I thank Louis Pols and Mirjam Ernestus for their comments on earlier versions of this paper.1 created by the interaction of more fundamental principles.licensing, underspecification, feature geometry, and OCP effects, are surface phenomenaand perception in phonology. Traditionally separate devices like the segment, spreading,many hitherto controversial issues if we are aware of the different roles of articulationminimization of articulatory effort and maximization of perceptual contrast. We can solvePhonological structures and processes are determined by the functional principles of Abstract (Please send any comments, criticisms, and challenges to [email protected]) January 1997 University of Amsterdam, The NetherlandsPaul Boersma THE ELEMENTS OF FUNCTIONAL PHONOLOGY1 R111111987654321 C eferences....................................................................................................................... 1675 Conclusion................................................................................................................ 1654 Empirical adequacy of Functional Phonology......................................................... 1553 Degrees of specification........................................................................................... 1442 Correspondence: segmental integrity versus featural autonomy.............................. 1231 Typology and phonologization: the local-ranking hypothesis................................. 1010 An example of acoustic faithfulness: vowel reduction.............................................. 91 Interaction between articulation and perception.......................................................... 87 Specificational and faithfulness constraints................................................................. 57 Perceptual distinctivity................................................................................................. 52 The emergence of finiteness......................................................................................... 40 Articulatory effort......................................................................................................... 28 Formalization of functional principles......................................................................... 27 Hybrid, articulatory, and perceptual representations.................................................... 20 Articulatory, perceptual, and hybrid features................................................................. 5 Functional principles...................................................................................................... 3 ontents principle of mimization of articulatory effort evaluates the articulatory implementationThe theory of Functional Phonology, introduced in this paper, claims that thetEnslistener may reconstruct .//(“_” = silence). With the help of the processes of categorization and recognition, the[[thEE)nts]]In a microscopic transcription (§3.3), this perceptual result can be written as _ (0.3) nasal+ F2max F1open mid noiseaspsibilant voicedsonorant coronalbursttr.sidebu. cont silence++ perceptual phonetic events, time-aligned with the articulatory score (0.2) (tr = transition):This will give rise to an acoustic output that we can translate into the following table of (0.2) lipsspread glottiswidenarrowwide velumclosedopenclosed tongue tipclosedopenclosedcritical runs from left to right):some varieties of English, a part of the dominant articulatory implementation is (timeSeveral articulatory strategies can be followed to implement the utterance (0.1). Ingrammar.communication has succeeded. This basic insight should be reflected in our theory oftEnsthe listener will recognize the utterance as , and a substantial part of the//the specified perceptual features in the specified order with the specified time alignment,I will take this to be the perceptual specification of the utterance: if the speaker produces tEns(0.1)// Consider the English utterance tense. Its underlying phonological form islanguage is communication, and that languages are organized in a way that reflects this.The functional hypothesis for linguistics maintains that the primary function of a Introduction: from speaker to listener 1 Paul Boersma the help of the distinction between articulation and perception (§14; Boersma fc. a-e).synchronic autosegmental phenomena like spreading and the OCP, can be solved with•Many recalcitrant issues in the study of segmental inventories, sound change, andperceptual features, and between high- and low-ranked specifications (§13).a theory of underspecification vanish if we distinguish between articulatory and•The degree of specification in (0.1) should actually be quite high. All the arguments forrefer to “vertical” and “horizontal” perceptual connections, respectively.•Both segmental and autosegmental faithfulness are visible in the grammar (§12); theyto a straightforward strategy for the phonologization of phonetic principles.and obstruent voicing will illustrate the typological adequacy of this approach. It leadsranked on a language-specific basis (§11). The examples of nasal place assimilationperception, determines which constraints can be ranked universally, and which must be•The local-ranking principle, rooted in general properties of motor behaviour andrankings of the constraints.and how phonetic and pragmatic circumstances influence the result by shifting theimplementation is determined by the interaction of two continuous constraint families,•As an example, §10 describes how the realization of vowel height in phoneticapproach to the interaction between faithfulness and articulatory constraints (§9).•An adequate account of phonological structures and processes needs a comprehensiveoutput faithfulness, which can be ranked individually in each language (§8).•Constraints against perceptual confusion (§7) branch into many families of input-properties of motor learning and perceptual categorization (§6).•The finiteness of the number of feature values in every language is a result of generalindividually in each language (§5).•Constraints against articulatory effort branch into many families that can be rankedthat grammar allows constraint violation (§4).•Both articulatory and perceptual principles can only be brought into the grammar ifspecifications and outputs, and articulatory implementations (§3).