Computers and the Humanities 17 (1983) 27-36 North-Holland

27

BOOK REVIEWS

All in Order: InJormation Systems Jor the Arts: A Report of the National Information Systems Project, Inchtding the National Standard Jor Arts InJormation Exchange, Mary Van Someren Cok in collaboration with Henry A. Bromelkamp, Ellen Thurston and Thomas Wolf, National Assembly of State Arts Agencies, Washington: 1981; 191pp. By Geraldine M. Otremba All in Order appeared early in 1982 and combines three somewhat differently focused tasks within a whimsically illustrated volume. Despite a multiplicity of arts-information gathering points throughout the United States today, the subject of this volume--the National Information Systems Project (NISP)--represents the first national effort to introduce standard definitions for reporting a variety of information about cultural activity in the United States. The project originated with a meeting of regional arts councils in 1977 that resulted in an initial inventory and analysis of automated management information systems currently in use within the states and concluded with a recommendation for a national project to standardize a variety of reporting data. tn 1978 the National Endowment for the Arts was increasingly concerned about the lack of reliable and comparable data about the activity of the state arts agencies network (funded in part through NEA), particularly the reporting of funded activities and the agencies' assessment of unmet needs. About to embark on the automation of its own grants management information system, the NEA agreed to support a three-year national inGeraldine Otremba's responsibilities at the Kennedy Center for the PerformingArts include information networks in the arts and liaison with such institutionsas the National Endowment for the Arts and the Library of Congress. Her academic background includesa doctorate in Englishliterature from the Universityof North Carolinaat Chapel Hill.

formation project under the auspices of the National Assembly of State Arts Agencies, with Mary Van Someren Cok and Henry Bromelkamp as project directors and systems consultants. By and large, the NISP has achieved the goals adopted in 1978, at a cost of just over $I million: --national compatibility in both the organization and labeling of informatiofl used by public arts agencies, through standardization of terms and definitions and standardization of methods of collecting, organizing, and disseminating information; --development of standard systems to be used by state arts agencies for their most basic information needs, i.e., mailing lists, grants management, and arts resource directories. To achieve these goals, NISP staff directly assisted state and regional arts agencies in implementing the NISP system (once it had been developed and approved by the NISP policy board) by making grants for consultants, staff training, software development, and hardware purchase. When compared to the advanced state of information management in the science fields, these goals may seem modest; from the inception of the project, they were not intended to produce an integrated automated data collection center but simply a set of compatible information-gathering and -reporting systems based on ~'a national system of working definitions." Compliance with the systems of definitions and reporting documents developed by NISP is considerably assisted by its partnership with NEA, which will require utilization of NISP definitions and systems for reporting on grant activity in FY t983. All state arts councils have agreed to adopt NISP, and the project director will continue as a consultant for the duration of FY t983 through the National Assembly of State Arts Agencies. All in Order 15erforms three functions: it recounts concisely the genesis and development of NISP;

28

Book Reviews

it contains the complete NISP-developed arts reporting data and key definitions; and it serves as a gentle primer to those staff members or readers reluctant to consider working under the threat of automation. In its whimsical approach to the widespread latent fear, the publication could serve as the basic text for an initial staff training session. Although NISP deliberately steered its course away from a central data collection center, the project results have encouraged a great deal of future compatibility. That more than half the regional and state arts agencies are reported to have utilized identical hardware and software makes possible data collection on a large scale. For the arts field, largely dominated over the past decade by the continual need to demonstrate the case for increased public funding, it is not surprising that the focus of the first major information systems project is arts funding--who receives money and who still needs funding. The recent pattern of large and steady increases for NEA has effectively been reversed by President Reagan's recent budget. As attention is increasingly directed to the question of quality of activity, some of the limits of NISP's goals may become more apparent. Of perhaps greatest interest for policy shapers, arts-producing and -presenting centers and artists themselves is the potential wealth of information that a compilation and analysis of the state-bystate arts resource directories could yield. If NEA stands by its commitment to enforce compliance with NISP reporting standards beginning this fiscal year and if ways are found for the collection and analysis of these data, the very visible field of the arts could at last have a mechanism for establishing a long-overdue information base. Information needs in the fields of the humanities may be indirectly assisted by NISP data. Many of the state arts councils also maintain responsibility for humanities funding within their states. The humanities fields are therefore included within the scope of NISP's data gathering. The extent and usefulness of these data remain to be explored and assessed for their utility. With NISP, the possibility for a national data network--the subject of a major meeting in New York in 1982 under the auspices of the Center for

Arts Information--is, for the first time, a topic for serious discussion and logical implementation. The Performing Arts Library, a joint information center on the arts maintained by the Kennedy Center and the Library of Congress, will be an active participant in the fall meeting and will work during the coming year with NASAA staff to see what national collection and access to NISP data might be possible.

