I volunteered to review this book because I found Dr. Markowitz's 1995 book, Using Speech Recognition, extremely helpful when I first became involved in medical applications of speech. In fact, I made all my students in that area read it. This book is very different, but in its own way, just as useful. In fact, it is a must-read for anyone contemplating a new speech application or enhancing a current one, as well as for anyone searching for new directions in dialogue research.This book is divided into four parts, followed by a separate conclusion. It is always hard to do justice to all the papers and authors in a collection like this—with twelve varied papers addressing many different parts of the complex problem of enabling a robot to carry on a spoken dialogue—but I will try to give at least a brief taste of each.Part I is about “Images.” In the first chapter of Part I, Steve Mushkin, the founder and president of Latitude Research describes an experiment in which 348 children aged 8–12 from six digitally-advanced countries were asked to draw and describe in words how they would like to interact with a personal robot in the future. In the second chapter Markowitz herself reviews the language capabilities and behavior of powerful cultural icons, such as Frankenstein, the Hebrew Golem, the Japanese karakuri, and Pygmalion's beloved statue, Galatea. This is highly relevant, because science fiction, whether on paper or in the movies, part of traditional mythology, or local social history, has largely shaped our expectations about what new technology promises or threatens. In the third chapter, David Dufty, author of the well-known book How to Build an Android: The True Story of Philip K. Dick's Robotic Resurrection, believes that robots are intrinsically part of art and entertainment, and that is the major reason why we should build them.Part II is called “Frameworks and Guidelines.” The first of the three chapters in this part, written by Bilge Mutlu and two of his students at the University of Wisconsin, Madison, discusses the requirements for designing a robot capable of sustaining productive dialogue. They test their framework in robot-human interactions; first using a robot as a teacher, then to assess linguistic factors that contribute to expertise. Director of Columbia University's Japanese Language Program, focuses on the knowledge that any teacher must have to be effective in teaching a foreign language. Based on her series of research studies on the teaching of Japanese, her concern is that a robot (or human) teacher must understand the bond between language and culture, which she sees as inseparable. She emphasizes the importance of using human-like facial expressions in the teaching machine, at first for teaching pronunciation, for communicating emotional reactions to the student's contributions, and making the student feel comfortable with the machine. In this excellent paper, chock full of valuable references, Nazikian unintentionally reminds us of the huge and inexplicable divide between the world of Computer-Aided Language Learning (CALL) and the world of AI in education. Art Graesser and his group at the University of Memphis (Link et al., 2001) have carried out a long series of experiments studying the effect of using expressive faces on the screen coordinated with the dialogue, but the work at Memphis is never mentioned in this paper and there are no references to papers in Discourse Processes or the Journal of the Learning Sciences. The third paper in Part II, written by Manfred Tscheligi and his student, Nicole Mirnig, at the University of Salzburg, focuses on the problems of Comprehension, Coherence, and Consistency. These qualities depend, they argue, on the use and transfer of mental models, which they find essential to effective dialogue with any robot capable of learning what human beings want and giving it to them.Part III is about constructing a speech-enabled robot capable of useful, real-world tasks. In Chapter 7, Jonathan Connell of IBM's T. J. Watson Research Center explains how he defined these capabilities for a robot named ELI (the Extensible Language Interface). First of all, such a robot needs to be able to accept and synthesize multimodal information. It needs to recognize speech, gestures, and object manipulations and put this information together into an algorithm for getting you a cup of coffee? It has to find your kitchen, pick up the right coffee pot, heat the water in a microwave safe measuring cup, and put in just the right amount of coffee, cream, and sugar. In the process, it will need to learn new nouns (the names that you use for objects and rooms in your house) and new verbs (for activities like opening bottles and boiling water). Their robot parser is a finite state machine, but it can handle new names and new activities. The emphasis is on the grounding of new language in a new environment. In Chapter 8, Alan Wagner, a Senior Research Scientist at Georgia Tech, outlines what a robot has to know before you can teach it to tell lies or recognize lies produced by others. Unlike ELI, this robot does not yet exist and I find it hard to imagine wanting to build it or even use it myself, but the description of deceptive language in this chapter has charm. Joerg Wolf and Guido Bugmann, at the University of Plymouth in the UK, recount their experience getting university students to teach their off-the-shelf robot how to play a simple card game. They began by collecting a corpus of sessions in which one university student taught another to play the game. They used this corpus to develop the rules for their robot, grammar rules, rules for anaphora, dialogue rules, rules for dealing cards, and then they carried out a series of experiments in which university students tried to teach the robot the game. One of many problems that they had not anticipated was that when the robot did not understand what the student wanted him to do, the student raised his or her voice and tried to simplify the explanation. The result was many out-of-grammar and out-of-vocabulary errors.Part IV contains two very different