You are on page 1of 47

Yale University Department of Music

Generative Music Theory and Its Relation to Psychology Author(s): Ray Jackendoff and Fred Lerdahl Source: Journal of Music Theory, Vol. 25, No. 1, 25th Anniversary Issue (Spring, 1981), pp. 45-90 Published by: Duke University Press on behalf of the Yale University Department of Music Stable URL: http://www.jstor.org/stable/843466 . Accessed: 01/05/2013 13:42
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Duke University Press and Yale University Department of Music are collaborating with JSTOR to digitize, preserve and extend access to Journal of Music Theory.

http://www.jstor.org

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

MUSIC THEORYAND ITS GENERATIVE

RELATIONTO PSYCHOLOGY

and FredLerdahl Ray Jackendoff

For a number of years we have been developinga formalizedtheory of tonal music, basedin part on the goals and methodology, but not the. substance, of generativelinguistics.' In addition to being of interestfor what it says about music, the theory also has certain propertiesthat recommendit as a model for more traditionalissues in psychology. The fragmentof the music theory present article describesa representative and then shows how it bearson the theory of visualperception.
GOALSOF A GENERATIVE THEORYOF TONALMUSIC

The overall goal of the theory is to account for the musical intuitonal musicalidiom. By tions that a listener experiencesin a particular "tonal" we mean music whose pitch organization includes one specified pitch, the tonic, which determinesrelativeconsonanceand dissonance. By "musicalintuitions" we mean the unconscious principlesby which the listener experiencedin the idiom organizeswhat he hears, beyond the simple registeringof such surface features as pitch, duration,volume, attack envelope, and timbre.It should be made clearthat we take musical intuitions as largelydistinct from anythingthe listenerhas been taught formally. The theory is thus explicitly psychological,in that it is 45

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

concerned not with the organization of music in and of itself, but that the listeneris capableof hearing. with the organization Like other music theories, our theory develops an analytic system within which relationshipsheard by the listener can be expressed.Unlike other theories, however, our theory poses a furtherquestion:if an analysismodels what organizationthe listener hears,what is it that the To listener knows that has enabled him to arriveat this organization? answer this question, we have tried to develop a formal musical grammar that models the listener's connection between the presentedmusical surface and the organizationor organizationshe attributesto the music. The grammartakes the form of a system of rules that assigns analyses to pieces. By contrast, previousapproacheshave left it to the piece. analyst'sintuition to decide how to fit an analysisto a particular Readers acquainted with generative linguistics will recognize the similarityof our researchgoals to those of the study of language.2Linguistic theory is an attempt to describe the linguistic intuitions of a native speaker of a human language. Generativelinguistics seeks this which models the speaker's description in terms of a formal grammar knowledge of his language. Because many people have thought of using generative linguistics as a model for music theory, it is worth pointing out what we regardas the significantparallel:the combination of psychologicalconcerns and the formalnatureof the theory. Formalism alone is to us uninteresting,except insofar as it serves to express psychologicallyinterestinggeneralizationsand to make empiricalissues more precise. Previous attempts to apply linguistics methodology to music have proven relatively uninteresting because they attempted to translate linguistic theory, more or less literally, into musical terms, looking for musical parts of speech, for instance, or deep structures, or transformations, or semantics.3 We believe that a generativemusic theory, beyond the general principles stated above, must be sought in purely musical terms, uninfected by the substance of linguistic theory. If substantiveparallelsbetween the two theories emerge (as they in fact have in a number of areas), they are to be regardedas simply an unexpected bonus. We are concernedabove all with developinga theory of music. mistake made in previousattempts to apply genAnother recurring as a device erative linguisticsto music is to regarda linguistic grammar to manufacture grammaticalsentences. Under this interpretation, a musical grammar should be an algorithm that composes pieces of music.4 In stating his criticisms and apprehensionsof a music theory based on generative linguistics,Babbitthas such a conception in mind.s However,it was pointed out by Chomskyand Miller,and it has been an unquestioned assumptionof actual researchin linguistics,that what is 46

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

is the structureit assignsto really of interest in a generativegrammar sensentences, not which strings of words are or are not grammatical tences.6 Our theory of music is thereforebased on structuralconsiderations; it reflects the importance of structure by concerningitself not with the composition of pieces but with assigningstructuresto already existing pieces. This view of grammarleads to a crucial difference between the research paradigmsof linguistics and music theory, revealinga way in which music is not very much like languageat all. In a linguisticgrammar, perhapsthe most importantdistinctionis grammaticality-whether or not a given string of words is a sentence of the languagein question. A subsidiarydistinction is ambiguity:whethera givenstringis assigned two or more structureswith differentmeanings.In music, on the other hand, grammaticality per se plays a far less importantrole, since almost any passageof music is potentiallyvastly ambiguous-that is, it is much easier to construe music in a multiplicity of ways becausemusic is not tied to specific meaningsand functions as is language.In a sense, music is pure structure,to be "played with" within certainbounds. Thus the interestingmusical issues usually concern what is the most coherentor must be able to expreferredway to hear a passage.Musicalgrammar a function that is largely press these preferencesamong interpretations, absent from generativelinguistictheory. Our music theory specifically addresses those aspects of musical structure that are hierarchicalin nature. It is therefore not directly concerned with undoubtedly important matters such as thematic development, although these often play a role in analysis. We have structureswhich are simultaneously identified four distinct hierarchical imposed on a musical passage to form its structuraldescription (or analysis): grouping strucutre, metrical structure, time-spanreduction, and prolongationalreduction. Briefly, groupingstrucutredescribesthe segmentationof the music into motives, phrases,and sections. Metrical structure describesthe regular,hierarchicalpattern of beats which the listener attributesto the music. Both reductions,while using different criteria,ascribe degreesof relativeimportanceto all pitch-events(notes or chords) of a passage.In time-spanreduction,importanceis measured with respect to the other pitch-eventsin the same time-span,where a time-spanis a rhythmic unit constructed from an interactionof grouping and metrical structure. Prolongationalreduction also develops a hierarchyof pitch stability, but in ratherdifferentterms. It emphasizes the connections among pitch-events, establishingtheir continuity and progression, their movement toward tension or relaxation. It is this most closely to a Schenkercomponent of our theory that corresponds ian analysis. Elsewherewe describe all four components of musical structurein 47

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

detail.' Here we will restrictourselvesto a discussionof groupingstrucit is also systure. This is musically the simplest of the four structures; tematically the most basic, in that the other components make heavy use of it in their rules.Moreover,the groupingcomponent appearsto be that describes of most generalpsychologicalinterest, for the grammar grouping structure seems to consist largely of general conditions for auditory pattern perception that have far broaderapplicationsthan for music alone. As such, the rules for groupingseem to be idiom-independent-that is, a listener needs to know relativelylittle about a musical idiom in orderto assigngroupingstructureto pieces in that idiom. comthe grouping Like the other componentsof the musicalgrammar, Rules ponent consists of two sets of rules. GroupingWell-Formedness (GWFRs)establish the formal structureof groupingpatterns and their relationship to the string of pitch-events that form a piece; Grouping PreferenceRules (GPRs) establishwhich of the formallypossible structures assigned to a piece correspondto the listener'sactual intuitions. Before discussinggrouping structure,we must make an important caveat. At the theory's presentstage of development,we are treatingall music as essentiallyhomophonic;that is, we assumethat a singlegrouping analysis suffices for all voices of a piece. Our theory is therefore inadequate for more contrapuntal tonal music where this condition does not obtain. We consideran extension of the theory to account for polyphonic music to be of great importance. However, we will not attempt to treat such music here except by approximation.
RULES WELL-FORMEDNESS GROUPING

This section defines the formalnotion "group"by statingthe conditions that all possible grouping structuresmust satisfy. We use slurs beneath the musical notation to representthe groupingof a piece. Example 1 shows the groupingfor the first four bars of melody in Mozart's Symphony in G minor, K. 550.8 The first rule defines the basic notion of a group. GWFR 1. Any contiguous sequence of pitch events (or drumbeats, and so forth) can constitute a group, and only contiguous sequences can constitute a group. GWFR1 permitsall groupsof the sort designatedin Example 1 andprevents certain configurations from being designated as groups. For example, it prevents all the eighth notes in Example 1 from being designatedtogether as a group, or the first six occurrencesof the pitch D. (This contiguity condition is what makes the slur notation a viable representationof grouping intuitions; if there could be discontinuous groups,some other notation would have to be devised.) 48

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

LI

Example 1

k- k.G

,2 k

G2

..

J
1.

Example2

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

The second rule expresses the intuition that a piece is heard as a whole ratherthan as merely a sequenceof events. GWFR2. A piece constitutes a group. The third rule providesthe possibility of embeddinggroups,evident in Example 1. GWFR3. A group may contain smallergroups. The next two rules state conditions for the embedding of groups within groups. GWFR4. Ifa group G1 containspart of a group G2, it must contain all of G2. This rule prohibits groupinganalysesin which groupsintersect, such as those shown in Example 2. G 1 in these examples containspart of G2, but not all of it. On the other hand, all the groupsin Example 1 satisfy GWFR4, resultingin an orderly embeddingof groups. Thereare cases in tonal music in which an experiencedlistener has intuitions that violate GWFR4. We turn to these phenomenain a moment. The second condition for embeddingis perhapsless intuitively obvious than the other GWFRs.It is however formally necessaryfor the derivationof the time-spanreduction. GWFR 5. If a group G1 contains a smallergroup G2, G1 must be exhaustivelypartitionedinto smallergroups. GWFR 5 prohibits grouping structureslike Example 3a, b, in which part of G1 is contained neither in G2 nor in G3. Note however that GWFR 5 does not prohibit grouping structureslike those shown in Example 4, in which one subsidiarygroup of G xis further subdivided and the other is not. In fact such a situation occursin Example 1 and is extremely common. Let us now consider the discrepancy between the predictions of to which groups GWFR4 and certainactual musicalintuitions according In this 5 a case. example, one hears overlap. Example presents typical the beginnings of the third and fifth measures as belonging simultaneously to two intersectinggroups, at variouslevels of grouping.This situation is a violation of GWFR4, since there are groupsthat contain part but not all of other groups. Such situtations are common in tonal music, especially in developmentalpieces such as sonata-formmovements, where they provide a sense of continuity: overlaps at major group boundariesprevent a piece from reaching a point of rhythmic completion. In addition to true overlap,in which an event or sequenceof events is shared by two adjoininggroups, there is another overlap situation 50

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

a.

