XENAKIS’ KEREN: A COMPUTATIONAL SEMIOTIC ANALYSIS

Christina Anagnostopoulou Faculty of Music, School of Philosophy, University of Athens, Greece chrisa@music.uoa.gr Chris Share School of Music and Sonic Arts esearch Centre, !ueen"s University #elfast cshare$%@&u'.ac.u( )arrell Con(lin )epartment of Computing, City University *ondon con(lin@city.ac.u( In Ma(is Solomos, Anastasia Georga(i, Giorgos +ervos ,eds.-, Definitive Proceedings of the International Symposium Iannis Xenakis (Athens, May 2 !", ....iannis/0ena(is.org, 1cto'er 2$$3.

Paper first pu'lished in A. Georga(i, M. Solomos ,eds.-, International Symposium Iannis Xenakis# $onference Proceedings, Athens, May 2$$4, p. 255/265. 7his paper .as selected for the Definitive Proceedings 'y the scientific committee of the symposium8 Anne/Sylvie #arthel/Calvet ,France-, Agostino )i Scipio ,9taly-, Anastasia Georga(i ,Greece-, #eno:t Gi'son ,Portugal-, ;ames <arley ,Canada-, Peter <offmann ,Germany-, Mihu 9liescu ,France-, Sharon =anach ,France-, Ma(is Solomos ,France-, onald S&ui''s ,USA-, Georgos +ervos ,Greece-

ABSTRACT 7his paper descri'es a computational semiotic analysis of %eren, a piece 'y >ena(is for trom'one solo. A (no.ledge representation method for pattern representation and discovery, previously applied to a large musical corpus, is here applied to the analysis of a single .or(. Patterns in %eren are found .ithin se&uences of musical properties, and a statistical model is used to test the significance of patterns .ithin these se&uences. 9n a second stage, follo.ing the process of syntagmatic analysis, a search is made for additional patterns, in .hich each element of the pattern is a musical segment. 7his reveals ho. the various segments are placed together in time to create larger scale semiotic structures, and facilitates the identification of the macrostructure of the piece. 9t is proposed that this method of analysis, sho.n previously to 'e effective for a large corpus, can 'e e&ually appropriate for the analysis of a single musical piece. 7he method is suita'le for the analysis of challenging post/tonal pieces, .here patterns .ithin the various musical dimensions and their intricate repetitions play a fundamental role in the overall music structure.

7hese decisions are primarily 'ased on the various music properties of a piece. and repetitions of patterns of segments.ith understanding pieces of music 'y identifying their constituent structures and ho. c. and segments are grouped into categories according to their similarity.ever has an a priori perception and interpretation of the piece and of the analytical method to 'e follo.or( is not to reveal or e0amine the compositional processes that >ena(is might have used. 'oth in a single piece as opposed to a homogenous corpus of pieces. its segments and its structure.ledge of the piece. ho. .or( on their o.here the music score is segmented. these properties can 'e represented in a general and a'stract .hich is analysed independently of compositional .processes . the piece is vie.Con(lin. as developed 'y DattieE . More specifically. . 2$$3. 7he aim is to loo( for melodic patterns .poietic level. is 'ased on the various musical properties.aesthesic.or(.Molino. features from e0isting ones.or perceptual .ithin a piece. 7his similarity.or(s.ay.een structures. our approach is paralleled to semiotic analysis. %eren ma(es full use of the instrument"s capa'ilities and stretches them to their limits.s for the e0pression of features of music o'@ects.ed as a se&uence of paradigmatic class la'els. 7his paper presents an approach to multi/levelled computational analysis of music . as opposed to more traditional tonal .hich e0plores the properties of structures . on the music o'@ect. for the piece? rather. the vie&point formalism . From a musicological point of vie. and ho..%6A4-.ay. '.een these properties in a transparent . INTRODUCTION %eren is a piece for trom'one solo. as produced 'y the composer. Analysts thus . For this analysis. %6A4-.ed.ay. to find musical patterns that reoccur . the focus is on the neutral level. 7he score. and provides constructors that can 'e used to 'uild ne. the analysis can 'e divided into the follo. 7he aim of the present . and ma(es choices that cannot truly 'e considered Cneutral". 7his formalism and methodologies are tested for the first time.ing stages8 a. the score. implicitly or e0plicitly. duration.ithin the piece in a statistically significant .2 1. such as melodic contour. . although not e0plicitly defined. 7he melodic representation and discovery level? 7he segmental representation and discovery level? 7he macrostructure level. has its o. and an analysis on the neutral level attempts to reveal internal relationships that e0ist 'et. At a second stage. intervals.ithin the musical properties of notes. rhythmic patterns and so on. composed in %653 for performer #enny Sluchin. to the degree that this is possi'le given the nature of this . *i(e other pieces composed 'y >ena(is during this period.is used to represent the music (no. an analyst ho. and in a contemporary atonal style. Although the intension might 'e scientific o'@ectivity.n poietic level. and thus it is crucial to 'e a'le to distinguish 'et.n e0istence. Bhile . one also ma(es related analytical choices? these are indicated and discussed in the paper. these are transformed in time.or(ing on the neutral level. 7his formalism allo. Formal music analysis at the neutral level is concerned .

