Search

This report describes a study of the recall of narrative prose. Subjects listened to repeated presentations of a tape recording of two pages from a history book, with verbal recalls collected after each presentation. The elements of the passage were organized according to a serial structure based on order in the passage and a story grammar structure based on casual relations. While the serial structure at first influenced which elements of the passage were remembered, as the subjects remembered more of the passage, the story grammar structure became the dominant influence over the elements remembered on subsequent recalls., Sponsored in part by the DARPA.

This paper describes a hypothetical computer program capable of emulating the behavior of a human tutor. Such a program must, ideally, display a number of characteristics. It must be mixed-initiative, meaning that either student or tutor may initiate an interaction. It must be capable of understanding natural language input and generating natural language output. It must be highly knowledgeable about its subject matter, and generative in order to be able to answer unanticipated questions. Finally, the system must be flexible enough to detect the needs of a student and alter its teaching strategy so as to best meet those needs. The first part of the paper discusses what the major parts of the system would be, the functions each part would carry out, and the means by which they interact. The goal is to develop a general system that can be used with a number of different teaching strategies. The second part of the paper analyzes the Socratic teaching method in order to determine the requirements for implementation within the system.

Work on this contract examined various applications of human information processing, focusing on the analysis of human error. As a result of the contract investigations, several areas of potential application were identified. One area of immediate possible application is in the design of human-machine interfaces, with particular concern for aspects of system design that lead to human error, taking into account what is known about human attentional and short-term memory limitations. The major focus of the research effort was devoted to the study of human error, especially those made by highly skilled operators in complex, high-demand systems. A classification scheme for these errors was developed.

Expert typists have acquired a highly practiced motor skill. A typical professional typist has accumulated over 10,000 hours of practice. Expert typists are much faster than novices, but in addition, their performance is qualitatively different in many ways from novice performance. During acquisition of typing skill, there is a general shift from cognitive to motor limits on performance. Expert typing is characterized by parallel mental processes that overlap in time, overlapped hand and finger movements, a decreased load on conscious cognitive resources, and reduced variability fo the interstroke intervals. (Author)

These experiments demonstrate that the perception of two distinct words in a briefly presented display can interact, causing perceptual migrations of letters from one word to the other. For example, when LINE and LACE are presented, subjects might report seeing LICE or LANE instead of LINE. Several properties of the letter migrations were revealed: Migrations are more frequent when the words are separated by smaller physical distances; a majority of the migrations are a result of letters being copied from one word to the other, not from the interchange of letters of the two words; migrations to a word are less frequent when subjects focus attention on that word; and migrations are far more frequent when the words share letters in common. This last result suggests that migrations are not caused by a loss of spatial information at the letter level, that is, by free-floating letters being wrongly combined. Rather, migrations occur because of structural limitations at a high level of the word-recognition process, perhaps during lexical activation. Implications for models of multiple-word perception are discussed. (Author)

Terzuolo and Viviani, in widely cited research, propose a central control model of timing in typing, in which keystroke times are generated in parallel from centrally stored, word-specific timing patterns. Differences in overall time to type a given word are attributed to a multiplicative rate parameter, constant for a given typing of the word, but varying from one typing to another. Three major lines of evidence are cited for this model: (a) keystroke times expand or contract proportionally when words are typed slower or faster; (b) in a word; (c) the times to type a given digraph exhibit word-specific differences. My analyses show that (a) keystroke times do not expand proportionally; (b) the apparent constancy of variances is an artifact of the method that Terzuolo and Viviani used to transform the keystroke times; (c) the effects of surrounding character context are sufficient to explain differences in digraph latencies and these effects cross word boundaries, showing that they are not word-specific. (Author)

The issue of temporal control of motor behavior was investigated using a rhythmic tapping task. It was found that: (1) subjects are better able to tap before a beat than after a beat; (2) the variability of tapping depends on whether the subject is attempting to tap on, before, or after a beat; (3) the control of rhythmic tapping is relatively central; (4) the starting times of taps are more variable than the ending times. These results are interpreted in terms of a model by Wing (1980), and their significance for theories of timing is discussed. (Author)

This report consists of two papers on MENUNIX, an experimental interface to the programs and files on the UNIX operating system. In the first paper, I discuss how the decisions about the design of MENUNIX were made: based on my intuitions and user comments, but also on psychological theory and data whenever available. MENUNIX presents both the programs and files of UNIX in two menus from which users can make selections with single keypresses. The FILE menu presents the UNIX file hierarchy that allows users to organize files into directories by subject (e.g. writing and programming). The PROGRAM menu presents UNIX programs in a hierarchy organized into workbenches according to the tasks for which they are used (e.g. writing and programming) much as files can be organized in directories. Special facilities are provided for: finding out about useful commands; using variables to set options, to save commands, and to avoid typing long strings; and for editing strings (including recent commands). The second paper is a tutorial user manual for MENUNIX, in which the features of the program are more fully explained. (Author)

