capability of digital electronics, electronic devices ... Developing a digital system that contains, as its ..... tionary, such as size and layout of keyboard, but the.
The pocket calculator is now an integral part of math education; a portable language-oriented device could have a similar impact on reading and writing instruction.
AUTIOMATD ilno
Mark S. Fox Donald
J. Bebel
I
Alice C. Parker Carnegie-Mellon University
U V X
As a result of the declining cost and increasing capability of digital electronics, electronic devices will probably replace some quite commonplace educational tools, at the same time expanding the way these tools can be used. The subject of this article, the dictionary, is one such item.' Developing a digital system that contains, as its data base, a standard dictionary of the English language poses some interesting educational questions and provides fertile ground for new ideas that can aid in vocabulary development and improvement of reading and writing skills. Two advantages provided by using an automated dictionary instead of a dictionary in book form are that the user-interface will provide a simpler means of accessing dictionary entries and that the device will provide a set of search capabilities and games enhancing the usability of the dictionary.
Although there are several technological issues to be resolved, a first-generation automated dictionary can be built with present technology, demonstrating the potential usefulness of such a device and pointing out the difficulties that can be expected in more sophisticated systems in the future. By starting the design process now, automated dictionary designs will exist by the time component costs decrease to the point where large automated dictionary systems become economically attractive. The purpose of this article is to present the results of some preliminary studies conducted at the request of the National Institute of Education. We determine the capability of an automated dictionary system built using present technology and estimate present and future costs based on trends in the cost and functionality of digital electronics.
access to and the display of entries. The "automation" of a dictionary raises several issues: * What information should be stored? * How does the automated dictionary's size determine the style of usage? * What is the interface between the user and the dictionary? * What are the automated dictionary's physical components? Each issue provides a rich set of alternatives from which to choose. Part of the design problem of an automated dictionary is picking feasible points in this multidimensional design space.
The dictionary entry. The primary purpose of the automated dictionary is to store and allow access to a dictionary data base. An initial problem is determining the type of dictionary information to be stored. Consider the definition of the word "flag," taken from Webster's New Twentieth Century Dictionary of the English Language2:
flag1: (flag) n.[LME.flagge< ? Flag4, inObs. sense "to flutter"] 1. A piece of cloth or bunting, often attached to a staff, with distinctive colors, patterns, or symbolic devices, used as a national or state symbol, to signal, etc.; standard; ensign. 2. [pl.] [Now Rare] Long feathers or quills as on a hawk. 3. The tail of a deer. 4. The bushy tail of certain dogs, as setters and some hounds. 5. Music any of the lines extending from a stem, indicating wIhether the note is an eighth, sixteenth, etc. -vt. flagged, flag'ging 1. to decorate or mark with flags. 2. to signal with or as with a flag; esp. to signal to stop (often with down). 3. to send (a message) by signaling -dip the flag to salute by lowering a flag briefly -strike the (or one's) flag. 1. to lower the flag. 2. to give up; surrender flag2: (flag) n. [ME. flagge < ON. flaga, slab of stone < IE. base *plak-, to spread out, flat, whence L. placidus,
flat] same as FLAGSTONE -vt. flagged, flag'ging to pave with flagstones
What is an automated dictionary? An automated dictionary is a computer-based device that holds all or part of a dictionary and allows July 1980
'
We list two separate entries for the word "flag," although the dictionary actually contains four
0018-9162/80/0700-0035$00.75
©
1980 IEEE
35
separate entries. The rationale for separate entries is that each entry represents a conceptually different definition of the word. Notice that each entry contains multiple definitions which are not necessarily related. The multiplicity of definitions presents problems in entry display. How many and which definitions should be displayed-all, one, or several of the most common? Of course, such considerations are dependent on display size. Each definition also contains a variety of information, including correct spelling, syntactic category, pronunciation (written), etymology, synonyms, and syntactic variations. Information such as antonyms, examples of usage, hyponyms, and pictures can also be found. Obviously, a dictionary definition is a rich source of information. What information is chosen for inclusion in the automated dictionary is dependent upon many variables, such as the type of user (child vs. adult, writer vs. physicist) and storage availability (such as 3000 "4complete" entries vs. 10,000 abbreviated entries). An automated dictionary can be designed as a specialized dictionary, the dictionary can be an easily replaced memory module, or it can have a network connection for changing the dictionary. Alternatively, the data base can be stratified to present different information to different users, according to user profile or command.