•The traditional hybrid phonological representations should be replaced with perceptualarticulatory gestures and perceptual features (§2).•The traditional hybrid feature system should be replaced with separate systems ofproduction (§2).•Phonology controls both the articulatory and perceptual specifications of speech•Functional principles control both speech production and speech perception (§1). articulation and perception is an integral part of the grammar:In the present paper, I will defend the hypothesis that the distinction betweenimplementation will actually be chosen to surface.result (0.3). Together, these principles will determine which candidate articulatoryevaluates the differences between the perceptual specification (0.1) and the perceptual(0.2) and its competitors, and that the principle of maximization of perceptual contrast The Elements of Functional Phonology2 principle can be responsible for segment merger in cases of dialect mixture.strategies, which might otherwise not have been invoked. Labov (1994) showed that thisutterance, she will take recourse to alternative (semantic, pragmatic) disambiguationdisambiguating utterances: if the listener accepts the phonological ambiguity of ancategory, successful recognition is actually helped by not trying to use this contrast forclasses is not reliable, i.e., if an acoustic feature is sometimes classified into an adjacentcategories than it is to divide it into five. Moreover, if a contrast between two perceptualthe acoustic input can be analysed: it is easier to divide a perceptual continuum into twodisambiguation of an utterance is facilitated by having large perceptual classes into whichOn the other hand, in a world of large variations between and within speakers, thebecause that will help her recognize the meaning of the utterance.The listener will try to make maximum use of the available acoustic information,recognition and minimization of categorization.On the part of the listener, we have the functional principles of maximization of 1.3 Functional principles of speech perception “put as many bits of information in every second of speech as you can”.Passy’s “quickly” translates into the principle of the maximization of information flow: 1.2 Functional principle of the communication channel contrast.articulatory effort and the listener-oriented principle of the maximization of perceptuallet themselves be disentangled into the speaker-oriented principle of the minimization ofperceptually motivated constraints are violated. Passy’s two composite principles easilyidea that articulatorily motivated constraints may be honoured unless strongeranything that is necessary”. His use of the terms superfluous and necessary expresses thesuperfluous”, and the principle of emphasis: “languages tend to stress or exaggeratePassy states the principle of economy: “languages tend to get rid of anything that is 1.1 Functional principles of speech production quickly and clearly as possible”.language itself: “language is meant to convey information from one person to another asPassy (1890), sound changes have the same cause that motivates the existence ofFunctional principles were first expressed in explanations for sound change. According to 1 Functional principles 3 Paul Boersma behaviour and perception.features and representations with systems based on general properties of human motorthe constraints will be defined. This will lead to a replacement of the traditional hybriddetermine the nature of the phonological spaces (§2) and representations (§3) on whichphonological community in the form of Optimality Theory. First, however, we mustframework of constraint-ranking grammars, which, fortunately, is now available to thebe violable. We can expect, therefore, much from formalizing their interactions within ainherently conflicting, the constraints, if stated in their naked, most general forms, mustfamilies of constraints, which will be identified in §5, §6, and §8. Since the principles arecan achieve this by translating each of the principles (a) to (d) directly into severalhypothesis, involves formalizing the various aspects of the functional principles (§4). WeMaking typologically adequate predictions about what is a possible language under this of gestures conflicts with the minimization of energy.•Conflicts also arise within the various principles, e.g., the minimization of the numberminimization of categorization (§8.6).•Maximization of information flow conflicts with both minimization of effort and•Minimization of categorization conflicts with maximization of recognition.•Minimization of effort conflicts with maximization of contrast. These principles are inherently conflicting: (e)The speaker and the listener will maximize the information flow.use the maximum amount of acoustic information.(d)The listener will minimize the number of mistakes in recognition, i.e., she will try toperceptual categories as possible.(c)The listener will minimize the effort needed for classification, i.e., she will use as fewmeanings.(b)The speaker will maximize the perceptual contrast between utterances with differentto get by with a small number of simple gestures and coordinations.(a)The speaker will minimize her articulatory and organizational effort, i.e., she will try will be honoured:the structure of sound inventories are built in such a way that the following natural drivesThus, I maintain that historical sound changes, synchronic phonological processes, and 1.4 Functional hypothesis for phonology The Elements of Functional Phonology4 Conclusion: the speaker can set her muscles to a specified length.innervating the muscle spindles simultaneously with the other fibres.oriented gesture probably involves the learning of an efficient mix of and activity,αγcontrol muscle length (Hardcastle 1976; Gentil 1990). The learning of a fast, shape-this -loop system, which does not go further up than the spinal cord, can be used toγmuscle to contract. Thus, while direct activity would cause an uncontrolled contraction,αexternal cause. Consequently, the reflex mechanism described above will cause thethat the afferents fool the spinal cord into thinking that the muscle itself is stretched by an efferent fibers, the muscle spindles can be actively stretched, soWith the help of the γmotor neuron then causes the stretch reflex: a compensatory contraction of the muscle.is