Electronic Analysis of Medieval Texts: The Case of Raoul de Soissons

by Ineke Hardy

Over the past century, the conception of medieval texts has gradually moved from that of a printed and thus "fixed" document to that of a fluid and essentially oral communication seeking to be understood across the ages in all its multidimensional aspects. The advent of the computer has acted as a sort of catalyst in this process of ontological reappraisal, creating a new form of communication that seems situated half-way between orality and print literacy in ways we are only just beginning to understand. Electronic text analysis still relies on the written text, however: text retrieval software compiles inventories of written symbols (encoded for electronic processing), not sounds. Given the oral nature of the transmission and reception of medieval texts, this limitation represents a serious drawback, and even more so in the case of lyric poetry, which was sung. In this article (an expanded version of a paper presented to the Southeastern Medieval Association conference held in Knoxville, TN, in October 1999), I propose to demonstrate how the computer can be taught to "hear", following a method developed in collaboration with Elizabeth Brodovitch, a graduate student at Simon Fraser University. Designed to transform graphemes into units of sound for the purpose of textual analysis, this method allowed us to draw up what we called "phonetic blueprints" of printed texts, with the aim of identifying the presence of anagrams in the songs of the troubadours and trouvères 1. I will show how this type of analysis makes it possible to single out songs that "deviate" from the norm established on the basis of the poet's total output, suggesting the presence of a hidden phonetic "agenda", and I will follow up on some of the clues thus obtained. The corpus chosen is the work of Raoul de Soissons, a trouvère from Northern France, whose songs were composed around the middle of the 13th century 2; this choice was governed by the fact that the texts are technically complex and of a high standard, and their number and size appropriate for the purpose of this research project.

Introduction

Roger Guiette has written of courtly poetry that "...jamais poésie ne fut plus ... totalement et plus consciemment calcul, mathématique et harmonie" 3 : poetry was never more completely and more intentionally calculation, mathematics and harmony 4. In the medieval view, as interpreted by Dante, courtly poetry was the art of treating words as items in an aural harmony. Dante saw it as consisting of three parts: the basic structure of the stanza, the proportioning of the parts, and the harmoniousness of lines and syllables 5. Unlike the modern conception of poetry, the grant chant courtois was composed on the basis of a prescribed message; the artistic challenge was not so much in what to say as in how to say it and in seeking to 'invent' (trouver or trobar) new forms, with the result that aesthetic quality of form and structure outranked cognitive merit. The medieval courtly audience, accustomed to the reception of oral texts and thus equipped with sharply honed audiovisual skills, would have been well qualified to perceive and appreciate nuances of structure, proportion and harmony, as well as special sound effects such as anagrams, operating at a level just below the surface of the text. The courtly song, intended to be performed for the enjoyment of an audience (members of the nobility), had an entertainment value as well as artistic merit, and as long as the poet stayed within the rules governing the art and the (ostensive) message, he was free to play the game of love, such as hiding an erotic contra-text beneath the stately exterior of the grand chant courtois. The clues would be there, to be grasped by those alert and knowledgeable enough to detect them - a process that created an extra degree of closeness between the poet and members of his public.

In today's print-based culture, the perception and decoding of many of these stylistic devices represent a considerable challenge. However, our lack of aural skills can to some extent be overcome by having recourse to the ubiquitous computer, itself a tool that relies on calculation, mathematics and the inherent harmony its system imposes. The courtly song, with its artful design incorporating the mathematically precise (and often numerically symbolic) construction of syllables, lines, stanzas and rhyme sounds lends itself particularly well to the computer's analytical capabilities. As we shall see, electronic text retrieval software can be adapted to create a database of sounds, rather than written symbols. Several text retrieval programs are available for this purpose, amongst them one called TACT 6 (an acronym for Text-Analysis Computing Tools), developed by the Centre for Computing in the Humanities at the University of Toronto. TACT compiles a random-access textual database comprising all words in a given text, permitting a sophisticated level of querying by means of various metacharacter operators capable of selecting frequency, similarity, phrases, co-occurrence, exclusion and so on. TACT offers a number of sub-programs, including one that identifies anagrams and another that generates statistics on letter frequencies.

It was the latter feature that prompted Elizabeth Brodovitch and me to attempt a statistical substantiation of the presence of anagrams in troubadour songs. The presence of anagrams, or hypophones as we prefer to call them 7 (since we are dealing with sound effects operating below the surface of a text), is notoriously difficult to defend: once one starts looking for them, one tends to find them everywhere. Ferdinand de Saussure himself wrestled with this question: "Are we, or are we not dealing with simple fortuitous coincidences?" he asked in one of his letters 8. The challenge was to develop an analytic method capable of ruling out coincidence and establishing the presence of minimal phonetic units able to function in hypophonic procedures.