papers, one about improving audition (so that the robot can understand what you are telling it, even when several people are talking at once or machines are clanking) and the other about how design factors, such as robot voices and gesture speed affect memory and engagement in both children and adults. The experts in audition, François Grondin and François Michaud from the Robotics Laboratory at the Canadian Université de Sherbrooke, have developed algorithms for localization (figuring out where a sound is coming from), tracking (following sounds as people or robots move), and separation (extracting sounds from a given person or machine when several are making noises at once). They have achieved remarkable success in a number of experiments run on a system with a bank of separate microphones and parallel hardware capable of processing all this information at once. Sandra Okita, then at Stanford, now at Columbia University, and Victor Ng-Thow-Hing from the Honda Research Institute USA in Mountain View teamed up to carry out a series of experiments with robot voices. The objective of their studies was to determine how design choices like robot voices and gesture speed effect how humans of all ages respond to a robot. They tried out two voices with young children, one robot-like (monotone) and one more human-like. Both robots followed exactly the same script, but the children remembered much more of what the humanoid voice said and they were much more likely to be willing to talk to that robot again. Teenagers preferred the humanoid voice, but there was much less difference between the two groups interacting with the different voices. Adults seemed to be affected even less. Similar experiments with gesture speed showed that robots who gestured rapidly were judged to be happier and more pleasant to spend time with. Again, small children were affected the most, but teenagers also had strong preferences here. Again, adults were affected less.We badly need more work of the kind described by Okita and Ng-Thow-Hing. Too many people believed Reeves and Nass's (1996) Media Equation, which is the claim that people interact with computers in the same way that they interact with people; and decided that they could substitute research on human interaction for research on interactions between people and machines. Our own experiments (Bhatt, Argamon, and Evens, 2004) have convinced us that the Media Equation is certainly not true for the students interacting with our intelligent tutoring system, who are constantly polite to human tutors, whether their professors or their peers, but often very rude to our computer tutor.The concluding chapter in the book, written by Roger K. Moore, Professor of Computer Science at the University of Sheffield and the Editor-in-Chief of Computer Speech and Language, takes us back to the future where the first chapter began. He argues that we have a long way to go to meet the goal of intelligent communication with machines. We need to build robots that understand human behavior and are capable of generating the language necessary to change it.Moore essentially focuses this chapter on his own view of the major conflict of opinion and approach separating the experts featured in this book. It will never be enough to simply tack on a multi-purpose language module to an existing robot, he argues. Spoken communication is a fundamental and integral part of ordinary human beings; and if we are to succeed in our goal of constructing “intelligent communicative machines,” we need to make the communication function a central part of the machine design and not simply an add-on function.There is a major gulf between Moore's argument and the experts quoted by Dufty back in the third chapter (page 55). Dufty quotes Sylvia Solon, deputy editor of Wired in the UK, as saying, “There is no point making robots look and act like humans.” Dufty adds that Martin Robbins, who blogs for the Guardian, under the name “The Lay Scientist,” wrote: “Humanoid robots are seen as the future, but for almost any specific task you can think of, a focused simple design will work better.” Of course, Dufty goes on to tell us about the use of robots in art and entertainment and about the workings of the Philip K. Dick android, so he may not agree entirely with Solon and Robbins.There is a cogent case on the opposite shore of this gulf in at least three of the other chapters (4, 7, and 9). In Chapter 4, Mutlu's group at the University of Wisconsin describe the simple semantic grammar used by their robot (a Wakamaru), along with more sophisticated dialogue and behavior models and algorithms to control gaze and gestures. Jonathan Connell begins Chapter 7 with “Suppose you buy a general fetch-and-carry robot from Sears and take it home” and continues in this vein. In Chapter 9, Wolf and Bugmann designed their system to operate on components made of production rules, but they still have a lot to teach us.Personally, I am on a raft drifting in the middle of this gulf. I think that we need thoughtful experiments with existing robots now, to help us discover the problems, but I believe that really satisfactory solutions will require much more research. I cannot agree with Moore, however, when he argues that it is unethical to attempt to imitate human communication by stitching together the limited technologies we have today. It seems to me that much of what we now know about human communication comes from attempting to do just that.In the process of reading this book I discovered that I had not heard about two other collections edited by Markowitz and published by Springer in 2013. Both of these were edited by Neustein & Markowitz jointly. Dr. Amy Neustein is founder and CEO of Linguistic Technology Systems and Editor-in-Chief of the International Journal of Speech Technology. The first is entitled Mobile Speech and Advanced Natural Language Solutions (hardbound ISBN 978-1-4614-6017-6; e-Book ISBN 978-1-4614-6018-3); the second, Where Humans Meet Machines: Innovative Solutions for Knotty Natural Language Problems (ISBN 978-1-4614-6933-9, eBook ISBN 978-1-4614-6934-6).