It is well known that the Sholes or 'qwerty' keyboard is deficient in designed, hard to learn, and awkward to use. Alphabetically organized keyboards would appear to be superior for novice typists, but this has not been demonstrated in previous studies of typing. These experiments confirm the lack of virtue for alphabetically organized keyboards over both a randomly organized one and over the standard Sholes keyboard. To take advantage of the alphabetic keyboard requires considerable mental processing on the part of the user, and this processing is neither easy to do nor does it appear to offer much savings over visual search of the keyboard. The novice is faced with a tradeoff between mental processing and visual search, and this tends to make different keyboard layouts equivalent. In addition, many people know at least something about the Sholes keyboard, and even this little knowledge is useful, with the result that their performance will usually be better on Sholes than on alphabetic keyboards (and certainly no worse). The conclusion is that it is not worthwhile to use alphabetic keyboards for novice typists, nor to change to the Dvorak layout for experts. Keyboards can probably be improved, but only though radical redesign of the present physical key configuration.

Six skilled typists were studied while they transcribed English text. The typists showed stable patterns of performance, but with significant individual differences among the typists. Inter-Keypress latencies for 2-finger digraphs (typed by two fingers on the same hand) were particularly variable among typists. Two typists showed large differences in 2-finger digraph latencies, but similar overall typing speeds. Finger movement trajectories, determined from analysis of videotapes of these typists, indicated that the differences in 2-finger digraph latencies correspond to differences in the independence of within-hand finger movements. A high-speed film of one typist showed that finger movements of this typist almost always overlapped. The starting times of movements were six times as variable as the ending times, suggesting that it is the completion rather than the initiation of the movements that is controlled in skilled typing. These studies demonstrate the importance of considering individual differences in constructing a theory of skilled human performance, even in a highly automatized task such as transcription typing. (Author)

We review the major phenomena of skilled typing and propose a model for the control of the hands and fingers during typing. The model is based upon an Activation-Trigger-Schema system in which a hierarchical structure of schemata directs the selection of the letters to be typed and, then, controls the hand and finger movements by a cooperative, relaxation algorithm. The interactions of the patterns of activation and inhibition among the schemata determine the temporal ordering for launching the keystrokes. To account for the phenomena of doubling errors the model has only 'type' schemata -- no token schemata -- with only a weak binding between the special schema that signals a doubling and its argument. The model exists as a working computer simulation and produces an output display of the hands and fingers moving over the keyboard. It reproduces some of the major phenomena of typing, including the interkeypress latency times, the pattern of transposition errors found in skilled typists, and doubling errors. Although the model is clearly inadequate or wrong in some of its features and assumptions, it serves as a useful first approximation for the understanding of skilled typing. (Author)

The major theme of the paper is that the primary role of attention is in the control of action. The basic idea is that human action sequences can run themselves off, efficiently, smoothly, without any need for deliberate attention. However, when modifications in a plan must be made, or when it is desired that some novel alternative action sequence be followed, or when it is desired to prevent some habitual act from occurring, then it is necessary for deliberate attentional intervention into the process. We argue that most attentional conflicts occur with the initiation rather than the execution of actions. We suggest two levels of control: a contention scheduling mechanism that selects from among competing schemas; a supervisory attentional mechanism that biases the selection process. We propose that the supervisory attentional system is required where the action sequences are ill-learned or novel, where the action is highly critical or dangerous, or where planning is required. In other cases, selection is by contention scheduling alone. The result is three modes of the control of performance: automatic, contention scheduling without deliberate direction, and deliberate conscious control. Will becomes the application of attentional resources to the control of action. (Author)

The schema theoretic view of comprehension is reviewed. Several techniques for the study of on-line measures of comprehension are developed. The techniques include a question probe technique and a reading time technique. Data were collected using each of these techniques and the results are compared with each other. The results are then interpreted in terms of a schema theoretical view of comprehension. (Author)

This paper examines the role of analogy and procedural representation in learning. Examples of analogical manipulation of knowledge schemata are presented from several domains, including turtle geometry, kinship terms, and the learning of a computer text editor. The view presented in this paper has a number of implications for instruction and for performance. In particular, the learner or user of a system should be presented with a conceptual model that has the following properties: (a) It is based on a domain with which the student is already knowledgeable and for which the student can reason readily; (b) The target and source domains should differ by a minimum number of spcifiable dimensions; (c) Operations that are natural in one domain should also be natural within the other domain; (d) Operations inappropriate within the target domain should be also by inappropriate within the source domain.