stretched by an external cause, a direct excitatory synapse of the afferent with the α(through the afferent fibers marked 1A) to the spinal cord or the brain stem. If the musclespindles (and the tension by the tendon organs), which send this information backvocal tract shape. The length and length change of a muscle are measured by the musclecontraction then results in a change in the shape of the human body, e.g., a change in neuron fibers from the spinal cord or the brain stem to the muscle fibers, whoseby the αdirect muscle command (every term set in italics can be found in figure 2.1) is conducted– Top right: length control. The speaker can control the tension of a muscle. For this, a specifications of the utterance, i.e., both the representations that we saw in (0.1) and (0.2).establish, is that phonology controls both the articulatory and the perceptualfollowing paragraphs, I will explain this figure. The main point that I am trying tointerface of the rest of the grammar to the phonologicalphonetic component. In the/phonology are integrated into speech production. The point labelled “start” marks theFigure 2.1 shows a simplified view of how the articulatory and perceptual aspects of 2.1 Articulation versus perception in speech production roles in speech production and perception.The difference between the two groups of features can be traced to their differentphonological debate.difference between them will solve many hitherto unsettled problems in several realms offeatures. These two groups play different roles in phonology, and an awareness of thethough they may be, can be divided into two large classes: articulatory and perceptualA thread of this work is the idea that features of speech sounds, language-dependent 2 Articulatory, perceptual, and hybrid features 5 Paul Boersma part of phonology specifies al these variables.locations and degrees of the constrictions in the vocal tract. Hypothesis: the articulatoryConclusion: the speaker can directly control muscle tensions, muscle lengths, and theThis system is called proprioceptive feedback.articulatory specification, and appropriate action is taken if there are any differences.back to the motor cortex, where it is compared to the intended shapes, i.e., thetactile receptors, and pressure receptors, sends the information about the realized shapesThe proprioceptive sensory system, consisting of muscle spindles, tendon organs,in the vocal tract as in normally articulated vowels, while having very different shapes.in such a way that the tongue muscles bring about approximately the same area functionspeakers immediately compensate for the constraints on the jaw, even before phonating,production, can be seen from bite-block experiments (Lindblom, Lubker & Gay 1979):specified. That the muscle lengths are not the target positions specified in speechFor the vocal tract, this means that the locations and degrees of constrictions aresufficient. Rather, the motor cortex specifies the actual position of the body structures.– Top left: control of position. For most gestures, the control of muscle length is not , , 1A = nerve fibers.αγEncircled minus signs = comparison centres. Arrows = causation.Rectangles = representations. Rounded rectangles = sensors.Fig. 2.1Integration of phonology into speech production. spectrum, intensitysonorant, nasalperiodicity, noise,voiced, fricative,specificationnAuditory featuresoPerceptual itazirogetac cortexkearsensorycabdeef ypressure receptorsrotiPhonologytactile receptorsd“sound”ua starttendon organs“vocal tract shape”muscle spindleskcabdeefcortexevitpeαcoimotorr1Aporpγcordmuscle commandsspinal stpressuresretcdegrees of constrictionshmuscle lengthsrlocations of constrictionseflexArticulatory specificationMuscle featurescoordination The Elements of Functional Phonology6 speech production. They cannot be the immediate (proximal) targets, because the auditoryFor a skilled speaker, the perceptual specifications must be the ultimate (distal) targets of 2.2 The two targets of speech production: levels of phonological specification perceptual specification of the utterance. Hypothesis: this is integrated into phonology.Conclusion: the speaker can compare the realized perceptual categories with thecheck and for maintenance.perceptual invariance and articulatory ease. She will use the auditory feedback only as aunderspecification, determined by economical considerations, i.e., the balance betweendegrees of constrictions and air pressures, with a language-dependent degree ofskilled speaker will also have highly organized articulatory specifications in terms ofultimately leads to a language-dependent categorization of perceptual features. TheBut the child learns to group perceptual events into categories. For speech, thisstill under development.sophisticated yet either, because the orosensory (proprioceptive) feedback mechanism isthe sake of faithful imitation and the reproduction of her own speech, are not veryan unsegmented gestalt. The articulatory specifications, which she is now constructing for(auditory feedback). At first, the perceptual specification (initially, the adult utterance), isthe little language learner will have to compare her own utterance with the model– Bottom left: speech perception. At the time she starts to imitate the speech she hears, coordinations (top left) and perceptual results (bottom right).Conclusion: speakers learn the forward relationship between articulatoryalternations of silence and sound (babbling).opening and closure gestures of the jaw on a background of phonation, thus getting nicemovements; like hitting the mills and bells that are within her reach, she will superponatesome agonistantagonist pairs, she will start exploring the benefits of repetitive/pleasing result of a periodic sound (voicing). A little later, when she has a command ofcertain combination of expiration and vocal-fold adduction gives her the auditorilydeltoid muscles gives her the visually pleasing result of a swinging object (her arm), a(i.e., her perceptual impressions), by pulling some muscles. Like the use of one of heruncoordinated. After a few months, the infant learns that she can control her environment(drinking). Other gestures, like the movements of the limbs, are still largelyspeech: breathing, vocal-fold adduction (crying), and repetitive jaw movementsThe infant is born with an innate control of some of the gestures that are also used inphonology in young children.will illustrate the perceptual part of speech production with the development ofnoisiness), spectrum (timbre), and intensity (loudness), all of them functions of time. Iarising from a speech utterance, into auditory features like periodicity (pitch and– Bottom right: auditory perception. The human ear will analyse any sound, perhaps one 7 Paul Boersma controls both.production, the perceptual specifications are the distal targets. Hypothesis: phonologyConclusion: the articulatory specifications are the proximal targets of speechafter some practice, using auditory feedback.