Even the best information retrieval systems miss a large amount of relevant material. This work describes an approach that adds some artificial intelligence to the task of improving the results. We combine shallow parsing and information extraction techniques with conventional information retrieval to build a richer database of events to support the retrieval effort. In our system each event has three main types of identifying factors: keywords that indicate that an interesting event has occurred, dates that specify the time of the event and proper nouns that specify the people, organizations, locations, products, etc., that are involved in the event. In This work we describe a system that uses these keys to find events, classify them, and save them in a database along with the identifier of the document that mentions the event. The retrieval process uses this information to provide the user with menus to form queries about the events; it then executes those queries, and finds the related documents.

In computer networks, fast routing is one of the main factors to speed up the transmission of the IP packets from the source to the destination computers. A router has a forwarding table (cache table) and a routing table. The forwarding table is derived from the routing table, which is maintained constantly by the router using one of the common routing protocols. There are two factors that affect the speed of the forwarding decisions. First, when a forwarding table holds a huge number of IP addresses, a complicated search algorithm is needed, which implies a high search time. If a complicated search algorithm is not used, then the forwarding table has to be set up in a manner to reduce search time; this implies that the table must be set up by a processor that has a substantial computing power. Second, the availability of the incoming IP address in the forwarding table (hit rate) reduces the extended search in the routing table. Our study examines the IP address streams. Two models were built to predict IP addresses. These models are plugged into the cache system simulation to prefetch the IP addresses prior to the router requests. The prediction scheme produces a higher hit ratio in comparison with the current cache schemes. Also, it reduces the cache table size, which implies low search time.

We have developed a bilingual interface to the Novell network operating system, called the Dialoguer. This system carries on a conversation with the user in Arabic or English or a combination of the two and attempts to help the user use the Novell network operating system. Learning to use an operating system is a major barrier in starting to use computers. There is no single standard for operating systems which makes it difficult for novice users to learn a new operating system. With the proliferation of client–server environments, users will eventually end up using one network operating system or another. These problems motivated our choice of an area to work in and they have made it easy to find real users to test our system. This system is both an expert system and a natural language interface. The system embodies expert knowledge of the operating system commands and of a large variety of plans that the user may want to carry out. The system also contains a natural language understanding component and a response generation component. The Dialoguer makes extensive use of case frame tables in both components. Algorithms for handling a bilingual dialogue are one of the important contributions of this paper along with the Arabic case frames.

The development of high-performance spatial access methods that can support complex operations of large spatial databases continues to attract considerable attention. This paper introduces QSF-trees, an efficient and scalable structure for indexing spatial objects, which has some important advantages over R*-trees. QSF-trees eliminate overlapping of index regions without forcing object clipping or sacrificing the selectivity of spatial operations. The method exploits the semantics of topological relations between spatial objects to further reduce the number of index nodes visited during the search. A series of experiments involving randomly-generated spatial objects was conducted to compare the structure with two variations of R*-trees. The experiments show QSF-trees to be more efficient and more scalable to the increase in the data-set size, the size of spatial objects, and the number of dimensions of the spatial universe.