\G2

G
_

b. kG1 G2 G23 G1 Example3 G3

G3

kG
Example4

51

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

more accurately described as elision. Consider Example 6. The first group, markedpiano, is interruptedby the newfortissimo group.One's sense is not that the downbeatis shared,as if the first groupwere heard as Example 7a. A more accurate descriptionof the intuition is that the last event of Example7b is elided by the fortissimo. A total abandonmentof GWFR 4 would additionally predict the existence of groupingstructuressuch as those in Example 8, which do not in fact occur in music. In Example 8a the intermediatelevel of grouping bisects one element of the smallestlevel of grouping;in Example 8b the intermediatelevel of groupingoverlapsand the smallest level does not; in Example 8c the small groups overlapand the intermediate ones do not. In order to make those and only those modifications appropriate to grouping structures (that is, to permit the groupings shown in Examples 6 and 7, but not those in Example 8), we propose to disa piece's groupingstructure.The tinguish two formalsteps in describing first, underlyinggrouping structure,is describedcompletely by means of the groupingWFRs given above; that is, it contains no overlapsor elisions. The second step, surfacegroupingstructure,containsthe overlaps and elisions actually observed.These two steps are identicalexcept where the surface groupingstructurecontains an overlapor elision. At points of overlap,the underlyinggroupingstructureresolvesthe overlappedevent into two occurrencesof the same event, one in eachgroup. At elisions, the underlyingstructurecontains the event understood as being elided. rule to state the exact relaThe theory requiresa transformational tionship between underlying and surface groupingstructure.Consider overlap first. The essential idea is that adjacent groups in underlying structure that have identical events at their common boundary may overlap those identical events in the surface structure. To make this more precise, safeguardsmust be added to the rule to ensure that all groupsmeeting at a boundary are overlappedin exactly the sameway. Such a condition prevents the rule from creating situations like Example 8b, c in which overlappingis not uniform from one level to the next. Here is the rule; the last two conditions are the desired safeguards. GROUPING OVERLAP RULE. Given a well-formed underlying groupingstructureG as describedby GWFR1-5, containingtwo adjacentgroupsg 1 and g2 such that 1. g ends with event e 1; 2. g2 beginswith event e2; and 3. el 1 e2; then a well-formedsurfacegroupingstructureG' may be formed, identical to G except that 52

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

Allegro

or 71
F-1:

ttF17-

,-------!J

Example5. Mozart:SonatinaK.279, I

r r I
((

T' . ' .. . ...

r - '

r
r ' . '

t
r " ' '

Example6. Hayden:Symphony No. 104, I, mm. 29-33

a.

.
-... .....

__

-4

fpI Fr_ r

,_I
.

..

r
Example7

, r

elided

53

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

a.

) k,

kI

K
b. k,

'L

)
)

\ C" k

JL r )

Example 8

el

e2

underlying

surface

Example9

54

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

1. it contains one event e' where G had the sequencee1 e2; 2. e' =1e = e2; 3. all groups endingwith e1 in G end with e' in G'; 4. all groupsbeginningwith e2 in G begin with e' in G'. When notating surface grouping structure, we designate grouping overlapsby overlappingslurs beneath the music. Whennotatingunderlying groupingstructure,we use braces to join events that come to be overlappedin the surface.These notations are shown in Example9. The formal rule for elision is almost identical to that for overlap. The only difference lies in the relationships of the boundaryevents e1, e2, and e'. For the type of elision illustrated in Example 6, e1, the underlyingevent to be elided, is harmonicallybut not totally identical to e2; typically, it has a lower dynamicand a smaller pitch rangeas well. The corresponding event in surfacegroupingstructure,e', is identicalto e2. The description of a groupingstructurecontainingan elision thus contains in its underlyinggroupingstructurea descriptionof the intuitively elided event. The rules for overlapand elision have the desiredeffect of expanding the class of well-formed grouping structuresto include the counterexamples to GWFR4. In doing so, they express the musicalintuitions behind these counterexamples,and they restrict the predictedrangeof to two specific relatedtypes. counterexamples If one were to look for analogues of overlap and elision rules in linguistic theory, two different parallelsmight come to mind. First, with respect to their place in the formal description, they resemble linguistic syntactic transformations,in that they increase the class of well-formed structures by applying certain optional distortions to underlyingstructures.However,in their substance,they do not particularly resemble syntactic transformations,in that the distortions they introduce do not include such things as movement of constitutents(as in the passiveor subject-auxiliary inversiontransformations of English). Rather, their effects are most like those of highly local phonological rules that delete or assimilatematerialat word boundaries,for example, the process that results in the pronunciationof only one slightly elongatedd in the middle of the phrasedead duck. To sum up, the five well-formednessrules plus the transformations define a class of grouping structures that can be associated with a sequence of pitch-events,but which are not specifiedin any directway by the physical signal, as pitches and durationsare. Thus to the extent that groupingstructurestruly correspondto a listener'sintuitions, they represent part of what the listener bringsto the perception of music. This will become cleareras we discuss the preferencerules in the next two sections. 55

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

PERCEPTUAL MOTIVATION FORTHE PREFERENCE RULE FORMALISM While the GWFRsexclude as ill-formed certain groupingssuch as those in Examples 2 and 3, they do not preclude the assignmentof groupingstructuressuch as those in Example 10a, b to the opening of the G minor Symphony. Although these groupingsconform to GWFR 1-5, they do not, we trust, correspondto anyone's intuition of the passage'sactual grouping. One might attempt to deal with this problem by refiningthe well-formednessand transformational rules;but in practice we have found such an approach counterproductive.A difWe call this type ferent type of rule turns out to be more appropriate. of rule a preferencerule, for reasonsthat will soon be obvious. To begin to justify this second rule type, we observethat nothing in the GWFRsstated in the previoussection refersto the actual content of the music; these rules describeonly formal,not substantive,conditions on grouping configurations. In order to distinguishExample 1 from Example 10, it is necessary to appeal to conditions that refer to the specific music. In working out these conditions, we find that a number of different factors within the music affect perceived grouping, and that these factors may either reinforce or conflict with each other. When they reinforceeach other, stronggroupingintuitions result;when they conflict, the listenerhas ambiguousor vagueintuitions. Some simple experiments that compare musical grouping with a visual analogue suggest the general principles behind grouping preference rules. Intuitionsabout the visualgroupingof collections of small shapes were explored in detail by gestalt psychologists9. In Example 11a, for instance, the left-hand and middle circles group together and the right-handcircle is perceivedas separate;that is, the field is most 1 naturallyseen as two circles to the left of one circle. In Example ib, circlesare seen as grouped on the other hand, the middle and right-hand together and the left-hand circle is separate.The principlebehind this groupingobviously involves relative distance: the circlesthat are closer together tend to form a visual group. The grouping effect can be enhanced by exaggeratingthe difference of distances, as in Example 12a; it can be weakened by reducingthe disparity,as in Example 12b. If the middle circle is equidistantfrom the outer circles,as in Example 12c, no particular groupingintuition emergesat all. Similareffects exist in the groupingof musical events. Considerthe rhythmsin Example 13. The perceptionsabout groupingfor these five examples are auditory analoguesto the visual perceptionsin Examples 11-12. The first two notes of Example 13a grouptogether(that is, the example is heardas two notes followed by one note); the last two notes group together in 13b; the grouping of the first two in 13c is very perceivedgrouping. strong,in 13d relativelyweak; 13e has no particular These examples show that on an elementarylevel the relativeintervals 56

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

i)

Example10

rT--% ir

b)
:
,.

L, I

i+

_.

r+,

__:

Example 11

a. O

b. O

Example 12

a. OO

b. O

-...