A Concordance to the 'Utopia' of St. Thomas More, Ladislaus J. Bolchazy, editor, Georg Olms Verlag, Hildesheim and New York : 1978 ; 388 pp. By Warren W. Wooden This concordance to the original Latin text of Thomas More's Utopia, based on the standard critical text found in the R.S. Sytvester-J.H. Hexter edition of Utopia in the Yale Edition of the Complete Works of St. Thomas More (1965), boasts a complete and comprehensive listing of every word in More's book, with each word keyed to the Yale text by both page and line number. In photographically reproduced printout Bolchazy lists each word alphabetically in its grammatical rather than its lexical form. For this organization, the editor's one-and-a-half page introduction does not provide a rationale. Thus the user of the volume is left to speculate on why the work is arranged in a form so inconvenient for checking irregular Latin forms, which are found not under a central lemma but rather scattered throughout the concordance in their various forms. Bolchazy's claim of comprehensiveness seems imprecise, since his concordance seems to include, strictly speaking, the text of More's Utopia, but not the complete original 1516 edition, with its full apparatus consisting of title page, map, Utopian alphabet, Utopian verses, extensive marginalia, and two epistles. While some of this material probably resulted from a collaboration between More and Warren W. Wooden, a professorof Englishat Marshall University, is the author of a half-dozenarticleson More's Utopia in such journals as The Southern Humanities Review, College Literature, and The Sixteenth CenturyJournal.

Book Reviews

Erasmus, who saw the 1516 edition through the press at Louvain, it occurs in all the early editions of the book as an essential adjunct to the text, and the user of this concordance needs to be advised that Bolchazy has excluded it from his listing. An additional feature of Boichazy's volume is an index verborum of 56 pages, separately paginated under the rubric "A Frequency Word List," which catalogues every word in the Utopia in descending order of frequency. Thus we learn the three most frequently occurring words in Utopia are 'in' (544 times for a frequency percentage of 1.9851), '&' (415 times for 1.5144 percent), and 'non' (340 times for 1.2407 percent). Despite the author's assurance that such information supplies us with 'important stylometric data," it is difficult to see how lists of articles, conjunctions, prepositions, and similar parts of speech are likely to deepen our knowledge of More's masterpiece. Surely this volume would be leaner, less expensive, and more functional were the index verborum restricted to the major parts of speech. Bolchazy seems, however, to have performed his primary task well; a random spot check reveals no errors or omissions in his listing. Despite such minor editorial idiosyncrasies, this is a very useful research tool. Today, More's Utopia has the distinction of being the only work of the flood of Latin treatises, proposals, satires, and miscellaneous literary productions of the Renaissance humanists still read by a non-academic audience. And as its popularity among readers seems proof against the fads of taste, so the scholarly controversy over the 'real' meaning of More's enigmatic little book, nee minus salutaris quam festiuus ("no less beneficial than entertaining") according to its title page, continues lively. In the debate over More's meaning, scholars are turning back increasingly to the original text, to More's Latin, for evidence of nuance, irony, submerged meaning, and linguistic patterns and echoes of all kinds. Bolchazy's timely gatherum should both stimulate and facilitate this area of examination,

Hexameter Studies, Rfidiger Grotjahn, Quantitative Linguistics 11, Studienverlag Dr. N. Brockmeyer, Bochum: 1981.

