Reviews
The Song of Roland: Formulaic Style and Poetic Craft, Joseph J. Duggan. Center for Medieval and Re-
the poem the unity it so obviously possesses. To shed more light on this question, Duggan proposes that we use Milman Parry and Albert Lord's conjecture that oral poetry, composed under pressure of public performance, has a greater number of set expressions - clich6s or "formulas" - than work produced by a literate person writing and correcting his work as he proceeds. If some measurement of formulaic density can be found to distinguish between these two modes of composition in Old French narrative poetry, we shall be able to establish to which mode the Roland belongs. With that goal in mind, Duggan defines as a detectable "formula" any hemistich "found two or more times in substantially the same form" (p. 10) within each work studied. In other words, the content words (nouns, verbs, attributive adjectives and adverbs) have to be the same although other parts of speech can vary. Moreover, if one of the content words comes at the end of a line - and is thus subject to rules of assonance - it too can vary and still count as a formula, inverted word order and paradigmatic changes are allowed but not, one gathers, changes in tense which require or delete auxiliary verbs already deemed content words. Formulas which cross the caesura are also, in theory, rejected. Although a computer might be programmed to detect formulas by that definition in a lemmatized text, the task of lemmatizing alone would take as much time as the method Duggan uses. Moreover, the programming of a computer to count syllables in Old French to find the caesura is not a job one would begin lightly. Duggan wisely prefers to work with original texts, and uses a computer only for assembling KWIC concordances. The number of formula tokens, i.e., hemistichs repeated elsewhere, compared to the total number of hemistichs, then gives an indication of the "formulaic density" of any given text. Duggan examined thirteen Old French narrative poems: three known to have been composed in writing, nine of which we are reasonably sure were composed orally, and the Roland. He then measured the
naissance Studies Publication No. 6. Berkeley, Los Angeles, London: The University of California Press, 1973. 226 pp. $8.50.
By John R. Allen This is an important work, the merits of which will long outlast any of the controversy it generates. Professor Duggan has used statistics in an attempt to answer a question raised more than a century ago, namely, whether the Chanson de Roland was an oral or literate creation. In so doing, he has generated much valuable information for specialists in medieval French literature. We are greatly indebted to him for that accomplishment, although we feel we must question his methodology. The major problem Duggan tries to solve concerns the origin of the Chanson de Roland and medieval romance epic in general. During the nineteenth century, most critics believed that the Roland was created by illiterate jongleurs who preserved the memory of the battle of 778 in songs gradually developed through the centuries until, toward the end of the eleventh century, one version was transcribed into the surviving Oxford manuscript. In the first half of the twentieth century, an alternate theory arose as scholars found it hard to reconcile the obvious unity of the Roland itself with the idea of multiple authorship. This new theory suggested that, toward the end of the eleventh century, a single genius composed, in writing, the archetype of the Oxford manuscript, using as his source the written records of the battle. Yet another theory sought to reconcile those extremes of traditionalism and individualism by seeing the Roland as the product of oral tradition reworded by a literate individual, a remanieur de gOnie, to give John R. Allen, an associate professor o f French at the University o f Manitoba, is an editor o f System. 36
REVIEWS
3"7
three written works to have a "formulaic density" of Duggan's definition leads, of course, to the discovbetween fifteen and eighteen percent,' while the nine ery of many "formulas" which might otherwise have oral works measured from twenty-three to thirty-nine been ignored in such a study. The decasyllabic lines percent "formulaic." That suggested to Duggan that of the Roland break into hemistichs of four and six his measurement could be capable of distinguishing syllables, so such simple expressions as li amiralz (the between the two methods of composition. He then emir), dist Baligant (said Baligant), li emperere (the measured the Roland by the same method and found emperor), en Rencesvals (in Roncesvaux), sun cumit to be 35.2 percent formulaic, thus apparently de- paignun (his companion), need appear only twice (in stroying both the individualist and remanieur de 4,002 lines), in any of the variant forms allowed, to gOnie theories. Moreover, those scenes which critics be counted as "formulas." When, by accident, a first have considered the greatest moments in the epic, the hemistich has six syllables, as in line 1763, Bel sire, scenes which could be attributed to a literate re- cher cumpainz, Duggan declares the line formulaic by manieur de gOnie of the Roland if he existed, are ignoring the content words which were presumably shown by Duggan to be the most "formulaic" of the added. However, while such methods may appear arpoem. Those sections are great then, not in spite of bitrary, when they are compared with previous scholthe formulas, but because it is here that the poet arship on formulas done without the help of computmade the most effective use of "the basic matter of ers, Duggan's methods seem a model of precision. He his creation" (p. 61). A strong argument of the indi- has reduced to a minimum the subjective element in vidualist critics, that oral creation is not compatible counting formulas, no matter how much other critics with the unity and beauty of the poem, can no longer may feel such criteria produce erroneous counts of be considered valid, and evidence from other litera- formulas. ture, not referred to by Duggan, also shows that oral More serious objection should be raised concerning poetry is not always hastily improvised under pres- some of Duggan's statistical assumptions. He correctsure of performance. The Scots Gaelic poet Donn- ly points out that in his measurements 'Where are chadh Bgm Macantsaoir (Duncan Macintyre), totally three absolute figures involved . . . : the length of the illiterate, composed his longest single poem Miladh sample, the number of formulas in the sample, and Beinn Dobhrain (Praise o f Ben Dorian) over a period the number of lines in the referent which was searchof fourteen years before dictating it for transcription, ed to find evidence for these formulas" (pp. 18-19). and Douglas Young describes it as "beautifully melo- Moreover, " I f samples are to be dependable, then the dious Gaelic, in a complicated stanza system, with number of lines on which the evidence for formulas is elaborate assonantal schemes. ''~ Macintyre's example based must be identical in each case. Otherwise inferjustifies Duggan's finding that oral composition is not ences drawn from the relative densities of the samples incompatible with high poetic quality. have no meaning" (p. 19) Unfortunately, when he One might want to question, however, Duggan's then compares total formulaic density of different definition of a detectable formula. It is rather general, poems, he uses different length referents, since with though not so vague as Parry's "a group of words each work studied, he makes the "referent" equal to which is regularly employed under the same material "the size of his population" (p. 20), usually the entire conditions to express a given essential idea. ''3 At first poem under consideration. When he compares relative Duggan describes formulas simply as "semantically formulaic density between different parts of the stylized hemistichs" (p. 7, n. 17), but this for his Roland he is on safe ground, since his referent for study, as indicated above, counts only content words each measurement is the fixed number of lines of the while allowing changes in order, paradigm, spelling, entire poem. But when he compares eleven different and content words at the end of a line. He nowhere poems and uses ten different length referents for states that, despite the paradigmatic freedom allowed, gathering evidence (in one case he compares two a change from a simple past to a compound past is equal-length extracts), his figures cannot escape disunacceptable, but one gathers such when he does not tortion. Duggan makes no attempt to adjust for this equate (p. 55) line 2322a (Jo l'en eonquis) with line error, for he feels there is no problem at all in making 2333a (Cunquis Fen ai) both translated as "I con- such comparisons as long as "the size of the populaquered them for him." His decision not to count for- tion and the size of the referent are identical, both mulas which cross the caesura also requires some fur- being equal to the entire poem" (pp. 20-21). He ther explanation since he equates line 32b .iii.c. muls seems aware that increasing the length of his referent eargez ("four hundred mules laden") with line will increase the number of formulas detected, since 18 5 a/18 5b: Quatre cenz muls / cargez del or d 'Arab e. he states " I f all the hemistichs of all twelve poems
38
REVIEWS