Example1l3)r)

b)

c)

d)'

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

of time between attack points of musical events strongly influence groupingperception. Examining simple visual perception again, we see that like shapes tend to be groupedtogether. In Example 14a, the middle shape tends to form a group with the two left-hand shapes, since they are all squares; in Example14b the middle shapegroupswith the two right-hand shapes, since they are all circles. Similarly,equally spaced notes will group by likeness of pitch. In Example 15a the middle note is groupedmost naturallywith the twoleft hand notes and in Example 15b with the two right-handnotes (assumingall notes are played with the same articulationand stressand are free of harmonic implications, since these factors can also affect weakereffects are producedby makgroupingintuitions). Considerably ing the middle note not identical in pitch to the outer pitches, but closer to one than the other, as in Example 16a,b. If the middle pitch is equidistant from the outer pitches, as in Example 16c, grouping intuitions are indeterminate. These examples have demonstrated two basic principles of visual and auditory grouping:groups are perceivedin terms of the proximity and similarity of the elements availableto be grouped. In each case greaterdisparityin the field producesstrongergroupingintuitions, and greateruniformitythroughoutthe field producesweakerintuitions. Let us next consider fields in which both principles apply. In Example 17a, the principlesof proximity and similarityreinforce each other, since the two circles are close together and the three squaresare close together; the resulting grouping intuition is quite strong. In Example 17b, however, one of the squaresis near the circles, so that the principlesof proximity and similarityare in conflict. The resulting intuition is ambiguous:one can see the middle squareas part of either the left or right group (it may even spontaneouslyswitch, in a fashion familiarfrom other visually ambiguousconfigurationssuch as the wellknown Necker cube). As the middle squareis moved still furtherto the left, as in Example 17c, the principle of proximity exerts an even stronger effect and succeeds in over-ridingthe principleof similarity; intuition now clearly groups it with the left-hand group, though one may still sense some conflict. Parallel musical examples appear in Example 18. Three important properties of the principles of grouping have emerged. First, intuitions about grouping are of variable strength, depending on the degree to which individual grouping principles apply. Second, different grouping principles can either reinforce each other, resultingin strongerjudgments, or conflict, resultingin weak or ambiguousjudgments. Third,one principlemay over-rideanotherwhen the intuitions they would individually produce are in conflict. The 58

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

Example 14

a. O

O 0

b. 0

a.
1 . -

b.
--

Example 15

a. Example 16

b.

Example17

a.O

O O O

b.O 0 "0

130

a.
(.A

b.
-f-e
1 , f- --T --- .. .

Example 18

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

formal system of preference rules for musical perception which we develop in the next section possesses these properties.The term "preference rule" is chosen because the rules establish not inflexible decisions about structure,but relative preferencesamong a number of logically possible analyses. We have illustrated with elementaryvisual examples as well as musical ones in order to show that the preference rule formalism is not an arbitrarydevice, invented solely to make musical analyseswork out properly.It is ratheran empiricalhypothesis about the nature of human perception. We return to this generalissue toward the end of this paper. With this background,we turn to statingpreferencerules for musical groupingin some detail. GROUPING PREFERENCE RULES Two types of evidence in a musical surface can determine what grouping an experienced listener hears. The first is local detail-the patterns of attack, articulation,dynamics, and registrationthat lead to perception of group boundaries.The second type of evidence involves more global considerationssuch as symmetry and motivic, thematic, rhythmic, or harmonic parallelism.We explore these two types of evidencein turn. Local Detail Rules. There are three principles of grouping that involve only local evidence. The first is quite simple. GPR 1. Stronglyavoid groups containinga singleevent. Perhaps the descriptiveintent of the rule would be clearerif the rule were stated as "Musicalintuition strongly avoids choosing analysesin which there is a groupcontaininga single event" or "One stronglytends not to hear single events as groups." Readers who may be initially uncomfortablewith our formulationmay find such paraphrases helpful as they continue throughthe rules. The consequence of GPR 1 is that any single pitch-event in the normalflow of music will be groupedwith one or more adjacentevents. GPR 1 is overridden only if a pitch-event is strongly isolated from adjacentevents, or if it has an independentmotivic function. Underthe former condition, GPR 1 is overriddenby another of the rules of local detail, which we will state in a moment. Under the latter condition, to be stated as GPR1 is overridden by the preferencerule of parallelism, GPR 6. But the rarityof clearlysensed single-notegroupsattests to the strength of GPR 1 as a factor in determiningmusical intuition. (An example of an isolated note functioningas a groupis the fortissimoC in m. 17 of the finale of Beethoven'sEighthSymphony. An exampleof a single element functioning motivically occurs at m. 210 of the first 60

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

movement of Beethoven'sFifth Symphony:in the precedingmeasures, elements of the motive are progressivelydeleted until one event stands for the originalmotive.) An alternative formulation of GPR 1 is somewhat more general. Evidencefor it will appearlater on. GPR 1 (Alternativeform). Avoid analyses with very small groupsthe smaller,the less preferable. The effect of this version is to prohibit single-notegroups except with very strong evidence, and to prohibit two-note groups except with relativelystrongevidence. By three or four-notegroups,its effect would be imperceptible.Put more generally, this rule preventssegmentation into groups from becomingtoo fussy: very small-scale groupingperceptions tend to be marginal. The second preferencerule involvinglocal detail is an elaboratedand more explicit form of the principle of proximity discussed in the previoussection. It detects breaksin the musicalflow that areheardas boundaries between groups. Consider the unmetered examples in Example 19. All else being equal, the first three notes in each example are heard as a group,and the last two areheardas a group. In each case the carat beneath the example marksa discontinuitybetween the third and fourth notes: the thrid note is in some sense closer to the second note, and the fourth is closer to the fifth note, than the third and fourth are to each other. In Example 19a the discontinuityis a breakin a slur;in 19b it is a rest; in 19c it is a relativelygreaterintervalof time between attack points. (The examples of proximity in the preceding section involveda combinationof the last two of these.) In order to state the preferencerule explicitly, we must focus on the way the music gets from one note to the next-the transitionsfrom note to note-and pick out those transitionsthat are more distinctive than the surroundingones. These more distinctive transitions are the ones that intuition will favor as groupboundaries. To locate distinctive transitions,the rule considersfour consecutive notes at a time, designated as n1 through n4. The sequence of four notes contains three transitions:from the first note to the second, from the second to the third, and from the third to the fourth. The middle transition, n 2n 3, is distinctive if it differs from both adjacenttransitions in particularrespects. If it is distinctive, it may be heard as a boundarybetween a groupendingwith n2 and one beginningwith n3-. There are two ways we measure the distance between two notes: from the end of the first note to the beginningof the next, and from the beginning of the first note to the beginningof the next. Both of these ways of measuringdistance contribute to groupingjudgments. The former is relevant when an unslurredtransitionis surroundedby 61

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

slurredtransitions,as in Example 19a, or when a transitioncontaininga rest is surroundedby transitionswithout rests, as in 19b. The latter is relevantwhen a long note is surroundedby two short notes, as in 19c. Thus the rule of proximity has two cases, designated as (a) and (b) below. GPR2 (Proximity). Considera sequence of four notes n1n2n3n4. All else being equal, the transitionn2n3 may be heardas a group boundaryif (a) (Slur/Rest Rule) The intervalof time from the end of n2 to the beginningof n3 is greater than thatfrom the end of n1 to n2 and that from the end of n3 to the beginning of n4; or if (b) (Attack Point Rule) The intervalof time between the attack points of n2 and n3 is greater than that between the attack points of nj and n2 and that between the attack points of n3 and n4. It is important to see exactly what this rule says. It applies in Example 19 to mark a potential group boundary where the carat is marked. However,it does not apply in cases such as Example20. Consider Example 20a. There are two unslurredtransitions,each of which might be interpretedas a potential boundary.But since neither of these transitions is surroundedby unslurredtransitions,as GPR 2a requires, the conditions for the rule are not met, and no potential group boundaryis assigned.This consequence of the rule correspondsto the intuition that grouping in Example 20a is far less definite than in Example 19a, where GPR 2a genuinely applies. Example 20b,c are parallel, illustratingthe nonapplicationof GPR 2 when rests and long notes are involved. Another rule of local detail is a more complete version of the principle of similarity illustrated in the previous section. 21a-d show to the earlier four cases of this principle;the first of these corresponds examples of similarity. As in Example 19, the distinctive transitions are heard between the third and fourth notes. Whatmakes the transitions distinctive in these cases is change in (a) register,(b) dynamics, (c) pattern of articulation, (d) length of notes. We state the rule in a fashion parallelto GPR 2: GPR 3 (Change).Considera sequence of four notes n1n2n3n4. All else being equal, the transition n2n3 may be heard as a group boundaryif a. (RegisterRule) The transitionn2n3 involvesa greaterchangein pitch than both n n2 and n3n4; or if b. (Dynamics Rule) The transition n2n3 involves a change in dynamicsand n1n2 and n3n4 do not; or if 62

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

b.
A

c.

Example 19

-i

b.

c.

Example20
S-

--.-"
I __

I!didz

a.

b.

c.

Example21

Example22
ON

o.
p

b.
,"

c.

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

c. (Articulation Rule) The transition n2n3 involves a change in articulationand n1n2 and n3n4 do not; or if d. (LengthRule) n2 and n3 are of different lengthsand both pairs n, -n2 and n3-n4 do not differ in length. (One might add further cases to deal with such things as change in timbre or instrumentation.) Again this rule relies on a transitionbeing distinctivewith respectto the transitionson both sides. Example 22, like Example 20, illustrates cases where transitions are distinctive with respect only to the transition on one side; groupingintuitions are againmuch less securethan in Example21. As observedin the previoussection, the variouscases of GPRs2 and 3 may reinforce each other, producinga strongersense of boundary,as in Example 23a. Alternatively, different cases of the rules may come into conflict, as in Example 23b-d. Here,each caratis labeledwith the cases of GPRs 2 and 3 that apply. In Example 23b,c,d, there is evidence for a group boundary between both the second and third notes and between the third and fourth notes. However, GPR 1 prohibits both being group boundariesat once, since that would result in the third note alone constituting a group. Thus only one of the transitions may be a group boundary, and the evidence is conflicting. This prediction by the formal theory correspondsto the intuition that the grouping judgment is less secure in Example 23b,c,d than in Example 23a, where all the evidence favors a single position for the groupboundary. Although judgments are weaker for Example 23b,c,d than for Example 23a, they are not totally indeterminate.Close consideration suggests that one probably hears a boundary in Example 23b and Example 23d after the second note, and in Example23c after the third note (though contextual considerations such as parallelism could easily alter these judgments if they occurred within a piece). These intuitions can be reflected in the theory by adjusting the relative strengthsof the different cases of GPRs 2 and 3, so that in these configurations the Slur/Rest Rule (GPR 2a) overridesthe Attack Point Rule (GPR 2b), the Slur/Rest Rule overridesthe RegisterRule (GPR 3a), and the Dynamics Rule (GPR 3b) overridesthe Attack Point Rule. In general, all cases of GPR 3, with the possible exception of the Dynamics Rule, appear to have weaker effects than GPR 2. As in the examples in the second section judgmentsshould changedependingon the degree to which different conditions are satisfied. For example, if the G in Example 23b is lengthened to four quarters,increasingthe disparity in time between attacks, the evidence for the Attack Point Rule becomes stronger than the evidence for the Slur/Rest Rule, and the G is heardas groupedwith the E and the F. 64

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

O.

b.
-

C. A 20 A 2b A 3a

Example23

v f

20 3(t, 5b

A20.