29

By James J. Helm This collection of miscellaneous studies on classical hexameter (both Greek and Latin) appears as vol. 11 in a series on Quantitative Linguistics, published by G. Altmann's Sprachwissenschaftliches Institute of the Ruhr-Universit~it at Bochum, W. Germany. Altmann's colleague Rtidiger Grotjahn has edited the collection of six papers and added critical notes to three of them. The first article, by R. Schmiel, treats the relationship between ictus and accent in Greek hexameter. Starting with Jackson Knight's distinction in Latin hexarr/eter between homodyne and heterodyne lines (lines which show coincidence of metrical ictus and word accent, and those which do not), Schmiel refines it by defining as homodyne a line in which ictus coincides with 60~ of the accents. Conversely, lines with 40~ or less coincidence are categorized as heterodyne, while the remainder are considered neutral. Schmiel applies this distinction to 12 books of Homeric hexameters, which he divides into speech and narrative passages. Using the chi-squared measure of significance he concludes that speeches are significantly more homodyne than narrative. Then, noting that the information thus produced is not very precise, he abandons the categories of speech and narrative and divides lines solely on the basis of whether they are 'calm', 'vivid', or 'average'. Here he finds a correlation between homodyne lines and vivid or dramatic passages on the one hand, and heterodyne and calm on the other. Next he extends his study to compare Vergil and Homer, and adds statistics on Apollonius Rhodius, Quintus Smyrnaeus, and Nonnus, which seem to support the general thesis. His final general conclusion is that stress, quantity, and accent are all interrelated in Greek and affect the 'texture' of a given passage. For a number of reasons, the approach seems flawed. First Schmiel totally ignores the grave accent, although including it in the calculations practically eliminates any correlation between type James J. Helm, an associate professor of Classics at Oberlin College and a member of the Computer Advisory Committee of the American Philological Association, has recently published a commentary on Plato's Apology.

30

Book Rev#ws

of passage and dynamic quality of the lines. Furthermore, the terms 'vivid' and "calm' denote fairly subjective qualities (as is especially apparent in the discussion of Argonautica 3), and even Schmiel himself considers the procedure imprecise (p. 13). Ultimately, the conclusion does not seem to be adequately supported by the evidence. All of this analysis is further vitiated by methodological criticism leveled by Grotjahn, who claims in his appended note that Schmiel has been both inconsistent and insufficiently demanding in setting the probability levels for identifying for a general discussion of tests of significance, particulitrly the "Hypergeometric Probability Distribution" (HPD), which can be used for predicting expected values of an event and for analyzing the correlation of two exclusive properties (e.g., accent/no accent vs. ictus/no ictus). Although rather complicated, the process is, according to Grotjahn, more exact than the chi-squared procedure widely used in studies of significance. He goes on to discuss several approximations to the HPD which may be used when the number of discrete phenomena exceeds 100 (as when a number of lines are considered together), including the normal approximation (z-test), and the chi-squared approximation, including which re-emerges as the most useful. He also presents the phi coefficient as a measure of the strength of an association. All in all, this first note is a very helpful discussion of a number of potentially useful statistical tests and analyses. Unfortunately, the arrangement and interdependence of the first article and accompanying note make them difficult to follow. Schmiel revises the article as he goes along in order to take into account the first part of Grotjahn's criticism (but not the last part). Consequently the reader would be well advised to read pp. 33--45 first to get the theoretical model, then pp. 1-32 for an application of this model, and finally pp. 46-74 for a criticism of this application and an extension of the model. But all of this would have been easier had the full methodological criticism been given to Schmiel prior to publication, so that he could have revised his procedure accordingly and thereby given the resulting analysis some validity.

The second essay (in French) is P. Tordeur's study of metrical word-types in Ausonius, an attempt to analyze the style of Ausonius's poetry on the basis of the frequency of the different metrical word-types in each poem. After applying the chi-squared test, he concludes that there are significant differences between the poems, arid in particular that works which he characterizes as 'polished' show a significantly higher proportion of long words, while the more 'carelessly written' shorter poems have a greater number of short words. Furthermore, because the figures which include the Technopaegnion (in which each verse ends in a monosyllable) are markedly different from those which exclude that work, it is clear that descriptions of the average or overall practice of a poet may obscure important unique features of particular works. Although the conclusions here seem reasonable, Grotjahn once more adds a note which challenges virtually all of them--even the most obvious that the metrics of the Technopaegnion are strikingly different from the others. Groqahn maintains--without, to my mind, adequate explanat i o n - t h a t when testing the significance of individual cells of a table presenting a family of simultaneous tests, a higher level of significance is required than when the data are unrelated. In addition, regardless of significance, sample size increases the magnitude of chi-squared or z values. As a result there is less significance to Tordeur's findings than he thought, although the Technopaegnion remains strikingly different from the rest of the poems. Then Grotjahn continues by introducing the Cramer coefficient test for the strength of relationship between two variables to show that the metrical word-types are not very strongly related to the individual poems in the works under consideration, not even in the Technopaegnion. Thus the value of Tordeur's study is almost completely undercut, and we are left knowing very little new about the phenomena under discussion. Given this damaging criticism, it is not clear why the authors of the first two articles were not allowed further time to revise their work before publication. There is, no doubt, some utility in observing the process of proposing a method for testing a hypothesis and then revising it after