[a]apart, though this would yield a much more -like sound; she will only learn this trickis not specified for horizontal lip spreading, she will not draw the corners of her mouth[a]clenched, she will make compensating tongue and lip movements; however, because [a]experiments. If you ask someone to pronounce a central (e.g. Dutch) with her teethspeech production, can be seen in a simple experiment that embroiders on the bite-blockThat the perceptual features, not the proprioceptive features, form the distal targets ofzand that the articulatory specification follows from the perceptual specification for .//constraint on bilabial closure after all (the lips must not be completely or nearly closed),position that enables voicing”. We see that the perceptual specification does require ashould be closed, the lungs should exert pressure, and the vocal cords should be in aconstrictions should be wider than this alveolar constriction, the naso-pharyngeal portachieves this, is: “make a near closure at the alveoli; meanwhile, the bilabial and dorsalnoise”. Speakers will learn that the only articulatory implementation (“task”) thatshould be in these terms: “make a periodic sound that will produce strong high-frequencyzspecifications, a part of the ultimate perceptual specification of (in some languages)//However, as will be apparent from our separation of perceptual and articulatoryunderspecification in the immediate targets of speech production.[z]a large variation in the degree of bilabial closure during . Therefore, there is some[z]Crucially, the task for does not specify bilabial closure at all; this is why there can be[z]complete closure with the lips”, and for it is: “make a near closure at the alveoli”.[b]Apparently (Kelso et al. argue), the immediate task for producing is: “make a[aza]bilabial closure will then be much larger than in the case of an unconstrained .apart, the required alveolar closure will still be attained. Crucially, however, the smallest[aza]. Thus, if the upper and lower teeth are externally constrained to be more than 1 cm[aba]bilabial closure in or the achievement of an appropriate alveolar near-closure ininduced perturbation of the movement of the jaw does not prevent the completion of therefers to learned motor behaviour. Kelso et al. notice, for example, that an experimentallycoordination of the muscles of speech production: specification of these tract variablesconstrictions, as functions of time. This approach explicitly focuses on describing theshould consist of specifications of tract variables, such as locations and degrees ofBrowman & Goldstein (1986, 1990), maintains that the input to an articulation modelThe task-dynamic approach advocated by Kelso, Saltzman, & Tuller (1986) andspindles, tendon organs, and joint receptors.targets can be monitored by the collective effort of tactile and pressure receptors, muscledegrees of constriction and the air pressures in the vocal tract. These proprioceptivefeedback loop is much too slow for that. The immediate targets are the locations and The Elements of Functional Phonology8 ballistic movement: plosives and nasals); critical (usually brought about by a controlledinstance, supralaryngeal degrees of closure can be: complete (usually brought about by atension. Though these scales are continuous, languages discretize most of them. Foreach of which is characterized as a vector measured along scales of degree of closure orimplementation of the utterance through this space is a voyage along many positions,epiglottis, laryngeal structures, vocal folds, and lungs. The trajectory of thetensions of the lips, cheeks, tongue tip, tongue body, velum, tongue root, pharynx walls,articulatory space, which consists of all the possible positions, shapes, movements, andThe functional principle of minimization of articulatory effort is evaluated in the 2.4 Articulatory specifications universal binary features.divided on a basis of equal perceptual distance rather than on a basis of maximum use of[E][e]have two vowels close to canonical and ; this shows that the height continuum is[a][i]approximately midway between and , and a language with four vowel heights will[i]highest vowel is , a language with three vowel heights will have an “e” whose height is[a]languages with three and four vowel heights (§6): if the lowest vowel is and thecan be most clearly seen from the different divisions of the height continuum for– No universal feature values. The language-dependency of perceptual feature values (2.1) nasal––––+––––––++ sonorant––––++–+–+––+ noise–++––––|+–+––|+–+ voiced––++++–+–++++ pfvbmwpHVhWub-u)v) perceptual contour, i.e., a temporal change in the value of a perceptual feature:where ‘+’ means ‘present’, ‘–’ is ‘absent’ (suggesting a privative feature), and ‘|’ is avoicing, friction, sonorancy, and nasality scales, can be read from the following table,perceptual specification of labial sounds for a language that has two categories along thethese scales into a language-dependent number of categories. An example of theAll these features are measured along continuous scales, but languages discretizenasality).