We took a statistical approach, based on the assumption that the phoneme count of a poet's combined output could function as a norm against which to measure phonetic profiles of individual songs, making it possible to identify individual songs containing unusually high numbers of phonemes forming a key word. This approach was based in part on a discussion by Paul Zumthor (op. cit, see note 7), outlining the difficulties in distinguishing the degrees of a poet's intentions on a phonetic, repetitive, and alliterative level or at a level at which words are hidden within words. His approach to the study of what he called "des paragrammes chez les troubadours" was to calculate statistical averages based on a phoneme count of three narrative passages that served as a norm against which to measure phoneme counts in five songs. In his conclusion, he suggested that the results emerging from this sort of inquiry would not really assume a positive value until the method had been applied to a much larger corpus, comprising "au moins plusieurs dizaines de chansons" (66).

When we began our efforts to establish what we called "a phonetic blueprint" of a corpus of several dozens of troubadour songs, we were immediately faced with the problem of how to convince TACT to count units of sound rather than written symbols. Courtly love songs were sung, after all, their audience did not read them but heard them. The challenge was to teach TACT to "hear" rather than to read. We were unwilling, on principle, to tamper with the texts by altering them, changing the hard "c" to "k", for example, and the soft "c" to "s." After much experimentation, we developed a method of marking the text with tags, since TACT requires the specification of alphanumeric symbols, diacritics, punctuation etc. to be scanned by the program so it will be able to recognise them. These metatextual characters (identified in the markup file forming an integral part of the program) permit the system to inventory the text and subsequently generate various statistics. We prepared a markup file that could handle both the inconsistency of spelling practices (due to the absence of orthographical standards, manuscript provenances, scribal practices, medieval typos, editorial decisions, changing editorial attitudes, and even modern editorial typos) and the problem of identifying grapheme/phoneme combinations (which we referred to as "sound units") that would be consistent and comprehensible for retrieval. This applies specifically to double letters (e.g. souffrir) and to diphthongs and triphthongs: without specific instructions, TACT would count all letters separately, so that the word oeil [eye], for instance, would be counted as having three separate vowels instead of one vowel sound unit, the tripthong [œ]. The word souffrir would be counted as having eight letters in stead of six phonemes, including one double "f" and the dipthong [u]. Our method was fully documented in a paper presented to the Ninth Triennial Conference of the International Courtly Literature Society in Vancouver in July 1998, to be published as part of the Conference's Proceedings.

The current article deals with the application of this method to Old French courtly poetry texts rather than those of the troubadours, detailing both problems and results. The corpus is relatively small (comprising only 13 songs, all by 13th century trouvère Raoul de Soissons) and since statistics take on significance in direct proportion to the size and variety of the data base, the research represents only a small step, given the lack of a means of comparison. The subject of this article should therefore be seen more as an approach rather than the presentation of a wealth of data. Following are a description of the method used to establish a database, some of the problems encountered, preliminary results, and an analysis prompted by the statistical evidence generated.

Text Encoding

For our Old Occitan corpus, Elizabeth Brodovitch and I developed a system of parentheses and curly brackets {} to enclose groups of letters for identification in the text retrieval program. We used parentheses to enclose all the spelling variants of individual sound groups, and curly brackets to indicate that the letters composing the enclosed grapheme should be retrieved together. This allows letters to be retrieved as one sound unit if enclosed in curly brackets or as separate units if left unmarked. For example, vowel combinations can be retrieved as diphthongs and triphthongs, or separated in cases of hiatus (these situations were determined by syllable count). The same system was used for the current project; the text was typed on a word processor, then the "Search and Replace" tool was used to insert the various marker tags.

It very soon became clear that whereas Old Occitan is a largely phonetic language (that is, in most cases, the spelling reflects the actual pronunciation 9), Old French is not. It therefore proved necessary to separate stressed and unstressed vowels, free and blocked vowels, nasal and non-nasal vowels, diphthongs and triphthongs (both nasal and non-nasal), and eliminate letters that are not pronounced (such as the case of a word ending in "s" preceding a word beginning in "s") by enclosing them in double brackets. Given the number of diphthongs and triphthongs and palatal consonants such as "ll" ([j]), the final list of sounds identified numbered fifty, and this by no means represents all the phonemes of Old French - merely those encountered in the corpus. It should be noted, inter alia, that the marking system is easily adjustable to accommodate differing views on the pronunciation of Old French; the system can also be used for modern 10 French texts. Appendix I shows the TACT "alphabet", and Appendix II represents an example of a marked up version of a portion of one of Raoul's songs.