12

3
V A

45

6
A

7 8

9
AIA
A

112

13

Example24
2b 2b 2o 3d 3a

20 2b

2b

Example25 ~
C

Ji.

b\

Example26

stt

0N Ul

~J~j

JJ

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

In order to make the theory fully predictive,it might be desirable actually to assign each rule a numerical degree of strength, and to assignvarious situations a degree of strength as evidence for particular rules. Then in each situation, the influence of a particular rule would be the of the measured as rule's intrinsic numerically strengthand product the strength of evidence for the rule at that point; the most "natural" judgment would be the analysiswith the highest total numericalvalue from all rule applications. However, we will not attempt such a quantification here, in part for principled reasons to be discussed below. Our theory is neverthelesspredictive,even at its present level, insofar as it identifies points where the rules are and are not in conflict; this will often be sufficient to carry the musical analysis quite far. Furthermore,the construction of simple artificial examples such as those in Example 23 can serve as a helpful guide to the relative strengths of variousrules, and these judgmentscan then be applied to more complex cases. We will often appeal to this-methodologywhen necessaryratherthan trying to quantify rule strengths. Before stating the remaininggroupingpreferencerules, let us apply GPRs 1, 2, and 3 to the opening of Mozart'sG Minor Symphony. Besides illustratinga number of different applicationsof these rules, this exercise will help show what furtherrules are needed. Example24 repeats the Mozartfragment.For convenience,the notes are numbered above the staff. Below the staff, all applicationsof GPRs 2 and 3 are listed as in Example23. Consider first the sequence from notes 2 to 5 in Example 24. The time between attack points of 2 and 3 is an eighth; so is that between 4 and 5, while that between 3 and 4 is a quarter.Thereforethe conditions of the Attack Point Rule (GPR2b) are met and a potential group boundary is markedat transition 3-4. Similarconsiderationsmotivate all the rule applications marked. On the other hand, one might be tempted to think that the Slur/Rest Rule (GPR 2a) would mark a potential boundary between 2 and 3. However,since there is no slur at transition 3-4, the conditions for the rule are not met and no boundaryis marked. Next observe that, with three exceptions, the potential group boundariesmarkedby GPRs 2 and 3 in Example24 correspondto the intuitively perceived group boundaries designatedin Example 1. The exceptions are at transitions 8-9, 9-10, and 18-19, where the rules mark potential boundariesbut none are perceived.Transition9-10 is easily disposed of. Because GPR 1 strongly prefersthat note 10 alone not form a group, a boundary must not be perceivedat both 9-10 and 10-11. Thus one of the rule applications must override the other. Example 23c shows a similarconflict between GPRs 2a and 3a. There the former rule overridesthe latter, even with a relativelyshorterrest. 66

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

Hence, GPR 2a should predominatehere too. Furtherweight is put on 10-11 by the Attack Point Rule (GPR 2b). Hence the applicationof GPR 3a at 9-10 is easily overridden. Let us ignore transitions 8-9 and 18-19 for the moment (we will return to them shortly). The remainingtransitionsmarkedby GPRs 2 and 3 are exactly the group boundariesof the lowest level of grouping in Example 1: 3-4, 6-7, 10-11, 13-14, and 16-17. Thus the bizarre grouping analysis in Example 10a, though permitted by the wellformedness rules, is shown by GPRs 1-3 to be a highly non-preferred groupingfor the passage. However,consider again the analysis in Example 10b, which has all the low-level boundariesin the rightplace, but whose largerboundaries are intuitively incorrect. GPRs 1-3 do not suffice to preferExample1 over Example 10b, since they deal only with placement of group boundaries and not with the organization of higher-level groups. Further preference rules must be developed in the formal theory to expressthis aspect of the listener'sintuition. Organization of higher-level grouping. Beyond the local detail rules, a number of different principles reinforce each other in the analysisof the higher-levelgroups in Example 1. The first of these dependson the fact that the largest time intervalbetween attacks, as well as the only rest, areat transition10-11. This transitionis heardas a groupboundary at the largest level internal to the passage. The most generalform of of this principlecan be stated as GPR4. GPR 4 (Intensification Rule). Where the effects picked out by GPRs 2-3 are relativelymorepronounced,a relativelylargerlevel group boundarymay be placed. A simple example that isolates the effects of GPR 4 from other preference rules is Example 25, which is heard with the indicated grouping.GPRs 2a and 2b (Slur/Restand Attack Point Rules) correctly mark all the group boundariesin Example 25, but they say nothing about the second level of grouping, which consists of three groups followed by two groups. GPR 4, however, notes that there is a rest at the end of the third small group, strongly intensifying the effects of GPR 2 at that particular transition.It is this more stronglymarkedtransition that is responsiblefor the second level of grouping. A second principleinvolved in the higher-level groupingof Example 1 is a general preference for symmetry in the grouping structure, independentof the musicalcontent:

67

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

GPR 5 (Symmetry Rule). Prefergroupinganalysesthat most closely approach the ideal subdivision of groups into two parts of equal length. GPR 5 alone is involvedin the higher-level groupingof Example26a, in which the smallergroups are furthergroupedtwo and two ratherthan, say, one and three. In Example 26b, where there are six small groups, GPR 5 cannot apply in the ideal fashion. The ideal can be achieved in the relation between the small and middle-levelgroups, or in the relation between the middle-level and large groups, but not both. The result is an ambiguous middle-level grouping, shown as analyses i and ii in Example 26b. In a real piece, the ambiguity may be resolved, for example, by metrical or harmonic considerations, but then the result is not due solely to GPR 5. In general, it is the impossibility of fully satisfying GPR 5 in ternary grouping situations that makes such groupingssomewhatless stable than binarygroupings. In Example 24, GPR 5 has two sorts of effects. First, it reinforces GPR 4 (Intensification Rule) in markingtransition 10-11 as a higherlevel boundary, since this divides the passage into two equal parts. Second, each of the resulting intermediate-levelgroups contain three groups, the first two of which are two quarternotes in duration and the third of which is four quarternotes. GPR 5 therefore groups the first two together into a group four quarter notes in duration, producing the ideal subdivision of all groups. (Note that GPR 5 does not requireall groups to be subdividedin the same way: it is irrelevant to GPR 5 that the first and third four-quarter-notegroups are subdivided,but that the second and fourth are not.) In addition to GPRs 4 and 5, a third importantprincipleis involved of in the higher-levelgroupingof Example 24: the motivic parallelisms events 1-10 and 11-20. We can isolate the effects of this principlein passages such as Example 27a,b. Other things being equal, one's intuition is that Example 27a is grouped in threes and Example27b in fours. Since both examples have uniform motion, articulation, and dynamics, the grouping preference rules make no prediction about their grouping. Hence a further preference rule, GPR 6, is necessary to describethese intuitions. two or more segmentsof the music GPR 6 (ParallelismRule). Where can be construedas parallel,they preferablyform parallelpartsof groups. The application of GPR 6 to Example 27 is obvious: the maximal parallelismis achieved if the "motive"is three notes long in Example 27a and four notes long in Example27b. 68

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

Example27

O.
_r

_r

tJ~~,

Example28

Ali

Example29
- --:-- +

.e,-.

-i

~6-

...