2. and this goal can conflict .ithin the piece. and a segmented melody has type Se#. 2.2 T$e %!o&'e("e re)re*e!t+t o! A central tas( of a (no. %656-.een the printed score and the music o'@ect data type @ust descri'ed. METHODS 2.descriptive vs.the score in various .ith descriptive aspects? uncovering and evaluating the repeated patterns .ledge representation.hen encoding %eren . since M9)9 does not have an e&uivalent num'er for these notes. computational models that can 'e used to classify or group o'@ects.. 9n the application of data mining techni&ues to an individual piece such as %eren.a se&uence Se#. discussing the vie. For e0ample. %65A-.ith a pitch and a duration-. . Score Structur !" 7o represent %eren.F Music analysis is often concerned .points.Coo(.ays. 9t .ere mainly on the pitch encoding? ho. A general tas( in data mining of music.or(s. is the discovery of patterns or features that are in some .of music o'@ects all of the same type X. or . to represent &uarter tones. 2$$%-.ledge representation scheme for music is the computational inference of a'stract properties of music o'@ects. ho.1.ithin an individual piece.ith the e0plication of the organisational principles . 7he challenges faced . predictive. M9)9 includes information on each note"s pitch and delta onset times.ith that of systematic musicology. vie. )ata mining is mainly concerned . 7his slight disconnect .as convenient to first encode the score in a M9)9 format using a professional se&uencer .recursively. shared .Note-. esults section presents 2.. a se&uence of notes has type Se#. to represent non/discrete pitches such as glissandi.hich are discussed in detail in Methods. 7he some preliminary findings and discusses their potential musical validity. there'y grouping o'@ects that may 'e different at the music surface into the same paradigmatic class. then moving on to music (no.#ro.Se#.X. 2. 9nformation on phrases. 'reaths and fermatas . a data type that permits a hierarchical structuring of a melody is used. For .segment. Data mining is the field of study in computer science concerned .<uron. can 'e productively e0plored .ith the discovery of interesting patterns in large data'ases.n and )empster. 7he field of comparative or systematic musicology .ith the goal of ma(ing generaliEations 'eyond the particular analysis piece .ith statistical approaches .as also added as te0t annotations.or( is presented.ith a corpus of .Cu'ase S> v. A music o'(ect is a Note .ever. 'eing concerned .ay outstanding in an analysis corpus as related to a comparison set of pieces.point formalism and the techni&ue for constructing ne. since they form an integral part of the composition.ith the o'@ectives of data mining has led to some interesting methodological issues . Stein'erg Media 7echnologies Gm'<-.Note--.e are content .ith inferring predictive models8 for e0ample. concerning the encoding of the piece to M9)9 format and the challenges faced.s8 first all the methodological frame. . and ho. and to structure .hich served as an intermediate representation 'et. 7he rest of the paper is organised as follo.

ithin se&uences.point . and 1pus 3A &uartets. Do.points? #ottom8 vie.is called a melodic vie&point.ithin unrelated pieces.0 P+tter! * "! / c+!ce 9n systematic musicology studies. e&)o !t co!*tructor* 7he vie. in <uron .or( for computing a'stract properties for o'@ects . )a. a vie.point elements constructs a segmental vie.K e0ample. % &uartet.points 2. Significant patterns are sought .point primitive duration vie. fI H' !%e(.point elements.ith e&ual pro'a'ility . )'I H!e&. a (no. A pattern occurs in a piece if it is contained . ) tc$ (ur+t o! -o(12 *$+)e H !ter. )I H*e'ecte(. 2.+'.point constructors used to create composite vie. . For e0ample. ).or(. Do.points formalism provides some primitive features.represented in terms of their e0pected count .points and constructors that have 'een used in this study of %eren.ith domain Se#. . the analysis corpus is the first movement of #rahmsG 1pus 4%. and the comparison set contains the first movement of #rahmsG 1pus 4%. 7he vie&points (no.points and constructors used for the analysis of %eren.point elements for a segment applies the function f to ) vie.o ) vie.here X is a varia'le that can refer to any type of music o'@ect. and can therefore 'e placed in an e&uivalence class. returning ne.point . and a vie.<uron. 7hese properties are called vie&point elements.ithin the analysis corpus? those that are over/ .ithin the corpus.point primitive modulo %2 function <uronGs .point elements if f is true 7a'le %8 Jie.Se#.points.melodic shape function su'traction of t.ledge representation scheme might infer that t. composite vie. 7a'le % provides a list of primitive vie. 'ut should not 'e so general or trivial as to occur .ithin a piece.points )a and )' #oolean measure of vie. it is necessary to identify those features that are in some sense salient and uni&ue to the particular analysis piece or corpus . Significant deviation .o segments have the same contour shape or melodic density.point elements the direction of movement in ) vie.ithin the se&uence of vie.point from a melodic vie.is called a segmental vie&point. Patterns should reoccur .points are created using functions called constructors? these are functions that ta(e vie. )I H' /t. 7he domain of a vie. and some 'asic mathematical operators such as modulo arithmetic. only over/represented patterns are e0plored-.ithin a comparison repertory.point is of the form Se#.X-. 7he e0pected count of a pattern can 'e 'ased on its relative fre&uency .%663.points.2$$%-.points as arguments. A pattern is a se&uence of vie.here.point ) the set of ) vie.ith domain Se#. 7op8 primitive vie. )I Hco!tour. such as ) tc$. 7he attainment of this o'@ective can 'e assessed using a statistical hypothesis testing frame.point elements of the piece.point ) change selects ) vie. fI primitive pitch vie.Note-.Note. vie. (ur+t o!. 7he count of a pattern is its num'er of occurrences . )I H*et. 2$$%-.or under/. ). From these primitive vie. For e0ample.ledge representation scheme provides such a frame.point elements pairs the elements of vie. )I Hco-)o*e.