Usage scenarios. The primary factor determining possible student usage is the size of the device. Constructing an automated dictionary the size of a handheld calculator would greatly restrict dictionary size and modes of access (e.g., simple query) and display. For example, it would be awkward to try to display a complete dictionary entry when the screen displays only one 20-character line. Neverthele!s, the small size of the dictionary would enhance its portability, possibly resulting in greater utilization. A device the size of a briefcase would still be portable, although not as convenient; however, the dictionary would be larger, and the possible modes of access and display would be richer. For example, a briefcase could have a 20 X 80 character display capable of displaying a complete entry and a menu-selection command interface. Portability would allow the student to take the device home and either use it with batteries or plug it into an electrical socket to access the dictionary and associated games. The device could be used either as a learning tool or as an aid in doing assigned homework. If desired, it could be set up to gather statistics about student usage. It could be brought back to school the following day for assignment to another student or for off-loading the statistics gathered during the home session. In all cases, it is important that the student become comfortable with the device so that it is viewed as another resource. As the automated dictionary grows in size, its portability is lost. The transition to a stationary device removes the constraints on dictionary size and modes of interaction. Then, its functionality is limited by cost and technology. Stationary devices can be divid36
ed into two categories: stand-alone and networked. A stand-alone unit would be self-contained in that dictionary and programs would be contained within the computer and peripherals. It would not communicate with other devices. Although the stand-alone device would not be portable, the fact that it is self-contained would allow it to be placed in any location. Stand-alone automated dictionaries could be placed in libraries, classrooms, or student rooms. One could be assigned per student or be used on a demand basis. The networked automated dictionary would have the ability to communicate with other automated dictionaries, computers, or datastores. There are no requirements that the complete dictionary be resident, nor must query processing be done by the dictionary. It could be a complete computer system communicating with others via a network, or it could be a "dumb" terminal connected to a timesharing system. Each school could have an automated dictionary timesharing system with terminals in classrooms, hallways, the library, and elsewhere throughout the school. In addition, the system could be accessed over dial-up phone lines from student homes-a possibility arising from the proliferation of home computers and terminals.
User-dictionary interface. Several functional considerations must be kept in mind throughout all stages of system design. Most of these deal with the interface between the user and the dictionary and, in turn, help to establish the physical capabilities of the device. First, the device must be simple to use and understand. The information must be displayed in a clear manner, and the device should respond coherently to invalid input (graceful interaction). The user interface can range in complexity from simple function selection (e.g., Speak and Spell, Simon Says) to menu selection3 or natural language interfaces.45 Besides providing information on request, the software controlling the system should contain programs that entice the user to explore the dictionary data base. The response of the system must be geared to the educational level of the user or provide information at several levels so as to be useful for a prescribed range of user interests and capabilities. Access to the device should be convenient and simple. If the automated dictionary is to be used as an educational tool, it should have the capacity to change over time so as to maintain user curiosity, and it should be able to monitor and record user progress. Below, we consider some of these interface issues in more detail.