We consider two questions related to student modeling in an intelligent tutoring system: 1) what kind of student model should we build when we design a new system; and 2) should we divide the student model into different components depending on the information involved. We consider these two questions in the context of a conversational intelligent tutoring system, CIRCSIM-Tutor. We first analyze the range of decisions that the system needs to make and define the information needed to support these decisions. We then describe four distinct models that provide different aspects of this information, taking into consideration the nature of the domain and the constraints provided by the tutoring system. Finally, we briefly discuss our experiments with enhancing the student model in CIRCSIM-Tutor and some general problems regarding building and evaluating different student models.

Stemming is one of the most important factors that affect the performance of information retrieval systems. This article investigates how to improve the performance of an Arabic Information Retrieval System (Arabic-IRS) by imposing the retrieval method over individual words of a query depending on the importance of the WORD, the STEM, or the ROOT of the query terms in the database. This method, called Mixed Stemming, computes term importance using a weighting scheme that uses the Term Frequency (TF) and the Inverse Document-Frequency (IDF), called TFxIDF. An extended version of the Arabic-IRS system is designed, implemented, and evaluated to reduce the number of irrelevant documents retrieved. The results of the experiment suggest that the proposed method outperforms the Word index method using the Binary scheme and the Word index method using the TFxIDF weighting scheme. It also outperforms the Stem index method using the Binary weighting scheme but does not outperform the Stem index method using the TFxIDF weighting scheme, and again it outperforms the Root index method using the Binary weighting scheme but does not outperform the Root index method using the TFxIDF weighting scheme.

We have put together a corpus of 242 abstracts of Arabic documents using the Proceedings of the Saudi Arabian National Conferences as a source. All these abstracts involve computer science and information systems. We also designed and built an automatic information retrieval system from scratch to handle Arabic data. The system was implemented in the C language using the GCC compiler and runs on IBM/PCs and compatible microcomputers. We have implemented both automatic and manual indexing techniques for this corpus. A long series of experiments using measures of recall and precision has demonstrated that automatic indexing is at least as effective as manual indexing and more effective in some cases. Since automatic indexing is both cheaper and faster, our results suggest that we can achieve a wider coverage of the literature with less money and produce as good results as with manual indexing. We have also compared the retrieval results using words as index terms versus stems and roots, and confirmed the results obtained by Al-Kharashi and Abu-Salem with smaller corpora that root indexing is more effective than word indexing. © 1997 John Wiley & Sons, Inc.

This article describes a series of experiments with an interactive Chinese information retrieval system named CIRS and an interactive relational thesaurus. Two important issues have been explored: whether thesauri enhance the retrieval effectiveness of Chinese documents, and whether automatic indexing can compete with manual indexing in a Chinese information retrieval system. Recall and precision are used to measure and evaluate the effectiveness of the system. Statistical analysis of the recall and precision measures suggest that the use of the relational thesaurus does improve the retrieval effectiveness both in the automatic indexing environment and in the manual indexing environment and that automatic indexing is at least as good as manual indexing. © 1997 John Wiley & Sons, Inc.

This paper introduces a Knowledge Engineering System, the Disorder Toolbox (DT), which assists physicians to build, test, and verify the knowledge base for the Medical Emergency Decision Assistance System (MEDAS). DT is designed using a hypertext/hypermedia system and gives physicians an intelligent and user-friendly tool to streamline the disorder pattern creation process. DT reads the Portable Patient File (PPF) generated by the Intelligent Medical Record Entry (IMR-E) System, and allows the medical expert to enter prior and conditional probabilities, and test the posterior probabilities for disorders during the knowledge creation time. After building the disorder patterns, the physician can load a PPF and produce a differential diagnosis using the multimembership Bayesian inference engine. The process of building the Disorder Toolbox and its use are described.>

A treatment risk analysis system was developed and presented by the MEDAS project group. Now, a risk analysis system for medical tests has also been developed to form a complete medical decision support system. Moreover, the current system (a combined treatment and test system) has been redesigned and developed using an object-oriented hypermedia programming language, called Spinnaker Plus, running on the MS-Windows environment. The system has already been used to analyze several neurological problems.>

This system is designed to support clinic operations using an MS Windows graphical user interface to provide a menu-driven and interactive operation requiring only mouse clicks to replace the traditional character-based environment for physicians to query from the database and to generate reports. The object-oriented design methodology has made it easy to implement the system and adapt it to changing user needs.>