Processing the data

After the file comprising all of Raoul's songs had been marked up, it was run through TACT's Makebase program to generate a database. This database was used, in turn, to obtain phoneme statistics through the TACTstat program. The same procedure was followed for each of Raoul's songs individually. Next, all these data were transferred to MS Excel, a spreadsheet program, in order to transform the numerical data into graphs (see Appendix IIIa and III b for graphs representing all phonemes found in all of Raoul's songs). The next step was to study the graphs, looking for large peaks and valleys representing significant discrepancies as compared to the norm. The reasoning is based on the principle that if in all of Raoul's songs taken together, the sound [k], for instance, occurs 562 times or 3.39% of the time, the fact that this percentage jumps to 4.65% in a particular song (an increase of 37%) is potentially significant. Songs thus identified as yielding potentially interesting statistics were then checked for other significant deviations to see if there was a pattern, and the songs themselves were examined to see if an obvious explanation presented itself, such as patterns formed as a result of rhyme words in a coblas unissonans construction (a single rhyme scheme applied to all stanzas).

One problem that arose was that working with percentages obfuscates the significance of absolute numbers. For example, while comparing sound inventories of different songs, a rarely used diphthong such as nasal "ei" (as in "einsi" and "feintise"), which occurs only eight times in the entire corpus, showed a huge percentual increase when it featured not at all in one song and three times in another. It is clear that two or three occurrences of a phoneme in a given text are highly unlikely to point to the presence of a major stylistic device and thus, the significance in terms of stylistic impact would be very small in these cases. Yet to leave these phonemes out of the analysis would result in a distortion of the data. I therefore decided that in the initial stage of data analysis, percentages below 0.5% should be ignored; this resulted in the elimination of 15 phonemes (most of them diphthongs and triphthongs).

The distinctive phonetic profiles of individual songs are well illustrated by the line graph in Appendix IV, showing song R211 Chanter m'estuet and R1978: Quant je voi et fueille et flor. The latter song appeared to show interesting features, when compared to the "norm" as represented by the entire corpus (see Appendix V). It is clearly rich in "o"'s and "r"'s and also shows an increased percentage of [k]'s. To pursue this further, I consulted another set of statistics, also generated by TACTstat, showing percentages of initial and final letters. I found that R1978 scored within the top three for the total percentage for [k] and "r", as well as for initial [k] and final "r", and it had the highest percentage overall of "o"'s. While some songs showed higher percentages in the individual categories mentioned, R1978 was the only song to score high in all of them (see Appendix VI). I highlighted these three phonemes in the song, giving the result as shown in Appendix VII. Clearly, the phonemes forming the word cor are liberally sprinkled throughout the text. It is true that the sequence -or is part of the rhyme scheme, but since the rhyme scheme is in coblas doblas, i.e. the rhyme sounds change with each two stanzas, this is not considered statistically significant. In fact, R1978 is not the only song with rhymes in -or: R 929 has 11 of them, as compared to eight in R1978, yet the percentage of "o"'s and "r"'s (and of [k]'s) in R929 is about average.

An examination of the text (see Appendix VIII) revealed four occurrences of the word cors (the form cor takes in the cas sujet form): once in the first stanza, twice in the second stanza, and once in the third, as compared to an average of 1.5 times for the corpus as a whole (and the text has a rather low incidence of the sound "s": only 77% of the average). The database shows 18 occurrences of cors (but none of cor) in the entire corpus, with the following distribution: none in two songs, six songs with one occurrence each, two songs with two occurrences each, two other songs with three occurrences each, and four occurrences in R1978. The word features as a rhyme word in two songs, including R1978. All these factors taken in combination point to the strong possibility that the poet deliberately constructed his song in such a way as to reverberate with echoes of the word cor - a word that carries a range of meanings deriving from different Latin etymons and is therefore particularly suitable for use in word play and double entendre. The song is preserved in seven manuscripts (Winkler 55), which is an indication that it may have enjoyed considerable popularity in its time (only three other of Raoul's songs have a more numerous manuscript tradition). All in all, the statistical evidence seemed strong enough to warrant further investigation of this song, with the purpose of discovering the nature and effects of Raoul's "phonetic agenda."