This paper is the second part of a two-part series introducing an interactive activation model of context effects in perception. In the previous part we developed the basic form of the model and showed how it accounts for several of the fundamental phenomena of work perception. In this part, we first present a number of new experiments and show how the model accounts for these experiments. Then we propose a number of extensions of the model to such cases as spoken input, pronunciation tasks, and words embedded in sentential context. Finally, we discuss the strengths and weaknesses of the model, pointing out further possible extensions of the model to account for aspects of word perception currently ignored. The new experiments all revolve around what we call the 'context enhancement effect.' These experiments are designed to assess the roles of direct and indirect evidence concerning the identity of a letter in an input string., See also Part 1, ADA086946. Sponsored in part by Grants NSF-BNS76-14830 and NSF-BNS76-15024.

This report is the first part of a two-part series introducing an interactive activation model of context effects in perception. In this part, is developed the model for the perception of letters in words and other contexts and a number of experiments in the recent literature is applied. The model is used to account for the perceptual advantage for letters in words compared to single letters and letters in unrelated strings. In the model, these word-superiority effects are produced by feedback. The visual input produces partial activations of letters, which in turn produce partial activations of words. These activations then produce feedback to the letter level, reinforcing letter sequences which spell words. The model can account for the basic findings on the perception of pronounceable nonwords as well as words. The account is based on the idea that pseudowords can also activate representations of words, even though they do not match any word perfectly. As with word displays, feedback from the activated words reinforces the letters presented, thereby increasing their perceptibility. The model also accounts for the role of masking in determining the magnitude of the various effects, the fact that expectations influence perception of letters in words, and for the fact that effects of contextual constraint and letter cluster frequency are obtained under some conditions and not others., Sponsored in part by Grant NSF-BNS-76-15024.

High speed film and key press latencies of a skilled typist show that finger movements usually overlap in time. The starting time of these movements is highly variable, even when comparing identical sentences. Often movements toward keys were initiated out of the order in which the letters were eventually typed. These results conflict with current models of rapid, well-learned motor movements. (Author)

SLIPS OR UNINTENDED ACTIONS HAVE BOTH THEORETICAL AND APPLIED IMPORTANCE, BY PROVIDING INSIGHT INTO POSSIBLE UNDERLYING MECHANISMS OF BEHAVIOR AND BECAUSE OF THEIR REAL-WORLD CONSEQUENCES. Several collections of slips are examined and an outline for a theory of action is proposed, based on selection of appropriate schemas, their activation in memory, and triggering by appropriate conditions, with errors possible at each stage. A categorization of slips is presented based on this outline. Finally, the role of feedback and cybernetic theories of human performance are discussed. (Author)

The Coach system, a computer simulation of a human tutor, was constructed with the goal of obtaining a better understanding about how a tutor interprets the student's behavior, diagnoses difficulties, and gives advice to a student learning a simple computer programming language. Coach is based on a hierarchy of active schemas representing the tutor's general concepts, more specific information represented in a semantic network, and the coordination of data-driven and conceptually-guided processing which enable it to interpret behavior, recognize errors, and give advice. (Author)

First the roles of a science of cognition and of psychology are discussed, and then the differences between animate and artificial cognitive systems. Twelve central issues for cognitive science are next presented and discussed. Finally, other possible issues are considered, and the argument made that a wide variety of behavior must be studied before a single class can be well understood. (Author)

This project examines the learning of complex topics. The research was aimed at two different aspects of the study of learning and instruction: (1) theoretical understanding of what is going on the learner's mind; (2) development of intelligence interactive computer-based instructional systems. Two computer-based tutorial systems, Coach and Instruct, were developed. The findings of our research are summarized. (Author)

A human tutor brings a wide range of knowledge to the task of instructing a student. The tutor must develop a model of the student and of the topic matter; he must have a plan of instruction, but be able to deviate from it when student behavior calls for changes. This paper discusses observations made of human tutors and describes the FLOW tutor system -- a computer-based simulation of a human tutor capable of giving advice to a student learning the simple 'FLOW' computer language. The tutor has a schema-based knowledge structure containing information about the programming language, the student's instruction booklet, and the student's developing knowledge. These schemas form the basis of a distributed intelligence system which uses conceptually guided and data-driven processing to interpret the student's behavior, update the model of the student, and give advice to the student.