This paper describes the architecture of the recent version (v.3) of a cardiovascular physiology tutor, called CIRCSIM-Tutor. The intended purpose of CIRCSIM-Tutor is to help first year medical students in developing a problem solving skill in the domain of cardiovascular physiology. This architecture was constructed using a systematic intelligent tutoring system (ITS) development methodology which combines the features of a systematic knowledge based system development methodology with an instructional system development methodology in a single framework. The architecture of CIRCSIM-Tutor (v.3) is flexible and is consistent with some important software engineering concepts. The architecture divides the functionality of CIRCSIM-Tutor into a set of subsystems which communicates with a message passing protocol. The shared resources such as domain knowledge base, lexicon, student model are kept separate as information stores which may be freely accessed by all subsystems of CIRCSIM-Tutor.>

Describes the patient simulation program that we created for the MEDASPC system. The program provides the patient file with different features that were randomly chosen by the computer or allows the user to set the features individually. We can use the program to create a multi-encounter patient file that corresponds to the way patients and physicians often interact. For the first encounter, we choose the setting of and the physical examination features. Then, for the second encounter, we set laboratory and radiology tests results, etc. By using simulated patient files, medical students can learn and make mistakes without taking risks with real patients.>

The Micro-AIRS System, a microcomputer system for Arabic Information Retrieval, was designed as an experimental system to investigate indexing and retrieval processes for Arabic bibliographic data. A series of experiments were performed using 29 queries against a base of 355 Arabic bibliographic records, covering computer and information science from the bibliographic databank at King Abdulaziz City for Science and Technology. These experiments revealed that using roots and using stems as index terms gives better retrieval results than using words. The root performs as well as or better than the stem at low recall levels and definitely better at high recall levels. Several different binary similarity coefficients were tried: the cosine, Dice, and Jaccard coefficients

A quantitative, multi-function treatment risk analysis system is being developed by the MEDAS project group. Two main purposes are: to provide a computer-based decision support tool in treatment decision making for practicing physicians; to provide an effective medical decision analysis training tool for medical school students as well as for physicians who are interested in acquiring the skills in this area. The fundamental methodology used in this system is based on Weinstein's quantitative approach (M.C. Weinstein, H.V. Fineberg, 1980) and the concept of value of clinical information. Three different function models in this system have been constructed. They are the automated consultation model, the user self-consultation model, and the training model.>

The authors describe a local area network being used in the Adult Emergency Department at cook County hospital in Chicago. Two modules of MEDAS, the LOG program and the electronic patient tracking board program, are running on this LAN. The network itself is composed of PCs, Macintoshes, Novell Netware, and an Oracle Server and is designed to be fault tolerant. The system handles about 250 visits per day well and more than 100000 patients have been logged in by the system. MEDAS (the Medical Emergency Decision Assistance System) is a medical expert.>

This paper describes a series of experiments to test whether there are significant differences between several different versions of the intelligent medical record-entry system (IMR-E). As the costs of including color graphics decrease, the use of multimedia technologies is becoming an important issue. The development of pen computers provides a whole new range of possibilities for data input. The authors are planning to measure the effects of color, machine speed, and the use of a mouse or a pen computer on input time and user satisfaction when the system is used to input patient data by physicians and medical students.>

This paper describes the building of the domain-expert for a cardiovascular physiology tutor called CIRCSIM-Tutor using a systematic development methodology. The domain-expert is a functional component of CIRCSIM-Tutor and provides the domain intelligence to the rest of the system. The authors have developed a multi-dimensional knowledge representation technique to capture expertise in the domain, which led them to the construction of a conceptual model of the domain-expert of CIRCSIM-Tutor. The authors describe how this conceptual model is transformed via an object-oriented methodology to yield the architecture of the domain-expert of CIRCSIM-Tutor. The conceptual model is at the epistemological level and has a real-world-orientation, whereas the architecture of domain-expert emphasizes system-orientation. The resulting architecture is query-driven and provides a fully transparent access to each chunk of domain knowledge, of different types, with optimal flexibility.>