Measuring interface quality. Acceptance of the automated dictionary in the educational environment depends upon user impressions. Is it easy to use? Or is it complex? Does it provide more functions than the dictionary in book form? How long do you have to wait for an answer? The answers to questions like these determine the device's acceptance. In their work in modeling human-computer interfaces, Card et al.6 have identified seven interface performance COMPUTER
factors that affect user perception. * Time: How long does it take a user to accomplish a given set of tasks? * Errors: How many errors does a user make, and how serious are they? * Learning: How long does it take a novice to learn how to use the system to do a given set of tasks? * Functionality: What range of tasks can a user do in practice with the system? * RecalL- How easy is it for a user to recall how to use the system on a task that he has not done for some time? * Concentration: How many things does a user have to keep in mind while using the system? * Fatigue: How tired do users get when they use the system for extended periods? The goal is to design an interface which can satisfy as many of these factors as possible. In identifying these factors, it is assumed that the user wants or has to use the machine. This is not necessarily the case in an educational environment, since students must want to use the device or it will fall into disuse. Therefore, there are four additional factors that must be considered when designing interfaces for educational environments. * Variety: Does the interface provide information in a variety of ways; for example, does it offer color vs. black and white, flashing images, speech output? * Curiosity: Is the user's curiosity to explore piqued by the interface? * Fun: Does the user find the device fun to use? * Adaptable: Does the automated dictionary adjust to the student's level of competence as the user becomes more experienced with the device? While satisfying these factors is a difficult design problem, it must be done if the automated dictionary is to be successfully introduced in the educational environment.
by the user. If spelled correctly, a positive response is given to the user. Functions that gather statistics on user performance could also be provid-
ed.
Word recognition: A word is randomly chosen and permuted according to standard spelling errors. The student must confirm or deny its legality as a word. A list of letters is provided to Anagrams: the user, who then tries to compose as many legal words as possible from them. A definition is displayed, and Word search: the user must provide the corresponding word. Picture naming: A picture is displayed, and the user must name it or parts of it. Synonym search: By following synonym paths, the user must determine whether two words' are related. The availability of these functions increases the automated dictionary's value as an educational device while hopefully sustaining the curiosity level of the user. The presence of a large variety of functions does not imply that the user can access them. Allowing access to these functions requires a control interface with which the user can confirm what functions he may execute in a particular state andproperly specify the desired function. Hand-held calculators deal with this problem by providing a key for each primitive function. Data-base systems often provide complex idiosyncratic interface protocols. In the former, key space limits the number of functions; in the'latter, the command protocols can be complex and opaque. One approach that allows the display of an arbitrary set of commands in a simple manner is to Functionality and its perspicuity. In its most basic display a menu of available commands. A menu is form, the dictionary would contain an ajphabetical dynamically generated on the display screen, and list of dictionary entries. A user would ask for a word, hence is not limited by key space. It clearly informs and the dictionary would respond with q definition. the user what functions are available for execution. Such a device could be readily implemented using a Graceful interaction. Providing a variety of funcprocessor, a large ROM, and an I/O device. However, tions which to access the dictionary does not the benefits of an automated dictionary can be real- assurewith Both naive and expert product ized when the full capabilities of the CPU are utilized. users often makeacceptance. when mistakes initiating functions. processor the Besides taking over the search task, errorful to input is a factor in reacts How the device can be used for such diverse tasks as spelling correcIt user satisfaction. important that determining tion, synonym and antonym search, voiced word pro- the dictionary react in a "natural"isway to incorrect nunciation, game playing, and perhaps gramm#r cor- requests. rection and sentence-building. Some examples of Typical interfaces to information systems require games are the exact specification of the query. Errors in spelling The user types in a word, and or command format are generally not acceptable. One Word retrieval: the corresponding dictionary potential use of the dictionary would be to provide information is displayed. the correct spelling of a word when the user does not The model presents to the user, know how to spell it exactly, but has a good idea of Spelling test: by voice output, a word at a what it should look like. If the automated dictionary level of difficulty chosen by the only retrieved correctly spelled words, then it would user. The word is then typed in be of little more value than a standard dictonary. A July 1980