Book Revwws

criticism, but the blind alleys could have been described briefly in the articles prior to drawing conclusions from methodologies which meet accepted criteria of validity. The next paper fares a bit better. Here D. Clayman seeks to verify Parry's contention that early Greek hexameter poetry consisted of shorter sentences and of sentences which coincide more often with line end than those in later Greek hexameter. She measures sentence length not only by number of words, but also by number of syllables and of phonemes. With a few exceptions, later works do have sentences which are longer by any of the above measures. Also, when the numbers are grouped and displayed by means of bar graphs, there is a single peak for length measured by words (sentences of 6-10 words), but four separate peaks when measured by syllables (11-20, 31-40, 61-70, and 91-100). Clayman attributes these peaks to the number of syllables per line being fixed between 12 and 17 (the number of phonemes having similar limits), so that the peaks represent the tendency to finish sentences at the end of hexameter lines. This discussion of peaks is a bit peculiar, however, since the presence or absence and the position of peaks are to some degree dependent on the ways in which the figures have been grouped for the bar graph. For the graph of sentence length by words, Clayman uses groups of 5 words (1-5, 6-10, etc.), whereas for length by syllables or phonemes, the groups are by tens (1-I0, 11-20, etc.). It might have been clearer to use a line graph representing the distribution of lines of each length rather than in groups, so that the placement and number of peaks could be determined by actual practice, rather than by the size of the groupings. In fact, fewer bars representing larger groupings will show fewer peaks. For example, grouping syllables by twenties yields a bar graph with only three peaks and no obvious explanation. Part of the problem here may be the additive nature of the phenomena under consideration: if the number of syllables in a line must be from 12 to 17, then the number of syllables in two lines is 24-34, in three lines 36-51, and in four lines 48-68. But here the groupings have begun to overlap; does an average of 50 syllables per sentence show a preference for three-

3I

line or for four-line sentences. In any case, with regard to Parry's hypothesis on sentence length and the tendency to conclude sentences at line end, this article makes a small advance: fuller statistics, but almost the same conclusions. However, in a similar test on the homogeneity of sentence length within a single author, the evidence is inconclusive. Nevertheless, Clayman attempts to provide some generalizations about sentence length and content, concluding (without much supporting evidence) that, for example in the Argonautica, "the difference in sentence length distributions can be linked to a change in subject matter within the poem and that passion demands greater contrasts in sentence lengths than heroic adventure." Again, a subjective estimation of this sort tends to fall flat. Altmann's article on the homogeneity of metric patterns in hexameter proposes a test of that quality between two populations which may be used as a convenient alternative to chi-squared. This is the "minimum discrimination information statistics", which approximates the chi-squared distribution, but is easier to calculate. He first uses this method to show that the various possible metrical patterns in hexameter verse are not distributed homogeneously (i.e., do not all have the same frequency in a given poem). He then goes on to show how it is also useful for testing the homogeneity of two authors (or texts), of several authors, or of groups of authors. We find, for example, that Horace's Satires and Epistles are homogeneous in metrical practice by means of this test, whereas a test of samples of Vergil, Horace, Lucretius, Manilius, Persius, Juvenal, and Lucan shows that they are not a homogeneous population, and therefore do not constitute (as has been claimed) a "Vergilian" group. The test is a useful and convenient one, and since the calculations are simpler to carry out than those for the chi-squared test, it is a welcome addition to the humanist's repertoire. In a more problematic move, Altmann also uses the statistic to show that samples from Latin and German authors are not homogeneous. This comparison is dubious, since the basis for metrical scansion is quite different in the two languages. Nonetheless it is comforting to find that by this