A re&uirement of the chi/s&uare significance test is that the comparison repertory is of sufficient siEe to include the pattern a sufficient num'er of times.ith a variance that depends on the period of the pattern . contri'ute to the segmentation. consecutive notes . 2$$%? Apostolico and Crochemore. 7his process can still produce a large set of patterns. for a specified score structuring and vie. in places. and in the studies 'elo.er3 7o find all significant patterns in %eren.for e0ample. . A23LA26-. <ere. and this implicitly limits the method to short patterns.Con(lin and Anagnostopoulou. 7he process resulted in KF segments of significantly different lengths. the position that this is far less of a pro'lem than reporting too many spurious patterns is adopted here.2$$3.noted that the 'ac(ground distri'ution of counts for a pattern may 'e appro0imated 'y a normal distri'ution .or( is the computation of the 'ac(ground pro'a'ility of a pattern.o'@ects. and there'y not reporting potentially significant patterns. 7he construction of the 'ac(ground model from the analysis corpus itself raises the possi'ility of overestimating pattern pro'a'ilities.hich guide the process of manual segmentation of the score into Se#.1 Se"-e!t+t o! 7here are several simple segmentation points in this piece. Central to this analytic frame.ay . it is necessary to use alternative procedures to estimate the e0pected counts of patterns. events 2%2L2%A. Con(lin and Anagnostopoulou . <o. 1ne . . many times one e0pects to see the pattern in an analysis corpus. For e0ample. From this distri'ution. 'oth of . 1ne should stress here that there can 'e a num'er of . a type of high/order Mar(ov model is used. the significance level has 'een slightly reduced. and . Bhere 'reaths and fermatas are scarce. and each one . Bithout a large comparison repertory. Patterns not meeting the significance level are discarded.point of the appropriate type. constructed from all su'/pattern counts in %eren. the melodic phrasing indications and the dynamics. significance levels as p/ values are computed for all patterns. 2. 7hese tend to 'e at every single note.hich are very meticulously notated 'y the composer. and all patterns meeting a specified significance level are reported to the analyst.Note. to reveal a larger set of patterns.point se&uence is computed. and all patterns occurring more than once are found using an efficient algorithm.4 from the e0pected count can 'e measured using the chi/s&uare statistic.ay is to design an analytic method to directly model the e0pected pattern counts . 2$$2-.ould yield different results.2 P+tter! ( *co. the vie.ays to segment the piece.hich naturally 'rea( the score into smaller units of diverse lengths. 7hese are not interpreted as indicative of segmentation points in our analysis.ever. 7here are a small num'er of 'reaths and fermatas used for articulation purposes. for a fe. For this analysis one musically sensi'le . 'ut there are certainly other meaningful segmentations.as chosen. 'ecause this pro'a'ility dictates ho. >ena(is has incorporated 'reath and fermata mar(ings.its amount of self/overlap-. 2.

this is translated to a num'er of melodic vie. intervals of pitch classes -o(12 and vie.points.and the longest significant .s an e0ample of the opening four/note figure of %eren.ould segment the score according to repetition.ill produce different patterns. 9n the follo. the segmental.hich is made possi'le 'y using the formalism. these stages are discussed in detail. . 7o illustrate some melodic vie. 7he follo.points.ith the pattern discovery algorithm shortest and longest significant patterns .$%. presenting a selection of the results.hich has three stages8 the melodic.een segments is e0plicitly defined. to constructed and gradually more a'stract ones. such as ) tc$ and (ur+t o!.ill illustrate some shortest and longest significant patterns . 9t should 'e noted that there are many representation possi'ilities of the musical surface using constructed vie. ANALYSIS AND RESULTS Using these methods an analysis layout is created . 7he melodic vie.are revealed.points considered ranged from 'asic ones.e' Paradigmatic analysis. and ho. the first part of a semiotic analysis. 'ut .1.point se&uences.are reported.3 therefore a further filtering is applied to the set of all significant patterns in several applications .e have found that the e0trema of chains of su'/patterns are of the most interest.hich appear to have the most apparent musical interest . 0. Me'o( c 'e. . and the macrostructural. melodic intervals. First. and each vie.independent of compositional processes-.points using the constructors ' !%e(.points and patterns .no other significant pattern is a su'pattern.o main diferences.ith real music. only the shortest significant .point representation and the a'straction .su'/patterns to no other significant pattern.ing results .point patterns. 7his paper reports only on a small num'er of vie.ell as some results that demonstrate the vie. 7herefore.point . Bhile the concept and rationale here is similar.ith a p/value threshold of $. . 7a'le 2 sho. and classify the resulting segments according to similarity.or segments. !e& and *e'ecte(. there are t. similarity 'et.hen using the pattern discovery algorithm . 0. as class e&uivalence segments share the same vie. the score is not segmented.ing. Second. such as pitch classes. as .