.....

aN
\.O

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

However, consider Example 28, in which a contrary articulationis applied to Example27a. One'sintuition is that a groupinginto threes is now relatively unnatural.The theory accounts for this as follows: the Slur/Rest Rule places ratherstrongpotential groupboundariesin Example 28 after every fourth note, where the slursarebroken;thus each of the three note segments so obvious in Example 27a comes to have an internal group boundary in a different place. Since motivic parallelism requires, among other things, parallel internal grouping, Example 28 cannot be segmented into three-note parallel groups as easily as can Example 27a. Thus GPR 6 either is overriddenby the Slur/Rest Rule or actually fails to apply. GPR 6 says specifically that parallelpassagesshould be analyzed as forming parallelparts of groups, ratherthan entire groups. It is stated this way in order to deal with the quite common situation in which groups begin in parallelfashion and divergesomewherein the middle, often in order for the second group to make a cadential formula. A clear case is Example 29, the opening of Beethoven's Quartet Opus 18, No. 1. GPR 6, reinforcedby the Slur/Rest Rule, analyzes the first four measuresas two two-measuregroups.The fifth measureresembles the first and third, but at that point the similarity ends. If GPR 6 demanded total parallelism,it could not make use of the similarityof mm. 1, 3, and 5. But as we have chosen to state the rule above, it can to help establishgrouping. use this parallelism In the Mozartexample (Example 24), GPR 6 has two effects. First, it reinforces the Intensification and Symmetry Rules in assigningthe major grouping division at the middle of the passage. Second, recall that GPR 2a (Slur/Rest Rule) marks a possible group boundary at transitions 8-9 and 18-19, and that these group boundaries were intuitively incorrect.Considertransition8-9; 18-19 is treated similarly. GPR 6 is implicated in the suppressionof this potential boundary by detecting the parallelismbetween the sequences of events at 1-3, 4-6, and 7-9. If a group boundary appeared at transition 8-9, parallelism would requireit at 2-3 and 5-6 as well. But, in violation of GPR 1, this would in turn make notes 3 and 6 form single-notegroups. Hence the only way to preserve parallelismis to suppress the possible group boundary at 8-9; indirectly, then, GPR 6 here overridesthe Slur/Rest Rule. The ParallelismRule is not only important in establishingintermediate-level groupingssuch as those in the previous examples: it is the major factor in all large-scalegrouping.For example, it recognizes the parallelismbetween the exposition and recapitulationof a sonata movement, and assigns to them parallel groupings at a large level, establishingmajorstructuralboundariesin the movement. Finally, a seventh preference rule concerns primarily large-scale 70

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

grouping. Different choices in sectionalization of a piece often result in interesting differences in the time-span and prolongationalreductions, and it is often the case that the choice cannot be made purely on the basis of grouping evidence. Rather, the choice of preferred groupingmust involve the relative stability of the resultingreductions. Without a full account of the reductions,we obviously cannot justify such a preferencerule here, but we state it for completeness. GPR 7 (Time-spanand ProlongationalStability). Prefera grouping structurewhich results in more stable time-span and/or prolongational reductions. Havingstated the system of GPRs,we proceed now to a few remarks on the notion of parallelism which was mentioned in GPR 6 and on the natureof the formalismwhich we used in statingour rules. Remarks on parallelism.The importance of parallelismin musical structure cannot be overestimated.The more parallelismone detects, the more coherent an analysis becomes and the less independent information must be processedand retained.However,our formulation of GPR 6 still leaves a great deal to intuition in its use of the locution "parallel." Two identical passagescertainly are parallel,but how different can they be before they no longer seem parallel?Among the factors involved in parallelismare similarity of rhythm, internal grouping,and pitch contour. Whereone passageis an ornamentedor simplifiedversion of another, similarityof corresponding levels of the time-spanreduction must also be invoked. Hereknowledgeof the idiom is often requiredto decide what counts as ornamentationand simplification.It appearsthat a set of preference rules for parallelismmust be developed, the most highly reinforced case of which is identity. But beyond this we are not prepared.to go, and we feel that our failure to flesh out the notion of parallelismis a serious gap in our attempt to formulatea fully explicit theory of musical understanding.For the present we must rely on intuitive judgments to deal with this area of analysis in which the theory cannot make predictions. The problem of parallelism,however, is not at all specific to music theory: it is a special case of the more generalproblem of how people recognize similaritiesof any sort, for example similarities amongfaces. This relation of the musical problem to the more generalpsychological problem has two consequences. On the one hand, we may take some comfort in the realizationthat our unsolvedproblemis reallyonly one aspect of a largerand more basic unsolvedproblem.On the other hand, the hope of developinga solution to the musical problem through the preference rule formalism suggests that such a formalismmay be of 71

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

more widespreadapplicabilityin psychologicaltheory. Recognizingthe generalityof the problem may provide a point of attack for a broader theory of cognition. Remarkson formalism. Some readersmay be puzzled by our assertion that GWFRs 1-5 and GPRs 1-7 constitute a formal theory of musical grouping.There are perhapstwo respects in which our theory does not conform to the popular stereotype. First, the rules are couched in ordinary English, not in a mathematical or quasimathematicallanguage. Second, even if the rules were translatedinto some sort of mathematical terms, they would not be sufficient to provide a foolproof algorithm for constructing a grouping analysis from a given musicalsurface.This seems an appropriate place to defend our theory againstsuch possible criticisms. The first point is easily refuted. As we mentioned at the outset, we are interested in stating as precisely as possible the factors leading to intuitive musicaljudgments. Mathematicization of the rules ratherthan in statement is useful insofar as it allows more precise only English interesting or more precise predictions. Consider for example the GWFRs,which together define a class of hierarchicalgroupingstructures connected in a simple way to musical surfaces. One could presumablytranslatethese rules into the mathematicallanguageof set theory or network theory. But it would serve little purpose; since there are no particularlyinteresting theorems about sets or networks that bear on musical problems, no empiricalcontent would be added by such a translation. In fact, such formalismwould only clutter our exposition with symbolic formulas that would obscure the argument. Hence we have chosen to state our rules in ordinaryEnglish,but with sufficient precision that their consequences, both for and againstthe theory, are made as clear as possible. The secondargument advancedagainstthe theory is more substantive. The reason that the rules fail to producea definitiveanalysisis that we have not completely characterizedwhat happens when two preference rules come into conflict. Sometimes the outcome is a vague or ambiguous intuition; sometimes one rule overridesthe other, resulting in an unambiguous judgmentanyway.Abovewe suggestedthe possibility of quantifying rule strengths, so that in a conflicting situation a judgment could be determinednumerically.A few remarksare needed to defend our decision not to attempt such a refinement. First, as we pointed out earlier,our main concernhere is to identify the factors relevantto musicalintuition, and to learnhow these factors interact to produce the richness of musical perception. To present a complex set of computationsinvolvingnumericalvaluesof rule applications would have burdenedour exposition with too much detail that did not involve purely musicalor psychologicalissues. 72

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

But our decision was not merely methodological. A little reflection suggeststhat the assignmentof numericalvalues to rule applicationsis not as simple a task as one might at first think. Patrick Winston, in developing a computer programfor certain aspects of visual pattern recognition, uses proceduresnot unlikepreferencerules."1 Becausethe computer must make a judgment, Winstonassignsrules strengthsand sets threshold values which must be attained in order to achieve a positive judgment. Winston himself notices the artificiality of this solution, for it allows only positive and negative judgments, not ambiguousor vagueones. Moreover,the choice of thresholdvaluesis to a certain extent arbitrary:should a value be, for instance, 68 or 72? A simple numerical solution of this sort creates an illusion of precision that is simply absent from the data. A more formidableproblemlies with the need for preferencerulesto balance local and global considerations.Whileit is not hardto imagine numericallybalancing,for instance, the length of a rest againstthe size of an adjacent change in pitch, it is much more difficult to balancethe strengthof a parallelism againstthe breakin a slur.Partof the difficulty lies in the present obscurity of the notion of parallelism,but part also lies in confusion about how to compare parallelismto anything else. Evenworseis the difficulty of balancinginter-componentconsiderations suchas those introducedby GPR 7, the rule of Time-span and Prolongational Stability. How much local instabilityin grouping,or loss of parallelism, is one to tolerate in order to produce more favorableresults in the reductions? Evidently, if we are to quantify strength of rule application, nothing short of a global measure of stability over all aspects of the structuraldescriptionwill be satisfactory.Thus, we feel, it would be foolish to attempt to quantify local rules of grouping without a far better understandingof how these rules interact with other ruleswhose effects are in many ways not comparable. To sum up, we acknowledge it as a failing that our theory cannot provide a computable procedure for determining musical analyses. However,achievingcomputabilityin a meaningfulway requiresa better understandingof many difficult musical and psychologicalissues than exists at present, so we have felt uninclinedto pursue it. In the meantime, we haveattemptedto make the theory as predictiveas possible,by stating rules clearlyand carefullyfollowing throughtheir consequences, avoiding ad hoc adjustmentsto make analyses work out the way we want. Webelieve that the insightsthe theory has been able to afford are sufficientjustificationfor this methodology. The performer'sinfluence on preferredhearing. The performerof a piece of music, in choosing an interpretation,is in effect decidinghow he hears the piece and how he wants it to be heard. Among the aspects of an interpretationwill be a preferredanalysis (either conscious or 73

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

unconscious) of the piece with respect to the grammaticaldimensions addressedby our theory. Since groupingstructureis the most crucial link between the musical surface and the more backgroundlevels of time-spanand prolongationalreductions, the perception of groupingis one of the more importantvariables the performercan manipulate. A performer'sprincipalmeans of influencing groupingperceptionis in the execution of local details, which affect the hearer'schoice of low-level grouping boundariesthrough GPRs 2 and 3 (the local detail rules) and of larger boundaries through GPR 4 (Intensification). For K. 331, example, consider the beginning of the Mozart Piano Sonata " The musical in Example30 with two possiblegroupings. whichappears surface is in conflict between these two groupings. Since the longest duration between attacks is after the quarternote, local detail favors grouping (b). But maximal motivic parallelismfavors grouping(a) (if the piece began with an upbeat eighth, parallelismwould favor grouping (b)). The variations that follow take advantageof this grouping ambiguity, tipping the balance in favor of grouping (b) in Variations 1, 2, and 5, and in favor of grouping(a) in Variations3, 4, and 6. A performerwho wishes to emphasizegrouping(a) will sustain the quarternote all the way to the eighth and both shortenthe eighth and diminishits volume. He thereby creates the most prominentbreakand change in dynamics at the bar line, thus enhancingthe effects of GPRs 2 and 3. On the other hand, a performerwho chooses grouping(b) will shorten the quarter,leavinga slight pauseafter it, and sustainthe eighth up to the next note. The effect of GPRs2 and 3 is then relativelygreater before the eighth note and less after it. A second, less noticeable alterationcould involvea slight shift in the attack point of the eighth note, playingit a little early for grouping(a) and a little late for grouping (b). This slight change in attack-point distancealso affects preferred groupingthroughits influenceon GPR2. Subtle variations in articulation such as these are typical of the strategiesperformersuse to influence perceivedgrouping.However,it is important to emphasize that the performer'sconscious awarenessof these strategiesoften does not go beyond "phrasingit this way rather thanthat way";that is, in largepart these strategiesare learnedand used unconsciously. In making explicit the effect of such strategies on musical cognition, we have suggested how our theory potentially addressesissues relevantto performanceproblems.
TWOMORE EXAMPLES

To support the claim that the rules of groupingare not limited to a single style, we analyze the grouping structure of the opening of Stravinsky'sThree Pieces for ClarinetSolo (Ex. 31). As in the example from Mozart's G minor Symphony, each note is numbered and 74

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

Example30
I _J k \ J . J k

b. b.c

(Clarinet in A) sempre p e molto tranquillo, MMJ = 52


1

10

Example31
(3d)
. ".