37
more "graceful" approach would be either to reply with the correct spelling or to present a list of words that resemble the requested word. If neither can be provided, then the user should be given access to a word index. With ill-formed commands, a graceful reaction would be to inform the user that an error was made and then display the correct form of the command. Another aspect of the gracefulness of interaction7 is the ability to "undo" commands or "back up." Often, a user places an incorrect request and overprints the current display, or grows tired of a particular line of inquiry and wants to go back to a previously displayed piece of information. Either way, the user wishes to undo a sequence of function requests so that he can return to a previous state. Whatever the error may be, the interface between user and dictionary must be robust enough to allow successful execution of the desired request, explanation of incorrect function specification, or reconstruction of previous states from which to start over. Physical components. Once size has been determined, a choice of physical components must be made. The most important choices available to the designer are display, memory, and processor characteristics, as well as extensibility to other devices. Some type of display, either hard-copy or electronic, is required to allow the user to communicate with the automated dictionary. Due to the cost and reliability constraints, the decision was made early to use an electronic display. Two major design choices remained-display size and technology. While the size of the display depends on both cost and portability constraints, a minimum number of characters was required so that an entire definition could be displayed. The display technology choice is also affected by cost and portability. Possible choices include raster scan video, vector CRT, LED, LCD, and plasma displays. An external connection for a home TV is another alternative. Memory design is dominated by size and technology issues. Since the cost of memory grows somewhat linearly with size, the amount of storage is limited by cost and portability constraints. The memory speed is not an issue for most of the possible implementations since the compute tasks are not complex, particularly in systems designed for single users.
The type of memory used depends on the current state of IC technology, cost, portability, and reliability. ROM is preferable to RAM, since densities tend to be higher and ROM is nonvolatile. While floppy disks are inexpensive, they do not enhance portability. Hard disks are a design choice reserved for the largest model of automated dictionary, due to price and portability constraints. Video disks provide a future replacement for hard disks, and magnetic bubble memories will probably be a cost-effective replacement for ROM storage in the 1980's. The choice of a processor for the automated dictionary mainly depends on how the processor affects
38
cost and portability. It must have an adequate address space to handle the complete dictionary. Response time and processor-memory bandwidth are not issues except for the networked-standalone device, which is timeshared. There are indeed other design considerations that might affect the ultimate utility of the automated dictionary, such as size and layout of keyboard, but the four items discussed above are the most important in terms of the performance capability that can be expected over the next 10 years.
Designing the automated dictionary Not only is it impossible to examine all potential designs when a new system is being created, but the range of feasible designs is also usually limited by constraints. For these reasons, several design decisions were made a priori with the intention of establishing a simple but adequate set of design choices on which the cost analysis could be made. These decisions involved the size of the automated dictionary (i.e., the number of entries), the amount of information contained in each entry, the manner of display and other I/O, and the functional capabilities of the dictionary. Perhaps the greatest restriction concerns the number of entries, which was limited to between 3000 and 30,000. Although this figure is low when compared with the number of entries in most published dictionaries, it is within the range that can be produced with present memory technology. Each entry is assumed to include the word, syntactic category, written pronunciation, and definitions. In addition, there are examples of usage, synonyms, antonyms, voiced pronunciation, and inverted lists. Illustrations are not included because the technology for lowcost graphics displays is not presently available and will not become available within the next 10 years. Computer-generated voiced pronunciation is included because this technology is available8-a feature that distinguishes the automated dictionary from standard book-form dictionaries. The size of each dictionary definition was deteimined by estimating the average number of characters in a standard dictionary entry.2 If each character is encoded in a bit string at the rate of 4.5 bits/ character9 and with the addition of the examples of usage, synonyms, antonyms, etc., the average dictionary entry would require approximately 2800 bits. Appendix 1 summarizes the storage requirements for each entry according to the list of items just described. The automated dictionary's functional capabilities create an interactive learning environment for the student. Some of the functions provided, such as the spelling test, distinguish between correct and incorrect input and provide assistance in correcting errors. For example, one approach to correcting spelling errors is to store both rules for common spelling errors (such as "i" before "e" except after "c") and diagram COMPUTER