This paper describes an automated Progress Note System running in the diabetic clinic of the VA Medical Center in North Chicago, Illinois. The organization of the authors progress notes is based on Weed's problem-orientated approach to medical records (L. Weed, 1970). The system helps the physician to formulate a problem list and then, for each problem, it assists in the collection of subjective and objective data and the generation of assessments and plans. The Progress Note System uses data downloaded from the VA's MUMPS database and it interfaces to the Oracle database maintained for research purposes by the diabetes clinic. The reminders are based on the American Diabetes Association's position statement on standards of medical care.>

Computers have proved to be an effective teaching tool for science education. Developments in the field of computer graphics have made it possible to communicate information visually and expedite the process of learning abstract concepts. This paper presents the design of a graphical user interface to an intelligent tutoring system and discusses its use, not only as a user-friendly screen manager but also as a visualization tool to convey information about the domain of learning. The authors also propose some experiments with different versions of the interface, to answer questions about how domain visualization may affect learning.>

Presents a surface generator for an intelligent tutoring system based on analysis of human tutoring sessions. The tutoring system is designed to help first-year medical students understand the negative feedback system that regulates blood pressure. The authors illustrate the generation of appropriate natural language output using Lexical Functional Grammar based on a study of the cardiovascular sublanguage.>

The authors introduce a novel patient severity measurement model using neural networks. A three layer, fully connected backpropagation neural network was used in the pilot experiment. The results are promising and demonstrate that the backpropagation neural network technique is capable of assessing the severity value by learning from raw data. The neural network is easy to improve and of relatively low cost. It saves the expert's valuable time used in assigning numerical values to variables.>

Describes the implementation methodology of an intelligent progress note system which is designed to support physicians in writing problem-oriented progress notes. The authors are currently developing their system on PCs in Microsoft Windows 3.0 by using Spinnaker Plus. The conversion of the system from a PC base to Macintosh or OS/2 has proven to be easy. The design applies cognitive models of memory as well as hypermedia to provide a creative and productive environment for the physician. The basis of the work is observation of practicing physicians writing progress notes in today's paper-driven world. The hope is to simulate some of the more routine parts of the thinking processes that physicians devote to producing the progress note, to save time, and to provide useful reminders, so that the users can concentrate on the more complex parts of the task.>

A computerized test selection module for a multimembership Bayesian diagnostic expert system built to recommend tests to further refine the differential diagnosis is described. The output of the program includes recommendations for individual tests, for groups of tests to be administered in sequence, and for groups of parallel tests. The original set of available tests is large. This generates a large number of groups of tests, any of which could possibly provide the most valuable information. Selection of the proper subset of tests from the original set is critical. A smaller proper subset of tests improves the systems response time by avoiding unnecessary calculation.>

A medical record input system (MEDRIS) that makes use of hypermedia technology to provide a clinical tool for physicians to input patient data in a simple manner is described. The design of MEDRIS focuses on providing a natural working environment for the physicians. Physicians can easily learn to use MEDRIS since it eliminates the use of command-level entries and special function keys. MEDRIS comprises the following modules: history (chief complaints, review of systems, family history, etc), physical examination, laboratory, and medication. The chief complaint and the physical examination modules are discussed in detail. The software techniques adopted in building this system are explained. MEDRIS produces an immediate report of the data entered and an encounter file containing all the patient data collected during the current visit, which is added to the portable patient file.>

CIRCSIM-TUTOR, an intelligent tutoring system (ITS), combines both the overlay approach (used to detect declarative errors) and a bug library (for recognizing misconceptions in problem solving). The student modes in most ITSs evaluate the student's current response and store an analysis. However, both earlier responses and immediate ones should be taken into account. It should also be recognized that students may show inconsistent behavior. Exploring initial implementation of CIRCSIM-TUTOR shows that this is a better approach to tutoring first-year medical students about the causal relationships between parameters in the cardiovascular system involved in regulating blood pressure using complex negative feedback process.>

An intelligent tutoring system (ITS) that will assist first-year medical students in learning the causal relationships between the parameters of the circulatory system and in solving problems of disturbances to the system is discussed. The central component of the ITS, the instructional planner, is responsible for determining what to do next at each point during a tutoring session. The approach taken is to build the planner by combining the capabilities of lesson planning and discourse planning in order to provide globally coherent and adaptive instruction. The planner consists of two parts: a lesson planner and a discourse planner. The lesson planner generates lesson plans, which the discourse planner carries out. A sample dialogue, extracted from the transcript of an actual human tutor-student interaction, is used to provide a framework for the development of the overall system, especially from the planner's point of view.>