(2a) (30) , '-22 23 24 21

(20) (2b) 3o0) 25 26


--'

,, 27

19

20

ii ul

(2o) 13d)

(2 a) (2b) (3g)

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

applications of GPRs 2 and 3 are marked at appropriatetransitions. We assume that the breath mark tells the performer, in effect, to produce a grouping boundary by using one or both of the strategies just discussed:shorteningthe precedingnote and leavinga space, providing evidence for the Slur/Rest Rule (GPR 2a), and perhapslengthening the time between attack points, providingevidence for the Attack Point Rule (GPR 2b). In addition to rule applications,the example shows the smallestlevels of groupingpredictedby the rules. The dotted lines in Example 31 requiresome explanation.Consider first transition 2-3; although there is weak evidence for a grouping boundary at this point due to the change in note values, one tends to hear events 1-4 grouped together and to suppressthe smallergroups. Above we suggested an alternativeversion of GPR 1: Avoid analyses with very small groups-the smaller,the less preferable.This versionof the rule would say that the weak evidence at transition 2-3 is insufficient to establish a group boundary because of the shortness of the resultinggroups. At transition 9-10 no local evidence supportsa groupboundary,but parallelism with transition 2-3 and its context would argue for a boundaryif one were chosen at 2-3. Similar(though weaker) parallelism, plus the changein register,indicate a boundaryat 15-16; finally, a number of relativelyweak rules apply at transition 18-19. Placementof a group boundary at each of these points results in one or more twonote groups, which the revised GPR 1 attempts to avoid. The overall effect of the revised GPR 1, then, is to suppress,or at least make less salient, all groupsrepresentedby dotted lines in Example31. On the other hand, at all other marked transitions, there are applications of the more influential preference rules of proximity. In general, these rule applicationscause little difficulty. However,at one point they do lead to a two-event group, notes 12-13. The analysis retains this group for two reasons: first, the local evidence for a boundary at transition 11-12 is relatively strong, and second, group 1-4 followed by group 5-7 is parallel motivically to group 8-11 followed by group 12-13. Thus, strong local evidence and motivic parallelismboth support a groupingboundary at transition 11-12 and override the preference of the revised GPR 1 against the two-note group 12-13. The result of local evidence interacting with GPR 1, then, is to establishthe small-scale groupingindicatedby solid lines in Example31. In attempting to establish larger-levelgrouping, we first observe that motivic parallelismof the groups beginning at 1, 8, and, to a lesser extent, 14 and 21 favors larger level boundaries at transitions 7-8, 13-14, and 20-21. In addition, the strongest local rule applicationsin the passageare at transitions7-8 and 13-14; the breath at 20-21 also 76

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

establishesit as a relativelystrongapplicationof GPR2a (the Slur/Rest Rule). So far, then, GPRs 6 (Parallelism) and 4 (Intensification) suggestthe groupingin Example32. There are two possible ways to construct still larger groups. Symmetry (GPR 5) suggeststhe groupingin Example33a. On the other hand, transition 7-8 has the strongestapplication of GPR2 in the passage, because of its rest and the precedinglong note; thus Intensification favors the groupingin Example 33b, where this transitionis the most important grouping boundary. Moreover,the strongest motivic parallelismin the passageobtains between events 1-4 and 8-11; since the rule of parallelismprefersthese to be parallelparts of groups,this rule too favors the grouping in Example 33b. (If, in addition, purely binary grouping is desired in Example 33b, to minimally satisfy the Symmetry rule, the relatively strong motivic parallelismbetween 8-10 and 14-16 favors an additional group, including events 14-27, as shown in Example33c.) The choice between Examples33a and 33b is the first point where the preference rules result in an ambiguous grouping. We personally favor Example 33b, treating the second large group, in effect, as an extended repetition of the first. The resultingasymmetryis characteristic of the piece's style, a style which deliberatelyavoids symmetry, thus blocking the maximal reinforcementof preferencerules. That is, the difference between this style and Mozart'swith respect to grouping is not grammatical per se, but resides in what structuresthe composer chooses to build. usingthe grammar In the two previous examples, the grouping preference rules have encounteredat least minor conflicts. Consideran example in which the preference rules encounter no conflicts and strongly reinforce each other at all points. Such an example would have stronglymarkedgroup boundaries; the major group boundaries would be more strongly marked than the minor ones; the piece would be totally symmetrical, have only binarysubdivisionsof groups,and would displayconsiderable parallelism among groups. The theory predicts that the grouping of such a passagewould be completely obvious. Example 34, part of the anonymous 15th-centuryFrench instrumentalpiece Dit le Bourgignon, represents such a case. As usual, applications of GPRs 2 and 3 are markedat relevanttransitions. This example deserveslittle comment. The repetition of phrasesis, of course, the strongest form of parallelism.The smallest groupsjoin by twos with adjacent groups of equal length; these intermediate groups again group by twos with groups of equal length. Furthermore, the intermediate-levelboundaries are marked both by rests and by greater duration between attack points, whereas the less important boundaries are marked only by the latter distinction, and to a lesser 77

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

1 -4 5 -7 8- 11 12-13 14--20
L J L J

21 -24 25 -27

Example32

a.

1-4

5 - 7 8 - 11 12-13 14 - - 20 21 - 24 25 - 27

b.

1 - 4 5 - 7 8 - 11 12-13 14 - - 20 21 - 24 25 - 27

c.

1- 4 5 - 7 8 - 11 12-13 14 - - 20 21 -24 25-27

Example33

78

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

rrr

SF

..2b

L _ . _2, ___L
------

, -_ 2.

2 _ 2b_

Example34

79

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

degree. Thus the rules of Intensification, Symmetry, and Parallelism are simultaneously satisfied by the grouping suggested by local evidence; there is no question of ambiguityor vagueness. Many folksongs and nursery rhymes also exhibit this sort of regularityin the application of groupingpreferencerules.Pieces of this sort are often thought of as having "stereotypical"groupingstructure, which means in terms of the present theory maximalreinforcementof grouping preference rules. And here lies a danger for research.Some attempts at a generativegrammarof music have treated such stereotypical groupingstructuresas basic and assumedthey could be extended to more complex structures.12 But if our theory is correct, the stereotypical structuresreveal little, since they representthe confluence of a great number of interacting factors whose individualeffects therefore cannot be identified. It is essential to investigate more sophisticated examples from the very beginning in order to arriveat any notion of what is going on. Closingremarkson musicalgrammar.This completes our exposition of the rules for groupingstructure. To sum up, the rules fall into two rules describea set of hierarchical subcomponents.The well-formedness and certainaspects of their connection to the musicalsurface. structures ruleswhich account This subcomponentalso contains transformational for certain permissible violations of the normal hierarchy. The preferenceruleschoose, from all possible well-formedstructuresassigned to a given musical surface, those which are most highly preferred by the listener. These rules can be dividedinto local detail rules(GPRs 1, 2, and 3) and rules with more global conditions (GPRs 4, 5, 6, and The other three components of the musical grammar-metrical structure,time-spanreduction, and prolongationalreduction-display a similarabstractorganization.In addition, there are variousrule systems which link together the structuresassignedby the four componentsboth well-formednessrules and preferencerules (of which GPR7 is an example). The effect is a highly interactivesystem in whichlocal choices made in one component of the musicalstructurecan have ramifications in all components. These effects have much to do with the richnessand subtlety of musicalcognition. Since the groupingcomponent proves to be independentof style, it is worth mentioning a few areaswhere idiom-specificrules do appearin the musical grammar,differentiatingClassicaltonal music from other rules, which define genres. One place is in the metricalwell-formedness the possible meters in the idiom. Another place is in the rules that define scale structure and establish a hierarchy of consonance and dissonance relative to the tonic; these rules are intimately involved in constructing both reductions. Third, Classicaltonal music appearsto 80

7).