ing.ill henceforth 'e referred to as section #. 0.A%-? '. 7his is a section comprising fast. 'ut that their co/ occurrences into patterns are significant. CK. containing these three notes. a large num'er of shortest and longest significant patterns.4A-. 7a'le 2 sho. CsK.3A-. %656.4$-.point therefore had slightly more general results than the ) tc$ vie.+'6 ) tc$: 9co-)o*e6 9 !ter.1 7 . GK.points are investigated. in addition to the a'ove.here more a'stract vie. GnF.M9)9 num'er 33-. Be o'served that all reported significant patterns made use of only very specific pitches8 a. disregarding octave information .ere reported. 7he large num'er of the patterns that consist of various com'inations of FsK.45-. A num'er of patterns . more patterns additionally made use of )F.point se&uences.-o(12-.3$-. AsF. 0. GK.32-? a fe. )K.s the opening motif of the piece.3%-.+ ) tc$ 7his 'asic vie. Score fragment reproduction from Mditions Sala'ert.+'6 ) tc$:6 -o(12: 9co!tour6 ) tc$: 9co!tour6 (ur+t o!: 45 5 u!(e/ u!(e/ u!(e/ u!(e/ 24 7 .point. GsF. comprising the opening motif.1.44-. a num'er of patterns. #F. = 22 8 . and . 9nterval and contour vie.A P tc$ Hco-)o*e6 ) tc$6 -o(12I 9 !ter. . Bhat these results convey is not that these pitches are important on their o. translated into a num'er of melodic vie. .n . 9n the follo.1 11 . 7his vie.point is more a'stract than ) tc$ in that it chec(s for patterns of pitch classes. also used GsF. made e0clusive use of the pitches FsK . AF.hich is encountered five times in the piece.as the opening motif of the piece8 CK. c. = 24 7 1 1 < < 7a'le 28 7he first four notes of %eren. Paris. and #K are all found in the section of event num'ers KF$/3K4. GF.though they might 'e-.46-. more occurrences are found that are e&uivalent to this motif.43-.1.> 9co-)o*e6 ) tc$6 -o(12: 7his vie. 1ne of the patterns reported .point returns the pitch of an event.points are undefined for the first element in the se&uence. Do other pitch patterns . isochronous events of these pitches. #K.

161: .261616. 0.16.ith the results of the ) tc$ vie. it is interesting to note that >ena(is realised most of the interval patterns to the same pattern of notes in each case.261616. .2: . in fact use the same three pitches. creates the interval pattern 9.2-.c 9 !ter. 9161: .ing interesting 9 !ter. . 9t is very interesting that no other intervals . Most patterns instances are distri'uted evenly throughout the piece.7a'le 2-.F%-.hich has 'een discussed in relation to the previous vie.161: . and several others.in various com'inations.262: . G and #.GI . M0amples of this in shortest significant patterns.as mentioned a'ove in relation to the previous vie.26. 7his is in accordance .1616. and a fe. e0amples of longest significant patterns are8 91616.162: .point a'ove.A and %% .2-. 7hese patterns .Af. 1ne realisation could 'e the pitches Fs.1616. and others.1. <o.2-. and to a much lesser e0tent N3 and N2. 926. 9. 9. G.26.%$-. N5. 916.point loo(s at patterns of pitch intervals in num'ers of semitones.points.Fs.ere reported. 9.for e0ample .s for transposition. and it is also found here as part of longer patterns .26261: . the most interesting 'eing a large group of shortest and longest significant patterns .16.1: . patterns .16.16.( 9co-)o*e6 9 !ter.261616.16.16.ere reported to 'e part of the significant results8 7he intervals NA. NK.%5-..161616.2: .16.1626. tend to appear in the same distinct fast section # that .2I .261616.+'6 ) tc$:6 -o(12: .ithin the glissando passage of the piece.ays the case in the score 'ecause this vie.1: @ust mentioned-.2626.16. #. 9161616161: .ere also found in the same distinct section # of the piece.1616. 0.%K-. 7he ma@ority of the interval patterns.ever.26.point allo. and '.hich made e0clusive use of the pitch classes 3. 7he results also contained the follo.count F-.1626.hen realised to actual instances of pitches.1.2626.hich is .16. 7he patterns reported for 'oth shortest and longest significant patterns made use of the intervals N%.FK-.26.count F.ith their count in %eren. ho.+'6 ) tc$: 7his vie.points .16. 9. N6 and a'ove did not appear at all in the set of significant patterns.5 . 7he longest patterns.4-. are8 916.1: . the pattern 9.16.ever.n.ard trend8 9.count K-.ith a do.+'6 ) tc$: patterns8 a.16. 7he initial motif HC. 91616. although this is not al.16. N4.1:.

