Constructing inferences during narrative text comprehension.

Describes a constructionist theory that accounts for the knowledge-based inferences that are constructed when readers comprehend narrative text. Readers potentially generate a rich variety of inferences when they construct a referential situation model of what the text is about. The proposed constructionist theory specifies that some, but not all, of this information is constructed under most conditions of comprehension. The distinctive assumptions of the constructionist theory embrace a principle of search (or effort) after meaning. According to this principle, readers attempt to construct a meaning representation that addresses the reader's goals, that is coherent at both local and global levels, and that explains why actions, events, and states are mentioned in the text. This study reviews empirical evidence that addresses this theory and contrasts it with alternative theoretical frameworks. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

AutoTutor: Incorporating Back-Channel Feedback and Other Human-Like Conversational Behaviors into an Intelligent Tutoring System

This paper describes our recent attempts to incorporate human-like conversational behaviors into the dialog moves delivered by an animated pedagogical agent that simulates human tutors. We first present a brief overview of the modules comprising AutoTutor, an intelligent tutoring system. The second section describes a set of conversational behaviors that are being incorporated into AutoTutor. The behaviors of interest involve variations in intonation, head movements, arm and hand movements, facial expressions, eye blinking, gaze direction, and back-channel feedback. The final section presents a recent empirical study concerned with back-channel feedback events during human-to-human tutoring sessions. The back-channel feedback events emitted by tutors are mostly positive (63%), mostly verbal (77%), and immediately follow speech-act boundaries or noun-phrase boundaries (83%). Tutors also deliver back-channelevents at a very high rate when students are emitting dialog, about 13 events per minute. Conversely, 88% of students' back-channel feedback events are head nods, and they occur at unbounded locations (63%).     

Who said what? Source memory for narrator and character agents in literary short stories.

One dimension of reading literacy involves the tracking of agents associated with the text. In a literary short story, there is a society of character agents and pragmatic agents. This study investigated the relative salience of different classes of agents in memory. Two experiments measured source memory as an index of agent salience in long-term memory. Patterns of source memory scores supported an invisible third-person narrator hypothesis and an agent amalgamation hypothesis, but not a structural prominence hypothesis: First-person narrator > nonnarrator character > third-person narrator > 0. Statement detection parameters did not significantly differ among the 3 classes of agents, so differences in source memory could not be explained by differences in the content of the speech acts. Source memory scores also could not be explained by surface features of the text, differences among readers, and sophisticated guessing on the basis of a story abstract. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

How does one know whether a person understands a device? The quality of the questions the person asks when the device breaks down.

Models of question asking predict that questions are asked when comprehenders experience cognitive disequilibrium, which is triggered by contradictions, anomalies, obstacles, salient contrasts, and uncertainty. Questions should emerge when a person studies a device (e.g., a lock) and encounters a breakdown scenario ("the key turns but the bolt doesn't move"). Participants read illustrated texts and breakdown scenarios, with instructions to ask questions or think aloud. Participants subsequently completed a device-comprehension test, and tests of cognitive ability and personality. Deep comprehenders did not ask more questions, but did generate a higher proportion of good questions about plausible faults that explained the breakdowns. An excellent litmus test of deep comprehension is the quality of questions asked when confronted with breakdown scenarios. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Verification of statements about story worlds that deviate from normal conceptions of time: what is true about Einstein's Dreams?

College students read chapters from a novel written by Alan Lightman (Einstein's Dreams) and later provided verification judgments on the truth/falsity of test statements. Each chapter described a different fictional village that incorporated assumptions about time that deviate from our normal TIME schema, e.g., citizens knowing exactly when the world will end, time flowing backward instead of forward. These novel assumptions about time provided interesting insights about life and reality. In two experiments, we examined whether readers could accurately incorporate these novel assumptions about time in the fictional story worlds, as manifested in the verification judgments for statements after story comprehension. The test statements included verbatim typical, verbatim atypical, inference typical, and inference atypical information from the perspective of mundane reality that meshes with a normal TIME schema. Verification ratings were collected on a 6-point scale in Experiment 1, whereas Experiment 2 used a signal-response technique in which binary true/false decisions were extracted at-.5, 1.5, 3.5, 5.5, and 10.0 s. The college students were measured on literary expertise, reading skill, working memory span, and reading time. Readers with comparatively high literary expertise showed truth discrimination scores that were compatible with a schema copy plus tag model, which assumes that readers are good at detecting and remembering atypical verbatim information; this model predicts better (and faster) truth discrimination for verbatim atypical statements than for verbatim typical statements. In contrast, fast readers with comparatively low literary expertise were compatible with a filtering model; this model predicts that readers gloss over (or suppress) atypical verbatim information and show advantages for verbatim typical information. All groups of readers had trouble inferentially propagating the novel assumptions about time in a fictional story world, but the slower readers were more accurate in their verification of the atypical inferences. A construction-integration model could explain the interactions among literary expertise, reading time, and the typicality of test statements.     