were to be compared to each other (66,750 possible formulas), more formulas would be detected" (p. 23). By then making his referent vary from epic to epic, he makes his inferences highly dubious. Indeed, some of the measurements he makes suggest that his method will never be able to detect a difference between oral and literate creation except by chance, since when his measurements are adjusted to compensate for length of referent, the difference between oral and literate works disappears. A rough idea of the amount of correction that is involved can be deduced by examining the evidence he furnishes (pp. 4 2 - 3 , n. 40) for the formulaic content of the "French Council Scene" (vv. 168-341). When measured against the whole of the Roland as a referent, the scene is 40.8 percent formulaic. When only the 804 hemistichs of the six "best scenes" are used as a referent, the formulaic density of the council scene becomes 20.7 percent, even though those six scenes are shown by Duggan to be the most formulaic of the poemr In other words, the reduction of the size of the referent by about ninety percent decreases the formulaic density measurement by about fifty percent. If Duggan were to reduce the size of his Moniage Guillaume referent by ninety percent, the resultant halving of the formulaic density would, however, remove that work from the "oral" group and make it more "literate" than any of the three "literate" songs he examined. Even if the size of the referent had been held constant, one would also expect that the length of the poetic lines being compared could similarly distort measurements, since dodecasyllabic lines have no four-syllable hemistichs which, as Duggan's evidence shows, have a higher probability of repetition. 5 It is not surprising, then, to see that of the three literate works he measures, the single dodecasyllabic one, Buevon de Conmarchis, measures the lowest "fonnulaic density" even though it had the longest referent of the literary works. Of the nine oral works, the two dodecasyllabic ones, the Pblerinage de Charlemagne and the Si~ge de Barbastre, also measure the lowest "formulaic density." The extent of this distortion is unclear, for one of the octosytlabic works, the Gormont et Isembart fragment, shows a lower formulaic content than six of its decasyllabic counterparts, although one would expect the contrary. The explanation probably lies in the fact that the Gormont fragment is the shortest of all of the works Duggan measures (659 lines). It shows the "formulaic density" of an oral work only because it is merely a series of repetitive single combats. Duggan also examines the authenticity of the
Baligant episode in the Roland. If, in the process of oral composition, that section had been added to a version that did not originally contain it, a different authorship might be detected in the formulas. No evidence can be found from formulaic density, which in the Baligant episode is 34.5 percent, compared to 35.2 percent for the whole of the Roland and 35.5 percent for only the non-Baligant parts of the poem. Duggan suggests, then, that if formula tokens show no difference here, perhaps formula types will. Of the 907 formula types he detects in the Roland, 526 appear exclusively in the non-Baligant sections; eightyone appear exclusively in the Baligant; and 300 are in both. Duggan concludes "this overall view of the abstract formula types turns up fewer formulas exclusively in the 'Baligant' than one might expect, for the episode constitutes just a shade under one quarter of the whole poem but retains exclusive possession of only slightly more than one-tenth of the types" (p. 84). Even apart from the distortions caused by comparing two different lengths of text, one three times larger than the other, the analogy is invalid. The eight-one exclusive types in the Baligant should be compared, not to the total number of formulas, but to the number of formulas exclusive to sections of the same length taken from the rest of the poem. The point would not be important were it not for the fact that Duggan bases his authorship test on the "lack of distinct formulaic repertory" in the Baligant. Although there are eighty-one formulas used exclusively in that section, he finds that "only" five are not related to the context, yet they express, in different words but with the same assonance, ideas found in formulas from the non-Baligant sections. From that, he concludes that "these alternative formulas show, by their paucity, the relative uniformity of formulas throughout the poem, albeit in a negative way" (p. 88). Given the restrictions on the choices - that they not be related to events found only in the Baligant section, that they be found at least twice exclusively in the 999 lines of that episode, that each express in different words a concept found in a formula which appears at least twice exclusively in the 3003 lines of the rest of the epic, that in cases of the second hemistich the assonating vowel be the same as that found in the equivalent formula of the other section - then the five formulas Duggan succeeds in finding would seem to argue if anything the exact opposite of his conclusion. Duggan attempts to reinforce his thesis by examining all the formulas used at least four times in the Roland. Eliminating those which are related to context, he finds only one exclusive to the Baligant and
REVIEWS
thirty-four exclusive to the rest of the epic, which, for him corroborates "the hypothesis that 'Baligant' does not possess a formulaic repertory of its own" (p. 97). Again, his figures are distorted by the different sizes of what he is comparing. Of course, there will be a smaller number of formulas repeated four times within 1000 lines than are repeated four times within 3000 lines, even without taking into account the requirement that they be exclusive to those sections. But the basic problem is that his method has no provision for calibrating the statistics he derives through prior tests on works of known origin. The result is that, without further information, the figures he cites are virtually meaningless. While his statistical analysis leads him to conclude that the Baligant section and the rest of the Roland "possess a uniformity of style" (p. 101), what he has really shown is that his method is not capable of distinguishing stylistic differences between the two parts. His subsequent efforts to reconcile that "uniformity of style" with a belief that the two sections still had different authorship are not really necessary. Statistical analysis in authorship problems such as these can never prove "a uniformity of style" without defining its features in sufficient detail to allow it to be distinguished mathematically from all other similar works. There simply is not enough surviving eleventhcentury narrative poetry in Old French to justify such a conclusion. In these circumstances, all that any statistical analysis can do is either detect a difference in style or fail to detect it and thus not prove (or disprove) anything. Duggan's conclusion on the matter, then, although he does not feel it supported by his formulaic evidence, is plausible: the Roland probably developed over a long period of time; an early eleventh-century accretion was the Baligant episode; by the end of the eleventh century the formulaic differences between that part and the rest of the epic were concealed as other poets reworked the legend. Perhaps a more sophisticated stylistic analysis would reveal differences imperceptible to Duggan's method. The rest of his book, two chapters plus conclusion, while of great value to students of medieval French literature, has little to do with computational analysis other than that Duggan here uses the lists of formulas his earlier work accumulated. One of these chapters shows how the Roland poet used different individual formulas to fit metrical requirements dictated by his text; the other chapter analyzes the way groups of formulas were combined to form different motifs repeated in the Roland and other epics. While one cannot always agree with Duggan's conclusions (e.g., the Roland "confirms the relevance of" the Danish folk-
39
lorist Axel Olrik's "epic law" that the maximum number of persons in a conversation is three,6 the views presented are stimulating and make Duggan's study a valuable contribution to medieval French studies. Indeed, his analysis of the literary aspects of formulaic style - unfortunately slighted in this review, which is aimed at a more general audience - is unexcelled. Despite flaws in his statistical reasoning, Professor Duggan has raised the quality of scholarship in this area and is to be highly commended. 7
NOTES 1.Duggan claims t he y are fifteen, sixteen, and seventeen percent, but if his figures on the n u m b e r of repeated hemistichs in the R o m a n d'En~as are correct (953 out o f 5390), t ha t w o r k is 17.68 percent rather t ha n 16 percent formulaic. 2. "Never Blotted a Line," Arion, 6 (A ut umn, 1967), 184. I am i nde bt e d to Professor D. MacMillan for bringing this article to m y attention. 3."Studies in the Epic Technique of Oral Verse-Making," Harvard Studies in Classical Philology, 61 (1930), 30. 4.Duggan states that for works of less t ha n 400 lines his m e t h o d " w o u l d be effective in inverse p r o p o r t i o n to the shortness of the t e x t " (p. 220), since "When a singer is reproducing a w o r k of a hundred verses or so, it stands to reason that he will, w h e t h e r he wishes to or not, depend m o r e upon the performance from which he has heard the w o r k than if he is singing a song of over a t hous a nd lines whose telling might even be spread over a period of several evenings" (p. 220). The 400 line m i n i m u m he suggests for effective me a s ure me nt applies, then, to the length of the w o r k itself, to the length of the extract analyzed. 5.The argument that the higher frequency of four-syllable formulas is due to the assonance requirement for the second half of t he poetic line is not valid, since in counting formulas Duggan ignores the assonating w ord, even if it is a c ont e nt word. 6. Cf. "Epische Gesetze der V o l k s d i c h t u n g , " Zeitschrift ffir deutsches Altertum, 51 (1909), 1--12; Duggan, p. 1131. 7. Minor errors, misprints, and infelicities of style include: p. 23, the figures giving formulaic de ns i t y for the R o m a n d'Endas and the Moniage Guillaume have not been rounded off correctly: Endas should be 18 rather than 16 percent formulaic while t he Moniage should be 25 rather t h a n 24 percent formulaic. (No justification is given for rounding percentage totals here and on p. 99, in contrast to the accuracy to one decimal place used elsewhere in the text.) P. 38, 1. 11: " 2 9 . 7 % " contradicts the earlier " 2 9 . 8 % " given on pp. 26, 28, passim (the correct figure, to t w o decimals is 29.78%). P. 42, 1. 169b quot e d from the Roland is formulaic and should be in italics; n. 40, 1. 14: " 1 8 6 a " should read " 1 8 8 a . " P. 83, 1. 13: " 3 4 . 5 % " should " 3 4 . 0 % " ( 3 4 . 0 3 4 0 3 4 . . . , assuming t ha t the figures 680 and 1998 are correct; p. 113, 1. 7, p. 183,1. 24, p. 208, 1. 28, p. 209, 11. 12 & 22, p. 212, 1. 8, p. 215, 1. 18: "P~lerinage" should be "P~lerinage" (correctly spelled in the eleven other references to the w o r k ) ; p. 41: no justification is given for including the " F r e n c h Council S c e ne " in the analysis; p. 23, !. 2 2 : " 6 6 , 7 5 0 possible form u l a s " w oul d be clearer if it stated " 6 6 , 7 5 0 possible occurrences of formulas," since it is not until p. 84 t ha t t ype s and t o k e n s are distinguished.