ith more . Furthermore. 9f .ith section #8 these patterns are used throughout the piece apart from that specific fast section.o or more consecutive up.here in the piece. as e0pected.6.point element O ..count 2$. 7he shorter versions of these patterns are found every.4 and A.6.e notice a distinction .6.ith the t.A.K.ithin the resulting patterns are mainly8 %.ith t. 7he intervals .:.ith e0clusive do.e have8 F8 C/to/)s 68 C/to/A.ere no patterns .6.. events 22F/F55 and KF$/3K4 .events KF$/3K4-.6. 7he interval )s/to/A is the tritone. 7his is in accordance .contiguous P elements-.:.ard motions .here in the significant pattern set. and they sho.o fast isochronous sections in the piece.6.%%.ever. .6. tends to avoid the intervals 2 and 3 .n. 1scillating te0tures O patterns .: .contiguous Q elements-. 7his is interesting 'ecause >ena(is.o intervals 565 appear only in the first and third sections of %eren. <o. 7hese . in the fast passage section # ..6.e call #-.: . t.hich .count F2./ 9co!tour6 (ur+t o!: Analogous to pitch contour. Another interesting o'servation here is that the intervals F and 6 do not appear any. 7hese are symmetric to the unison.. 0. 7here .in this piece.een each rhythmic value .hereas the longer versions.K. 3 and 6 are omitted completely from the isochronous fast section #. 7his vie.ere part of shortest and longest reported patterns respectively.n motion at every vie.e 9co!tour6 ) tc$: 7his vie.6 7his vie.ith interchanges of a melodic contour of up and do. there are more occurrences in this case. Another interesting result is that patterns 'eginning .6. and only very fe.een pitch classes in num'er of semitones. 0. .ere found throughout the piece.%%.and 9. Again here .point pointed out the t.5. as e0pected.6. Patterns .o symmetric structures8 %. .1. as reported in the interval results a'ove.ith its predecessor.. starting from pitch class C.ere.point descri'es melodic contour.ard trend also appeared8 9. instances of the long pattern 9=6=6=6.ith the results on intervals a'ove8 'ecause contour is more a'stract representation than intervals.o intervals..ere very prominent in the results and . the three intervals.point loo(s at interval patterns 'et.e ta(e a possi'le realisation of these t. F.5.4. .tritone.6.ith same motion .6.1. a very long pattern 9=6=6. duration contour involves the comparison 'et.

ere constructed and considered. contour points.ould involve the order in .ere8 9' !%e(6 ) tc$6 (ur+t o!: 9' !%e( 9 !ter. . 7a'le F sho. 7he shortest significant pattern 9=6=: appears 44F times throughout the piece.point se&uences. 7he vie.points lin( constructedvie. . . .as manually segmented as descri'ed in Section 2.K.ay only in the second fast section. the melodic shape as this . there are K e0tra occurrences pre/#. there is only one occurrence. all patterns reported are from the fast section # mentioned a'ove. For the segmental level.ith (ur+t o!.n as syntagmatic analysis. 9n the second one. Several segmental vie.hose elements are not single notes.points . using score indications for segment 'oundaries. 7he second step.points 'ased on ) tc$ .ould 'e the score segmentation and the categorisation of the segments. this is translated to a num'er of segmental vie.hich is outside that # section. 9n the first one.+' ) tc$: (ur+t o!:: 9' !%e(6 9co-)o*e 9 !ter.%$ than A$ elements. 0.hich is (no.points reported here are8 the set of pitch classes. in one of the patterns.2 Se"-e!t+' 'e.as introduced 'y <uron .hich the segments appear. and lifted selected contour at ne. . 'ut segments of notes. the first step of a semiotic analysis .%663-. the score . occurs %A times in an overlapping . Some other tested vie.point.points .+'6 ) tc$:6 -o(12:6 (ur+t o!: 7hese vie.hile in the third vie. 7he algorithm then loo(ed for segmental patterns? patterns .o segments of %eren. contour of density. and 2 post/# . and ho.e' As mentioned a'ove. contour of duration.s an e0ample of the first t.the rest 'eing found at section #-.