(frication, aspiration), silence, burst, continuancy, and frequency spectrum (place,perceptual space. Perceptual features include periodicity (voicing and tone), noiseThe functional principle of maximization of perceptual contrast is evaluated in the 2.3 Perceptual specifications 9 Paul Boersma [b]consider the articulatory and perceptual features and specifications of in the utterancevocalic opening without affecting the perceptual features too much. As an example,have a lot of freedom: they can be anywhere between the approximant closure and a wideinstance, if the lips are completely or almost closed, the coronal and dorsal constrictions– Articulatory underspecification. There is a lot of underspecification in (2.2). For (2.2) supralar. tension+–––– glottis opening2-32-31111311121311 pharynx opening2-52-52-52-52-53-52-52-53-52-52-52-53-54-53 velum opening000040-100-140000-10-10-2 tongue body opening2-52-52-52-52-532-52-532-52-52-5334 tongue tip opening2-52-52-52-52-53-52-52-53-52-52-52-53-54-55 lip opening0110030|2-52300|2-50334 pfvbmwpHVw)b8b-∫hWuç from 2 to 5):1 = critical, 2 = approximant, 3 = narrow, 4 = open, 5 = wide, | = time contour, 2-5 =would faithfully implement the perceptual features of (2.1), is given in (2.2) (0 = closed,An example of the articulatory specifications of some labial sounds in a language thatthe independence of articulatory features in phonology.independence of these tiers represents the independence of the articulators, and reflectsspace is organized in tiers, with one tier for every degree of opening and tension. Thesame articulator twice (e.g. no clicks with dorso-palatal front closure). The articulatory[Ó][…W](Swedish , Holland Dutch syllable-final <l> ), but there are no sounds that use thein doubly articulated sounds (labial-velars, clicks) or even triply articulated soundsThe labial, coronal and dorsal articulators can be used independently to a large extent[vain][v´t][vil]( ‘vine’), Afrikaans ( ‘white’), and French ( ‘city’).[v][vaen]constriction are concerned, are the -like sounds of German ( ‘wine’), EnglishVi…]([ ‘wheel’); in between those two, as far as noisiness and, therefore, degree of[vi…]contrasts a noisy voiced labiodental fricative ( ‘fell’) and a noiseless approximantsubdivision of vowel height, here is an example with non-vocalic closures: Dutchsubdivision. Besides the obvious articulatory implementation of the language-dependentthe same articulator. Still, there is nothing canonical, preferred, or universal about thispair of successive labels is found somewhere in the world to contrast two phonemes onI classified these degrees of closure according to perceptual differences, i.e., every; spread lips, advanced tongue root).vocalic); or wide (4 - 15 cm2; high vowels, glides, liquids, retracted tongue root); open (1 - 4 cm; neutral1 cm22fricatives); approximant (strong secondary articulation, pharyngealization); narrow (0.3 -movement, which makes it precise enough to maintain friction noise or vibration: The Elements of Functional Phonology10 intervocalic position: widening or constriction of the glottis, raising of the larynx,Likewise, active gestures are sometimes needed for voiceless obstruents, especially in[pharyngeally expanded], though the term “expanding” might be more correct).formulated articulatory feature [obstruent voicing] (or Steriade’s (1995) suggestionspecific language, we can supply plain voiced obstruents with the implementationallyalways easy to find out which trick (other than implosion or prenasalization) is used by awith a more comprehensive model (Boersma 1993, 1995, in progress). Since it is notsimulations with a simple model of the vocal tract (Westbury & Keating 1986) as well asthe stop may be pre-nasalized. The effects of all of these tricks have been confirmed inwalls of the pharynx, the cheeks, or the velum may be expanded passively or actively, orlowered, the width of the glottis or the tension of the vocal folds may be adjusted, theoften needed to maintain voicing, especially in initial position: the larynx may be[b]If the passage is obstructed, as in , active laryngeal or supralaryngeal gestures areby the interarytenoid muscles; sufficient airflow is then guaranteed.freely, as in sonorants, there is spontaneous voicing if the vocal folds have been adductedvary depending on the degree of constriction above the larynx. If the air is allowed to exitsufficient velocity. The articulatory settings needed to implement the voicing feature,folds to vibrate, they must be close enough and air has to flow through the glottis with aThere is no single articulatory gesture that can be associated with voicing: for the vocalthe perceptual feature [voice], which refers to a high degree of periodicity in the sound.– Voicing. If we define voicing as the vibration of the vocal cords, we are talking about generative phonology.articulatory features. In the following, I will disentangle the hybrid features used inwith most place features, there are several sources of asymmetry between perceptual andThough it is often the case that similar articulations produce similar perceptual results, as 2.5 Perceptual versus articulatory features perceptual invariance.Conclusion: articulatory underspecification is constrained by faithfulness toargued in §2.2, the articulatory specifications follow from the perceptual specifications.quite wide so that the vowel does not sound more open or centralized. Thus, as alreadyso that it will not sound like a front vowel, and the pharyngeal constriction should also be[u]however, the lips are less closed, as in , the coronal constriction should be quite wideback again. The muscles of the lips will also make a closing-opening movement. If,[a][b]The jaw will travel a long distance in going from the position to the position and[aba]by the hyoglossus muscle. This state will last during the whole of the utterance .[aba][a]. During the pronunciation of , the tongue will be low in the throat, pulled down 11 Paul Boersma (dorso-uvular). From the perceptual standpoint, however, we could divide the continuousprefer locations with small holes (labio-dental, corono-interdental) or unstable structuresperfect closures are easy to make (bilabial, corono-postdental, dorso-velar), and fricativessame place of articulation; as so happens, plosives tend to occur at locations wherefricatives, on the other hand, often alternate, or are historically related to, plosives at thecommunication in a world with distances and background noise; the non-strident& Halle 1968, p. 327), and are, therefore, on the average more suitable for human(make more noise) than their non-strident counterparts on the same articulator (Chomskysuggests, this division is based on distributional grounds: the strident fricatives are louder∏TxfsSXstrident (, , ) and strident (, , , ). In contrast with what the label//////////////– Noise. In the phonological literature, fricatives are economically divided into non- from the CN case. The question how such