40
REVIEWS
Shakespeare's Grammatical Style: A ComputerAssisted Analysis of "Richard H" and "Antony and Cleopatra," Dolores M. Burton. Austin: University
texts under discussion; and (2) a more abstract level, the level of the theory of literature.
of Texas Press, 1973. Pp. xviii, 364. $12.50.
By Barron Brainerd In this work Professor Burton seeks to achieve "two ends: to determine the role played by syntax in Shakespeare's stylistic development and to formulate a theory of style and a method of applying that theory to works of literature" (p. xiii). Richard II, from Shakespeare's middle period (written 1595), and Antony and Cleopatra, from the late period (1608), are the basis for a discussion of stylistic development "because they are generally considered to be good examples of distinct stages in Shakespeare's career" (p. viii). Shakespeare's syntax is discussed in terms of systemic grammar (a theory of language developed by M.A.K. Halliday) as exemplified in Sentence and Clause in Scientific
English. 1 Fundamentally, the author's approach to the problem is to apply scientific methodology to the analysis of the two texts, and I will concentrate for the most part upon her use of this methodology, commenting upon her handling of the data, the use of statistical evidence, and her overall theoretical position. More particularly, she develops certain formal theories which she tests against data (usually the text of the two plays). In her own words, A formal theory may be adequate or inadequate to account for the data it was constructed to explain. In either case, it will prove illuminating. If the theory is adequate, it will explain the data to which it is applied. If it is inadequate, it will not fit certain aspects of the data, but one can state more precisely than otherwise what those aspects of the data are and why they defy definition. (p. 11) Professor Burton applies this methodology upon two levels: (1) the level of the individual text, where the formal model is developed in terms of oppositions (in the Prague School sense) which can be tested to see whether or not they are characteristic of the difference between the Barron Brainerd is a professor o f mathematics and linguistics at the University o f Toronto.
The discussion at level (1) is for the most part careful and thorough, while at level (2) it is more diffuse and metaphorical because there is as yet no scientific theory of style and rhetoric. Indeed, Professor Burton's is one of the first steps in that direction. The author's approach to the data is, as it should be, careful - almost puritanical. She has made a conscious effort to be explicit in regard to the conventions by which she selects the passages to be discussed, 2 the decisions taken regarding the assignment of items to particular grammatical categories, and the conventions she adheres to when she performs the counts. 3 These decisions and conventions are explicated in the two Appendices: A. "Supplementary Tables and Notes on Procedure," B. "Citations to Richard H and Antony and Cleopatra." With the aid of these indications, the reader can replicate her steps in order to test whether he arrives at the same conclusions as she does. In order to manifest "a . . . sense of responsibility to the entire text and to the complex properties of language" (p. 144), Professor Burton has had to resort to quantitative methods and hence to computerization of at least some of her data analysis. In particular, she has produced and used extensively (f.n. p 18) a computer-generated function-word concordance of the two plays. Since most of her computer work was done prior to 1968, her concordance overlaps that of Spevack published in that year (A Complete and Systematic Concordance to the Works of Shakespeare). For some of her work, she uses Spevack's concordance as well as her o w n . 4 Her texts are from the Kittredge edition s of the Complete Works as opposed to the G.B. Evans texts (yet to be published) used by Spevack. To render the texts as nearly comparable as possible, the prose parts of Ant. were discarded so that both texts under consideration were in blank verse and of almost equal length: 21,809 words for R H and 21,777 (verse) words for Ant. In addition to this overall equilibriation of the material to be compared, she manifests, whenever possible, a conscious effort to control, in the statistical sense, the affect of extraneous variables and error. For example, in her section "Constant Topic and Varied Style: Adjectives," she endeavors
REVIEWS
to compare passages in both plays which have "constant semantic content." Further attempts to control extraneous variables can be seen on pp. 181, 2 0 5 - 6 . The judicious use of random samples in studying group 13 clauses (p. 56) constitutes another effort to control error. Professor Burton's use of statistics to provide evidence of significant differences between the two texts is not quite as clear and decisive as her primary treatment of the data. First, her propensity to view the data in terms of oppositions leads her to rely heavily upon the fourfold breakdown and hence upon the chi-squared test for homogeneity with one degree of freedom ( f . n . p . 33). Unfortunately for the reader, she discusses neither the rationale of the test, the degrees of freedom it allows, nor for that matter the notion of level of significance. She mentions in passing on p. 63, after the test has been applied five times, that she is rejecting the null hypothesis at the 5% level of significance. In order that the statistical evidence she obtains not be lost to readers of this review, I will attempt, in as short a space as possible, to fill in some of the gaps. The test for homogeneity involved can be summed up as follows: For the sake of concreteness, consider the information concerning Antony and Cleopatra given in f.n. 23 of page 38, where she attempts to show that Antony's excess of imperatives over interrogatives is significantly greater than is Cleopatra's. The following table shows the data: Interrogatives
Imperatives
obs.
(exp.)
obs.
(exp.)