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

have a preferencerule that establishesa normativeform for the structure of prolongational reductions.This rule, which is relatedto Schenker's notion of Ursatz, influencesgreatlythe way tension and releaseare structured, from the largestto the smallest levels of structure.Certain other idioms appeareither to lack this rule or to have a different normative prolongational-structure. Other differences among idioms, such as overall thematic form, motivic vocabulary,and timbraleffects, concerneither non-hierarchical is aspects of structureor different choices in how the musicalgrammar is basicallyunchanged exploited. Thus the claim that musical grammar from Bach to Brahms parallels the claim that English grammaris to the present;what has changed basically unchangedfrom Shakespeare is the artisticuse made of the grammar. We close the paper by discussing the relevance of this generative music theory to other cognitive domains.
RELEVANCE TO LINGUISTICS AND PSYCHOLOGY

As we promisedat the beginning,the fragmentof a generativemusic theory presented here does not look much like generativelinguistics. The differencein methodologybetweenthe two theoriesis symptomatic: where linguistic theory is highly concerned with grammaticality, music theory is more heavily concernedwith preferenceamofiga considerable number of competing well-formed (that is, grammatical)structures. The closest analogue to linguistic grammaticalityin the theory of grouping is well-formednessof grouping structure. The rules defining grouping well-formednessresemble linguistic rules in that they either establish a nested segmentation of the surface string (like phrase structurerules in syntax) or characterizepermissibledistortions of the hierarchical structure (like transformations). This suggests that the bulk of a linguistic grammarconsists of well-formednessrules-from the phonology through the syntax to the semantics. Even the lexicon can be considereda part of the well-formednesscomponent, in that it establisheswell-formedmatchingsbetween phonological,syntactic, and semantic form at the terminal nodes of hierarchicalstructures. The well-formednessrules for music, even consideringall four components of the musical grammar, hardlyapproachthe linguisticwell-formedness rules in complexity. This reflects the much greaterrole of grammaticality in languagethan in music. The question thus arisesas to whetherlinguistictheory contains any rule systems comparableto preferencerulesin music. Webelieve that a number of phenomena have been discussedin the linguistic literature that have appropriateproperties to be governed by preference rules. One of these is the notion of markedness conventions: conditions that establish preferredforms for rules of grammar, but which may be 81

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

violated by particulargrammars.A second area is parsing strategies: heuristic procedures for assigning syntactic structures to potentially ambiguous strings of words in the course of processing a sentence. These are perhapsclosest in spiritto the preferencerules of the musical grammar.A third area is in the theory of pragmatics,which is concerned with the relation between a sentence's literal meaningand how the sentence can be used in a verbal or nonverbal context. The balancing of conflicting preferences, characteristicof preference rule interactions, appears quite clearly, for example, in the applicationof Grice'sconversationalmaxims.13 Finally, the use of preferencerulesas part of the internal structureof word meaningsappearsto account for "family resemblance" and stereotype phenomena that have plagued standardaccounts.14 In short, the notion of preferencerulesmay prove useful in the treatment of several disparate aspects of linguistic competence that have resisted formalization along more traditional lines. Turning to psychology, recall the parallels between visual and auditory grouping pointed out above. That such parallels exist was observed by Wertheimer,Koffka, and K6hler;is K6hler conjectured that they are a result of similarphysiologicalorganizationin the two domains. Lashley also arguedthat essentiallysimilarmental representations serve for both spatially and temporally sequenced memory.16 However,none of these psychologistssaid little systematicallyabout the structureof temporalorganizationper se, beyond enumerationof some highly interestinganecdotalevidence. Havingdevelopeda formaltheory for one type of temporal organization,we are in a position to turn the parallelismon its head and see what the theory of musical grouping has to offer to the study of visualperception. In Koffka's detailed and wide-ranging investigationof the organization of the visual field, there appearexamples (p. 135) almost exactly analogous to our illustrations of visual grouping.17 Koffka suggests are involvedin these judgments:(1) greater three general characteristics proximity and greater similarity enhance visual groupingjudgments; (2) judgments are less certain when the principlescome into conflict; (3) undercertainconditions, one principlecan overridethe other. These we have used to motivate the preferenceare the exact characteristics rule formalism. Similarobservationswith a number of other principles occur throughout Koffka's discussion of the perception of shapes, figure-groundopposition, and three-dimensionality.He demonstrates, in addition to proximity and similarity, the influence of such factors as symmetry, continuation of lines along regularcurves, ability of a patternto enclose space, and (significantly)the viewer'sintention. Koffka's work is intended as a demonstration of the fundamental claim of gestalt psychology: perception, like other mental activity, is a 82

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

dynamicprocessof organization,in which all elementsof the perceptual field may be implicatedin the organizationof any particular part. He is at pains to point out and prove two crucialaspectsof this claim. First, perception is not simply a product of what is in the environment: the viewer plays an active, though possibly unconscious, part in determining what he perceives. Second, the totality of the field as perceivedcannot be built up piecemeal, as a mere accumulationof the perceptionof its partseach taken in isolation. The general underlying principle that forms the basis of the gestalt account of perception was formulated by Wertheimeras the Law ofPrdgnanz.Koffka describesit as the following: Psychologicalorganizationwill alwaysbe as "good" as the prevailing conditions allow. In this definition the term "good" is undefined. It embraces such properties as regularity, symmetry, simplicity and others which we shall meet in the course of our discussion.18 In other words, the variousprinciplesof visual perception that Koffka demonstratesare taken as explications or elaborationsof the notion of "good" organization. The preferencerule component of the generativetheory of musical grouping resembles Koffka's account of vision not just in the way individualrules apply and interact with each other. It also has overall properties that make it a gestalt theory in the sense just described. First, it is mentalistic: it claims that perceived grouping is not present in any direct way in the musical surface. In fact, the events of with the groupingstructureare not even in one-to-one correspondence musical surface, since in grouping overlap one surface event appears twice in underlyinggroupingstructure,and in groupingelision an event appearsin grouping structure that has no counterpartin the musical surface. These non-correspondences are made possibleby claimingthat grouping structure is a mental construct, associated with the musical surfaceby unconsciousrules. But the theory of groupingis not just mentalistic: it is specifically gestalt, in that it claims that grouping structures cannot be built up strictly from their parts. Three of the preferencerules are indeed concerned with local details, and the information they provide is of undeniableimportance.But the rule of Intensification(GPR4) involves comparison of scattered potential group boundaries; the rule of Symmetry (GPR 5) involves the total internal structureof groups;and the rule of Parallelism(GPR 6) can invoke the mutual influence of regions spaced arbitrarilyfar apart. Furthermore,we have seen a casetransition 8-9 in the G minor Symphony passage-in which the of a low-levellocal detailwas overridden interpretation by considerations of parallelism.Other components of the musical grammarshow even 83

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

more clearly that global considerationsare essentialfor musicalperception. For example, rhythmic syncopation is a situation in which preference rules for local metrical detail are overridden by global on metricalwell-formedness. requirements The overall function of the preferencerules is to choose a structure that is maximallystable;that is, they define what assignmentsof grouping structure to a musical surface are perceptually "good." Thus the grouping preference rules in effect constitute an explicit statement of the Law of Prignanzas it appliesto musicalgrouping. Havingshown that the theory of musical groupingis an instance of gestalt theory, what are we to make of this relationship?This question must be approached with a certain amount of care, in view of the present status of the gestalt tradition: though its importanceis widely acknowledged,it has hardlybeen prominentin the literatureof the past twenty-five years. If we are to attach any significanceto the similarity of musical grammarand the gestalt theory of vision, it seems wise, if only for self-protection, to investigate the reasons for the decline of gestalt psychology. In fact, we think that generativelinguisticsin conjunction with the work presentedhere actually providesa key to overcoming some of the difficultiesthat the gestalt traditionencountered. The sociological (that is, irrational)difficulties have already been largely overcome. One gets the impressionthat gestalt theory could not withstand the powerful anti-mentalisticbias prevalent in (American) psychology during the 1940s and 1950s, and it seems to have been written out of existence by the more "scientific" behavioristschool. Now the shoe is on the other foot: the successof generativelinguistics has playeda largerole in rekindlinginterest in mentalistictheories, while behaviorist psychology has been to a great extent discreditedby arguments rather similarto those advancedby the gestaltists forty to sixty years ago.19 Thus the atmosphere is now conducive to a more sympatheticreadingof gestalt work than it has been in the past. More substantive difficulties with the gestalt tradition arose from the problem of how to couch a mentalistic theory in a rigorousand explanatory fashion. First, it was felt, perhaps partly in response to behavioristic arguments, that no mentalistic theory could be worthwhile without an account of its mechanism, and in this respect gestalt theory (like all psychological theories past and present) was unquestionably deficient. Koffka compares the Law of Prignanz to physical principlesminimizingenergy (for example, of surfacetension) at boundaries between substances. In his later work, K6hler tries to make this sort of analogy into a theory, by claiming a direct correspondence between the stabilization of perceptual fields and the stabilization of electrical fields in the brain.20But this physiological 84

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

reduction is far too crude for the finely-tuned observationsit is meant to explain. Generativelinguisticshas again preparedthe way for a resolutionof this problem, for it has demonstrated the virtues of dividing a psychological theory into description of knowledge (competence model), description of mechanism (performancemodel), and physiological reduction. It has come to be widely acceptedthat a theory may addressjust competence, or just competence and performance,and still be of great explanatoryvalue. Thus gestalt theory may be freed of the stigma of its weak attempts at physiological explanation, leaving a residueof generallyfar more attractivework. The other stumblingblock to gestalt theory was a lack of formalism. Generativeformalization modeling mental representationhad not yet been developed, so a physicallybased formalismwas probablythe only way available for going beyond the informality of most gestalt theorizing.Needlessto say, this too no longer presentsa problem. While it is not apparent how to adapt the formalismsof transformational grammarto the purposesof visual theory, we have seen that our music theory is a close analogue of the gestalt theory of vision. With this analogy in mind, it is not hard to construct rough outlines of a formal theory of visualperceptionon similarprinciples. The theory would be divided into two major components. A set of well-formednessrules would describe visual structures-structural descriptions that can be assigned to visual fields. Like the WFRsfor musical grouping,the visual WFRswould establishcertainbasic aspects of the correspondencebetween the visual surface and visual structure (the visual equivalentsof "A contiguoussequenceof events may form a group" and "The whole piece constitutes a group").On the other hand, many options would be left open by the well-formednessrules, and it would be left to a component of preferencerulesto determinewhich of the possible structuraldescriptionsassignable to a given field is the most highly preferred. In the case of an ambiguous field, more than one preferredstructurewould be selected. The preferencerules, like those for music, would involve aspects of the structures themselves (for example, symmetry), as well as aspects of the correspondenceof the given field to possible structures (for example, relative salience of different boundaries in the field). It is this component of the theory that would most closely reflect the work of the gestalt psychologists. Some recent work on the psychology of vision seems to fit quite well into this overallview of how visualtheory might be structured.2' The theory of musical groupingsuggestsa furthersubcomponentof the visual well-formedness rules. Earlier we showed how musical overlap and elision are situations in which a structureis attributedthat is not in one-to-one correspondence with the musical surface. In 85