details are phonologized, is answered in §11.pharyngeal cavity facilitates the maintenance of voicing; the exactly reverse situation(1995): the velum goes on raising even after the moment of closure, so that the enlarginggiven for the absence of CN voicing. An articulatory explanation was advanced by Hayeslike to share a non-redundant voicing feature with their neighbours. No explanation isTheir resolution of the paradox entails that nasals, because they are redundantly voiced, the unity and integrity of the distinctive feature [voice].” (Itô, Mester & Padgett 1995, p. 581)“the trouble lies not with [voice], (...) the challenge is to resolve the paradox without destroying tempted Itô, Mester & Padgett (1995) into the following remarks:approach, NC voicing (i.e., the phenomenon that plosives tend to be voiced after nasals),without a struggle: the one phenomenon that seems immune to a simple functionalsimply solve the mysteries associated with the voicing problem. However, this will not goperceptual voicing feature common to sonorants and voiced obstruents, would quitewhich only applies to obstruents because sonorants are spontaneously voiced, and avoicing feature for sonorants. A distinction between an articulatory voicing feature,analysis that sonorants were not voiced at all underlyingly; a late rule would insert the(with respect to spreading) of voicing in sonorants, was considered evidence for thedo not contrast voiced and voiceless sonorants, combined with the phonological inertnessthis feature: the fact that many languages do contrast voiced and voiceless obstruents butunderlying and surface representations resulted in the underspecification of sonorants forphonological processes (§13). Thus, an early decision to posit a single feature [voice] forfeatures, will lead to a rejection of the main argument for underspecification incannot. Acceptance of the distinction between articulatory and perceptual voicingtiming, we expect that obstruents can trigger voice assimilation, and that sonorantsSince assimilation processes are normally associated with changes of articulatorythis, we can similarly imagine a goal-oriented articulatory feature [obstruent devoicing].stiffening of supralaryngeal walls, or active narrowing of the supralaryngeal tract. For The Elements of Functional Phonology12 value of a loudness feature, or as a derived feature based on the presence of formant structure. This raises the question whether [sonorant] can be considered a primitive feature at all: it can be seen as a4[s] The reader may verify that she cannot produce a faithfully sibilant with a finger between her teeth.3[sTsTsTsTsT] Which the reader may verify by saying .2 be added.This condition is not sufficient, of course. Vocal-fold adduction and lung pressure have to supralaryngeal opening “approximant”(2.3)≥ needed for spontaneous voicing is:articulatory to perceptual features. For instance, the supralaryngeal articulatory settingThese derivative features can help as intermediaries in formulating the mapping from •Supralaryngeal opening. Minimum of suprapharyngeal and pharyngeal openings.•Suprapharyngeal opening. The maximum of the oral opening and the nasal opening.•Oral opening. This equals the minimum of the labial, coronal, and dorsal openings. with each articulator, we can derive the following abstract openings:[sonorant] implies [voice]. Its implementation is as follows. From the openings associated. Thus,high degree of loudness and periodicity that allows us to hear a formant structure4often based on auditory properties of sounds”, I will simply take [sonorant] to refer to aplay a role in phonology. Since, as Ladefoged (1971) states, “the rules of languages are302). This is neither an articulatory nor a perceptual definition, and, as such, not likely toproduced with a vocal tract configuration in which spontaneous voicing is possible” (p.– Sonorant. Chomsky & Halle’s (1968) definition of sonorants is that they are “sounds on the labial articulator as proposed by Yip (1988).for [sibilant] in base and affix (§14.2, Boersma fc. b), not to a missing stridency contrastThe epenthesis of a vowel in English fishes versus myths is due to the equal specifications with a precise specification for jaw height (see the discussion below for vowel height)., and sibilants are the only English soundsbetween the lower and upper teeth is critical3[T]. According to Ladefoged (1990a), the distancerespect to the normal strident 2sSstructure at another place (teeth): [], []; this causes a 15 dB intensity increase with•[sibilant]: a jet of air generated in one place (alveolar) and colliding at a roughfT]X•[strident friction]: airflow along sharp edges ([], [) or loose structures ([]).∏x•[mellow friction]: resulting from airflow through a smooth slit ([], []).[h][pH]•[aspirated]: as in , , and so-called “voiceless sonorants”. arbitrary, especially for the non-peripherals):noise scale into four levels of a combined loudnessroughness nature (which is rather/ 13 Paul Boersma 1[i][u], or a vowel with the tongue shape of and the lip shape of ?minimal F1[u][y]and are vowels with minimal F, what does the IPA symbol mean? Is it a front rounded vowel with[v)][i] If we take a perceptual definition for . The IPA is a hybrid notation system, and often ambiguous: if 6[z] You can check this by pinching your nose, making a “nasal” , and then suddenly releasing your nose.5 determines the articulatory gestures.tensions and positions from normal fricatives. Again, the perceptual specificationto come up with a very precise setting of your tongue blade, with different muscle[v)] in table (2.1) are so rare in the languages of the world; to make one, you’ll havelike 6. This is why nasalized fricativeslowered: the air will follow the path of lowest resistance5articulatory setting for a labial fricative will normally lose its friction when the velum isoral airflow does not necessarily reflect the degree of closure. A sound made with theIn another respect, oral stricture works just like velar stricture: the degree of perceivedcounterparts [+oral] and [–nasal].