Analysis

The form cor that derives from the Latin cornus, symbolising fertility and abundance, means "horn", "vigour", "power" or "strength" and has long been used as a metaphor for the male sexual organ 11. It is one of a number of quasi-homophones (at least in the cas sujet form) that include the invariable noun cors (from the Latin corpus) meaning "body" or "person", and cor, an early form of cuer meaning "heart" (from the Latin cor, cordis) - a key word in courtly poetry. Other meanings are "race" (from Latin cursus), "court" (from Latin curia), and "leather" (from Latin corium), with the related derivative cous (from late Latin coleus: "leather bag" > "testicles"). The three meanings "heart", "body" or "person", and "horn" form a sort of triptych, encompassing a wide range of human aspects: the spiritual, the individual and the sexual. The interplay between them and the (apparent) emphasis placed on one or the other infuses the text with a tension that fluctuates depending on the listener's interpretation at any given time, especially in a text that is so carefully ambiguous that one can never be quite sure of what is taking place, as we shall see.

The word cors in Raoul's song certainly appears to be open to interpretation. Given the phonetic echo of cor functioning as a signal, and quite possibly also the gestures and facial clues provided by the song's performer, the meaning of cor / cors becomes suspect, forcing listeners to consult their semantic "encyclopaedias" 12 for the various possibilities of decoding these words. In the first stanza (line 11), one would normally expect cors to mean the courtly lover's "person": as Philippe Ménard put it, the body was perceived as the carnal image of the individual, in its role as the seat of active life and affectivity 13. But other, less convential meanings are possible: as a cas sujet, it could mean the courtly lover's "heart" that is suffering the figurative pain of love, or it could refer to his "body" in the physical sense. Or, could it be the lover's "horn" that is causing him physical pain? In that case, an alternate reading of lines 9-12 presents itself:

                                                                Car souffrir
                                                                Ne puet, sanz morir,
                                                                Cors qui sent
                                                                Tel mal longuement

Because my "horn" cannot suffer such pain "while it is long" without "dying."

(The "pain" suffered by the "horn" is reminiscent of a line in a song by Guillaume d'Aquitaine cited by Kendrick (132): e venc m'en trop malaveg [or mal a veg]: and the pain of it / the pain to my penis was too much). It is of interest to note that the Manuscript C version reads tent instead of sent, so that with a change in the punctuation (which is of course absent in the manuscripts), the text can be read as: [Because my "horn", which is tensed/raised, cannot suffer such pain for long without "dying"]. If we accept that the mouvance of a manuscript tradition reveals more about medieval reception than any amount of theorising could, with various interpretations made evident by way of manuscript variants, then this case seems to suggest that the possibility of an erotic contra-text was recognised by the medieval audience. This also applies to the last two lines of Raoul's first stanza: Manuscript C reads souvent moul /mon lit [I often wet / my bed]. An alternative reading for souvent mueil / mon vis ... then presents itself: [And at night] ... I often wet "my screw" ... In addition, the phonemes of the words vis and lit yield the word vit, meaning penis, adding further fuel to the fire, as it were. The word vis appears four times in this song, and it is significant to note that it occurs only five more times in the corpus: once in each of five other songs. It can be seen that once the possibility of double entendre is recognised, the entire text becomes unstable. Even the song's conventional Natureingang (opening lines referring to aspects of nature) suggests erotic associations: the flower symbolising virginity, the leaf the usual biblical cover for the private parts.

In stanza II.3, cors is in the cas régime and given the ending in "s", it should mean "body" - a sense that fits in well with the context. In the same stanza, line 12, we find another occurrence in the cas régime suggesting either the body in the sense of "person" or in a literal sense. The latter interpretation clearly contravenes courtly convention: the lover is not really supposed to confess his "desire" to the lady's "body." Yet it could well be that this is exactly what he means to say, although he obfuscates the meaning sufficiently to keep up courtly appearances, as it were. The last occurrence of cors is in the third stanza, first line. Another cas régime, but the meaning again is fluid: "mercy," as a quality and a concept, is more likely to dwell in the lady's heart than in her body. Again, however, the meaning may well be that of the lady's "person." To complicate matters, we find two occurrences of the word cuer (III.14 and V.13) meaning "heart", undercutting the possible meanings of cors earlier in the text. Here we may note that the scribe who penned the Manuscript U version, or one of his predecessors in this branch of manuscripts, appears to have been aware of the ambiguities and sought to resolve them, if not clean up the act, by writing cuers in Stanza I, line 11 (meaning "heart", without any ambiguity). He writes cor in Stanza II, line 12, which is somewhat problematical given the cas régime; it could mean "heart", or represent an ungrammatical form of cors = body/person.