the length of the segment. 7op8 segment 3? #ottom8 segment K2. Paris.4II in %eren. 7he p/value threshold for all segmental pattern e0periments .ice or more .ere8 pattern 9956162:: occurred at events 3 and K2? pattern 9915611:: occurred at events 5 and%3? pattern 99867:: occurred at events %F and F4. 'ut segments can differ in any other property8 the order and the repetition of the pitches in the segment.o segments of =eren. translated into segmental vie. %656. 7his vie. 7his .s the set of pitch classes encountered in a segment. and so forth. .point . .s interesting musical relations. is short and there .o instances of the segmental pattern HH$. Mditions Sala'ert.2. Score fragment reproduction from %eren. %656.point sho.6<: u!(e/ 95616867615: co!.points are undefined for the first segment in the score.as ignored? any pattern that appeared t. Mditions Sala'ert.+ 9*et6 9co-)o*e6 ) tc$6 -o(12:: 7his vie. #elo.e u!(e/ 9.point sho. rhythm.as reported. Score fragment reproduction from =eren.here pitches in a segment are shared.e? < 9<6. 7he only repeated patterns found here .ere of length %8 no longer pattern appeared more than once in %eren# Figure %8 7. )ensity contour and duration contour vie. the most interesting patterns ta(en out of the segmental stage are reported8 0.ere not enough segments to @ustify a significance threshold.ith this vie.point se&uences.as 'ecause the piece. All results o'tained .: < 7a'le F8 7he first t. .%.hen segmented.%% 9*et6 9co-)o*e6 ) tc$6 -o(12II *$+)e 9co!tour6 (e!* t3: 9' /t69*e'ecte(69co!tour6) tc$:6 9!e&69co!tour6) tc$:::: 9co!tour6 (ur+t o!: 956867: co!c+. the register. Paris.

7his result is in line . and a different segmentation might have resulted in different results altogether. as defined 'y <uron .s a classification of all segments according to melodic shape. .e: .point e0amined a'ove.e?6 (e*ce!( !"6 co!.as 9(e*ce!( !"6 co!c+.ith %3 occurrences.ith the melodic contour vie.e6 co!. 9t is important to note here that these results are 'ased on a particular chosen segmentation. . A set of rules relating the first and last pitches in a segment to the mean of the internal pitches leads to a segment 'eing in one of nine shape classes. .2. 0. 7a'le K sho.as reported . 7he most common pattern that is met is 9(e*ce!( !":. occurring at segments K and 23.as a num'er of descending patterns found.point descri'es the overall melodic shape of a segment. 9n 7a'le K. .%663-.ere also important. a representation of the .e?: and 9co!c+. 7he longest pattern that .> *$+)e 7his vie.%2 Figure % illustrates the t.o instances of the pattern 9956162::8 segments 3 and K2.ith %$ and 5 occurrences respectively.here there .e?:.hole piece in terms of segment num'ers and their pitch class sets is displayed. 7he one/element patterns 9co!.

( 9' /t6 9*e'ecte(6 9co!tour6 ) tc$:6 9!e&6 9co!tour6 ) tc$:::: 7his vie.hen there is a change of contour.o segmental vie.e $or A(e* (e*ce!( !" co!. .ards and 'eginning of the fast section #.hich has 2 occurrences.e? (e*ce!( !" 7a'le K8 A syntagmatic analysis of all segments in %eren.hich appears at locations 23 and 2A.e co!c+.e (e*ce!( !" +*ce!( !" +*ce!( !" +*ce!( !" +*ce!( !" +*ce!( !" (e*ce!( !" co!c+.e? co!.e co!.of the contour value .as 99<6.6<6. 1ne of the patterns discovered here is 9<6<6<:.e co!.c 9co!tour6 (e!* t3: 7his vie.s the segment se&uence .num'er of notes divided 'y length of segment.:69.e (e*ce!( !" (e*ce!( !" co!c+.e co!.e? (e*ce!( !" (e*ce!( !" co!c+. in segments 24L26.e? co!c+.points8 set of all pitch classes and shape.ere found? 1ne of them .e? (e*ce!( !" co!. 0.ere %4 repeating segments .2.e? (e*ce!( !" +*ce!( !" (e*ce!( !" co!c+. 7his means that the longer pattern 9<6<6<6<: occurs only once in the piece.%F Segmen t % 2 F K 4 3 A 5 6 %$ %% %2 %F %K %4 %3 %A %5 %6 2$ 2% 22 2F 2K 24 23 2A 25 26 F$ F% F2 FF FK F4 F3 FA F5 F6 K$ K% K2 KF Pitch $lass Set )ie&point 956867: 95616867615: 95616867: 9162616468676@615611: 915: 956162: 9162616468676@: 915611: 95615611: 91626@615: 956264611: 91626268676@615: 9867: 9561626468676@611: 95611: 915611: 9562616264: 9156162626468676@615611: 956162626468676@: 95646867611: 91626468615: 9162676@611: 956162646867615611: 956162616468676@: 9164615: 956164611: 9561676@: 956468611: 9564686@611: 90626468611: 95616468676@: 9561626162646867615: 91626468676@615611: 956162616268676@615611: 9867: 95686@: 956061626@615611: 926268: 95616264686@611: 926468611: 956162616264686@611: 956162: 91626468676@615: Shape )ie&point co!c+. . at segments F/K and %A/%5.e? (e*ce!( !" +*ce!( !" (e*ce!( !" co!c+.lifted.2. in terms of t.een successive segments.'et.point compares the density .e? co!. 7his is interesting 'ecause this 'uilding of density appears at the place leading to.hich featured in a num'er of other results in the melodic points a'ove.e co!.e? (e*ce!( !" co!.e? +*ce!( !" (e*ce!( !" co!c+. .::.e co!.point sho. 7here . 0. .