statistics10 to analyze incorrect input and suggest legal alternatives. The dictionary machine development and analysis were based on these guidelines. A set of design choices was subsequently made, resulting in an analysis of three models of the automated dictionary. Below, we present and briefly discuss the uses of each of these three models, including a cost estimate for building a prototype. Afterwards, we give an example of a user-dictionary interface that could be used by one or more of these models. Model 1. The portable dictionary. Some of the advantages of a compact, portable device are lower cost, a potentially large market, and mass production. Of course, this means that the packaging must be small enough and light enough to be hand-held or handtransported. Therefore, the number of entries is limited by the packaging constraints, which also limit display size and the number of functional features that can be included in the design. For purposes of cost analysis, the number of dictionary entries is assumed to be limited by the amount of memory that can fit into a briefcase-size carrying case (approximately 50 integrated circuits on one circuit card). The dictionary size is limited to 2300 entries with present 128K-bit ROM circuits, but as discussed later, this number should easily increase by an order of magnitude over the next eight or 10 years. There is a keyboard containing alphanumeric characters and some special-function keys. Two possible display techniques selected for this design are LED alphanumeric displays of approximately 200 characters and an interface to a television set, similar to presently marketed video games. Primarily because of the large address space available, the processor should be one of the presently marketed 16-bit microprocessors, such as the Motorola 68000, Intel 8086, or Zilog Z8000. Other advantages of these units over 8-bit processors are their increased speed and enhanced instruction sets, which can more efficiently implement the functional requirements of the automated dictionary. The recommended software monitor for the system is a command interpreter with available functions continuously presented in a menu-selection format. For example, associated with each function is a selection character or a special-purpose key on the keyboard that iminediately evokes a response when entered. The types of functions included are definition search, spelling correction, synonym and antonym search, spelling test, anagrams, and a variety of other word games.
Figure 1 is a collection of representative curves of memory cost as a function of the number of dictionary entries for a variety of memory technologies. From the figure, it is clear that ROM has or will soon have a distinct cost advantage over the other types of memory systems. As in the first design, voiced pronunciation is included as well as the software monitor and the software functions described for Model 1. Model 3. The timeshared system. Included as an example of a system that requires no hardware development, this model is built around a single minicomputer and centralized disk storage device. The minicomputer services a large number of dumb terminals in an arrangement similar to a language lab where, for example, an entire class of students could use the system at once. The advantages of this system are the low cost per terminal (assuming a large number of terminals are connected to the system), the security of the facility, and the ability to handle class-oriented exercises. Furthermore, this system has the advantage of being able to support centralized peripherals, such as line printers or magnetic tape storage devices. Disadvantages of the system include limited access to the machine by individuals because of the centralized nature of the system and the loss of convenience available with stand-alone distributed devices.
Model 2. An intelligent terminal. The second de-
sign is best described as an intelligent terminal with
its data base stored in an external mass-storage peripheral. Unlike the portable model, this is a desktop device having a standard video terminal with keyboard data entry. The restriction on physical size of the memory is removed, and the number of en- Figure 1. Memory cost as a function of the number of dictionary entries tries is determined by utilization or by a limit on the (2800 bits/entry) in a comparison of several different types of storage system cost. media. July 1980
39
While the software included in this design is essentially the same as that discussed for the first model, it has the additional capability of handling some programming to aid instructors in the development of exercises. Interface example. It is assumed that the user would be, at best, a casual computer user; hence, the interface should display the graceful interaction characteristics described earlier. The following example describes a menu-selection style interface for displaying both available functions and individual dictionary entries. The interface requires a standard character video display, 24 X 80 characters, which would be available on the second and third models (and on the first model, if it has a TV connection). A small concept demonstration system was constructed using the Zog system.3'11 The goal was to investigate how entries could be displayed and a possible style of interaction. Storage requirements to support this type of interface were not included in the cost analysis.