Row Totals
Antony
60
(76.75)
251
(234.25)
311
Cleopatra
92
(75.27)
213
(229.73)
305
Col. Totals
152
464
616
Assuming as a (null) hypothesis that Antony's and Cleopatra's propensities to use interrogatives relative to imperatives are in any given instance the same, their combined propensity for using interrogatives would best be estimated as 152/616 = 0.2468, i.e. their combined interrogative count divided by the total number of sentences of both kinds used. Their combined propensity to use imperatives would then be 464/616 = 0.7532 (= 1-0.2468). Put another
41
way, the probability of an interrogative sentence (given that the sentence is either interrogative or imperative) is estimated at 0.2468 and the probability of an imperative under the same circumstances is 0.7532. One would then expect (under the null hypothesis) that 311 (the total of both kinds of sentences spoken by Antony) times 0.2468 (the estimated probability) would be the number of such interrogative sentences uttered by Antony, i.e. 76.75 interrogative sentences. In the above table, these expected values are entered in parentheses beside the corresponding observed counts. As a measure of the deviation of the observed frequencies from these expected or theoretical frequencies, we use g~ (i.e. chi-squared with one degree of freedom). It is the sum of terms of the form (observed - expected) 2 / expected, one for each of the four cells in the table and in the present case, it is X~ = 9.8, or if we use a Yates correction for continuity, 9.2. The probability of obtaining such a value under our null hypothesis is 0.00242 (.00175 using the correction), a highly unlikely event. Therefore we reject the null hypothesis in favor of the hypothesis that Antony and Cleopatra use different proportions of imperative versus interrogative sentences. The rationale just outlined is followed in all other statistical tests used in the book. In a study such as this, chi-squared with one degree of freedom is a relatively primitive tool to use. Surely this data deserves a more sophisticated statistical treatment. Even the use of the full possibilities of x2-tests would improve matters rather a lot. For example, the treatment of the total problem of the use of imperatives and interrogatives by Antony, Cleopatra, and Caesar (f.n. 23, p. 38) suffers from a triple application (Ant. vs. Cleo., Ant. vs. Caesar, Cleo. vs. Caesar) of the x2-test to essentially the same data, which somewhat clouds the results. This can be obviated in two ways. (1) Consider the data (Table A-2) as a whole and apply the homogeneity test to the proportions for Caesar, Antony, and Cleopatra simultaneously. This involves a chi-square with two degrees of freedom, 6 in this case X~ = 19.3, which can occur only one time in 100,000 under the null hypothesis
42
REVIEWS
of identical propensities among the three characters. An inspection of the differences between the observed and the expected values would show that the Cleopatra terms contribute most of the difference, 9.44 or 49% of the total chi-square observed, so she must be the inconsistent one. (2) Consider the results for Antony and Caesar together and observe as in f.n. 23 that they are not significantly different. Then using their pooled proportions as the expected proportions for the population to which they belong (0.2468 for interrogatives and 0.7532 for imperatives), invoke the null hypothesis that Cleopatra falls in the same population. Then Cleopatra's expected counts would be (0.2468) (105) (0.7532) (105)
= =
25.91 75.09
which, when compared with her observed counts, yields a X~ = 10.02 (without the continuity correction) which has a probability of less than 0.0016 of occurring. Therefore her usage is significantly different at approximately the 0.1% level. This result still leaves one a little uneasy because we have not been told the total number of sentences for each character. If we had, then we could have had three categories of classification (interrogative, imperative, other) as well as three samples (Antony, Cleopatra, Caesar), thus affording four degrees of freedom which would render the test more sensitive. Although the above results do not differ much from those appearing in f.n. 23, they are rationally and technically easier to defend. A more serious discrepancy occurs (f.n. 44, p. 63) when Professor Burton discusses whether or not the proportion of two-syllabled to four- and five-syllabled adjectives is greater in R H than Ant. Here she finds the difference not significant. However if she had considered all the evidence in Table 6 the 1-syllabled, 2-syllabled, 3-syllabled, 4-5-syllabled and the compound adjective counts for both plays and tested the entire table simultaneously, she would have had four degrees of freedom. Then X42 = 19.6, which has a probability of less than .00063 of occurring. This presents rather a different picture than that given in f.n. 44. To compare only the 2-syllable counts to the 4-5-syilable counts, leaving off the information about 1- and 3-syllable counts and compoundadjective counts, is to sacrifice real information. The discrepancy between the 2-syllable counts accounts for 48% of the final X4z whereas the 4-5-syllable
count accounts for only 6%. The 1-syllable discrepancies on the other hand account for 28% of the X42, which is clearly a more forceful difference than that produced by the 4-5-syllable counts. Clearly, R H has lower than expected 1-syllable counts. Professor Burton occasionally draws unwarranted conclusions on the strength of relative frequencies that in the light of the actual frequencies are not significant. For example, she remarks (p.131) that Ant. contains three times as many qualifiers among constituents affected by nominal-group inversion as does R H (15 instead of 5). This can be shown to be nonsignificant under the null hypothesis of equal propensity to use such qualifiers (• = 0.94). This lack of significance arises because the total number of instances involved is small. The author can seldom be faulted in regard to her choice of indices to be measured, but on some occasions, as the above remarks indicate, she falls short in her awareness of the possibilities for statistical analysis. The use of statistical methods in literary research has yet to begin on a serious scale and so in this area her book is in the nature of a pioneering work. 7 In order that such studies as this realize the full potential of statistical methodology, consultation between the authors and statisticians, as well as a passive knowledge of statistics among such authors, will ultimately be required. A number of Professor Burton's results are especially interesting, and often aesthetically pleasing from a methodological point of view. For example, in her attempt to discover linguistic indicators of a conscious "Roman style" on the part of Shakespeare, she displays (Table41) nine major syntactical features and their relative frequencies s in the speeches of Antony, Cleopatra, and Caesar. In all but one, Antony and Cleopatra resemble each other closely while Caesar appears rather different. In that one, the use of group /3's (the subordinate elements of hypotactic group complexes), Antony and Caesar resemble each other while Cleopatra is the outlier. Thus she concludes that high group /3's signify Roman style. She backs this up (p. 56) by noting that in a random sample of sentences (12% of the text in Ant.) there are 23 group/3's, eighteen of which were spoken by Antony, Caesar, Pompey, Ventidius, Agrippa and Enobarbus while Cleopatra speaks only two of them. The last chapter "From Stylistics to Poetics" contains no overt quantitative content. However, the quantitative results of the previous chapters (where a commendable responsibility is maintained
REVIEWS
to the texts in their entirety) tend to fortify our confidence in the skilful analysis displayed therein. On level (2), the theory of literature, Professor Burton approaches the problem of style by means of a style function, whose domain is the class A of all objects of which style is predicated (as she sees it, this should be "the set of all well-defined objects", p. 5). 9 As to the range (p. 7), "the style function will associate with [an element of A] a specific style." Thus the range of the style function is the set of possible specific styles. She also presents the notion of a style space (pp. 6, 223, 231) which is "an intermediate set of structural components with n dimensions." Each element a of A is associated with a point (Xl, x2, . . . . Xn) in this space. Each component x i is a number measuring the strength of some feature of the object a, e.g. the relative frequency of imperative sentences in the object. With this structure space it is possible to assign a topology to A, i.e. to place a structure upon A in terms of which we can discuss nearness. Now for the style function to be continuous, which she requires it to be (pp. 5, 21), it is necessary that the range of the style function also have a topology so that nearness of two styles can be discussed as well. Professor Burton does not touch upon this point. Indeed, she does not consider the nature of this space of possible styles at all. Since the nature of this space is a fundamental problem of stylistics, the discussion, fruitful as it may be, turns out to be only metaphorical in the end. However the work on the text level presented by Professor Burton can be integrated into another model which has at least some chance, I think, of becoming the basis of a concrete, viable theory of style. The program I propose would treat the problem of the style space as a classification problem and develop a taxonomy of style analogous to the taxonomy of biology. In recent years much progress has been made in this area using numerical methods, 1~ and a number of the philosophical and theoretical questions considered by Professor Burton have already been aired in a biological setting. For example, questions of independence of characters (p.225) has been considered in some detailJ 1 Following this program, stylistic species, genera, etc. could be defined and their diagnostic characteristics, hopefully, agreed upon by stylisticians in general. The sort of development presented by Professor Burton in her book could form a basis for such sets of diagnostic characteristics. The discussion on pages 2 2 2 - 5 shows she is heading independently in this
43
direction already. Professor Burton's book is stimulating, provocative and overall a useful addition to the library of anyone wishing to approach literary problems from a responsible scientific point of view.