Figure 28 7. N4.ards the cohesion of the piece.n. 7he distinction of this section from the rest of the piece results in a loose A#C.o instances of the segmental pattern HHP.ing features8 Pitch patterns ma(e use of pitches FsK. is indicated 'y the results in the other t. NK. . At the same time. %656.here all the compositional material is first presented? *ect o! B6 'et. also seen in the melodic vie. 7he same goes for pitch class patterns.hich ma(es use of the intervals N%. . Finally. .here the piece .hole section is represented as one segment in the segmental level. some of the more a'stract patterns are found throughout the piece. the longest significant patterns that ma(e use of the intervals N%.een section # and the rest of the piece.o instances have an interesting similarity 'et. and the second 'eing this do. contour duration patterns single out this section.een them. .here the music material is transformed8 certain features or patterns disappear.n. . and all of the melodic contour patterns. N4 appear again in this section. 'et. GK and #K.hich has 'een named # a'ove. 'et.hile others 'ecome more prominent and intensify.0 M+cro*tructure 7he third level of the analysis.point results a'ove.ith duration longest significant patterns. the melodic and the segmental. Mditions Sala'ert. hythmic values are e&ual and very fast./. 9n terms of intervals.%K Figure 2 demonstrates this pattern and its instances.een events 3K3LA55.here8 *ect o! A. Score fragment reproduction from %eren. 7hese results seem to suggest that there is one section in the piece .P. Paris.o levels./I. so the . as do the lin(ed pitch . . some of the shortest significant pitch and pitch class interval patterns.here there is a clima0. 7his section. 7op8 segments F and K? #ottom8 segments %A and %5. some of the shortest pitch class patterns.een events KF$L3K4. . .H/II in %eren.een events %LK26. 0. 7here are no 'reath mar(s or fermatas to 'rea( the intensity. the first pattern in each 'eing the characteristic melodic pattern . NK. has the follo.inds do. the macrostructure. . 7he t.ard characteristic movement.hile most of the results differentiate 'et. and thus contri'ute to.ith very specific persistent patterns? *ect o! C. as opposed to the rest of the piece. 7hese include some of the shortest and longest significant pitch patterns.

melodic patterns are found .K.or(. and its symmetric structure of A. 'ut rather much more comple0 structural units.e have called #.n. 7he vie. 7his means that the composer himself did not follo. also ma(es use of this interval structure . <o. are heard in this section. 'ut are diluted . the concept of symmetry has 'een a very important one in his .ards.ithin the various musical dimensions.5. the findings from the segmental level . At the same time. using different articulation accents. do. 235-.n.hose elements are multiples of a unity" . 9n %eren. starting from the 'asic ones.hich entails an additive operation and .ever.hich in fact is specified to certain pitches. and especially to any group . lin(ed intervals . . contour of duration.ith his compositional material? not so much of specific pitches or classes of pitches. For >ena(is. 9n the rest of the piece those interval structures are also very noticea'le. p.or(. >ena(is has 'een very selective . For e0ample. the do.ed that this section . most of the vie.point formalism and the pattern discovery method .Papaioannou. As a result. 1. 7he very characteristic opening motif of the piece is a statement of this interval structure. 'ut of intervals and interval structures. or even the simple melody. section # all formed part of a single segment.s one .ere not as interesting as the ones from the melodic level.ing pro'lems8 first. although these are salient too. %66K-.ay of formalising symmetry8 sieves are symmetries in Cany set of characteristics of sound or of . As a second step.e o'served that some of the melodic patterns .ards and up.4. the 'rea( points of the score to enclose his compositional ideas. As a first step. 7he manual segmentation in the segmental stage a'ove.ing the composer"s indications for 'rea(s. . and gradually progressing to more a'stract ones such as pitch classes. .ing the composer"s indications.as not as interesting as section # in terms of patterns and their significance.ith some repetition of pitches. since the level of representation can 'e tuned at any level of a'straction.ard movement.ith section #8 fast isochronous rhythmic values. .%% are very prominent in this . and no 'reath mar(s.e use supports the discovery of some of these structures and comple0ities.points presented in this paper sho. pitch and duration. Bith the theory of sieves he sho.ell/ordered structures. the piece is segmented follo. ta(ing as 'asic information pitch and duration. All permutations and com'inations.e can also o'serve a su'section 'et. segments .e discovered spanned across segments. melodic contour.een events 22FLF55. 9n the section that .ith other compositional material.for . and various segmental patterns are discussed. . Another characteristic pattern used in this piece. 7he interval structure of %. pitch intervals. although follo. A characteristic of the music of >ena(is is that the unit of composition is not the note.ere too une&ual in length to 'e a'le to demonstrate repetition in the segmental level.e hear. %662.ith duration. DISCUSSION 7his paper has presented a computational analysis of %eren 'ased on representation of music properties and discovery of interesting patterns.%4 9t is . had the follo.hich can differ from piece to piece .hich has at a first glance a lot in common . this structure is all .orth noting that in section A .>ena(is.