Zog is a rapid response, large information (frame) network, menu-selection system used for manmachine communication. A Zog user sits in front of a terminal displaying a frame consisting of some text and a set of options. At the discretion of the user, an option is selected and a new frame is displayed almost "instantaneously." Then, the process starts again. Zog also allows communication among the user, net, and background jobs. Option selection may cause background jobs to execute functions not provided directly by Zog. The basic philosophy of Zog is that a menuselection system can be an effective communication system if the user can move around in the system quickly and if there is a large network available to meet the user's needs. A menu-selection system allows the user to have almost complete knowledge of what is occurring in the system whenever he selects an option. By placing a link between related frames, it also allows for related information to be located nearby. Menu-selection systems normally have as a disadvantage the time it takes to move from one frame to
another, but the Zog system has a fast respofise nature. Another characteristic of many menuselection systems is that the same information is provided to all users, both expert and naive, regardless of their needs. In Zog, different paths would be provided for different levels of users. For example, the naive user would get more information about what he is doing, while the expert user would be presented with only the frames needed to perform the task. The example DICTIONARY-NETWORK is composed of frames representing dictionary entries, indices to the entries, and functions for accessing the entry frames. Each dictionary entry is represented by a primary frame containing pronunciation and etymology as text and definitions as options (see Figure 2a). Options are preceded by a selection character followed by a period. For example, in Figure 2a, the selection character for the first definition "A piece of cloth. .. " is the number "1." Selecting a definition option causes transfer to a secondary frame (see Figure 2b) containing the definition, synonym, antonym, hyponym, and examples of usage.
The frame represented in Figure 2b was reached by selecting the option text on the screen or hitting the "1" key on the keyboard. After selecting one of the synonym, antonym, or hyponym options, the primary frame for that word is displayed. Hence, synonym, antonym, and hyponym path searches are achieved simply through option selection. The frame for "banner" (Figure 2c) is reached by selecting item "1,," the synonym "banner," in Figure 2b.
The NEXT WORD (>) and PREVIOUS WORD ( > = => > ==>
30 words in the definition 6 characters per word 180 characters 4.5 bits per character 810 bits/definition
Estimations of the lengths of word usage examples assumed an average of three examples for each word and 10 words for each example-that is, 30 words used for examples for each dictionary entry. Combining this figure with the other estimates of six characters per word and 4.5 bitp per character, 810 bits are required for usage examples. For each word there is a set of synonyms. It was assumed that the average number of synonyms for each entry was five, and at six characters per synonym a total of 135 bits of storage are re-
quired. The representation and display of pictures were not considered, due to the display and storage requirements.
Appendix II. Notes on price estimates (1) A 16-bit microprocessor is used because of the need to address the extremely large memory. Furthermore, these processors have special functions that can more efficiently handle complex processing tasks, such as spelling correction. (2) The 128K-bit ROM and the 256K-bit ROM are not yet available, although Texas Instruments manufactures a 128K-bit ROM for use in the Speak and Spell learning aid.8 It is assumed that the price of these circuits will approach the cost ofthe 64K-bit ROM that is presently available. (3) Actual quantity prices were used whenever available (LED displays, D/A converter, memory circuits). Prices for items not yet available were based on prices of similar items that are currently available. (4) For the disk drives used in these estimates, it was assumed that 10 percent of available disk space is taken up by overhead. 47
3. G. Robertson, A. Newell, and K. Ramakrishna, ZOG: A Man-Machine Communication Philosophy, Techni-
cal Report, Computer Science Dept., Carnegie-Mellon University, Pittsburgh, Pa., Aug. 1977. 4. F. Hayes-Roth, D. J. Mostow, and M. S. Fox, "Understanding Speech in the Hearsay-I1 System," in Speech Communication with Computers, L. Bolc, ed., Springer-Verlag, New York, 1978, pp. 9-42. 5. E. D. Sacerdoti, "Language Access to Distributed Data with Error Recovery," Proc. Fifth Int'l Joint Conf. Artificial Intelligence, Cambridge, Mass., Aug.