The similarities and differences between four computer-based medical record systems are compared, and their progress note facilities are examined. They are: COSTAR (Computer-Stores Ambulatory Record), TMR (The Medical Record), RMRS (Regenstrief Medical Record System), and STOR (Summary Time-Oriented Record). An intelligent problem-oriented progress note system is proposed for MEDAS (the medical emergency decision assistance system).>

LITREF was developed to run on a microcomputer to support MAIESTRO, an expert system for the diagnosis and management of stroke cases. The architecture of LITREF uses an inverted file structure with a bitmap strategy. The evaluation process uses the cosine function to measure the similarity between a query and an abstract. The Salton interpolation process is used in the computation of recall and precision values. The experiment involves applying alternative suffixing algorithms to both index terms and query keywords. When the results are evaluated by comparing precision values at each recall level, it is found that the word-stem index method is superior to the full-word index method.>

In order to support explanations in a knowledge-based system, a method is proposed to extend an existing knowledge base: adding more knowledge and including relations and their semantics among the different types of knowledge. The criterion followed to organize a knowledge base is to classify and represent knowledge according to different levels of application and abstraction. This criterion also facilitates future extension of a knowledge base for explanations of the procedure and heuristic knowledge.>

A knowledge acquisition procedure is presented that is an alternative to the interview process, the traditional method used in knowledge engineering. In an effort to produce a knowledge base in a more timely and efficient manner, an automated procedure is developed. This process examines historical patient cases and generates expert system production rules from them. These rules form an initial knowledge base which can then be honed by the domain expert into the final knowledge base for an expert system. It is found that this automated procedure generates production rules that are equal in value to those produced through the interview process.>

The authors describe an attempt to construct a knowledge base automatically by parsing a natural language text containing the needed information. After preprocessing using a series of programs developed locally, the text is parsed using Sager's linguistic string parser. The parse trees are converted into information formats, and then the information formats are transformed into frames in the knowledge base. The knowledge base is destined to serve as an essential resource for an intelligent tutoring system being developed as a joint project between Illinois Institute of Technology and Rush Medical College. The text is a chapter on cardiovascular physiology written for medical students.>

The feature dictionary of a Bayesian diagnostic system is expanded to include feature interactions to support test selection and treatment protocols. The feature interactions required to provide this flexibility are drug side effects and allergic reactions, contraindications to tests and treatment, and drug interactions. Although the primary use of the dictionary is to store the feature information to be used by the MEDAS (Medical Emergency Decision Assistance System) diagnostic system, adding further functionality allows other applications, both inside and outside of MEDAS, to be supported. Generic and brand names of drugs appear as synonyms in the dictionary. This widens the ability to translate data from one system to another.>

An online bibliographic retrieval system, called the literature reference (Litref) system, is presented. It is designed to run on a high-end microcomputer to support the stroke consultant system (Maiestro). Litref interprets a Boolean command language derived from knowledge-index to provide fast, efficient browsing of abstracts from the stroke literature. User feedback on keywords derived from a relational thesaurus is used to enhance the original query. The system is designed to work in a microcomputer environment. Features and goals, design principles, and system architecture are discussed.>

We have analyzed definitions from Webster's Seventh New Collegiate Dictionary using Sager's Linguistic String Parser and again using basic UNIX text processing utilities such as grep and awk. This paper evaluates both procedures, compares their results, and discusses possible future lines of research exploiting and combining their respective strengths.

Relational database techniques have been used to create knowledge bases for a medical diagnostic consultant system. Known as MEDAS (Medical Emergency Decision Assistance System), this expert system, using disorder patterns consisting of features such as symptoms and laboratory results, is able to diagnose multiple disorders. Database technology has been used in MEDAS to develop knowledge engineering tools, called the TOOL BOX, which permit domain experts to create knowledge without the assistance of a knowledge engineer.

We are building a question-answering system to assist physicians obtaining information from a patient knowledge base. We were motivated to build this system by evidence that physicians are much more willing to use systems that provide interactive responses. Sager’s Linguistic String Parser is being used to generate responses from information formats. These responses combine information retrieved from the knowledge base with inferences made from that information.