Factors that Influence the Perception of Feedback Delivered by a Pedagogical Agent

Three experiments were conducted to assess the relative importance of speech parameters and facial expressions in the delivery of feedback by a pedagogical agent. In Experiment 1, we manipulated linguistic form (i.e., positive, negative, or neutral terms), rate, pitch, pause, and emphasis. In Experiment 2, we manipulated eye size, mouth curve, brow height, and brow curve. In a third study, both speech parameters and facial expressions were manipulated. In all three experiments, the participants were asked to indicate how positive or negative the agent's feedback seemed to be. Across the studies, the variables collectively accounted for a significant amount of the variance. More specifically, the linguistic expressions and mouth curve emerged as significant predictors of the participants' ratings. This suggests that these two features should be implemented by developers wishing to provide appropriate feedback in their pedagogical agents.     

Toward Spoken Human–Computer Tutorial Dialogues

Oral discourse is the primary form of human–human communication, hence, computer interfaces that communicate via unstructured spoken dialogues will presumably provide a more efficient, meaningful, and naturalistic interaction experience. Within the context of learning environments, there are theoretical positions supporting a speech facilitation hypothesis that predicts that spoken tutorial dialogues will increase learning more than typed dialogues. We evaluated this hypothesis in an experiment where 24 participants learned computer literacy via a spoken and a typed conversation with AutoTutor, an intelligent tutoring system with conversational dialogues. The results indicated that (a) enhanced content coverage was achieved in the spoken condition; (b) learning gains for both modalities were on par and greater than a no-instruction control; (c) although speech recognition errors were unrelated to learning gains, they were linked to participants' evaluations of the tutor; (d) participants adjusted their conversational styles when speaking compared to typing; (e) semantic and statistical natural language understanding approaches to comprehending learners' responses were more resilient to speech recognition errors than syntactic and symbolic-based approaches; and (f) simulated speech recognition errors had differential impacts on the fidelity of different semantic algorithms. We discuss the impact of our findings on the speech facilitation hypothesis and on human–computer interfaces that support spoken dialogues.     

AUTOMATIC DETECTION OF LEARNER'S AFFECT FROM GROSS BODY LANGUAGE

We explored the reliability of detecting learners' affect by monitoring their gross body language (body position and arousal) during interactions with an intelligent tutoring system called AutoTutor. Training and validation data on affective states were collected in a learning session with AutoTutor, after which the learners' affective states (i.e., emotions) were rated by the learner, a peer, and two trained judges. An automated body pressure measurement system was used to capture the pressure exerted by the learner on the seat and back of a chair during the tutoring session. We extracted two sets of features from the pressure maps. The first set focused on the average pressure exerted, along with the magnitude and direction of changes in the pressure during emotional experiences. The second set of features monitored the spatial and temporal properties of naturally occurring pockets of pressure. We constructed five data sets that temporally integrated the affective judgments with the two sets of pressure features. The first four datasets corresponded to judgments of the learner, a peer, and two trained judges, whereas the final data set integrated judgments of the two trained judges. Machine-learning experiments yielded affect detection accuracies of 73%, 72%, 70%, 83%, and 74%, respectively (chance = 50%) in detecting boredom, confusion, delight, flow, and frustration, from neutral. Accuracies involving discriminations between two, three, four, and five affective states (excluding neutral) were 71%, 55%, 46%, and 40% with chance rates being 50%, 33%, 25%, and 20%, respectively.