might have 'een to use the longest significant patterns of certain vie.% . is that the piece %eren itself .intervals %.hich point out a loose structure of A#C. An 1'@ect/1riented model for the >ena(is Sieve for algorithmic pitch. String pattern matching for a deluge survival (it. 9n Proceedings of the International $omputer Music $onference.as used to construct an analytic 'ac(ground model for evaluation of the statistical significance of patterns. Melodic analysis . .A. since these are so carefully notated 'y the composer. editors. especially related to pitch classes and intervals. REBERENCES Apostolico.K. especially related to durations.o residual classes.hile compositional intentions . . 2.e selected a small num'er of results.%3 e0ample. Segments in %eren seem to 'e te*tural.earning. dynamics and musical te0tures? cthe macro/strucure revealed .points to denote segment 'oundaries.K. has 'een segmentation.points. etc.5.the use of dynamics in the e0periments.er Academic Pu'lishers.-.2$$2-. 7his suggests that the compositional material had 'een distri'uted across segment 'oundaries. C. Another pro'lematic issue. #oston. Machine . G. . A feature of this . Pardalos. .ay for the segmentation. +and'ook of Massive Data Sets.%%-.A.ith our results here could 'e further e0plored 'y studying the musical material in sections A and C in more depth. .e segmented the piece at places indicated 'y the composer as 'rea(s.e only considered a small num'er of vie. 7he se&uence of intervals %. .ere not considered. At the same time. and it is o'vious that their role in the overall structure of the piece should not 'e neglected? 'the use of sieves in >ena(is" music might not only stop at pitches and intervals. 5. . as mentioned a'ove.or(. pages 3F/A$.4. AriEa. CONCLUSIONS AND BURTHER CORK 7he results presented in this . A more relia'le .. and Mauricio.ith segment classes. and out of those .. Miami. and also to do inter/opus pattern discovery .. 9n the futre .intervals 4. events %AL24.e are currently e0ploring some melodic vie. to appear. 9n A'ello.ithin each piece. Further . F.%% could 'e thought of as a sieve that comprises the union of t. ande aim to investigate relations 'et.2$$K-.points.points that might reveal changes in te0ture.as o'vious that this segmentation did not yield the 'est results in the segmental patterns. Con(lin. rhythm and parameter generation. this paper has demonstrated that this generalised method of representing music and discovering patterns might 'e a suita'le method for analysing single atonal pieces.or( are still very preliminary8 . it .ould li(e to use a . discussed in Methods.2$$3-. and Crochemore.e . pages %4%/%6K.2 . ).or( also includes the follo. A.ithin the piece.. 6%L65.and F. M. Although . Be have demonstrated that there are interesting music relations . MA. 4.een values and patterns in various vie. =lu.ider set of >ena(is pieces to create 'ac(ground models. P. and re&uire a special approach to auto/segmentation8 .ing aspects8 a. given the compositional style of >ena(is.

and Anagnostopoulou.%A Con(lin. 7he melodic arch in Bestern fol(songs. %$8F/2F. Sieves. . Perspectives of -e& Music. <avana.2$$3-.%663-. Bhat is a musical featureR ForteGs analysis of #rahmsGs 1pus 4%.point patterns. .%6A4-. . Coo(. >ena(is.. ed. 9annis >ena(is. Molino. revisited. Paris.ason )i0on. pages KA6/K54./0MS 1ournal on $omputing. . Sig0roni Mpo0i. >ena(is. ).%-845/A5. . epresentation and discovery of multiple vie. %656. Mnglish translation 'y . .%662-.%653-. ). . DT. I-. %5. Con(lin. %. DattieE. 8na afieroma tou 8thnikou Metsoviou Politehniou pros enan apofito tou. Computing in Musicology. in Dational 7echnical University of Athens. A. 10ford University Press.2$$%-. evised Mdition Pendragon Press. . ). Union gSnSrale d"Sditions. ).ondements d5une s6miologie de la musi4ue. Fait musical et sSmilogie de la musi&ue.%66$-. 9. 9n Proceedings of the International $omputer Music $onference.ormali7ed Music. Paris. .. Cu'a. Papaioannou. C. 25. Musi4ue en 1eu.K-. Ac%!o&'e("-e!t*: Special than(s to Ma(is Solomos and . %A8FA/3%..%6A4-. %eren pour trom'one solo.. Athens. and Anagnostopoulou. <uron. Segmental pattern discovery in music. 9. 9. . .F-8254/26F. A 2uide to Musical Analysis. ahn. . C. >ena(is. . 9. Do. . . Iannis Xenakis. Mditions Sala'ert.. D. <uron.2$$%-.%65A-. Music 3heory /nline.%66K-.