It seems apparent that the hypophones in this song function as a signal to alert the listener to the presence of a hidden subtext, much as do the ribald illustrations in some of the troubadour song books (Kendrick 105). As the audience puzzled over intended meanings, topics and isotopics, echoes of the word "cor" continued to edge their way into their subliminal conscience. Do these reinforce the key word "heart", or allow an erotic reading, or both? The performer may possible have tipped the scales in favour of one interpretation or the other, but both meanings remain possible throughout the song. The ambiguities are real enough to invite the double reading, yet slight enough to avoid giving offence and breaking the courtly code, in keeping with Huizinga's tenet that all games have rules that must not be broken: "... as soon as the rules are transgressed, the whole play-world collapses. The game is over." 14.

The three words cors, cor and cor ("body", "heart/spirit" and "horn" also exist in Occitan, the language of the troubadours, who appear to have taken full advantage of the opportunities for semantic games offered by this ambiguous set of courtly vocabulary and others like it. Laura Kendrick offers a convincing study of this phenomenon: songs that appear innocuous if not insipid at first sight, including apparently religious songs, in fact offer themselves to erotic readings 15. The courtly song normally invokes various elements of social and cultural significance, including the courtly code of ethics, the concept of vassalage, and the Marian cult. The contra-text arrived at by an erotic reading produces a textual interference that destabilises the song's meaning and forces the listener to determine semiotic strategies, hunting for clues to decode and choosing which textual level or levels to activate 16. The degree to which a text succeeds in turning its receivers into active participants in its decoding and interpretation can be said to be a measure of its success. Thanks to the computer's analytic capabilities, we may become active listeners and participants ourselves, some six hundred years later, and speculate upon the semiotic processes that allow us to understand a text on two levels at once - a topic that unfortunately exceeds the scope of this article.

Returning to Raoul's song, it may be noted that the song's metre and rhyme scheme are unusual: it is, in fact, the only one of its kind found in the entire body of trouvère songs, according to the statistics compiled by Mölk and Wolfzettel 17. The short, clipped lines which often do not provide enough space to accommodate syntactic units in their entirety and thus lend themselves to enjambement (as in Stanza IV lines 3/4), create a breathless, almost physical effect, conjuring up the heartbeat of a body in the throes of passion.

Another feature that merits investigation within the hypophonic context (and one that would be interesting to research with respect to the entire trouvère corpus) is the numerical harmony of the song's construction: each stanza contains four times four lines and nine times nine syllables - the latter adding up to a total made up of the digits 8 and 1, which themselves add up to nine. In fact, this property is a function of the number nine, since the sum of the digits of every one of its multiples always adds up to nine and thus, the number reproduces itself. For that reason, it was considered "circular" in medieval times 18. Nine, of course, is the square of the number three; squaring numbers was seen as giving them extension. The sum of the digits of the number sixteen, the total of the lines per stanza, is seven, or three plus four.

Number symbolism pervaded all aspects of medieval life including music, astrology and architecture. Medieval philosophy recognised number as a major key to cosmic secrets, and number theory was treated in detail by such notable authors as Capella, Boethius, Casiodorus, Bede, Alcuin and Hugh of St. Victor. Vincent Hopper, in his seminal work on number symbolism, pointed out that the Church accepted number theory in all of its forms, coordinating all previous number sciences and pressing them into the service of the True Faith (Stevens 25-6). Given the courtly poets' meticulous attention to all aspects of structure, the numerical disposition of their songs cannot be rejected as being accidental or incidental - as Stevens points out, it is very much part of the proportion, balance and harmony that characterised courtly poetry. It is of particular interest in the case of Raoul's song, taking into account the different interpretations the song offers: one on the courtly, spiritual level - the quasi-religious adoration of the unreachable lady, the plea for mercy, the lover's resolve to control his impulses (Et bonement atendrai / Com fins amis) - and the other, frankly erotic. This duality of meaning appears to be symbolised by the song's construction on the basis of the numbers of three and four, considered in medieval times as representing the spiritual and the corporeal, respectively. Twelfth-century philosopher Honorius of Autun put it thus: man has seven voices (the tones of the scale) since he is seven (4 = body, 3 = soul), and as a microcosm, he thus reproduces the celestial music (Hopper 95). Of Raoul's corpus of thirteen songs, seven are isometric; of the five heterometric songs, only R1978 has lines of four different lengths, and it is the only song featuring both tetrasyllabic and trisyllabic lines. The metre and rhyme scheme are as follows:

a b a b a b a b / c c d d / e e e e
7 4 7 4 7 4 7 4 / 3 5 3 5 / 7 4 3 7

In terms of the rhyme scheme, the stanza is constructed on a tripartite basis19: an opening section of eight lines (four plus four) consisting of 44 syllables with alternate rhymes in a and b, an interior section of four lines consisting of 16 (four times four) syllables with paired rhymes in c and d, and a final section of four lines consisting of 21 syllables (2+1=3) with rhyme e. The numerical symbolism extends to the word play based on the words cor and cors, numbering three and four phonemes respectively, and vis or vit, which number three.

Finally, it may be noted in passing that according to the statistics compiled by Frank 20 , the troubadour corpus contains only a single song with a metre and rhyme scheme similar to that of Raoul: a song by Peire Vidal 21, whose poetic activity took place between 1180 and 1250 22. While the rhyme scheme of Peire's song is similar to that of Raoul, his song is shorter by two lines, and the syllable count is different (7 5' instead of 7 4, and the last six lines are decasyllabic), so that we cannot classify it as a metric contrafactum 23. All the same, the rhyme schemes of the two songs are rare enough and there are enough similarities between them to support the assumption that Raoul may have known it and that it may have served him as a source of inspiration in writing his own song. For one, Raoul uses some of the rhyme sounds in Peire's song (-or, and -en / -ent), and many of the sentiments expressed are remarkably similar, even taking into account the predictable and prescribed content of the courtly love song. However, the most interesting feature of Peire Vidal's song in relation to Raoul's is a line in the first stanza that is conspicuous in its dense assonance and alliteration: Que.l cors e.l cor de mi e la valor / A... (She has my body and my heart). It seems quite plausible to suppose that Raoul was struck by the metre and rhyme scheme of Peire's song, took his inspiration from the opening line, and went on to do some creative sound engineering which may well have culminated in a stirring of the calm courtly waters.

Conclusion

I have described how the computer can be made to "listen" to oral texts, that is, instructed to compile an inventory and generate statistics of phonemes rather than letters through the application of a markup system forming part of the text retrieval program TACT. With the poet's full corpus serving as a norm, the statistics thus obtained were converted to graphs, to permit a visual inspection of significant deviations from the norm. An analysis of these data led me to conclude that at least one song appears to have a specialised "phonetic agenda", an assumption that seems substantiated by a closer inspection of the text. The hypophone identified on the basis of the statistical evidence appears to function as a decoding clue, as a marker signaling the existence of a hidden contra-text. The use of ambiguous vocabulary, distributional word frequency statistics, the nature of various manuscript variants and the numerical structure underlying the song all lend weight to this assumption. It should be acknowledged that in this sort of stylistic analysis, one risks succumbing to a scholar's worst temptation: that of trying to pummel facts long and hard enough so they finally fit the theory. While the presence of hypophones in a text can never be absolutely "proven", the statistical approach would appear to at least furnish reasonable grounds for drawing informed conclusions, and point the way towards stylistic phenomena that may otherwise not be apparent. The carefully engineered courtly song is a natural candidate for statistical analysis, and although one might consider that the research required seems like a vast amount of work in preparation for the analysis of one song, it has yielded a full data base which, in time, will bear a great deal more fruit.

The writers of the TACT manual observe that "Knowledge grows as the tools to acquire it improve" (Lancashire et al. 138). The quest for knowledge is, of course, self-perpetuating: the more we learn while using the latest tools, the keener we become on developing yet better tools to allow us to learn more. Given the development of voice recognition software, for instance, it should be possible to develop sound retrieval software, along the lines of text retrieval programs. Yet tools in and of themselves, and data per se, do not provide conclusions - it remains up to the researcher to analyse data and to formulate theories based on these data. And the broader the database, the more valid the data, the easier and the more objective the analysis. It is to be hoped that what Peter Ricketts and F.R.P. Akehurst have done for the troubadour corpus in compiling an electronic database of all known songs, will be done for the trouvère corpus in the not too distant future. We have the tools - all that is lacking is the database to tackle it with. What I have done, in a small way, is to give a hint of the range of the possibilities.