1.
2.
3.
4.
5.
6. 7.
8.
9.
NOTES R o d n e y Huddleston, R. A. Hudson, E. O. Winter, and A. Henrici, Report of the Research Project on "The Linguistic Properties of Scientific English," Communication Research Centre, London: University College, 1968. On rare occasions, we are seemingly let down in this regard. In particular on p. 42, where she discusses "representative passages," there is no indication h o w these are chosen. Presumably, it is by using the critic's intuition, which is a departure from the m e t h o d o l o g y of science. This lapse may be only apparent, a "fishing trip of the i m a g i n a t i o n " in order to generate hypotheses. In most sciences h y p o t h e s e s are presented and tested, and it is bad form to discuss h o w those hypot he s e s were generated. This is certainly not the case in at least some of the language sciences where s ome t i me s the generating of h y p o t h e s e s is indulged in to the e x t e n t of disregarding the data as a whole. On p. 275, for example, she tells us t ha t under certain circumstances the lexical item let is c o u n t e d twice when it appears only once in the text. The reasons for this are well laid o u t and an explicit algorithm is given for finding such circumstances. A Complete and Systematic Concordance to the Works o f Shakespeare, Hildesheim: Georg Olms. On p. 282, Professor Burton notes some differences between her concordance and Spevack's, due to the difference in the editions used by each. Burton remarks ( f . n . p . 18): "The Folio was not Chosen as the basic t e xt , because it seemed that decisions a bout the definition of a sentence or a particular ma rk of p u n c t u a t i o n were be t t e r made by a good editor than by one w ho lacked specialized knowledge of c o m p o s i t i o n practice, t e xt ua l history, and Renaissance theories of p u n c t u a t i o n . " The existential status of the data is very i m p o r t a n t to studies such as this, and in the ease of Shakespeare it poses a n u m b e r of very c ompl e x problems, often best left to experts. In some sense, this deferral to Kittredge is analogous in a more highly developed science to performing further analysis upon data already gathered by a not he r researcher. If a table contains n rows and m columns, the test for h o m o g e n e i t y has ( n - l ) ( m - l ) degrees of freedom. The w o r k of Gustav Hedan n o t w i t h s t a n d i n g , for he lacks Professor Burton's i ni t i ma t e c o n n e c t i o n w i t h the data, one of the essentials for the f o r m u l a t i o n of realistic formal models. Because these are major syntactic features and the actual counts will t e nd to be high, we can, insofar as this is the case, believe in the significance of the relative frequencies. However, a significance test w oul d ma ke the evidence more forceful. Such a broad c ons t ruc t i on of the doma i n may lead to set-theoretic paradoxes, because parts of an object as well as the object itself are included in A. Further i f B is a well-defined subset of A, it is also an element of A. In some sense, this paradox-prone situation is an integral part of the problem: We w oul d certainly like A to contain, for example, all Renaissance plays and we w oul d like to compare the style of A n t o n y ' s speeches w i t h t ha t of Cleopatra's, both parts of one of the plays, A n t o n y and Cleopatra, an element of A. Further we w o u l d like to compare the style of Shakespeare's middle period
44
REVIEWS
with that of his mature period, i.e., to compare the style of the set of plays in the middle period with the style of the set of mature plays. One solution of this paradoxical situation is to sacrifice generality and compare o n l y items of like type, i.e., compare plays with plays, characters with characters and sets of plays with sets of plays. I0. See, for example, Peter H. A. Sneath and Robert R. Sokal, Numerical Taxonomy (San Francisco: W. H. Freeman, 1973). 11. Cf. for example, Sneath and Sokal, pp. 103--6, 384; also E. C. Olsen and R. L. Miller, Morphological Integration (Chicago: U. of Chicago Press, 1958).
A Concordance to the Works of Sir Thomas Malory, Tomomi Kato, ed.Tokyo: University of Tokyo Press, 1974. Pp. 1659. $120.00. By Robert W. Ackerman This weighty and extremely expensive volume is a printout of the entire vocabulary, including proper nouns, of Malory's Le Morte Darthur as represented in Eug6ne Vinaver's edition of the Winchester Manuscript (The Works o f Sir Thomas Malory, 2nd ed., 1967). Every occurrence of even the "common" words-articles, the verb to be, and other function words-is here recorded. Thus, of the total of 1600 pages, 60 set forth some 13,000 entries of the, 56 are devoted to and, and 22 to a. The entries are presented according to the KWlC system of a brief contextual passage of six or so words. For ready reference, each entry cites not only page and line numbers in the Vinaver edition but also Caxton's book and chapter divisions. Words used in direct discourse, as opposed to exposition, are marked with "C" (conversation). The two scribal hands of the Winchester Manuscript are recognized, as are entries from sections of the work which Vinaver filled out from two forms of the Caxton edition. A fifty-page summary at the end of the volume records for each word the total occurrences, with attention also to those used in dialogue and the frequency of the word expressed as a percentage of the total vocabulary. Kato, in the Preface, says little on methodology of the project, although for technical details he refers readers to a rather uninformative note in Computers and the Humanities. That a Malory concordance will contribute to our understanding of fifteenth-century English and to diachronic and dialect studies in general is selfevident. Linguistic study in the service of literary Robert If. Ackerman is a professor of English at the University of Wisconsin, Madison, and professor emeritus of Stanford University.
history may offer a means of assessing the usefulness of the new concordance. William Matthews, in The Ill-Framed Knight: A Skeptical Inquiry into the Identity o f Sir Thomas Malory (1966) conjectures that the author of Le Morte Darthur was not the Malory of Warwickshire, as has nearly always been assumed, but rather a somewhat younger Malory of Yorkshire. To a large extent, the case rests on the frequency of Northemisms in the Caxton print and the Winchester Manuscript, both of which are perhaps thrice-removed from Malory's lost original. Among the forms he deems significant, Matthews cites the frequency of Northern their as against the more southerly hir and of them as against hem. It may be said that Matthews' rough estimates of these frequencies are fairly well corroborated by the facts presented in the new concordance. The exact data now available about such matters and also about dialectally significant vocabulary items should greatly facilitate a close examination of Matthews' theory and may lead to new suggestions as to Malory's identity. The truth of Kato's admission in his Preface that his treatment of homographs and variants "leaves something to be desired" becomes apparent upon even a cursory examination of the concordance. For the pronoun hir mentioned above, for example, the concordance does not segregate the plural possessive form hir from the feminine hit. Moreover, the contextual passages are sometimes too brief to permit distinguishing one from the other. Again, one who may be interested in searching out Northern ,and present participles, for example, will find the concordance of little help in view of the lack of any marking of grammatical forms. The omission of cross-references for variant spellings is likewise a limitation, as Kato acknowledges in his Preface. In the treatment of proper nouns, which assume considerable importance in literary research, the lack of a cross-referencing system results in particular confusion and misrepresentation. For variations in the spelling of names in Malory, as in much other Middle English literature, are probably greater and more unpredictable than may be found in the vocabulary at large. One would be able to collect all references to many persons and places from this concordance only with the aid of prior knowledge of the sometimes very wide inconsistencies in form and spelling. F o r e x a m p l e , Castle Carboneeke or Carbonek, the mysterious abode of the Grail keeper, is just as often called Corbyne or Corbyn, all of which forms are, of course, separately listed. The same is true of references to persons called indifferently
REVIEWS
Uwayne and Ewain, Arnold and Raynold, Vyllars and Wyllyars, Royens and Ryens, and numerous others. It is notoriously true that a number of Arthurian characters bear the same name, variously spelled, or closely similar names, yet lumped together here are references to two and even more than two different personages because of an occasional identity in the orthography of their names. One must be prepared, then, not only for the separate listings of the variant spellings of the name Elayne but further for the inclusion under each variant of references to four different Elaynes. The same is true of the three knights named indiscriminately Uwayne or Ewain, the two Launcelots, the two Isouds, and the several distinct Galahads (Galaad, Galehaut, etc.). The contextual passages given in the concordance are seldom sufficient to determine what character is intended, and one has no recourse but to turn to the text of Le Morte Darthur for enlightenment. One amusing consequence of the mechanical sorting of identical spellings of names is the listing along with the numerous invocations to God of a certain Sir God of Cardal, Constable of London, who in other Arthuriana is normally called Do, Du, or even Dieu. The new Malory concordance has much to recommend it, with particular reference to the frequency tables at the end, its completeness, the distinction it makes between words used in conversation and in exposition, and its accuracy. According to a statement in the Preface, considerations of cost account for most of the limitations that have been mentioned above. At the same time, one wonders why the special nature of Malory's proper names should not have been recognized, perhaps by deleting them from the work altogether.