This paper presents an analysis and model of the cognitive processes underlying complex learning situations. A theory is proposed that attempts to specify particular internal knowledge structures generated and modified during instruction, and to use them to explain specific difficulties that the learner experiences and also the overall progress being made. The theory states that (1) the underlying process reflects largely the prior knowledge structures of the student interacting with the information implied by the instruction; (2) the learning mechanisms involved are mostly simple and automatic; (3) the key information within the knowledge structures which allows complex learning to occur is the similarity between higher-order structures. This theory was applied to novices learning to use a computer text editor by reading a basic instruction manual and completing some exercises. After each sentence of instruction, they were asked to describe their current understanding, any difficulties they were aware of, and their expectations about what would follow. Their protocols were analyzed for evidence of underlying cognitive processes. These learning processes and the associated knowledge structures on which they operate were then modeled in terms of the theory at several levels of detail. Several issues of knowledge representation related to the model are discussed and possible solutions proposed., Sponsored in part by DARPA.

Is some appropriate level of prior knowledge a prerequisite for asking questions about a new topic matter. The hypothesis that learners should ask the most questions when their knowledge is well matched to the level of presentation was tested using two levels of background knowledge and two levels of difficulty of material to-be-learned. Subjects said aloud their questions and thoughts while learning. Novice learners asked more questions on the easier material than learners with some training; trained learners asked more questions than novices on the harder material. Accordingly, theories of question asking in learning should address interactions between knowledge levels of askers and material., Sponsored in part by DARPA and by Grant PHS-MH-15828.

Subjects were asked to think aloud while attempting to recall the names of their high school classmates of from 4 to 19 years past. Yearbooks of each subject's graduating class were available to verify the accuracy of subject recalls. The retrievals were characterized by overshoot, systematic hypothesizing, fabrications, the establishment of search contexts, self corrections, and the use of a number of basic search strategies. Several subjects were still retrieving new names after 10 hours at the task. These phenomena, as well as an array of traditional memory phenomena, can be understood from an information processing analysis which is based on interpreting retrieval as a problem solving process. The characterization of retrieval which results is that of a reconstructive process. Information about the target item is used to construct a description of some aspect of the item. The description is used to recover a fragment of information about the item which is added to what is known. From this information a new description can be formed to retrieve still more information, until the particular piece of information sought can be recovered. The characterization identifies three subprocesses: FIND A CONTEXT, in which a proper environment for conducting a search is recovered, SEARCH, in which bits and pieces of information appropriate to the context are recovered until an adequate description can be formed within the search context, and VERIFY, in which the record recovered is checked to confirm that it is about the target item. Each of the three stages has embedded within one or more recursive calls to the retrieval process.

This paper postulates that the processes of memory retrieval operate upon descriptions which provide an initial specification of the information being sought, guide the memory search process, and help determine the criteria for verification of whatever information is retrieved. Descriptions provide a basis for a set of iterative processes for both memory acquisition and retrieval. In acquisition, memory descriptions and records can be elaborated in ways intended to aid later retrieval. In retrieval, the initial description of the information sought can be modified as intermediate information becomes available during the retrieval cycle. Two important aspects of memory descriptions are discriminability and recoverability: their ability to discriminate among all possible records in memory and the likelihood that they can be recovered at the time retrieval is required. An analysis using descriptions provides a consistent interpretation of a number of different memory phenomena, including depth of processing, some interference phenomena, accessibility and encoding specificity, mnemonics, and some properties of naturalistic recall.

Learning is not a simple unitary process. In this paper we identify three qualitatively different phases of the learning process. In one phase, the learner acquires facts and information, accumulating more structures onto the already existing knowledge structures. This phase of learning is adequate only when the material being learned is part of a previously understood topic: the appropriate memory schemata already exist. In a second phase, the learner must devise new memory structures to interpret the material that is to be acquired. This is the most difficult and the most significant form of learning, for it marks the acquisition of truly new conceptualizations about a topic matter. The third phase of learning involves a continual process of modification: both constraining and generalizing the knowledge within the schemata of memory. This stage of learning does not increase the formal content of one's knowledge, but it makes the use of the knowledge more efficient. Thus, although a beginner and an expert might both perform a task with perfect accuracy, there is a marked qualitative difference between the performance of the two. We propose three different mechanisms that seem to be responsible for the different phases of the learning of complex topic matters: accretion, restructuring, and tuning.