32

Book Reviews

measure, the authors lack homogeneity. Unfortunately, several numerical errors in the data tables (pp. 141 and 148; see below) undermine our confidence in the correctness of the calculations. N. Greenberg starts with an empirical observation: words which begin one of the first four feet in the hexameter line have a higher proportion of initial vowels than words elsewhere in the line. The study focuses on Aeneid 12, but the observation seems to apply generally to Latin hexameter, particularly with regard to the beginning of foot 2, where a disproportionate number of initial vowels appear. Since they occur in these places whether there is elision at that point or not, their occurrence is apparently not due to elision per se. While a large number of elisions occur at these points, they do so more frequently when the preceding foot is a dactyl and no elision occurs: about three fourths of the words following a dactyl begin with a vowel. A census of metrical word-types which end with feet 1-5 shows that dactyllic words occur in large numbers only in feet one and five, where they constitute over half of the words which end here. Because bisyllabic words in the final foot more often begin with a consonant, correct scansion requires that they be preceded by words ending in vowels. Hence dactyls ending in consonants, displaced from the fifth to the first foot, thereby require words at the beginning of the second foot to have vocalic initials. This, then, is the explanation Greenberg gives for the curious statistic with which he began, but he concludes with the rather enigmatic observation that despite this kind of constraint, Vergil's creativity remained free. Unlike the other studies in this volume, this one involves primarily descriptive statistics. Greenberg adds a postscript suggesting ways in which inferential statistics might be brought to bear on the material. For example, a chi-squared test for the independence of word-ending and position shows that it is extremely unlikely to be the result of random distribution. The final study (in French) likewise involves juncture in hexameter verse, this time when it occurs between the short syllables of a dactyl. J. Gerard challenges Soubiran's thesis that the Latin poets prefer words ending with consonants

at this position when a pause occurs. The main interest in this issue of juncture with pause between short syllables arises from an interest in liaison between clauses. Since in hexameter verse two short syllables require that only one of the two words have a vowel at this juncture (more than one produces elision, and no vowels would make for length by position), it is an interesting place to study preference for consonant/vowel juncture (CV) between clauses, as opposed to vowel/ consonant (VC) juncture. Gerard refines Soubiran's definition of pause by distinguishing between weak and strong pauses, and by separating function words from content words. He then shows that the type of juncture depends rather on the constraints of vocabulary (and meter), since function words most often end in consonants and must be followed by words beginning with vowels, thus producing CV, whereas content words more often yield VC. Aside from this constraint, the poets seem indifferent to the choice of CV vs. VC, and their practice is equivalent to prose usage. No test of significance is made, and so Grotjahn provides several in the note which follows, applying to Gerard's data the inferential statistics set forth in his first note. The problem with this whole approach lies in the assumption that Latin liaison (the connecting by pronunciation of one word with the following word) follows the practice of French in making that connection a close and phonetic one. But French treats the ends of words quite differently from, say, English, and there is no reason to assume that Latin followed the French pattern. Thus the basic interpretation is moot. The last contribution to this volume is a handy annotated bibliography on the statistical study of hexameter verse since 1866, compiled by U. Job. The introductory remarks indicate that the list has been restricted fairly severely by the requirement that each entry include empirical data of a particular sort (e.g., sample size and absolute frequencies or percentages given precisely), and there is a running assumption that hypotheses may have claim to validity only if they are supported by quantifiable evidence. Despite this dubious restriction, there are 34 pages of titles on the topic, a testimony to its popularity and the

Book Rev#ws

amenability of metrics to quantitative methods. Finally, the book is not easy reading for the humanist. It is loaded with tables and charts and sometimes formidable mathematical formulas. Familiarity with mathematical notation is assumed throughout, and many complex statistical concepts are used without explanation. Yet the concepts with their attendant notation are essential to the intent of the volume, and the tables are necessary to permit the reader to evaluate the interpretations given. The formulas, while sometimes complicated are not really advanced: they are for the most part within the reach of one who has studied simple algebra, although they could have been made a bit more comprehensible to the novice. What seems to be most clear from the volume is that students of literature need to be more conversant with statistics. In particular, the first two articles demonstrate how vulnerable classical peltasts are to the hoplites of the mathematician. Even minimal familiarity with statistical methods would not only open to humanists the potential of quantitative studies, but also make us aware of their limitations. We cannot ignore the power of such an important tool, and we ought not be overwhelmed by it. There are features of the humanities which c a n - - a n d should--be counted; they may not always be the most central, since humanistic concerns are more often qualitative than quantitative. Nevertheless, numerical evidence is often relevant, and in an area like metrical practice it can even be crucial. In other areas, too, where certainty cannot be attained, likelihood may be tested by rules of statistical probability. We also need to be able to distinguish good evidence from bad. In general, humanists should be ready and able to read and evaluate such mathematical applications to their texts. One might have hoped that this set of papers would help make this process easier and provide clearer examples of the utility of quantitative methods, but it is insufficiently systematic and clear, and the applications do not all represent helpful paradigms. The book has more than its share of typographical errors, some of which are serious. Here are the more important ones which caught my eye: p. 83, first table: for '+7.34' read ' - 7 . 3 4 ' (cf. bottom ofp. 99, where ' - 7.3' r e p r e s e n t s ' - 7.34');