45
Oakman (pp. 411-426). The first article follows step by step the Center for Editions of American Authors' Statement of Editorial Principles and Pro-
cedures: A Working Manual for Editing NineteenthCentury American Texts, proposing ways that electronic technology can assist the recovery and preservation of the author's intent as has been done by laborious hand procedures in past years. The article does propose new technology, but merely aims to show places where computers can relieve editors of drudgery in their customary work of collating versions, choosing a presumptive copy-text, emending that text, and publishing their final work in an accurate way. Howard-Hill proposes that an interactive system, with the editor working directly with an optical display of the electronic record, will combine the best capacity of the human mind with the best potentialities of the machine. He admits that no very accurate estimate of cost efficiency of such systems is possible, especially for the pioneer applications; but he thinks that electronic procedures would be more accurate in collating and printing texts and that preserving the final edition on electronic tape, so that all subsequent printings of the text would be produced from that tape, could prevent the degeneration of the text after its modern edition had been established. He quotes the alarming information that in the four republications of the carefully edited Centenary The Scarlet Letter, there are already evident serious degenerations of the text, errors and departures from the edition ostensibly copied. Electronic storage would be a more permanent record of the editorial work than a printed text. In the second article, Oakman examines especially A Concordance to the Poems of Emily Dickinson by S.P. Rosenbaum, A Concordance to the Poetry of Robert Frost by Edward C. Lathem, and An O'Neill Proof" The Yearbook of American Bibliographical Concordance by J. Russell Rearer, and places them in and Textual Studies. Vol. III, ed. Joseph Katz. the context of the historical development of comColumbia, S. C.: University of South Carolina Press, puter-assisted literary indexes and concordances. No doubt written at least two years ago, this article tends 1973. Pp. 595. $27.00 to look backward rather than forward. It properly praises the page composition of Howard-Hill's Oxford By Todd K. Bender concordance to A Midsummer Night's Dream as Two articles in this volume are of particular superior to various examples of photo-offset printing interest to the readers of Computers and the Human- from chain printer or other primitive output devices, ities: 1) "A Practical Scheme for Editing Critical but does not foresee radical new ways of handling Texts with the Aid of a Computer" by T. H. input such as direct optical scansion or innovations in Howard-Hill (pp. 335-356), and 2) "Concordances the conception of a literary index. As a result, the from Computers: A Review Article," by Robert L. judgment passed on the works in question seems uncritical or undemanding at times. In both of these articles, the authors conceive of Todd K. B e n d e r is a professor o f English at the University o f Wisconsin, Madison, electronic technology merely as a convenient way to
46
REVIEWS
imitate traditional hand processes. Oakman takes as an ideal model of a concordance essentially the same kind of index which has been prepared and printed by hand in a format unchanged since the birth of printing. He then measures the value of a computerassisted index by how well or ill it app'roximates the traditional hand-made model. Howard-Hill likewise sees the computer merely as a tool to assist the traditional tasks of an editor, but does not see electronic technology as changing in a radical way our conception of what a literary work is or what the duties of a literary scholar are. Recent developments in literary computing suggest, however, that the scholar of the future will see the computer as opening a whole new concept of bibliography in which the "real" concordance or the "real" critical edition resides not in the printed pages, but in electronic storage. Howard-Hill's suggestion that the master electronic tape is less subject to degeneration in reprinted versions than paper copy directs our attention to the day when libraries will be built on electronic data banks rather than printed paper and scholars will have an infinitely more sophisticated method of representing authorial intent or describing features of a text than is possible so long as we are limited to thinking of the work as the same as a printed text. Any single printed text is not the same as the work, but merely a representation of it. Electronic models can represent the full complexity of a literary work better than any printed version, no matter how carefully prepared. In the future, the literary scholar will use the computer to set up data banks which take full advantage of the potentialities of electronic technology for storage and analysis of data, rather than using the computer mainly to help set up printed information storage. Meanwhile, these two articles sketch an intermediate step by which literary scholars may be induced to see how electronic data processing can help them.
A Concept Dictionary of English, by Julius Laffal. Essex, Connecticut: Gallery Press, and New York: John Wiley/Halsted Press, 1973. Pp. xiv, 306. $13.50.
By Jeffrey F. Huntsman In the common history of Western Europe, meaning in language has generally been thought to reside centrally, if not solely, in the isolated word. In spite of strong and occasionally dominant linguistic theories that included meaning within language
proper, the practical delineation of what language means has often been left to lexicographers while the theoretical explanation of how and why it means has been the concern of philosophers. The historical causes of this curious division of labor are well known, of course, and, given the reluctance of many earlier language specialists to confront meaning directly, it is scarcely surprising that the linguistically naive today should continue in this ancient half-truth. But it is rather distressing when contemporary researchers dealing with language ignore the known complexities of linguistic meaning in the name of simplistic and illusory notions of psychological taxis. For a quarter-century, psychologist Julius Laffal has studied word-association behavior and methods for isolating the natural cognitive sets to which words seem to belong. A Concept Dictionary of English (henceforth CDE) is his latest version of a computational sorting apparatus which sifts the vocabulary of English into 118 psychologically cogent categories according to the criteria of synonymity, similarity, and relatedness. After a preface (vii-xi) and an introductory first chapter, Laffal describes the categories (chapter 2, 19-30), lists each of the 23,500 coded words twice, first alphabetically (chapter 3, 31-154) and then by category (chapter 4, 155-299), and ends with a bibliography (300-302) and an index (303-5). With its "implicit orientation toward understanding the psychology of individual speakers" (vii), the book purportedly has a "broad" range of applications, but was especially designed to allow comparisons of different language corpora. Although it has chiefly been applied to so-called abnormal language, Laffal claims it may ultimately be useful for normal folks because "it offers the opportunity to discover one's preferences and deficiencies in language for the available conceptual areas" (xi). The CDE had its beginnings fifteen years ago in Laffal's fulgent insight that even the language of "psychotic" people was indeed human language, that is, it has an internally self-consistent organization: "It seemed to me that if one had a broad sample of a patient's language, it might be possible to shed light on his more obscure statements by observing how the words in these passages were used in other contexts" (vii). What was needed, he believed, was a simple method which Jeffrey F. Huntsman is an asMstant professor o f English and Director o f the Survey o f Medieval English Lexicography at Indiana University.