[–oral] and [+nasal] play more visible roles in phonological processes than theirare oral but not nasal, commonness considerations (§8.5) lead us to expect that the valuessymmetry between the nasal and oral pathways. However, because most speech sounds[m][a)]and are not, while is both oral and nasal. This move reflects the articulatory[f][h][a][p]audible oral airflow, I will replace it with [oral]; thus, , , and are oral, and be able to treat nasal and “oral” stops as a natural class. As a perceptual feature for– Continuant. This feature has been used to distinguish plosives from fricatives, and to in high vowels, and of no nasality in low vowels.Thus, the same small amount of velum lowering may give rise to a perception of nasalityvowels, because it can hardly be heard if the oral opening is large (Van Reenen 1981).[ç][u]specification for than for . A slightly open nasopharyngeal port is allowed in lowerfeature [lowered velum]. But not precisely. Table (2.2) shows a less restricted nasal– Nasality. The perceptual feature [nasal] more or less coincides with the articulatory [fricative].which would force the listener to distinguish between the categories [aspirated] andvoiceless fricatives and voiceless approximants only differ in their degree of noisiness,not; their voiceless counterparts cannot occur together in such a language, because[v][V](e.g., Dutch and ), because voiced fricatives are noisy and voiced approximants arewill. So, voiced fricatives and approximants can easily occur together in such a languageproduce friction, but a voiceless (aspirated) articulation with the same degree of closure,[f][hW]not if they are voiceless ( and ). This is because a voiced approximant will not[v][V]contrast between fricatives and approximants, but only if these are voiced ( and ),fricatives (aspirates). The binarily categorizing language of table (2.1) shows a perceptual– Fricatives versus approximants. So-called voiceless sonorants are just very mellow The Elements of Functional Phonology14 1). Simplistically, the muscles used ininner ear (the higher the vowel, the lower its Fthe perception of the first peak in the excitation pattern of the basilar membrane in the1), i.e., that the phonological effects of vowel height correspond toto the first formant (F(1971, 1990a) and Lindau (1975), I will assume that vowel height inversely correspondsspeakers adopt to implement this feature (Ladefoged 1990). Therefore, with Ladefogedvowel height with jaw height is weak, regarding the highly varying strategies thatactivities: no proprioceptors for non-low tongue height are known; the correlation ofa standpoint testifies to a view that speech is organized very differently from other motor[i]u]neutral point”. However, since different tongue muscles are involved in and [, suchinstruction the brain sends to the vocal apparatus to raise the tongue body above the– Vowel height. According to Kenstowicz (1994, p. 20), “we may interpret [+high] as the was preserved, which is a sign of the independence of the silence cue for plosives.as being constant. In the above example of place assimilation, the duration of the silenceof duration presupposes the recognition of another feature (the presence of sound, timbre)– Duration. Duration could be called a derived perceptual feature, because the perception of spreading phenomena.Boersma fc. a) that this kind of phonetic detail can be expressed directly in the grammaroutweighs the perceptual loss of losing the remaining place cue. We will see (§11,[ap|˘pa]](_), because the articulatory gain of not having to perform a blade gesturefor the recognition of [coronal]. In English, this may lead to place assimilation[[at|˘pa]]notable exception), so that the result _ shows the demise of the main place cueatpalike , is pronounced with overlapping gestures in most languages (with French as a//[t] merges with the ambient stillness, thus giving up its identity. A cluster of plosives,[[thEE)nts]]plosive. In _, both release bursts are heard, but silence associated with the firstplace of articulation, the silence and the burst together signal the presence of a voicelessspeech. While the formant transitions are shared with most other consonants at the samehas been a gigantic literature about the importance of all these cues in the perception of[[t|]][[]][[t]]formant transition + silence _ + release burst + formant transition. Their[ata]– Plosives. The intervocalic plosive in is perceptually marked by a sequence of implemented with the same ballistic articulatory gesture, e.g., [complete labial closure].labiodental nasal forms a natural class with its corresponding plosive because both arelabiodental plosives are very rare, but labiodental nasals quite common. Every non-labiodentals, even a ballistic movement often results in an incomplete closure; so,because it involves a ballistic instead of a controlled movement (Hardcastle 1976). ForAt most places of articulation, a complete closure is easier to make than a critical closure,If two articulations produce the same sound, the easier one is more likely to be used. 15 Paul Boersma strong perceptual peak near 4000 Hz.weaker impression on the inner ear than the third and fourth peaks, which tend to conspire to build a very[i]find five peaks between 0 and 5000 Hz. For , this second peak (at 2500 Hz or so) usually incurs a muchmeasured from 0 Hz upwards. This peak is commonly determined by a computer program that is forced to22 Known in the phonetic literature as F´, the usual definition of F being: the second spectral peak,7 space, ranging from [bilabial] to [glottal], and does not respect the discrete articulatoryThe perceptual place feature is a rather continuous path through a multidimensionalneighbouring sounds. Other cues must be found in noises (fricatives and release bursts).For consonants, place cues can be found in the formant transitions from and toimplied here: the two gestures just implement the same perceptual feature symmetrically.proposed by Stevens, Keyser & Kawasaki (1986) for reasons of lexical minimality, isan allegedly distinctive feature [back] by an allegedly redundant feature [round], as1; this is most faithfully implemented as a rounded back vowel. No “enhancement” ofF2 givenfront vowel with lip spreading. The value “min” specifies a minimum value of F21 should be at a maximum given F; this is most faithfully rendered by producing athat F2. Specifying the value “max” for F meansstrongest spectral peak above the first formant72); I will take it to specify theback distinction is represented by the second formant (F[round]; all these colour features assume [sonorant]. In the auditory spectrum, the front-can imagine