the basis of human judgments and LSA. Both human judgments and LSA were remarkably similar and indicated that students who were not complying with SERT tended to paraphrase the text sentences, whereas students who were compliant with SERT tended to explain the sentences in terms of what they knew about the world and of information provided in the prior text context. The similarity between human judgments and LSA indicates that LSA will be useful in accounting for reading strategies in a Web-based version of SERT.

Identifying reading strategies using latent semantic analysis: comparing semantic benchmarks.

We explored methods of using latent semantic analysis (LSA) to identify reading strategies in students’ self-explanations that are collected as part of a Web-based reading trainer. In this study, college students self-explained scientific texts, one sentence at a time. LSA was used to measure the similarity between the self-explanations andsemantic benchmarks (groups of words and sentences that together represent reading strategies). Three types of semantic benchmarks were compared: content words, exemplars, and strategies. Discriminant analyses were used to classify global and specific reading strategies using the LSA cosines. All benchmarks contributed to the classification of general reading strategies, but the exemplars did the best in distinguishing subtle semantic differences between reading strategies. Pragmatic and theoretical concerns of using LSA are discussed.

QUAID: A questionnaire evaluation aid for survey methodologists

QUAID (question-understanding aid) is a software tool that assists survey methodologists, social scientists, and designers of questionnaires in improving the wording, syntax, and semantics of questions. The tool identifies potential problems that respondents might have in comprehending the meaning of questions on questionnaires. These problems can be scrutinized by researchers when they revise questions to improve question comprehension and, thereby, enhance the reliability and validity of answers. QUAID was designed to identify nine classes of problems, but only five of these problems are addressed in this article: unfamiliar technical term, vague or imprecise relative term, vague or ambiguous noun phrase, complex syntax, and working memory overload. We compared the output of QUAID with ratings of language experts who evaluated a corpus of questions on the five classes of problems. The corpus consisted of 505 questions on 11 surveys developed by the U.S. Census Bureau. Analyses of hit rates, false alarm rates,d′ scores, recall scores, and precision scores revealed that QUAID was able to identify these five problems with questions, although improvements in QUAID’s performance are anticipated in future research and development.

Computerizing reading training: Evaluation of a latent semantic analysis space for science text

The effectiveness of a domain-specific latent semantic analysis (LSA) in assessing reading strategies was examined. Students were given self-explanation reading training (SERT) and asked to think aloud after each sentence in a science text. Novice and expert human raters and two LSA spaces (general reading, science) rated the similarity of each think-aloud protocol to benchmarks representing three different reading strategies (minimal, local, and global). The science LSA space correlated highly with human judgments, and more highly than did the general reading space. Also, cosines from the science LSA spaces can distinguish between different levels of semantic similarity, but may have trouble in distinguishing local processing protocols. Thus, a domain-specific LSA space is advantageous regardless of the size of the space. The results are discussedin the context of applying the science LSA to a computer-based version of SERT that gives online feedback based on LSA cosines.

Automatic classification of dysfunctional thoughts: A feasibility test

The identification of dysfunctional thoughts is a central effort in cognitive therapy. This paper describes the first version of a computer module that classifies dysfunctional thoughts automatically. It is part of COGNO, a system we are developing to give automatic feedback on dysfunctional thoughts. The system uses rules that were developed from language markers identified in a sample of 149 dysfunctional thoughts. The system was tested with an independent set of 112 example thoughts. The system detects the majority of dysfunctional thoughts, but works reliably only for some thought categories. Automatic thought classification may be a first step toward developing natural dialogue systems in cognitive therapy.

Abstract noun classification: using a neural network to match word context and word meaning

Psychologists have used artificial neural networks for a few decades to simulate perception, language acquisition, and other cognitive processes. This paper discusses the use of artificial neural networks in research on semantics—in particular, in the investigation of abstract noun meanings. It is widely acknowledged that a word’s meaning varies with its contexts of use, but it is a complex task to identify which context elements are relevant to a word’s meaning. The present study illustrates how connectionist networks can be used to examine this problem. A simple feedforward network learned to distinguish among six abstract nouns, on the basis of characteristics of their contexts, in a corpus of randomly selected naturalistic sentences.