REVIEWS
would permit him to "explore content without t r i p p i n g e n d l e s s l y over problems of word equivalence" (viii). If simplicity were the only metric of success, the CDE would be a marvel; unfortunately its simplicity reflects more the author's desires than the condition of the world. His system in actuality is a farrago whose individual members have no apparent relation to each other and, as a result, the system gathers everything but explains nothing. This failing is easier to document than to comprehend. Although Laffal refers to earlier studies in lexical semantics like those by Jost Trier (1934) and Hallig and Wartburg (1963), he seems to have misunderstood both their organization and their significance. Their classifications, like those of Roget (1852) and Buck (1949), are termed "logical" (4), while his are "conceptual" and "psychological." Yet he states that Roget "provided the earliest model of a conceptual scheme of language" (2). Laffal apparently means here to make a distinction based on method: "psychological models use group responses to derive norms of individual use" (3) while "logical" models represent the imposition of a "unifying viewpoint and guiding plan" (4). Nevertheless, although he claims the categories were "empirically derived by grouping words related in meaning" (ix), he admits that, in actuality, "the author's own judgments were a major factor in the evolution of the category system, and the dictionary in its present form is best viewed as a product of subjective intuitions, interactions with others, and empirical applications" (3). It therefore seems highly improbable that, with the possible exception of a relatively greater reliance on the last activity, his method of analysis differed much from that of Roget, Buck, and the others. Indeed, he asserts that "if someone who had never seen his dictionary undertook to divide the language into 118 concepts, it is highly likely that he would come up with a group which resembled the present one" (ix) and refers for support to Buck's Dictionary and Spurgeon's Shakespeare's Imagery (1935). Most notably missing from his discussion of these earlier studies is any indication that the lateral and hierarchical interrelations among the categories of a conceptual system are ultimately of greater import than the mere distribution of individual lexical items into categories. And, not surprisingly, those interrelations are likewise absent from his own treatment. A short review is hardly the place to relate the state-of-the-art in semantic theory, nor is such a
47
discussion necessary for most readers of this journal. However, it will come as no surprise to those familiar with the typically strabismic output of the spiritual descendants of John Broadus Watson that Laffal is seemingly quite ignorant of contemporary views on linguistic semantics. Only structuralists, who in principle exclude the study of meaning from the domain of linguistics, are mentioned in his Pathological and Normal Language (1965); only Katz' and Postal's transformationalist study (1964) appears in CDE and there only in the bibliography (301), not in the text; and no linguists at all are referenced in his 1970 article in Computers and the Humanities which is expressly concerned with linguistic semantics. Completely missing are mentions of other current approaches like the stratificational grammar of Sidney Lamb (1966) and others; semantically-based structuralist grammars like that of Wallace Chafe (1970); and case grammars of various types, including those proposed by Charles Fillmore (1966 and 1968), Geoffrey Leech (1969), and John Anderson (1968 and 1971). Whatever their differences or inadequacies, all these latter theories agree in their inclusion of meaning in language and in their acknowledgment that the organizing principles are complex and subtle. Nor is it true that the integrated view of language exemplified by these different approaches dates only from 1957. In 1936, Vil6m Mathesius wrote about Otto Jespersen's Essentials of
English Grammar: Jespersen, in his praiseworthy adherence to what is really accessible to our immediate observation, does not seem to realize enough the possibility of a deeper and more hidden reality . . . . The relative importance of a linguistic fact within the grammatical system of a given language can be ascertained only from the point of view of the whole system. (307) Nothing in the last thirty-five years has altered the truth of Mathesius' formulation. More startling is Laffal's apparently deliberate neglect of the psychological literature on wordassociation. Although Laffal's doctoral dissertation and earliest publications concern word-association behavior, he ignores the complex and asymmetrical networks of words described by Pollio (1966), Kiss (1968 and 1972), McNeill (1963), and Deese (1965). Indeed, Deese even points out, in one of the rare mentions of Laffal's work in the literature, that simple correlations among words are inadequate without reference to other aspects of language:
48
REVIEWS
The major deficiency in the tables presented so far [which include the results reported in Laffal and Feldman 1962] is that the grammatical structure of the language from which the words are drawn is ignored. There are many reasons for believing that there should be well-defined relations between the associative structures and the grammatical structures of the language. (1965:96) Laffal could confidently disregard this criticism by the author of the standard text in the field only if his system were sufficiently inclusive and sophisticated to deal with the complex structure of language in some other way. Unfortunately, his system is disappointingly simplistic and crude. Although lexical meaning is the locus of semantic factors of many differing types and strengths, Laffal treats each word as if only two broad cognitive labels were applicable and both with equal force: There is no assumption that concepts assigned to a word reflect totality [sic] of word meaning. For a word to be classified within a given concept, it must contain as a component of its meaning, the experiential phenomena addressed by the concept . . . . There is at present no evidence concerning the relative strength of evocation of the various cognitive concepts for given words; the assumption is therefore made that all words which evoke a concept evoke it in equal strength. (9) The second assumption is contrary both to the evidence adduced by the word-association researchers mentioned above and others, such as Kjellmer, who studied the semantic relations in a group of similar Middle English words (1971), and to the common sense of anyone who has reflected on any natural language for more than one minute. But to document, let me turn to some examples of Laffal's categories and some of the words that he classifies. The 118 concept categories, which are supposed to "capture a significant component of the meaning of a word" (15), are labeled by two- to fourcharacter mnemonic designators like AIM, BODY, EVER, GLAD, HEAR, LADY, MOTV, NO, PLAC, SOLE, TIME, VAPR, and WRIT. The words bride, mistress, virgin, hustle, and wife are classified under LADY AND SEX, while prostitute, whore, lecher, procure, and callgirl are BAD SEX. Fornicate is BAD SEX but screw, intercourse, copulate, and commerce are JOIN SEX. Seduction and pederasty are BAD SEX, miscegenation is JOIN SEX, and
incest is CRIM SEX. Entrail is simply SOMA (relating to internal anatomy and physiology), while
bowel is SOMA DIRT. Burp is SOMA HEAR and fart is SOMA VAPR. The VFW is a DAMG PAST GRUP, an aphid is a DIRT BUG (although a roach is just a BUG), and a Dalmation is a PLAC ANML. In the area of bilateral conviviality, we find that petting is FOND SEX, necking is BODY SEX, and orgy is PLAY SEX. And finally, on a more elevated topic, polyphony is MUCH MUSC but monody is PANG (pain) MUSC. Poor Handel. Curiously, Laffal can't seem even to make up his mind whether or not the assignment of category designations is especially significant. Although he admits that "the dictionary user will undoubtedly differ with me in many instances over the concepts assigned to a word" (ix), he claims that, "for conceptual content analysis, accurate placement of a word within a category is a crucial matter" (4). Apparently, categorization of a word is significant only if one disagrees with the CDE. Feeling that considering these citations in isolation might have done an injustice to Laffal's system, I analyzed a number of psalms by Laffal's system and found that the significant categories are SOLE and WE (i.e., the word I and its related forms), LEAD and HOLY (i.e., Lord, God, and related forms), and AID. (Psalms, it would seem, are private prayers for help.) Laffal himself doesn't often deign to interpret his results. After profiling Dylan Thomas' "Over Sir John's Hill" and a selection from Gulliver's Travels (1970:184-5), he avoids any conclusions: "I will not attempt to make anything of these distinctions, since the point is simply to demonstrate that two texts (differing in some obvious ways) are radically different in a conceptual analysis" (185). It would seem to me, as a student of English language and literature, that it doesn't take a very high level of reading competence to realize that texts on different topics involve different concepts, but I suppose that judgment is not really empirical. However, I did find solid empirical support for my impressionistic view in an article in the Journal of Verbal Learning and Verbal Behavior (1967), entitled "Characteristics of the Three Person Conversation." The researcher found, by coding conceptually the words used in conversations, that "the study did not reveal significant individual differences in predilection for content across all conversations. This suggests that topic of conversation and social conformity override individual differences in content" (555). Noting that "subject