This paper describes an interpreter for ABF, a language for constructing legal documents. The ABF language defines both commands to develop documents and commands to manipulate those documents. The language has typical programming constructs for sequence, selection, and repetition but also allows the user other capabilities to aid in document construction; the user can modify the program while it is running (change the document while it is being processed), type the variables during the initial processing of a document, and then change the type during later processing, enter commands that require a change of context during the processing of a document, output a document with variables still undefined. A document is constructed using conditional and repeat statements and incorporation of other documents by reference. Data types include documents, strings, numbers, dollar amounts and logical values. Arrays can be formed for all types except documents. There are three forms of subprograms: documents can be incorporated by reference in other documents at any point at which a variable name is permitted; rules (subprocedures) for computing values can be attached to any variable; and alternate questions can be attached to variables in a similar manner. The ABF system queries the user for any values needed to complete a document; questions are generated from variable names.

Natural language processing systems need large lexicons containing explicit information about lexical-semantic relationships, selection restrictions, and verb categories. Because the labor involved in constructing such lexicons by hand is overwhelming, we have been trying to construct lexical entries automatically from information available in the machine-readable version of Webster's Seventh Collegiate Dictionary. This work is rich in implicit information; the problem is to make it explicit. This paper describes methods for finding taxonomy and set-membership relationships, recognizing nouns that ordinarily represent human beings, and identifying active and stative verbs and adjectives.

article Free Access Share on Multikey retrieval from K-d trees and QUAD-trees Authors: D. A. Beckley AT&T Bell Laboratories Naperville, Ill. AT&T Bell Laboratories Naperville, Ill.View Profile , M. W. Evens Computer Science Department, Illinois Inst of Technology Chicago, Illinois Computer Science Department, Illinois Inst of Technology Chicago, IllinoisView Profile , V. K. Raman Stuart School, Illinois Inst of Technology, Chicago, Illinois Stuart School, Illinois Inst of Technology, Chicago, IllinoisView Profile Authors Info & Claims ACM SIGMOD RecordVolume 14Issue 4May 1985 pp 291–301https://doi.org/10.1145/971699.318925Online:01 May 1985Publication History 9citation765DownloadsMetricsTotal Citations9Total Downloads765Last 12 Months11Last 6 weeks5 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteeReaderPDF

Abstract This article describes the design and development of a new type of thesaurus based on lexical‐semantic relations. Relational thesauri have been constructed to perform a new type of term classification. These relational thesauri are generally applicable to any document collection and their maintenance is relatively simple. A series of experiments to evaluate thesauri of this new type have been run on an information retrieval system called IRS at Illinois Institute of Technology. The results of experiments with queries enhanced using thesauri based on several different groups of relations have been compared against performance with the original queries. Thesauri from most groups, except antonyms, made improvements in recall as well as in precision. The best results come from a set of ill‐formed queries with few index terms. These results have been analyzed with both precision‐recall graphs and statistical tests. While thesauri that combine a number of relations together have been most effective in a batch environment, there is reason to believe that individual relations will be more useful in an interactive retrieval system that presents index terms to the user and allows him to choose those that best convey his meaning.

Donald Norman has claimed that UNIX has cryptic and inconsistent command names. As Michael Lesk has remarked, the lack of objective data makes it difficult to evaluate the significance of Norman's criticisms. In an effort to explore this controversy we taught one group of novice users the UNIX command language and another group an English-based command language (NUIX). The number of errors and calls for on-line assistance were compared. The subjects in this study were 22 high school women with no formal exposure to computers. The results reveal that the UNIX group made fewer errors than the NUIX group in two training sessions a week apart. Although calls for on-line assistance for the two groups in the first session were comparable, the UNIX group made over twice as many calls for on-line assistance in the second session as the NUIX group. Our findings suggest that even though the UNIX command language may not be harder for novice users to learn, it is probably more difficult for them to use.

A compiler for a distributed system should both run on that system and generate distributed code for that system. The compiler must be organized so that its tasks will run in parallel on the nodes of the system. New methods of compile time error recovery are necessary. The compiler should include a partitioning routine which can divide any source program into separate run time modules automatically. The problems of control of the run time environment and run time error recovery also require new solutions in a distributed system.