1977, pp. 196-202. 6. S. K. Card, T. P. Moran, and A. Newell, The KeystrokeLevel Model for User Performance Time with Interactive Systems, Report SSL-79-1, AIP Memo 122, Xerox Palo Alto Research Center, Palo Alto, Calif., Mar. 1979. 7. P. Hayes and R. Reddy, An Anatomy of Graceful Interaction in Spoken and Written Man-Machine Communication, Technical Report CMU-CS-79-144, Computer Science Dept., Carnegie-Mellon University, Pittsburgh, Pa., Sept. 1979. 8. R. Wiggins and L. Brantingham, "Three-Chip System Synthesizes Human Speech," Electronics, Aug. 31, 1978, pp.109-116. 9. L. E. McMahon, L. L. Cherry, and R. Morris, "Statistical Text Processing," BellSystem Technical J., JulyAug. 1978, pp. 2137-2154. 10. C. Y. Suen, "N-Gram Statistics for Natural Language Understanding and Text Processing," IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. PAMI-1, No. 2, Apr. 1979.
Record of Proceedings-The 13th Annual Simulation Symposium March 19-21, 1980 Emphasizing discrete digital simulation, this annual conference also addresses continuous and analog techniques. Included in this year's proceedings are papers such as "Discrete Digital Simulation of a Genetic Control Theory," "GAP-A GPSS/Fortran Package," "Interactive Graphics for Enhancement of Simulation Systems," and "General Purpose Tactical Missile Simulation Program."
Non-members-S22.00
KS "M"
Mdmbers-S16.50
337 pp. Order No. 289
11. G. Robertson, D. McCracken, and A. Newell, The ZOG Approach to Man-Machine Communication, Techni-
12.
13.
14.
15.
16. 17.
cal Report CMU-CS-79-148, Computer Science Dept., Carnegie-Mellon University, Pittsburgh, Pa., Oct. 1979. M. S. Fox and A. J. Palay, "The BROWSE System," Proc. Ann. Conf. Am. Soc. Information Sci., Minneapolis, Minn., Oct. 1979. M. Mantei, Disorigntation Behavior in Man-Computer Interfaces, PhD Thesis, Annenberg School of Communications, University of Southern California, Los Angeles, Calif., 1980 (to appear). M. S. Fox and A. J. Palay, The BROWSE System: Phase II and Future Directions, ZOG Memo, Computer Science Dept., Carnegie-Mellon University, Pittsburgh, Pa., 1979. R. N. Noyce, "Microelectronics," Scientific American, Vol. 237, No. 3, Sept. 1977, pp. 63-69. T. R. Blakeslee, Digital Design with Standard MSI and LSI, Wiley, New York, 1975. A. S. Hoagland, "Storage Technology: Capabilities and Limitations," Computer, Vol. 12, No.5, May 1979, pp. 12-18.
Mark S. Fox is completing his PhD in computer science at Carnegie-Mellon University. His current research interests include knowledge acquisition, representation, and utilization in ar-
tificial intelligence; interface and database aspects of man-machine communication; and software organization. He was a member of the Hearsay-II speech understanding group, and has been associated with the USC/Information Science Institute and the Rand Corp. Fox received his BS from the University of Toronto.
Donald J. Bebel is a faculty member of the Department of Electrical Engineering at Carnegie-Mellon University and consults in the area of digital signal processing. His interests include signal processing, communications, optics, and visual information processing. In 1979, he completed his PhD thesis in electrical engineering at CarnegieMellon on the modeling of the processes of image noise perception. He received the BSEE in 1975 from Villanova University and the MSEE in 1976 from Carnegie-Mellon University. Alice C. Parker is an assistant professor of electrical engineering at Carnegie-Mellon University. Currently, she is supervising research and development of programs for automatic synthesis and optimization of digital hardware, and is designing a behavioral language for computer interconnection description and sixmulation. Parker received an MSEE degree from Stanford University in 1971 and a PhD in electrical engineering from North Carolina State University in 1975. Her professional memberships include IEEE, ACM, and Sigma Xi. A Use order form
on p.
96C.
COM PUTER