that it is a multi-valued perceptual feature, encompassing [front], [back], andThus, place information for vowels is restricted to the upper part of the spectrum, and weindependent perceptual feature; it disappears in the transitions to neighbouring obstruents.is in the lower part of the spectrum and represents the degree of closure, seems to be anmade on the basis of the associated auditory spectra. For vowels, the first formant, which– Place. The perceptual distinction between the various places of articulation is primarily pharynx and the cheeks, which can play a role in the voicing of obstruents.relation between articulation and perception is found with the tension of the walls of theloudness (spectral slope and intensity) and pitch of the perceived sound. A rather indirecttension of the lung walls determines the subglottal pressure, which influences theof the vocal cords, which is the main determiner of the pitch of voiced sounds. The– Tensions. A direct relation between articulation and perception is found in the tension Eeisame articulator, as in _ . Apparent exceptions are treated in Boersma (fc. a)./→assimilation of vowel height is expected to occur only if all the vowels involved use theçoiçOitongue root; the rule _ is relatively rare (as compared with _ ), and//→→confined to the more articulatorily tractable features of rounding, backness, and advancedperceptual categorization, see §6), but vowel harmonies and assimilations are largelyVowel height does define natural classes in inventories and rule targets (as a result of(higher back vowels), and hyoglossus (low vowels).implementing vowel height are roughly: genioglossus (higher front vowels), styloglossus The Elements of Functional Phonology16 [N˘˘][wawawa]analogous experiment with and shows velar excitation of a closed front cavity.[n˘˘][wiwiwi] Try saying and superpose the lip movements of . The colour does not change. An9two CV transitions.[ana][ala] You can hear them if you record or , create a backward copy of this sound, and compare the8 occurrence of the feature values in the average utterance: the presence of a perceptualappropriate to explain privativity directly by arguments that start from the frequency ofSo, instead of invoking a mysterious speech-neutral position, it seems moreneutral position, we now have the most common position.sounds, and voiceless sounds will occur less often than voiced sounds. Instead of alouder sounds. Thus, nasal sounds will occur less often in an utterance than non-nasalnasals, and voicing allows us to produce tone contrasts, better formant structures, andperceptual contrast: we can produce better spectral contrasts for non-nasals than forThis speech-neutral position can only be explained with reference to requirements ofwhich involve active muscular effort (interarytenoid and levator palatini).The alleged speech-neutral position would have glottal adduction and a raised velum,the neutral position for breathing, which involves a wide glottis and a lowered velum.The only real neutral position is the one in which most muscles are relaxed, namely,speech-neutral position (Chomsky & Halle 1968, p. 300).qualitatively different from its absence, because it constitutes a deviation from thefeatures. The presence of an articulatory gesture like [lowered velum], she argues, isSteriade (1995) provides an articulatory explanation for the existence of privativethought to be able to spread.1987, Van der Hulst 1988, 1989, Avery & Rice 1989). For instance, only [+nasal] isvalue can be phonologically active (Anderson & Ewen 1987, Ewen & Van der HulstSome features must be considered privative (mono-valued, unary), because only a single 2.6 The speech-neutral position and privative features [n].contrast, lip rounding does not influence at all the stationary part of the sound of 9[s][S]languages, lip rounding contributes to the perceptual contrast between and . ByVocalic place cues can be used with stops and fricatives to a certain extent: in many.and laterals8visual cue overrides the auditory cue. Release cues without noise occur for nasal stops[n][m]closures will sound as in the dark, and as if the listener sees the speaker: thebranches is very weak: an isolated nasal stop produced with simultaneous lip and bladeconstrictions, too. For nasals, the place information contained in the various oral sideshows no formant transitions to surrounding vowels because these have glottalincluded in the set of values of the [place] feature (adjacent to [epiglottal]), though itsimilar, and so do corono-postalveolars and dorso-palatals; perceptually, [glottal] must bedistinctions between the articulators: labiodental and corono-dental fricatives sound quite 17 Paul Boersma near independence of the articulators:The implicational geometry for articulatory gestures is extremely flat, because of theperceived on voiceless syllables (as in Japanese), etc.implications have to be taken with a grain of salt, as it is not unthinkable that pitch is[˙]easy (), but a sibilant sonorant would be much harder to produce. Some of thecannot co-occur because of articulatory constraints; for instance, an aspirated sonorant isThis figure only shows perceptual dependencies, so it does not show which features (2glottalepiglottalmin (back)pharyngeal2Fmax (front)uvularplacemax (low)velarmid1Flateralpalatalmin (high)LnasalalveolarsibilantMpitchdentalstridentHlabiodentalnoisespectrumsonorantmellowvoicedbilabialmurmuraspiratedweakloudnesslateralstrongtransitiondurationnasalreleaseclosedburstsoundingstaticspreadlipsdynamicsilenceround heardseen perceived disjunctions by horizontal branches):presence of perceptual features (conjunctions are shown by vertical branches,The above story gives rise to the following partial geometry of implications for the 2.7 Feature geometries specifications.values like [–nasal] are not absent, but only relatively invisible because of their weakthe corresponding perceptual features. I will argue there and in §13 that the commondirectly to the listener’s adaptation of recognition strategies to frequency differences indifferences in the phonological activities of various articulatory gestures can be relatednot signal any deviation from the more common non-nasality. In §8.5, I will show thatfeature like [nasal] is quantitatively different from its absence, because the latter would The Elements of Functional Phonology18 .4 )