33

p. 83, second table: all signs seem to be reversed (for ' - ' read ' + ' , and vice versa; compare the first table on the page); p. 100, lines 2, 6, and 11 below the table: for '74' read '84' (although .0006 is correct); p. 141, Table 4: for '33' (RF:6) read '38', and for '42' (HD :7) read '142' ; p. 148, Table 10: for '131' (Voss:16) read '161'; p. 170, line 1: for 1968, read 1966b(?); p. 220, line 2 below the table: for "Metamorphoses' read 'Lucan'. Typographical errors in words are often obvious, but wrong numbers are extremely diffficult to catch without making the original observations or doing the calculations over again. Thus it is even more crucial in statistical studies to undertake the laborious task of checking the figures once more before publishing. In summary, then, despite a few positive contributions, this is a rather disappointing collection. Very little specific advance is made in our understanding of classical hexameter verse, although there are some interesting figures and potentially useful procedures described or utilized. Some of Grotjahn's methodological criticism is informative and acts as a caution to tfiose venturesome humanists who combine statistical methods with the study of literature. The specialist in hexameter verse or metrics may want to acquire this book, but the amateur would better look for a volume which is more accessible to the uninitiated.

LISP, Patrick Henry Winston and Berthold Klaus Paul Horn, Addison Wesley, Reading, Mass.: 198t. By Denis L. Baggi Among the several high-level languages for computer programming, LIsP--an acronym for LISt Processing language--occupies a unique Denis L. Baggi's doctoral dissertation was about the realization of the unfigured bass by digital computer, i.e., a system of programs in LISP which constructs the upper three voices o f a piece of which the bass is given. His current position is with Mettler Instrumente AG in Greifensee, Switzerland, where he is working, inter alia, on speaking machines.

34

Book Rev~ws

position. The language was defined originally as a mathematical formalism to express recursive functions of symbolic expressions [4] and implemented in the early sixties as a fullblown computer language. Since then it has become the ideal medium to express algorithms for projects in artificial intelligence and symbolic manipulation--as opposed to numeric computations--as well as the first choice of artificial intelligence researchers, who view usP as the king of languages. Because of its high potential appeal to people interested in computer techniques applied to the humanities, some of the peculiar features of LISP will be briefly mentioned before the review of this textbook. The several reasons for the success of LISP derive from its concise mathematical definition, based on merely five elementary functions, and from the elegance of the programming style for problem solving, which is claimed to be closer to the human way of thinking than the style of any other programming language. For instance, in any algorithmic language, from FORTRAN t o PASCAL, the work of the program is almost invariably carried out by assignment statements, which are just high-level implementations of LOADand STORE machine instructions; that is, statements are executed for their effect. On the other hand, in LISP almost every statement is executed for the value it returns, as in the case of a mathematical function. Consequently, a LISPprogram consists of functions, generally recursive, which typically call each other recursively and construct the result, which is eventually returned to the top level. No loops, assignment statements nor variables (except parameters) need to be used. Thus, programming in LISP is equivalent to the design of a language of functions to solve the specific task, while, in a sense, the program itself generally consists of one single statement, the top-level call activating the system. Another remarkable feature is that, though LISP is considered a very high-level language, it allows some of the flexibility generally reserved to machine language. Since both programming code and data have the same format, that of an Sexpression (short for symbolic expression), a program can easily construct, modify and even execute other programs, including the LISP interpreter itself. In addition, S-expressions are a very power-