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

overlap, the underlyinggroupingstructurecontainstwo occurrencesof the same boundary element, one in each of two adjoininggroups;in elision, a boundary element of one group has been obscuredby one in the adjacentgroup. These are musicalanaloguesof the visualintuitions concerningExample35a and 35b respectively. In the theory of musical grouping,the relation between the underlying grouping structure and the musical surface is established by a transformationthat combinesor deletes underlyingstructuralelements. It is much more common in visual experience for one unit to partially occlude another than in musical grouping, and the possible types of overlapandelision are more varied;but a similarformalaccount appears not inappropriate. This outline of a generative theory of visual form perception obviously just scratchesthe surface.We hope to have shown, however, how the analogy with music theory may lead to a more comprehensive view of visual theory, and in particularto a more productiveway of viewingthe work of the gestalt school. Conclusions.This articlehas shown how to translatethe overallgoals and methodology of generativelinguistics into terms appropriatefor music theory and has suggested that such an approach to music is divides into two major feasible and productive. The musical grammar rule systems, well-formedness rules (including transformations)and preference rules; interaction of these two systems is responsible for interpretationof the musicalsurface. Previousresearchin other domains has not recognizedthe need for both of these rule systems in cognitive theory. The theory of language has emphasizedthe role of well-formedness rules;while in the theory of vision, most researchhas concentratedon the equivalentof preference rules. Thus, in addition to providinga new way of studyingour musical understanding,the present approachbegins to bridge the gap between linguistics and other branchesof psychology, by showing the relationship between the two types of rule systems. Such a result seems to us extremely important, for it demonstrates that the study of musical cognition is not merely a curious side issue in psychology, but a source of crucialevidencetowarda more organictheory of the humanmind.

86

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

a.

O0
b.

seenas

0s

o-t

se n

as

not

as

or

Example35

87

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

NOTES
NB: "Temporal Gestalt Perception in Music," by James Tenney with Larry Polansky, Journal of Music Theory 24 (1980): 205-239, appeared too late for reference in this article. Beneath differences in terminology, their theory of "temporal gestalt-units" corresponds in many respects to our theory of grouping preference rules. Their system includes an equivalent of GPR 1, though they make the rule a prohibition of single-element groups rather than an avoidance. They have a rule of proximity, in essential respects identical to our Attack Point Rule (GPR 2b), and a rule of similarity close to our Register and Dynamics Rules (GPRs 3a and 3b). Furthermore, they make clear that grouping perception relies on the interaction of the various principles. On the other hand, the more global preference rules (GPRs 5-7) and the transformational rules for overlap and elision do not receive treatment in their theory. Tenney and Polansky's system goes beyond ours in one respect: they quantify strengths of rule application, so that grouping judgments can be simulated by a computer program. We discuss above why we have not taken this step. 1. This article consists primarily of excerpts from our forthcoming book A Generative Theory of Tonal Music (Cambridge: MIT Press). An exposition of certain aspects of the theory appears in Lerdahl and Jackendoff, "Toward a Formal Theory of Tonal Music," Journal of Music Theory 21 (1977): 111-71. We are grateful to Jeanne Bamberger, Morris Halle, Richard Held, Elise Jackendoff, Jay Keyser, George Miller, and James Lackner for useful discussion and indispensable hints. The writing of this article was supported in part by a Fellowship for Independent Study and Research from the National Endowment for the Humanities to the first author. Finally, we must thank Leonard Bernstein, whose 1973 Norton Lectures provided the impetus for our collaboration, and who was wiser than we in speculating that a generative music theory might someday prove of use to linguistics. 2. See, for example, Noam Chomsky's Aspects of the Theory of Syntax (Cambridge: MIT Press, 1965), Language and Mind (New York: Harcourt, Brace, 1972), and Reflections on Language (New York: Pantheon, 1975). 3. The most prominent example of such work is Bernstein, The Unanswered Question (Cambridge: Harvard University Press, 1976). However, while we disagree with most of the specifics of Bernstein's work, we concur in our overall goals. See Jackendoff's review of Bernstein in Language 53 (1977): 883-894. 4. An example of this approach is Johan Sundberg and Bjorn Lindblom, "Generative Theories in Language and Music Description," Cognition 4 (1976): 99-122. 5. Milton Babbitt, "Contemporary Music Composition and Music Theory as Contemporary Intellectual History," in B. S. Brook et al., eds., Perspectives in Musicology (New York: Norton, 1972), pp. 151-184. 6. Noam Chomsky and George Miller, "Introduction to the Formal Analysis of Natural Languages," in Luce, Bush, and Galanter, eds., Handbook of Mathematical Psychology 2 (New York: Wiley, 1963): 269-322. 7. "Toward a Formal Theory of Tonal Music" and A Generative Theory of Tonal Music.

88

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

8. In this example and those following, we include the rest between groups as part of the earlier group. This is not merely a convention, but corresponds, we believe, to the intuition that a group normally begins with its first note. To make the intuition clearer, consider for example mm. 124-127 of the Eroica Symphony, where one clearly has the opposite sense: the rests on the downbeats of these measures are heard as part of a group that includes the two following beats. Such a striking exception by contrast makes the intuition in the normal case more evident. Apparently, then, the assignment of rests to groups is rule-governed. However, we have not investigated the phenomenon in any detail and will not pursue it here. 9. See in particular Max Wertheimer, "Laws of Organization in Perceptual Forms," in W. D. Ellis, ed., A Source Book of Gestalt Psychology (London: Routledge & Kegan Paul, 1938), pp. 71-88; Kurt Koffka, Principles of Gestalt Psychology (New York: Harcourt, Brace, 1935); Wolfgang K6hler, Gestalt Psychology (New York: Liveright, 1929). 10. Patrick Winston, Learning Structural Descriptions from Examples (Cambridge: M.I.T. Project MAC, 1970). 11. Grouping (a) is probably preferable, but grouping (b) has not been without its advocates. For discussion, see Leonard Meyer, Explaining Music (Berkeley and Los Angeles: University of California Press, 1973), pp. 26-43. 12. See, for example, Sundberg and Lindblom, "Generative Theories in Language and Music Description." A related use of "stereotypical" metrical structure appears in Arthur Komar's Theory of Suspensions (Princeton: Princeton University Press, 1971), and is subject to similar objections. 13. Paul Grice, "Logic and Conversation," in Cole and Morgan, eds., Syntax and Semantics, vol. 4 (New York: Academic Press, 1975), pp. 37-58. 14. The notion of "family resemblance" is introduced by Wittgenstein in the Philosophical Investigations (Oxford: Blackwell, 1953); stereotypes are discussed, among other places, in Hilary Putnam, "The Meaning of Meaning," in K. E. Gunderson, ed., Language, Mind, and Knowledge, Minnesota Studies in the Philosophy of Science, vol. 7 (Minneapolis: University of Minnesota Press, 1975). For full discussion, see Ray Jackendoff, Semantics and Cognition (Cambridge: MIT Press, in preparation). 15. See references in note 9. 16. Karl Lashley, "The Problem of Serial Order in Behavior," in L. A. Jeffries, ed., Cerebral Mechanisms in Behavior (New York: Wiley, 1951), pp. 112136. 17. Koffka, Principles of Gestalt Psychology, p. 135. 18. Ibid., p. 110. 19. See, for instance, Noam Chomsky's review of B. F. Skinner's Verbal Behavior, in Language 35 (1959), pp. 26-58. Reprinted in J. Fodor and J. Katz, eds., The Structure of Language, (Englewood Cliffs, N.J.: Prentice-Hall, 1964). 20. Wolfgang Kbhler, Dynamics in Psychology, New York, Liveright, 1940. 21. Leon Shiman, Grammar for Vision (M.I.T. doctoral dissertation, 1975) and "The Law of Perceptual Stability" (Proceedings of the National Academy of Sciences 75.4 and 75.5 (1978), pp. 2049-2053 and 2535-2538) present an account of well-formedness rules for the visual field. Roger Shepard, "Psychophysical Complementarity" in Kubovy and Pomerantz, eds., Percep-

89

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

tual Organization, (Hillsdale, N.J.: Erlbaum, 1980), presents a theory very close in spirit to the sketch of visual theory above, though his theory lacks an equivalent for the nontransformational part of the well-formedness rules. To this should be added work in computer vision such as Winston (op. cit.), where the use of preference rule-like formalism emerges quite clearly.

i,

Can mboOeeamb-...oo _-d" ?-a --

90

This content downloaded from 193.136.173.41 on Wed, 1 May 2013 13:42:25 PM All use subject to JSTOR Terms and Conditions

You might also like