This research has centered upon the study of the representation of knowledge in memory and of the learning and teaching processes for complex topic matters. An active structural network, a type of semantic network has been used to represent both knowledge and procedures within the same data base. An event schema representation has been developed to study the acquisition and organization of complex events. In particular, the schema representation has been used to study students learning American Civil War history or how to program in the computer language FLOW. The studies of the learning and teaching processes have been primarily concerned with developing models of the student and with building automated tutorial instructional systems. The automated systems have given us a controlled environment in which to observe learning and teaching, and the two contrasting topic areas of history and programming have been used to help insure the generality of our models. By the technique of incremental simulation in which the computer performs whatever aspects of the tutorial project are fully understood and human tutors enact the roles of other system components, we have been extending our understanding of the interactions among the components of the overall tutorial system. The goal is to develop a theoretical model of the learning of complex material. (Author), Sponsored in part by DARPA.

This paper provides a review of work on the representation of knowledge from within psychology and artificial intelligence. The work covers the nature of representation, the distinction between the represented world and the representing world, and significant issues concerned with propositional, analogical, and superpositional representations. Major controversies within psychology -- such as distinctions between declarative and procedural representation, propositional and analogical representation, and the nature of visual images -- are analyzed and found not to reflect fundamental disagreements. (Author)

A framework called 'a function-mechanism hierarchy' was developed to capture a course of understanding in the sewing machine interactions. According to this framework, understanding proceeds from global, functional understanding to local, mechanistic understanding by descending 'levels.' The subjects' conceptual point-of-view was related to this course of understanding. Point of view shifted more when the subjects felt they were not understanding, and this shift appeared to help them descend the levels. People corrected more errors when they reflected the current level of understanding; errors were not corrected when they belonged to already known levels. For both statistics and sewing machine interactions, the issue of 'focus' was identified to be important in understanding. When a schema needs to be changed, it seems necessary to have a global understanding of the old schema as well as attention to the place where the change is to take place. Two person interactions have virtues because the participant can take different roles in the interaction and divide the labor: While one person leads the interaction by engaging in the local task, the other can observe and provide help by criticising and bringing up new motions. (Author)

This report consists of five brief papers on different aspects of human-machine interaction. The first paper, 'Some Observations on Mental Models,' discusses the role of a person's mental model in the interaction with systems. The second paper, 'A Psychologist Views Human Processing: Human Errors and other Phenomena Suggest Processing Mechanisms' discusses the differences between conventional digital processing structures (the Von Neumann machine) and the mechanism of the human. The third paper, 'Steps toward a Cognitive Engineering,' shows how analysis of error can lead to design principles. The fourth paper, 'The Trouble with UNIX,' is an informal critique of the UNIX operating systems. The final paper, 'The Trouble with Networks,' describes some of the computer interactions that resulted from the distribution of the fourth paper. (Author)

The timing of keystrokes during errors in skilled transcription typing suggests central control of timing for two-letter sequences. I find that letter transpositions are performance errors, and the timing of interstroke intervals in these errors is closely correlated with the pattern found in the same words typed correctly. This and other aspects of the timing argue against both a distributed processing model with single-character units and a pure metronome model. For these two-character sequences, timing seems to be controlled by signals issued generally, rather than to a specific finger. (Author)

We examined the difficulty of tapping different sequences of five finger taps, one tap per finger, with only the order varying for different sequences. Different sequences have different levels of difficulty, depending upon their structure. The same sequence can be easy or difficult to perform, depending upon how it is encoded. (Author)

These studies use both discontinuous and continuous typing paradigms in investigating digraph frequency effects in skilled typing. Previous studies that controlled for hand use showed no digraph frequency effects. With more rigorous controls we found very reliable digraph frequency effects, suggesting that, with experience, typists develop multi-character response units. Videotape analysis indicated at least one way such response units can potentially coordinate and facilitate performance. (Author)

This report describes a study of how maps are learned. Subjects studied a simple map under various conditions and then answered questions about the map and drew the map from memory. It was found that: (1) Local relations are learned before large-scale relations; (2) Different types of map information must be presented simultaneously for effective integration; (3) Reading a story relevant to the map is more helpful in remembering the map than spending the same amount of time copying the map., Sponsored in part by DARPA.

The report describes a computer-based system, called the FLOW system, used in experimental studies of human learning. The student learns a simple computer language from printed instructions and can run his programs interactively on the FLOW system. An automated tutor simulates a human tutor who watches over the student and gives help when the student has difficulties. The system also records detailed protocols of the interactions among the student, the computer, the automated tutor, and the human tutor for later analysis.

The authors examine the process of learning large, complex subject matters, asking about the ways by which teacher and student communicate the necessary knowledge structures to each other, and asking how a student comes to select an appropriate paradigm for solving a problem and to revise inappropriate ones. They examine protocols from several different learning situations, and have made a first attempt to put together the pieces into a cognitive theory of learning and teaching.