REVIEWS
matter is a more important determinant of content than individual differences," the author reports that subjects correlate in content more highly with each other within conversations than they do with self across conversations. On the other hand, where topics judged psychologically similar are offered as stimuli for conversation, the contents are more similar to each other than for psychologically dissimilar topics. (558) The experiment proves that people really do talk about the same thing in conversation, and that's a comforting revelation in these days of mindwrenching discoveries like black holes. Obviously, this experiment seriously challenges Laffal's contention that his conceptual categorization system will "provide an adequate basis for discriminating the language of one speaker from another" (7) and it is rather surprising that he did not address himself to the apparent discrepancy, especially since he lists the article in the References to the CDE. By the bye, the scholar who probed the conceptual organization of conversations is Julius Laffal. Sadly, the level of sophistication of the computational techniques described in the CDE appears to be quite on a par with that of Laffal's linguistic theory. The core of the program is a dictionary look-up with a fall-back procedure that defmes "a dictionary equivalent of a text word as the first dictionary entry which matche[s] most initial letters of the text word" (viii), marking, for example, a nonagenarian NO. Even this relatively simple system requires a considerable amount of pre-editing by someone who knows the system well, since classifiable words must replace proper nouns, idiomatic expressions, pronouns, abbreviations, and titles, and compound nouns must be separated into component parts. I suspect the computational simplicity is a result of Laffal's unimaginative approach to language and does not reflect on the competence of the programmers, whose names compassion forbids me to mention. In sum, then, the CDE is a silly book which extends our knowledge of language or individual psychology not one bit. Its simplistic approach to the perplexing and profound organization of lexical semantics levels the very distinctions that make language interesting and its analysis revealing of the intentions of its speakers. Julius Laffal seems to have been willfully working in a self-imposed isolation from the major currents in language research, both linguistic and psychological (in the narrow
49
sense of those terms), and it is therefore scarcely surprising that the CDE, the fruit of his professional endeavors so far, should be stunted and unnourishing. Ten years ago Laffal called an earlier version of his category system a "lexically unsophisticated approach" (1964:153); a decade has not diminished in the s l i g h t e s t the force of that lucid understatement. REFERENCES 1. Anderson, J. M. "Ergative and Nominative in English," JL, 1968, 4:1--32. 2. Anderson, J. M. The Grammar o f Case. Cambridge: University Press, 1971. 3. Buck, C. D. A Dictionary o f Selected S y n o n y m s in the Principal Indo-European Languages. Chicago: University of Chicago Press, 1949. 4. Chafe, W. L. Meaning and the Structure o f Language. Chicago: University of Chicago Press, 1970. 5. Deese, J. The Structure o f Associations in Language and Thought. Baltimore: Johns Hopkins Press, 1965. 6. Fillmore, C. J. "Toward a Modern Theory of Case," The Ohio State University Project on Linguistic Analysis, 1966, 1 6 : 1 - 2 4 . 7. Fillmore, C. J. "The Case for Case," Universals in Linguistic Theory, ed. E. Bach and R . T . Harms. New York: Holt, Rinehart and Winston, 1968. 8. Hallig, R. and W. yon Wartburg. Begriffssystem als Grundlage fiir die Lexikographie. Vemuch eines Ordnungsschemas. Berlin: Akademie-Verlag, 1963. 9. Katz, J. J., and P. M. Postal. A n Integrated Theory o f Linguistic Descriptions. Cambridge, Massachusetts: MIT Press, 1964. I0. Kiss, G. R. "Words, Associations, and Networks," Journal o f Verbal Learning and Verbal Behavior, 1968, 7:707--713. 11. Kill, G. R., C. Armstrong, R. Milroy, and J. Piper. "An Associative Thesaurus of English and its Computer Analysis," The C o m p u t e r and Literary Studies, ed. A. J. Aitken, R. W. Bailey, and N. Hamilton-Smith. Edinburgh: University Press, 1972. 12. Kiellmer, G. C o n t e x t and Meaning. A S t u d y o f Distributional and Semantic Relations in a Group o f Middle English Words. (Gothenburg Studies in English 22). GSteborg: Acta Universitatis Gothoburgensis, 1971. 13. Laffal, J. "Linguistic Field Theory and Studies of Word Association," Journal o f General Psychology, 1964, 71:145--155. 14. Laffal, J. Pathological and Normal Language. New York: Atherton, 1965. 15. Laffal, J. "Toward a Conceptual Grammar and Lexicon," Computers and the Humanities, 1970, 4:173--185. 16. Laffal, J. and S. Feldman. "The Structure of Free Speech," Journal o f Verbal Learning and Verbal Behavior, 1962, 1:54--61. 17. Lamb, S. Outline o f Stratificational Grammar. Washington: Georgetown University Press, 1966. 18. Leech, G. Towards a Semantic Description o f English. Bloomington, Indiana: Indiana University Press, 1969/1970. 19. Mathesius, V. "On Some Problems of Systematic Analysis of G r a m m a r , " Travaux du Cercle Linguistique de Prague, 1936, 6795-107. Quoted from A Prague School Reader in Linguistics, ed. J. Vachek. Bloomington, Indiana: Indiana University Press, 1964, 3 0 6 - 3 1 9 .
50
REVIEWS
20. McNeill, D. " T h e Origin o f Associations w i t h i n the Same Grammatical Class," Journal o f Verbal Learning and Verbal Behavior, 1963, 2 : 2 5 0 - 2 6 2 . 21. Pollio, H. R. The ~tructural Basis o f Word Association Behavior. T h e Hague: M o u t o n , 1966. 22. Roget, P. M. Roger's Internaffonal Thesaurus (1852).
New York: Croweli, 1960. 23. Spurgeon, C. F. E. Shakespeare's Imagery and What It Tells Us. Cambridge: University Press, 1935. 24. Trier, J. "Das Spraehliche Feld," Neue Jahrbiicher fiir Wissenschaft und Jugenbildung, 1934, 1 0 : 4 2 8 - 4 4 9 .
Continued from page 29
News and Notes Two Michigan State University professors have programmed the university's main computer to talk. The synthesizer produces voice sounds like those of a man speaking with the enunciation of a long-distance telephone operator. He lisps his th's but otherwise he's as clear as a Bell employee. Dr. John B. Eulenberg, visiting assistant professor of linguistics and Oriental and African languages, Dr. Morteza Amir Rahimi, associate professor of computer science, have been experimenting with the voice synthesizer as part of an on-going project in artificial language development. The digitally controlled voice synthesizer, manufactured by Federal Screw Works, is about the size of a breadbox, much smaller than earlier voice synthesizers, which sometimes occupied whole rooms. Eulenberg and Rahimi see their project as leading to improved Computer-Assisted Instruction generally, and particularly for blind persons. Employment opportunities for blind persons in the field of computer programming also could be extended. A visually disadvantaged person trained in touch-typing could enter questions and requests using the keyboard, with instantaneous response coming both vocally and visually from the computer. The blind would thus be able to prepare and proofread letters, reports, and class assignments which have been fed to the computer in machine-readable form. Additional use could be for teaching semi- or pre-literates, guiding individuals working in the dark, and providing automatic navigation for bus drivers or plane pilots. Now programmed to speak English with the voice of a male adult, the synthesizer can be modified to speak like a woman or a child, and that it can be made to speak a language other than English.