ful generalization of list structures, including trees, graphs, and the like, and since they are dynamically allocated at execution time, they are excellently suited to searches into structures whose depth is not known in advance, as in the case of the tree of possible moves in a game. The program can 'grow itself' at will. In order to look like a complete computer language, and to allow FORTRAN-like style for algorithms of difficult implementation by recursive functions, a LISP system typically implements programming features such as assignment statements, free (read 'global') variables, loops and arrays. The LISPenthusiast should, however, de-emphasize these features, which at any rate are too prominent in the thinking of programmers educated into FORn~AN, assembly languages, and the like, and which therefore would lead one into writing improper programming style rather than good, LISPlike ways of coding. In spite of its brilliance, LISP has a reputation of being difficult to master, probably because of the conciseness of the first LISP manuals [5]. Its syntax, heavily dependent on the exact placement and matching of parentheses, is not very fortunate. For. more than a decade, however, excellent books have existed which dispel the myth that acquiring fluency in this language is difficult [3, 7, 8]. The book LISP by Winston and Horn, the object of this review, is the latest of the series. It is addressed to students with no previous background who would like to both study the language and learn about those applications in artificial intelligence for which LISP is famous. It is divided in two parts: Chapters 1 to 12 attempt to cover all features of the language, while the second part, Chapters 13 to 23, describe instances of its practical use. As they state in the preface of this new effort, the book is an extension of parts of the previous textbook in artificial intelligence by one of the authors [101. The main problem with the text, as with its predecessor, is that, in spite o f its title, it does not deal with the standard version of LISP as it has been known for years to the scientific community, but with some usP dialects, especially MACLISP, which, the authors seem to imply, have by general consensus replaced standard LISP; however, this is

Book Rev~ws not so. They probably believe--not without some arrogance--that since the leading institutions in artificial intelligence research, such as M.I.T., where they work, make large use of their LISP dialect, the rest of the world should follow their steps. However, artificial intelligence analogies notwithstanding, it is not always true that an artificial language for coding algorithms must 'evolve' like a natural (spoken) language and that it is in need of revision every so often; on the contrary, quite a few programmers, including this writer, believe that LISP was defined once and for all, is in no need of improvement and is elegant at its highest when used as originally described. The question is not about minor syntactical details per se, which could be safely ignored; they can be no longer ignored when the programming style deriving from this redefinition of LISP is significantly altered. In order to dispel the myth of the difficulty of LISP (a laudable enterprise), the book describes the 'friendliness' of M.I.T.'s interactive system with a few examples in Chapter 2. While doing so, the authors introduce at this stage the use of a free variable (see examples with SET, p. 14). Now, one of LISP'S beauties is the lack of reliance upon side-effect programming. In other words, whoever designs, reads or debugs code is relieved of keeping track of endless values, stored as variables, since he is interested in only the value of the statement; the examples on p. 14 encourage, unfortunately from the very beginning, the use of values hidden within the system itself. Free variables can, of course, occasionally be needed in standard LISP, but there, at least, they are flagged by the use of CSET rather than SET, whereas in MACLISPthe nonexistent distinction renders the example INC R E M E N T on top ofp. 46 difficult to read without comments. The behind-one's-back changing of system variables accounts for the poor readability of quite a few examples in the book, such as MATCH on p. 230, which is difficult to follow, although the function is explained in more than ten pages, because MATCH, though it returns 'true' or 'false', also sets some variables while it is being recursively used in the predicate part of a condition. Also because of system variables, the ex-

