dir@cbosgd.UUCP (Dean Radin) (04/14/84)
Since so many expressed interest, I'm posting a BRIEF version of my paper. Please keep in mind that I could not address all relevant points in this small an article. I'll be glad to discuss any comments you might have. EFFECTS OF COMMAND LANGUAGE PUNCTUATION ON HUMAN PERFORMANCE Dean Radin AT&T Bell Laboratories Columbus INTRODUCTION Nearly all computer command languages specify some form of punc- tuation to delimit elements of the language.... How does punctuation affect the efficiency of these languages? While some design guidelines can be found in the literature, few empirical studies have focused directly on human performance effects of command syntax (cf Cooper, 1983), thus when designers must choose among various syntax styles, there is little guidance other than opinion and intuition. For example, in Smith and Aucella's (1983) extensive compilation of user-computer design guidelines, only two of over 500 recommendations refer to punc- tuation in command languages: a) "Insofar as possible, the user should not be required to provide punctuation in command entries" (p.140), and b) "If punctuation is needed, perhaps as a delimiter to distinguish optional parameters, or the separate entries in a stacked command, one standard symbol should be used consistently for that purpose ..." (p.140). In an experiment, Ledgard, Whiteside, Singer, and Seymour (1980) investigated effects of surface syntax in two, semantically identical text editor command languages. One language used punc- tuation such as colon and semicolon as command delimiters, the other was English-like and used spaces as the primary delimiter. Among other results, they found that both experienced and naive users were faster and more accurate with the English editor than the punctuated one. Nearly twice as many errors were made with the punctuated editor. The present experiments were designed to directly address the effects of punctuation in command languages and to empirically test the guidelines offered in the literature. EXPERIMENT_1:_SPEED_AND_ACCURACY Eighteen adults familiar with a variety of command languages par- ticipated voluntarily; their average typing rate was 60 words per minute. All phases of the experiment, including a typing test, presentation of stimuli, collection of responses, and analyses of data were controlled by computer and displayed on a video display terminal (VDT). The procedure was as follows: A command in one of several dif- ferent syntax styles was displayed at the bottom of a VDT screen. As the subject copied it by typing, the characters were echoed the command by pressing the carriage return, the screen erased, and two seconds later the next command appeared. The computer timed each trial with 16 msec resolution from when a command was initially displayed to when the subject hit the carriage return key. Instructions were to type as quickly and as accurately as possible. Examples of the eight command formats and results of the experi- ment are shown in Table 1. The general command format was "command:key=value,key=value". An underscore (_) in Table 1 indicates a blank space. Dependent measures were characters per second (cps) typing rate and count of syntax errors in each con- dition. Subjects typed ten commands in each of eight syntax styles for a total of eighty commands per subject. Syntax styles were counterbalanced between subjects and individual commands were randomized within subjects. Data were analyzed by analysis of variance. TABLE 1. Syntax styles and results of speed and accuracy experi- ment. Syntax Syntax Example Typing Detected Undetected Name Style Command Rate* Errors+ Errors+ P3 :=, ENT:MA=1PN,OR=3 2.68 123 10 P2 :=_ ENT:MA=1PN OR=3 2.78 122 14 P2 :_, ENT:MA 1PN,OR 3 2.89 114 5 P2 _=, ENT MA=1PN,OR=3 2.98 102 8 P1 :__ ENT:MA 1PN OR 3 3.12 125 9 P1 _=_ ENT MA=1PN OR=3 3.43 101 5 P1 __, ENT MA 1PN,OR 3 3.48 73 8 P0 ___ ENT MA 1PN OR 3 3.87 64 1 * Average characters per second + Total errors, summed across subjects Results In summary, a) syntax styles resulted in different command entry rates (p <~0.001); b) P0 syntax was faster than any other syntax (p <~0.01) and P3 syntax was slower than all but the ":=_" syntax (p <~0.01); c) adding punctuation from 0 (P0) to 3 (P3) decreased entry rate (p <~0.001) and increased detected syntax errors (p <~0.001); and d) as detected errors increased, undetected errors also increased (r=0.67). EXPERIMENT_2:_READABILITY Thirteen subjects were asked to find mismatches between pairs of commands displayed in either P3 or P0 syntax. Commands were presented on a VDT with the correct command near the top of the screen and the incorrect command near the bottom. The mismatch occurred in one of two ways: Two characters of the correct com- mand were transposed or one character was changed (see Table 2). TABLE 2. Readability experiment commands and results. Name Style Example RT* Errors+ Correct ENT:MA=1PN P3 Transpose EN:TMA=1PN 4.04 18 Character ENT:MB=1PN 3.90 9 Correct ENT MA 1PN P0 Transpose EN TMA 1PN 3.02 11 Character ENT MB 1PN 3.42 5 * Average characters per second + Total errors, summed across subjects Trials proceeded as follows: Subjects scanned for the first mismatched character in the bottom command. When found, they pressed the carriage return key to stop a clock and erase the screen, then they typed in the mismatched character. Each subject received 40 randomized trials containing 20 pairs each of P3 and P0 commands. Results P3 commands took longer to scan for mismatches than P0 commands (p <~0.001). Of 43 total errors, 27 occurred in the P3 condi- tion, 16 in the P0 condition. A mismatch-type by syntax interac- tion revealed that subjects found character mismatches faster than transpose mismatches in the P3 condition and vice versa in the P0 condition (p <~0.002), because transpose mismatches tended to be more visually confusing than character mismatches. Con- sider, for example, P0 commands "ENT..." and "EN T..." versus P3 commands "ENT:..." and "EN:T...." CONCLUSION Minor differences in command syntax can apparently result in large differences in human performance. In particular, punctuated commands show lower entry speed, accuracy, and readability than semantically identical commands. Punctuated commands are prob- ably more difficult to read since punctuation tends to obscure word boundaries, which are an important redundant feature of written language (Fisher,1976). Would practice overcome the effects of punctuation? Clearly peo- ple can and do adapt to heavily punctuated command languages, but people can also learn to transmit Morse code at rates comparable to typing speeds. Surely no one would then argue that keyboards should be replaced with telegraph keys. People can adapt, but at what cost in job performance and satisfaction? To conclude, when command language typing efficiency is impor- tant, the space should be used as the primary command delimiter. REFERENCES Cooper, William E. (Editor), 1983. Cognitive aspects of skilled typewriting. Springer-Verlag, New York. Fisher, D. F., 1976. Spatial factors in reading and search: The case for space. In: R. A. Monty & J. W. Senders (Editors), Eye Movements and Psychological Processes. Erlbaum, Hillsdale, N.J. Ledgard, H. Whiteside, J.A., Singer, A. & Seymour, W., 1980. The natural language of interactive systems. Communications of the ACM, 23, 10, 556-563. Smith, S. L. & Aucella, A. F., 1983. Design guidelines for the user interface to computer-based information systems. MITRE Cor- poration Report MTR-8857.