35

planation of EVAL, top of p. 31, becomes rather obscure compared to examples in standard LISP with LAMBDA expressions. And on p. 48 it is not clear at first sight which is the variable and which the function. One can only guess that the liberal use of such programming techniques comes from the experience, by at least one of the authors, with large data bases for image processing [9]. However, in a book about programming the enthusiasm for globally stored values should be more restrained and more discipline should have led to an accepted style of LISP programming. The authors have eliminated any reference to EVALQUOTE, described as 'obsolete' (p. 12), the understanding of which is indeed a stumbling block for beginners. But it should not have been necessary to completely bypass the concept of dotted pairs. In spite of its clumsiness, standard LISP syntax represents exactly the way S-expressions are stored in the virtual LISPcomputer, and such knowledge is a helpful LISP tool. It could be argued that, syntactically, most programming examples in this book appear strange. Since D E F U N (p. 33), as opposed to LISP'S DEFINE, is not based on L A M B D A expressions (i.e., functions returning one value), one can write obscure function code in regard to the value of the expression (see F-TO-C, bottom of p. 36). This, too, accounts for the difficulty of translating these examples in LISP, including the above-mentioned MATCH, while in standard LISP,functions PROG2 (or PROGN) leave no doubt about the value they return. Thus, contrary to what the authors state on p. 43, the syntax of C O N D is not peculiar, but consistent with standard LISP syntax, though not with their illustrations. When finally LAMBDA is introduced much later (Chapter 6), it is in the context of examples which depend on M A P C A R and are hard to follow. The conventional way of defining forms, and complex forms, recursively, would have made the example on top of p. 83 clearer; as it stands, it is rather difficult for a beginner, in spite of the accompanying explanations. One Of LISP'Smain stylistical features is recursion, described in Chapter 4, p. 51. Since even programmers are only casually acquainted with this technique, recursion should be explained more

36

Book Reviews

formally, i.e. with emphasis on terminal condition and recursive step--which are often the source of errors, i.e., endless recursion. It is hard to ascertain what conclusions about the extreme importance of recursion in LISP would be drawn by a reader with no previous exposure to LISP; in any case, it is not hard to see that he would gain by being forced to think recursively in standard LISP. Hence one wonders whether recursion is not just a neat trick, devoid of fundamental importance, to maintain some semblance of compatibility between MACLISP and LISP, and whether, in such a case, examples in the book notwithstanding, this technique may not be conveniently put aside. The whole idea about LISP is to describe recursive functions of symbolic expressions. There are plenty of examples in this book with a programming style quite unlike what is normally regarded as good LISP. Let us briefly mention the 'strange' definitions of GET, P U T P R O P and DEFPROP, pp. 71-72, the introduction of arrays on p. 75 (a bad idea, except in rather exceptional cases such as table look-ups or hash coding), the poor explanation of ASSOC, p. 73, the description of I M P L O D E and EXPLODE, pp. 90-91, the fact that P R I N T returns T rather than its argument, p. 88 (and thus cannot be used for debugging as in standard LISP), the obscure discussion on top of p. 99, the need for DO loops, hardto-read examples on binary images, such as C E N T E R on p. 123, and many more, the listing of which would add nothing of interest to the reader of this review. It should be clear by now, therefore, that the main problem with this text is with its title, rather than with the material it contains, and this is too bad considering the beauty of the LISP language. The book contains several examples in MACLISPwhich

teach this dialect in detail and which appear correct and well designed; its organization in chapters of progressing difficulty is most welcome and the material of the second part very interesting. All this, unfortunately, has little to do with LISP. The authors are entitled to their own opinion as to what LISP is and which programming style should be encouraged. They should not, however, use book titles in a misleading fashion and they should realize that there are quite a few schools and research centers which still enjoy using standard LISP, and for which, therefore, the lack of generality, and the stress u p o n a particular version of LISP render their text useless. References [1] Allen, John. Anatomy of LISP. McGraw-Hill, New York, 1978. [2] Berkeley, E.C., and Daniel G. Bobrow (editors). The Programming Language LISP: Its Operation and Application. M.I.T. Press, Cambridge, MA, 1966. [3] Friedman, D.P. The little LISPer. Science Research Associates, Palo Alto, CA, 1974. [4] McCarthy, John. Recursive Functions of Symbolic Expressions and their Computation by Machine. Communications of the ACM 3 (No. 4) (1960), pp. 185-195. [5] McCarthy, John, et al. LISP 1.5 Programmer's Manual. M.I.T. Press, Cambridge, MA, 1962. [6] Steele, G.L., and Gerald J. Sussman. Design of a LISPBased Microprocessor. Communications of the ACM 23 (1 I) (1980), pp. 628-645. [7] Syklossy, Laurent. Let's Talk LISP. Prentice-Hall, Englewood Cliffs, NJ, 1976. [8] Weissman, Clark. A LISP 1.5 Primer. Dickenson, Belmont, CA, 1967. [9] Winston, Patrick Henry. The Psychology of Computer Vision.McGraw-Hill, New York, 1975. [10] Winston, Patrick Henry. Artificial Intelligence. AddisonWesley, Reading, MA, 1977.